{
  "ctfidf_model": {
    "bm25_weighting": false,
    "reduce_frequent_words": false
  },
  "vectorizer_model": {
    "params": {
      "analyzer": "word",
      "binary": false,
      "decode_error": "strict",
      "encoding": "utf-8",
      "input": "content",
      "lowercase": true,
      "max_df": 1.0,
      "max_features": null,
      "min_df": 2,
      "ngram_range": [
        1,
        5
      ],
      "stop_words": "english",
      "strip_accents": null,
      "token_pattern": "(?u)\\b\\w\\w+\\b",
      "vocabulary": null
    },
    "vocab": {
      "modelling": 72595,
      "users": 129087,
      "intentions": 55918,
      "structure": 113802,
      "spoken": 111962,
      "dialog": 30543,
      "outline": 82629,
      "utterances": 130623,
      "dialogs": 30622,
      "interpreted": 56280,
      "using": 129303,
      "partial": 85284,
      "order": 82262,
      "logic": 64988,
      "exploit": 41403,
      "capability": 15178,
      "talk": 117556,
      "truth": 125878,
      "status": 113214,
      "formulae": 45269,
      "define": 28481,
      "notion": 80403,
      "coherence": 18290,
      "explain": 41254,
      "relation": 97937,
      "serve": 107287,
      "construction": 21683,
      "andor": 6097,
      "trees": 125680,
      "represent": 99097,
      "segmentation": 104570,
      "bdi": 12594,
      "model": 70492,
      "formalize": 45195,
      "basic": 12512,
      "assumptions": 9667,
      "cooperative": 23255,
      "behaviour": 12679,
      "participants": 85310,
      "provide": 93747,
      "basis": 12542,
      "inferring": 54266,
      "speech": 111640,
      "acts": 3139,
      "relations": 98103,
      "attitudes": 10136,
      "prove": 93695,
      "useful": 128856,
      "determining": 30145,
      "segments": 104670,
      "defined": 28497,
      "completing": 19789,
      "expectations": 40387,
      "finally": 44140,
      "sketch": 109964,
      "explicit": 41304,
      "cue": 25203,
      "phrases": 87406,
      "covered": 24527,
      "lexicalized": 63851,
      "tree": 125574,
      "adjoining": 3865,
      "grammar": 48621,
      "english": 37061,
      "document": 32946,
      "describes": 29383,
      "sizable": 109907,
      "written": 135112,
      "tag": 117333,
      "formalism": 45184,
      "implemented": 51958,
      "use": 127876,
      "report": 98973,
      "described": 29374,
      "earlier": 34375,
      "1995": 466,
      "technical": 119741,
      "based": 11495,
      "extended": 41801,
      "include": 53218,
      "lexicalization": 63849,
      "unificationbased": 127100,
      "feature": 43247,
      "structures": 114055,
      "range": 95537,
      "syntactic": 116359,
      "phenomena": 87218,
      "handled": 49406,
      "large": 61005,
      "includes": 53250,
      "including": 53260,
      "inversion": 56705,
      "copula": 23282,
      "raising": 95464,
      "small": 110127,
      "clause": 17688,
      "constructions": 21703,
      "topicalization": 122597,
      "relative": 98341,
      "clauses": 17696,
      "adjuncts": 3872,
      "pro": 90793,
      "modifications": 74441,
      "sequences": 107110,
      "negation": 76910,
      "nounverb": 80450,
      "contractions": 22663,
      "sentential": 106554,
      "imperatives": 51920,
      "corresponds": 24318,
      "release": 98428,
      "continuously": 22649,
      "updated": 127801,
      "addition": 3396,
      "new": 78761,
      "analyses": 5438,
      "modification": 74440,
      "old": 81641,
      "ones": 81678,
      "online": 81740,
      "version": 131603,
      "web": 132219,
      "page": 83408,
      "separating": 106881,
      "dependency": 29129,
      "constituency": 21521,
      "rewriting": 102692,
      "paper": 83694,
      "present": 89357,
      "called": 15049,
      "variant": 130865,
      "synchronous": 116338,
      "tags": 117465,
      "approach": 7286,
      "coordination": 23264,
      "linguistic": 64416,
      "distinguished": 32586,
      "explicitly": 41359,
      "distinguishes": 32587,
      "dependencies": 29061,
      "gives": 48177,
      "better": 13496,
      "formal": 45151,
      "understanding": 126794,
      "representation": 99158,
      "compared": 19319,
      "previous": 90376,
      "approaches": 8037,
      "systems": 116705,
      "conflate": 21028,
      "issues": 57034,
      "incremental": 53721,
      "parser": 84904,
      "generation": 47286,
      "grammars": 48670,
      "parse": 84879,
      "tables": 117264,
      "parsing": 85055,
      "languages": 60379,
      "algorithm": 4685,
      "presented": 89774,
      "handles": 49409,
      "input": 55290,
      "updating": 127807,
      "generated": 47045,
      "far": 43067,
      "lazy": 61825,
      "parsers": 85009,
      "created": 24652,
      "need": 76773,
      "generator": 47763,
      "responds": 101186,
      "built": 14911,
      "processing": 91609,
      "unknown": 127359,
      "words": 133797,
      "hpsg": 50691,
      "lexical": 63724,
      "acquisition": 3045,
      "incrementally": 53743,
      "updates": 127804,
      "properties": 92434,
      "inferred": 54261,
      "surrounding": 116153,
      "context": 21996,
      "sentences": 106189,
      "german": 47879,
      "employ": 36249,
      "gradual": 48614,
      "concept": 20559,
      "providing": 94097,
      "uniform": 127136,
      "treatment": 125566,
      "completely": 19778,
      "known": 58283,
      "maximally": 67502,
      "entries": 38155,
      "information": 54351,
      "viewed": 131713,
      "generalizable": 46752,
      "takes": 117521,
      "place": 87598,
      "requires": 100239,
      "modified": 74445,
      "lookup": 65227,
      "pieces": 87496,
      "identified": 51461,
      "access": 1900,
      "paths": 85686,
      "mechanism": 67939,
      "corresponding": 24274,
      "places": 87615,
      "actually": 3152,
      "provides": 93995,
      "revising": 102647,
      "type": 126184,
      "union": 127177,
      "required": 100209,
      "example": 39772,
      "demonstrates": 28938,
      "inferential": 54257,
      "capacity": 15228,
      "translating": 124584,
      "nearsynonyms": 76742,
      "possibilities": 88364,
      "preferences": 89242,
      "interlingua": 56119,
      "argues": 8932,
      "interlingual": 56120,
      "parts": 85569,
      "meaning": 67617,
      "situation": 109886,
      "necessary": 76749,
      "definite": 28515,
      "components": 20015,
      "constraints": 21591,
      "enable": 36347,
      "analysis": 5478,
      "nuance": 80822,
      "faithful": 43000,
      "translation": 124612,
      "furthermore": 46141,
      "especially": 38429,
      "crucial": 25128,
      "specifies": 111624,
      "nuances": 80824,
      "convey": 23112,
      "contexts": 22376,
      "choosing": 16930,
      "word": 132900,
      "typical": 126393,
      "cooccurrence": 23222,
      "network": 77118,
      "presents": 89814,
      "solution": 110544,
      "component": 19964,
      "problem": 90912,
      "choice": 16884,
      "synonym": 116344,
      "expected": 40388,
      "apply": 7156,
      "statistical": 113074,
      "representing": 100040,
      "networks": 77497,
      "implementation": 51938,
      "trained": 123059,
      "evaluated": 38947,
      "corpus": 23628,
      "results": 101481,
      "inclusion": 53408,
      "secondorder": 104477,
      "improves": 52943,
      "performance": 86101,
      "program": 92111,
      "language": 58802,
      "modeling": 72370,
      "develops": 30448,
      "uses": 129190,
      "extract": 42065,
      "meaningful": 67710,
      "history": 50545,
      "enabling": 36408,
      "long": 65059,
      "distance": 32402,
      "assigns": 9564,
      "probability": 90856,
      "joint": 57256,
      "sequence": 106912,
      "headword": 49614,
      "annotation": 6276,
      "operates": 82049,
      "lefttoright": 63306,
      "manner": 66937,
      "usable": 127853,
      "automatic": 10488,
      "recognition": 96800,
      "probabilistic": 90796,
      "parameterization": 84732,
      "set": 107338,
      "experiments": 40751,
      "meant": 67772,
      "evaluate": 38780,
      "predictive": 89199,
      "power": 88628,
      "improvement": 52659,
      "standard": 112182,
      "trigram": 125794,
      "achieved": 2589,
      "structured": 113989,
      "parametrization": 84799,
      "flexible": 44653,
      "shallow": 107912,
      "text": 120620,
      "support": 115953,
      "efficient": 35051,
      "development": 30365,
      "nl": 79507,
      "orthogonal": 82584,
      "methods": 69270,
      "currently": 25394,
      "pursued": 94451,
      "emphasis": 36136,
      "reusable": 102485,
      "general": 46630,
      "linguistically": 64583,
      "motivated": 74841,
      "surface": 116056,
      "realization": 96112,
      "simple": 109353,
      "taskoriented": 118880,
      "templatebased": 120080,
      "techniques": 119824,
      "argue": 8907,
      "perspective": 87158,
      "benefits": 13002,
      "limited": 64199,
      "improve": 52329,
      "suggest": 115301,
      "associated": 9592,
      "increased": 53624,
      "flexibility": 44648,
      "close": 17807,
      "connection": 21134,
      "ontologies": 81829,
      "supports": 116035,
      "quick": 95383,
      "adaptation": 3205,
      "tasks": 118909,
      "domains": 33721,
      "reuse": 102486,
      "resources": 100938,
      "method": 68577,
      "designed": 29576,
      "generating": 47194,
      "reports": 99079,
      "variations": 130932,
      "fallacy": 43029,
      "attack": 9699,
      "generative": 47706,
      "lexicon": 63871,
      "largely": 61331,
      "argument": 8934,
      "rests": 101354,
      "traditional": 122788,
      "implausible": 51926,
      "view": 131703,
      "effectively": 34784,
      "content": 21854,
      "stands": 112358,
      "line": 64323,
      "explaining": 41273,
      "means": 67755,
      "symbol": 116311,
      "typographic": 126479,
      "share": 107946,
      "wrong": 135160,
      "belief": 12687,
      "correspond": 24265,
      "entity": 37902,
      "turn": 125968,
      "semantic": 104969,
      "rules": 103377,
      "pustejovsky": 94468,
      "novel": 80453,
      "features": 43346,
      "wellestablished": 132409,
      "artificial": 9243,
      "intelligence": 55821,
      "tradition": 122787,
      "reference": 97504,
      "mention": 68415,
      "assigned": 9548,
      "occur": 81495,
      "proximity": 94153,
      "argued": 8931,
      "foundation": 45342,
      "proved": 93709,
      "practical": 88697,
      "worth": 135073,
      "computational": 20355,
      "justification": 57477,
      "high": 50030,
      "points": 87841,
      "probably": 90876,
      "understood": 127022,
      "links": 64677,
      "world": 135018,
      "relationship": 98286,
      "cultural": 25226,
      "representational": 99476,
      "supervised": 115722,
      "induction": 54019,
      "training": 123341,
      "data": 25547,
      "constituent": 21534,
      "corpusbased": 24074,
      "generally": 46846,
      "relies": 98641,
      "learn": 61982,
      "unfortunately": 127077,
      "cost": 24344,
      "building": 14819,
      "annotated": 6134,
      "corpora": 23396,
      "prohibitively": 92201,
      "expensive": 40410,
      "work": 134348,
      "aims": 4495,
      "strategy": 113497,
      "labels": 58576,
      "informative": 55128,
      "constituents": 21553,
      "higher": 50157,
      "nodes": 80031,
      "typically": 126408,
      "denoting": 29009,
      "complex": 19797,
      "noun": 80421,
      "account": 2019,
      "20": 487,
      "inducing": 54013,
      "sparsely": 111243,
      "labeled": 58423,
      "higherlevel": 50213,
      "propose": 92523,
      "produces": 91985,
      "induced": 54006,
      "fully": 45940,
      "replace": 98927,
      "human": 50732,
      "annotators": 6486,
      "able": 1589,
      "automatically": 10709,
      "base": 11442,
      "particles": 85373,
      "japanese": 57183,
      "distinct": 32530,
      "central": 15896,
      "roles": 103228,
      "mark": 67222,
      "arguments": 8975,
      "functional": 46054,
      "straightforward": 113428,
      "matching": 67391,
      "functions": 46068,
      "ga": 46332,
      "subject": 114674,
      "object": 81047,
      "adjunct": 3871,
      "sentence": 105761,
      "cooccur": 23220,
      "verbal": 131528,
      "eliminated": 35297,
      "omitted": 81653,
      "proper": 92412,
      "make": 66616,
      "possible": 88374,
      "empirical": 36147,
      "investigation": 56859,
      "800": 1291,
      "dialogues": 30819,
      "hierarchy": 50025,
      "syntax": 116529,
      "verbmobil": 131545,
      "cascaded": 15570,
      "markov": 67251,
      "models": 72632,
      "contextfree": 22359,
      "layer": 61697,
      "resulting": 101429,
      "represented": 100026,
      "output": 83046,
      "lower": 65419,
      "passed": 85620,
      "evaluation": 39112,
      "yields": 135394,
      "good": 48458,
      "chunking": 16957,
      "newspaper": 79405,
      "texts": 121448,
      "nlp": 79557,
      "dialogue": 30624,
      "priority": 90776,
      "programme": 92122,
      "technology": 120030,
      "research": 100394,
      "aiming": 4489,
      "alternative": 5221,
      "natural": 76239,
      "modules": 74518,
      "developed": 30249,
      "parallel": 84583,
      "grammarbased": 48667,
      "conventional": 22875,
      "rulebased": 103356,
      "module": 74476,
      "dataoriented": 26693,
      "memorybased": 68402,
      "stochastic": 113353,
      "dop": 33953,
      "compare": 19219,
      "carried": 15534,
      "years": 135241,
      "start": 112396,
      "procedure": 91387,
      "performs": 86981,
      "comparison": 19522,
      "learning": 62300,
      "transformation": 124260,
      "grammatical": 48687,
      "relationships": 98315,
      "important": 52095,
      "level": 63411,
      "trainable": 123054,
      "errordriven": 38350,
      "finds": 44316,
      "core": 23308,
      "groups": 49149,
      "bypasses": 14993,
      "phase": 87212,
      "test": 120417,
      "achieves": 2727,
      "636": 1173,
      "recall": 96333,
      "precision": 88787,
      "fscore": 45909,
      "698": 1203,
      "mapping": 67126,
      "multilingual": 75194,
      "hierarchies": 50023,
      "relaxation": 98423,
      "labeling": 58482,
      "explores": 41642,
      "knowledge": 57737,
      "preexisting": 89232,
      "robust": 103013,
      "linking": 64651,
      "existing": 40041,
      "lexicalsemantic": 63861,
      "used": 128374,
      "constraint": 21578,
      "satisfaction": 103648,
      "select": 104687,
      "candidate": 15097,
      "translations": 125450,
      "proposed": 93163,
      "bilingual": 14019,
      "dictionary": 30872,
      "right": 102826,
      "wordnet": 133762,
      "synset": 116357,
      "sense": 105647,
      "taxonomy": 119660,
      "derived": 29344,
      "spanish": 111146,
      "monolingual": 74547,
      "average": 11171,
      "15": 337,
      "connections": 21145,
      "accuracy": 2060,
      "80": 1286,
      "ways": 132146,
      "technique": 119759,
      "applied": 7042,
      "enrich": 37555,
      "databases": 26649,
      "unified": 127101,
      "examplebased": 39802,
      "lexicalist": 63848,
      "machine": 65738,
      "combines": 18676,
      "ideas": 51347,
      "methodologies": 69239,
      "theoretical": 121788,
      "frameworks": 45743,
      "graphs": 48939,
      "framework": 45403,
      "multidimensional": 75067,
      "recent": 96387,
      "acyclic": 3157,
      "offer": 81559,
      "powerful": 88662,
      "incorporating": 53515,
      "overlap": 83328,
      "motivate": 74832,
      "illustrate": 51735,
      "discourselevel": 32109,
      "annotations": 6405,
      "drawn": 34130,
      "callhome": 15070,
      "trains": 123961,
      "schemes": 103945,
      "help": 49693,
      "domain": 33423,
      "specialists": 111380,
      "constructed": 21652,
      "hybrid": 51171,
      "multilevel": 75188,
      "fragment": 45371,
      "university": 127347,
      "radio": 95446,
      "following": 44958,
      "levels": 63525,
      "segment": 104558,
      "breath": 14586,
      "tobi": 122233,
      "tilt": 121974,
      "treebank": 125624,
      "coreference": 23338,
      "named": 76026,
      "derive": 29336,
      "diverse": 32793,
      "file": 44019,
      "formats": 45217,
      "facilitates": 42801,
      "substantive": 114919,
      "multiple": 75488,
      "single": 109705,
      "signal": 108681,
      "different": 30983,
      "discussion": 32317,
      "shows": 108543,
      "open": 81878,
      "door": 33949,
      "wideranging": 132612,
      "integration": 55807,
      "tools": 122424,
      "chunks": 16968,
      "dividing": 32914,
      "preprocessing": 89321,
      "step": 113242,
      "extraction": 42246,
      "retrieval": 102378,
      "marcus": 67178,
      "introduced": 56570,
      "convenient": 22869,
      "converting": 23104,
      "tagging": 117370,
      "task": 117809,
      "examine": 39732,
      "seven": 107866,
      "representations": 99482,
      "recognizing": 97067,
      "phrase": 87329,
      "minor": 70270,
      "influence": 54301,
      "equipped": 38238,
      "suitable": 115393,
      "chunker": 16956,
      "best": 13294,
      "published": 94347,
      "application": 6832,
      "realtime": 96135,
      "englishspanish": 37427,
      "translate": 124522,
      "closed": 17840,
      "captions": 15267,
      "discuss": 32231,
      "tradeoffs": 122782,
      "available": 10932,
      "introduce": 56375,
      "increase": 53584,
      "coverage": 24495,
      "maintaining": 66506,
      "space": 110968,
      "time": 121976,
      "efficiency": 35020,
      "indicate": 53831,
      "effective": 34621,
      "selective": 104853,
      "magic": 66375,
      "implementing": 51981,
      "advantages": 4067,
      "dynamic": 34294,
      "bottomup": 14501,
      "advanced": 3936,
      "topdown": 122487,
      "control": 22800,
      "allows": 5124,
      "user": 128965,
      "compilation": 19708,
      "specific": 111404,
      "result": 101355,
      "processed": 91593,
      "dynamically": 34338,
      "fashion": 43114,
      "state": 112419,
      "art": 9030,
      "deal": 27854,
      "remaining": 98778,
      "various": 131033,
      "aspects": 9370,
      "concerning": 20689,
      "recently": 96647,
      "scheme": 103910,
      "evaluating": 39032,
      "avoids": 11248,
      "shortcomings": 108280,
      "current": 25255,
      "encodes": 36666,
      "heads": 49603,
      "dependents": 29267,
      "publicdomain": 94286,
      "naturally": 76635,
      "occurring": 81515,
      "relate": 97842,
      "extant": 41777,
      "bootstrap": 14453,
      "transfer": 124031,
      "equivalences": 38247,
      "rule": 103344,
      "templates": 120082,
      "instantiating": 55646,
      "template": 120077,
      "provided": 93957,
      "belong": 12713,
      "appropriate": 8411,
      "categories": 15726,
      "creating": 24692,
      "inventory": 56695,
      "generate": 46894,
      "simpler": 109551,
      "consists": 21465,
      "extracting": 42198,
      "finite": 44532,
      "sample": 103559,
      "hand": 49314,
      "coded": 18164,
      "directly": 31857,
      "process": 91414,
      "abstracting": 1786,
      "initial": 55206,
      "terms": 120262,
      "correspondences": 24273,
      "involving": 56909,
      "phrasal": 87324,
      "mapped": 67118,
      "sets": 107643,
      "aid": 4431,
      "way": 132051,
      "infinite": 54271,
      "recursively": 97290,
      "instantiate": 55643,
      "definition": 28518,
      "enables": 36372,
      "creation": 24710,
      "multiword": 75941,
      "noncompositional": 80172,
      "cardinality": 15502,
      "architectural": 8599,
      "considerations": 21283,
      "conversational": 22984,
      "experience": 40441,
      "project": 92209,
      "design": 29512,
      "goals": 48417,
      "architecture": 8602,
      "incrementality": 53742,
      "achieve": 2405,
      "degree": 28566,
      "robustness": 103088,
      "scalability": 103681,
      "flow": 44680,
      "ability": 1481,
      "restrictions": 101348,
      "considering": 21315,
      "operational": 82057,
      "fall": 43025,
      "1996": 467,
      "followed": 44945,
      "engineering": 37042,
      "focussing": 44931,
      "symbolic": 116312,
      "led": 63272,
      "generalization": 46759,
      "pass": 85603,
      "beam": 12596,
      "search": 104292,
      "queries": 94911,
      "fruitful": 45905,
      "query": 94947,
      "pipelined": 87561,
      "aimed": 4477,
      "restricted": 101338,
      "want": 132014,
      "represents": 100070,
      "key": 57537,
      "issue": 56987,
      "likely": 64133,
      "illformed": 51731,
      "factors": 42879,
      "noisy": 80088,
      "environments": 38200,
      "low": 65343,
      "quality": 94582,
      "succeed": 115056,
      "partially": 85301,
      "regular": 97721,
      "approximation": 8469,
      "discussed": 32294,
      "construct": 21618,
      "automaton": 10848,
      "given": 47979,
      "lead": 61840,
      "subsets": 114842,
      "original": 82501,
      "refined": 97591,
      "form": 45067,
      "respect": 101072,
      "literature": 64741,
      "performed": 86907,
      "hypotheses": 51248,
      "recognizer": 97062,
      "filtered": 44077,
      "resolution": 100749,
      "verb": 131513,
      "ellipsis": 35305,
      "expressions": 41746,
      "examples": 39804,
      "verbs": 131550,
      "recover": 97140,
      "purposes": 94445,
      "experimented": 40744,
      "ellipses": 35304,
      "obtained": 81339,
      "rate": 95770,
      "73": 1231,
      "66": 1183,
      "japanesetoenglish": 57215,
      "tense": 120179,
      "aspect": 9314,
      "modality": 70484,
      "similarity": 109190,
      "ends": 36864,
      "knearest": 57716,
      "neighbor": 76982,
      "exclude": 39953,
      "effects": 34975,
      "noise": 80041,
      "wrongly": 135164,
      "tagged": 117346,
      "tenses": 120183,
      "modalities": 70474,
      "accurately": 2375,
      "toplevel": 122673,
      "mt": 74954,
      "software": 110512,
      "market": 67240,
      "does": 33322,
      "require": 100107,
      "handcraft": 49333,
      "deduction": 28186,
      "passage": 85607,
      "documents": 33172,
      "horn": 50656,
      "abstraction": 1788,
      "locate": 64954,
      "passages": 85614,
      "strategies": 113446,
      "applicable": 6822,
      "fields": 43990,
      "outside": 83193,
      "inference": 54108,
      "outlined": 82637,
      "detailed": 29749,
      "description": 29452,
      "exploiting": 41458,
      "thesis": 121846,
      "attempt": 9730,
      "improved": 52590,
      "merges": 68479,
      "shiftreduce": 108186,
      "maximum": 67528,
      "likelihood": 64117,
      "belonging": 12719,
      "class": 17029,
      "expectationmaximization": 40385,
      "algorithms": 4823,
      "employed": 36291,
      "wall": 132000,
      "street": 113561,
      "journal": 57405,
      "switchboard": 116278,
      "broadcast": 14685,
      "news": 79281,
      "perplexity": 87060,
      "error": 38275,
      "lattice": 61667,
      "rescoring": 100388,
      "significance": 108712,
      "lies": 63974,
      "presenting": 89806,
      "hierarchical": 49922,
      "vocabulary": 131864,
      "refinement": 97594,
      "inspired": 55564,
      "hidden": 49891,
      "complement": 19726,
      "ngram": 79425,
      "parameters": 84734,
      "potential": 88534,
      "twopass": 126150,
      "complementing": 19757,
      "decoding": 28069,
      "wsj": 135177,
      "ppl": 88692,
      "wer": 132448,
      "requirements": 100234,
      "lexicons": 63922,
      "expand": 40352,
      "scope": 104036,
      "larger": 61352,
      "discriminating": 32197,
      "senses": 105714,
      "incorporate": 53448,
      "ordinary": 82450,
      "dictionaries": 30856,
      "contain": 21731,
      "contents": 21985,
      "reveal": 102494,
      "satisfied": 103654,
      "developing": 30333,
      "primitive": 90662,
      "concepts": 20603,
      "procedures": 91407,
      "satisfying": 103660,
      "illustrated": 51748,
      "frames": 45398,
      "examining": 39768,
      "definitions": 28532,
      "change": 16356,
      "hypernym": 51214,
      "usage": 127854,
      "measure": 67776,
      "importance": 52044,
      "making": 66818,
      "sublanguages": 114718,
      "message": 68494,
      "classification": 17096,
      "center": 15889,
      "customer": 25436,
      "care": 15503,
      "increasingly": 53689,
      "email": 35336,
      "communication": 19025,
      "allowing": 5107,
      "reproduction": 100092,
      "approved": 8450,
      "solutions": 110571,
      "identifying": 51579,
      "customers": 25450,
      "timeconsuming": 122143,
      "changes": 16381,
      "products": 92070,
      "launched": 61679,
      "requests": 100106,
      "assistance": 9571,
      "agents": 4317,
      "commercial": 18834,
      "setting": 107728,
      "variable": 130836,
      "ngrams": 79459,
      "occurrence": 81509,
      "varies": 130946,
      "despite": 29677,
      "observation": 81163,
      "usually": 130403,
      "assumption": 9660,
      "constant": 21515,
      "investigate": 56710,
      "modelled": 72592,
      "poisson": 87895,
      "distribution": 32640,
      "continuous": 22610,
      "mixture": 70416,
      "estimating": 38645,
      "frequencies": 45833,
      "taking": 117539,
      "prior": 90689,
      "occurrences": 81514,
      "discounting": 32034,
      "smoothing": 110268,
      "considered": 21285,
      "reduction": 97437,
      "10": 118,
      "discusses": 32303,
      "television": 120067,
      "broadcasts": 14692,
      "particular": 85376,
      "concentrate": 20552,
      "names": 76140,
      "entities": 37741,
      "attribute": 10177,
      "second": 104381,
      "wordword": 134347,
      "transitions": 124515,
      "common": 18853,
      "formulation": 45293,
      "identification": 51359,
      "characterized": 16553,
      "relatively": 98388,
      "sparse": 111220,
      "related": 97843,
      "reported": 99057,
      "north": 80368,
      "american": 5307,
      "combination": 18545,
      "combining": 18706,
      "idea": 51315,
      "classifiers": 17595,
      "voting": 131950,
      "van": 130809,
      "etal": 38745,
      "1998": 469,
      "manage": 66894,
      "performances": 86869,
      "arbitrary": 8578,
      "bagging": 11365,
      "boosting": 14441,
      "roughly": 103310,
      "gain": 46334,
      "fmeasure": 44727,
      "doubling": 33969,
      "size": 109909,
      "reveals": 102512,
      "inconsistent": 53440,
      "penn": 85855,
      "suggesting": 115345,
      "semiautomatic": 105569,
      "finding": 44264,
      "diversity": 32865,
      "popularity": 88149,
      "applying": 7227,
      "linguistics": 64603,
      "problems": 91294,
      "produced": 91947,
      "supply": 115951,
      "array": 9026,
      "offtheshelf": 81623,
      "downloadable": 33980,
      "code": 18069,
      "independently": 53784,
      "observed": 81216,
      "errors": 38355,
      "tend": 120138,
      "distributed": 32610,
      "concerned": 20685,
      "community": 19066,
      "outputs": 83157,
      "interdependent": 56057,
      "classifications": 17506,
      "address": 3646,
      "discrepancy": 32154,
      "switch": 116276,
      "constructing": 21670,
      "intention": 55913,
      "perform": 85938,
      "collaborative": 18356,
      "ensemble": 37576,
      "producing": 92013,
      "collaborate": 18351,
      "members": 68273,
      "underlying": 126674,
      "complementary": 19732,
      "loosely": 65231,
      "constrained": 21564,
      "chosen": 16938,
      "turning": 125986,
      "scripts": 104266,
      "interfaces": 56106,
      "semiautonomous": 105575,
      "transform": 124251,
      "successive": 115198,
      "final": 44094,
      "executable": 39962,
      "scripting": 104264,
      "equivalent": 38248,
      "subset": 114825,
      "stage": 112143,
      "transformed": 124282,
      "byproduct": 14995,
      "metaoutput": 68561,
      "nature": 76651,
      "consistent": 21370,
      "outputmetaoutput": 83154,
      "distinction": 32551,
      "permits": 87048,
      "perspicuous": 87180,
      "apparently": 6787,
      "topics": 122612,
      "pronouns": 92347,
      "correction": 24141,
      "misconceptions": 70295,
      "optimization": 82186,
      "concretely": 20742,
      "realized": 96121,
      "prototype": 93683,
      "interface": 56097,
      "simulation": 109647,
      "personal": 87112,
      "assistant": 9573,
      "cle": 17697,
      "people": 85868,
      "develop": 30169,
      "intended": 55865,
      "broadcoverage": 14693,
      "goal": 48325,
      "mind": 70143,
      "covering": 24530,
      "developers": 30328,
      "wish": 132743,
      "driven": 34147,
      "main": 66392,
      "types": 126243,
      "tends": 120170,
      "happen": 49436,
      "working": 134895,
      "starts": 112416,
      "thinking": 121866,
      "makes": 66752,
      "efforts": 35186,
      "labelled": 58557,
      "marginal": 67203,
      "ignored": 51673,
      "virtually": 131748,
      "stop": 113371,
      "practice": 88729,
      "bring": 14627,
      "things": 121860,
      "focus": 44733,
      "look": 65210,
      "doing": 33417,
      "point": 87795,
      "cases": 15637,
      "missing": 70318,
      "offered": 81574,
      "suggests": 115369,
      "filling": 44032,
      "holes": 50598,
      "turns": 125990,
      "symmetry": 116328,
      "picture": 87488,
      "extent": 41972,
      "lacks": 58773,
      "conversely": 23082,
      "offers": 81580,
      "duplicate": 34266,
      "thing": 121859,
      "relevant": 98522,
      "quite": 95396,
      "easy": 34471,
      "port": 88173,
      "fact": 42815,
      "involve": 56875,
      "mechanisms": 68052,
      "entropybased": 38176,
      "pruning": 94163,
      "backoff": 11313,
      "criterion": 24799,
      "entropy": 38160,
      "pruned": 94160,
      "shown": 108438,
      "computed": 20473,
      "exactly": 39724,
      "efficiently": 35136,
      "expressed": 41710,
      "leads": 61916,
      "threshold": 121946,
      "removed": 98879,
      "productionquality": 92064,
      "lm": 64858,
      "reduced": 97369,
      "26": 837,
      "increasing": 53656,
      "heuristic": 49870,
      "experimentally": 40735,
      "similar": 109071,
      "85": 1322,
      "exact": 39713,
      "giving": 48212,
      "marginally": 67217,
      "marked": 67224,
      "domainspecific": 33886,
      "attributes": 10191,
      "produce": 91870,
      "serves": 107313,
      "baseline": 12174,
      "individually": 53943,
      "determine": 30118,
      "flights": 44671,
      "air": 4584,
      "travel": 125540,
      "optimality": 82177,
      "theory": 121813,
      "frank": 45754,
      "karttunen": 57495,
      "gradient": 48595,
      "finitestate": 44543,
      "compact": 19109,
      "syllabification": 116299,
      "prince": 90664,
      "smolensky": 110261,
      "1993": 464,
      "case": 15575,
      "studies": 114176,
      "challenges": 16123,
      "arise": 9000,
      "receive": 96352,
      "telephone": 120064,
      "conversations": 23045,
      "undergone": 126667,
      "independent": 53760,
      "transcriptions": 124004,
      "additional": 3487,
      "separate": 106844,
      "projects": 92241,
      "dispersed": 32374,
      "geographically": 47856,
      "chronologically": 16948,
      "received": 96354,
      "variety": 130962,
      "managed": 66900,
      "group": 49133,
      "propagation": 92389,
      "consistency": 21353,
      "references": 97557,
      "integrate": 55744,
      "having": 49551,
      "addressed": 3780,
      "successfully": 115174,
      "wordnets": 133784,
      "structural": 113756,
      "candidates": 15131,
      "node": 80017,
      "target": 117581,
      "bests": 13485,
      "matches": 67377,
      "source": 110710,
      "map": 67103,
      "nominal": 80139,
      "16": 374,
      "ambiguity": 5278,
      "controlled": 22832,
      "anomaly": 6508,
      "detection": 29862,
      "past": 85630,
      "study": 114302,
      "autonomous": 10852,
      "dynamical": 34336,
      "needed": 76869,
      "distinguish": 32566,
      "contextdependent": 22352,
      "productions": 92065,
      "applications": 6894,
      "drives": 34159,
      "interpolation": 56204,
      "yield": 135328,
      "accurate": 2324,
      "pattern": 85710,
      "capabilities": 15164,
      "languagebased": 60357,
      "promising": 92262,
      "early": 34383,
      "anomalies": 6506,
      "faulty": 43212,
      "interfacing": 56110,
      "constraintbased": 21589,
      "principle": 90676,
      "major": 66530,
      "semantics": 105394,
      "vary": 131240,
      "declarative": 27971,
      "derivation": 29327,
      "implicitly": 52022,
      "interpretable": 56236,
      "plausible": 87678,
      "cause": 15818,
      "severe": 107878,
      "disco": 32022,
      "accounts": 2049,
      "adding": 3369,
      "comparing": 19499,
      "finders": 44263,
      "obtain": 81252,
      "desired": 29663,
      "gr": 48582,
      "difference": 30938,
      "effect": 34583,
      "factor": 42852,
      "length": 63350,
      "appears": 6811,
      "suited": 115421,
      "modifier": 74452,
      "partitioning": 85512,
      "tests": 120607,
      "differences": 30949,
      "testing": 120587,
      "values": 130790,
      "metrics": 69916,
      "like": 64016,
      "balanced": 11401,
      "commonly": 18956,
      "underestimate": 126654,
      "detect": 29793,
      "exist": 40016,
      "underestimation": 126658,
      "comes": 18774,
      "independence": 53757,
      "violated": 131733,
      "randomization": 95522,
      "outperformed": 82775,
      "individual": 53897,
      "learners": 62296,
      "majority": 66592,
      "vote": 131947,
      "ranks": 95699,
      "pairing": 83476,
      "stateofthe": 112557,
      "psycholinguistic": 94201,
      "insight": 55520,
      "units": 127230,
      "lot": 65315,
      "cpu": 24574,
      "cycles": 25484,
      "considerable": 21240,
      "double": 33960,
      "brought": 14710,
      "derivations": 29331,
      "convincingly": 23132,
      "backbone": 11285,
      "reinterpret": 97833,
      "tries": 125777,
      "maximize": 67508,
      "substructures": 114934,
      "emphasizing": 36144,
      "away": 11264,
      "opening": 81992,
      "optimizations": 82221,
      "retaining": 102361,
      "acceptable": 1893,
      "enhanced": 37501,
      "leftcorner": 63302,
      "examines": 39765,
      "programming": 92126,
      "contrast": 22687,
      "depthfirst": 29324,
      "parses": 85053,
      "connected": 21122,
      "spanning": 111185,
      "entire": 37700,
      "left": 63295,
      "kind": 57674,
      "nonlocal": 80233,
      "interpretation": 56260,
      "read": 95928,
      "viable": 131640,
      "enhancement": 37526,
      "substantially": 114881,
      "selectional": 104848,
      "imposed": 52305,
      "capture": 15269,
      "limitations": 64168,
      "proven": 93717,
      "frequently": 45867,
      "disambiguation": 31951,
      "anaphora": 6061,
      "value": 130774,
      "explore": 41514,
      "assuming": 9655,
      "reader": 95952,
      "familiar": 43049,
      "employs": 36326,
      "hpsgs": 50694,
      "background": 11297,
      "referential": 97563,
      "indices": 53884,
      "blocks": 14364,
      "readings": 96040,
      "violate": 131732,
      "theoretically": 121802,
      "satisfactory": 103651,
      "particularly": 85467,
      "estimation": 38648,
      "attributevalue": 10206,
      "complexity": 19899,
      "parsed": 84894,
      "circumstances": 16983,
      "material": 67443,
      "experimentation": 40740,
      "gaussian": 46541,
      "reduce": 97311,
      "overfitting": 83318,
      "lexicalised": 63847,
      "overlapping": 83333,
      "minimal": 70153,
      "situations": 109894,
      "number": 80833,
      "recovery": 97145,
      "packed": 83405,
      "computationally": 20445,
      "measuring": 67904,
      "highaccuracy": 50149,
      "little": 64785,
      "attention": 9787,
      "paid": 83413,
      "proposes": 93593,
      "metric": 69867,
      "allow": 5073,
      "comparisons": 19588,
      "architectures": 8774,
      "events": 39560,
      "measures": 67848,
      "calculated": 15020,
      "singlepass": 109837,
      "multistage": 75798,
      "demonstrate": 28652,
      "usefulness": 128953,
      "differ": 30927,
      "fundamental": 46083,
      "metonymy": 69863,
      "handbuilt": 49331,
      "database": 26637,
      "instead": 55651,
      "advantage": 4034,
      "interpret": 56206,
      "bunsetsu": 14966,
      "analyzing": 6034,
      "previously": 90585,
      "machinelearning": 66320,
      "decision": 27934,
      "list": 64687,
      "highest": 50221,
      "location": 64961,
      "category": 15781,
      "retrieve": 102449,
      "submitted": 114764,
      "contest": 21995,
      "held": 49680,
      "japan": 57182,
      "1999": 470,
      "scored": 104142,
      "teams": 119725,
      "22": 770,
      "participated": 85333,
      "comparative": 19199,
      "changed": 16377,
      "confirmed": 21023,
      "effectiveness": 34863,
      "foundations": 45352,
      "quantitative": 94860,
      "rigorous": 102848,
      "mathematical": 67454,
      "algorithmic": 4820,
      "conceptualized": 20677,
      "clear": 17718,
      "logical": 65001,
      "sound": 110700,
      "complete": 19760,
      "subjective": 114690,
      "weights": 132371,
      "weightbased": 132333,
      "introduces": 56604,
      "loglinear": 65044,
      "proof": 92370,
      "incomplete": 53419,
      "possibility": 88366,
      "defining": 28511,
      "parameter": 84709,
      "programs": 92135,
      "empirically": 36214,
      "smallscale": 110249,
      "experiment": 40449,
      "correct": 24089,
      "calculation": 15035,
      "approximately": 8463,
      "solve": 110586,
      "approximate": 8452,
      "searching": 104370,
      "probable": 90875,
      "czech": 25487,
      "discover": 32112,
      "prague": 88764,
      "label": 58381,
      "ar": 8483,
      "88": 1343,
      "unseen": 127515,
      "introduction": 56652,
      "shared": 107964,
      "syntactically": 116504,
      "nonoverlapping": 80260,
      "socalled": 110320,
      "overview": 83361,
      "taken": 117504,
      "briefly": 14619,
      "consensus": 21168,
      "minimization": 70192,
      "confusion": 21052,
      "distilling": 32527,
      "lattices": 61674,
      "string": 113596,
      "path": 85668,
      "posterior": 88479,
      "acoustics": 3029,
      "optimal": 82160,
      "decoder": 27991,
      "necessarily": 76744,
      "minimize": 70194,
      "minimizing": 70198,
      "probabilities": 90839,
      "replacing": 98947,
      "global": 48225,
      "local": 64903,
      "improving": 53067,
      "wordlevel": 133725,
      "confusions": 21059,
      "format": 45206,
      "compression": 20304,
      "spotting": 112045,
      "confidence": 20982,
      "reevaluation": 97488,
      "sources": 110876,
      "supplement": 115936,
      "amounts": 5327,
      "boost": 14428,
      "learner": 62290,
      "basque": 12552,
      "deals": 27873,
      "exploitation": 41449,
      "bases": 12493,
      "extracted": 42134,
      "focuses": 44885,
      "synonymy": 116353,
      "hypernymy": 51219,
      "adjectival": 3853,
      "treated": 125555,
      "morphological": 74674,
      "inflection": 54286,
      "suffixes": 115300,
      "prepositions": 89315,
      "analyser": 5437,
      "successful": 115148,
      "highly": 50285,
      "inflected": 54279,
      "effort": 35160,
      "write": 135080,
      "actual": 3149,
      "leaving": 63267,
      "extremely": 42595,
      "utilizing": 130565,
      "wide": 132486,
      "encyclopedia": 36785,
      "term": 120204,
      "descriptions": 29475,
      "semistructured": 105587,
      "pages": 83411,
      "utilize": 130494,
      "patterns": 85726,
      "html": 50698,
      "fragments": 45377,
      "containing": 21779,
      "discard": 32006,
      "extraneous": 42585,
      "clustering": 17932,
      "summarize": 115596,
      "resultant": 101422,
      "ir": 56929,
      "resembling": 100729,
      "public": 94232,
      "valuable": 130759,
      "retrieved": 102458,
      "favors": 43223,
      "retrieving": 102469,
      "workshop": 134996,
      "count": 24412,
      "aim": 4440,
      "maximal": 67499,
      "oriented": 82486,
      "counts": 24451,
      "leading": 61882,
      "tested": 120560,
      "isolate": 56972,
      "neglect": 76967,
      "contribute": 22745,
      "articles": 9165,
      "article": 9106,
      "japaneseenglish": 57210,
      "classified": 17507,
      "generic": 47789,
      "indefinite": 53756,
      "estimated": 38639,
      "clue": 17912,
      "conflict": 21033,
      "total": 122696,
      "score": 104043,
      "selected": 104720,
      "established": 38597,
      "adjusted": 3877,
      "scores": 104147,
      "succeeded": 115058,
      "reducing": 97406,
      "adjust": 3873,
      "cc": 15852,
      "tried": 125773,
      "characteristics": 16518,
      "newspapers": 79408,
      "locational": 64971,
      "got": 48530,
      "portable": 88176,
      "free": 45755,
      "je": 57222,
      "ee": 34577,
      "determined": 30136,
      "weight": 132321,
      "robustly": 103086,
      "expansion": 40370,
      "augment": 10250,
      "communicative": 19049,
      "encompasses": 36739,
      "nlg": 79511,
      "referring": 97571,
      "expression": 41734,
      "aggregation": 4356,
      "bridge": 14590,
      "generators": 47786,
      "intent": 55891,
      "associates": 9616,
      "concrete": 20738,
      "planned": 87633,
      "inferences": 54251,
      "communicates": 19022,
      "discourse": 32037,
      "planning": 87636,
      "treeadjoining": 125621,
      "ltag": 65719,
      "connect": 21116,
      "modal": 70473,
      "resource": 100827,
      "pragmatics": 88761,
      "simultaneously": 109669,
      "assess": 9469,
      "progress": 92138,
      "translates": 124580,
      "welldefined": 132399,
      "methodology": 69240,
      "designing": 29635,
      "conceptual": 20656,
      "behavior": 12645,
      "specified": 111620,
      "highquality": 50367,
      "kinds": 57686,
      "man": 66891,
      "woman": 132891,
      "vs": 131963,
      "great": 48992,
      "demonstrating": 28968,
      "gone": 48457,
      "quantifying": 94858,
      "person": 87095,
      "direction": 31818,
      "conditional": 20752,
      "compares": 19494,
      "maximizing": 67520,
      "distributions": 32753,
      "learnt": 63248,
      "somewhat": 110657,
      "surprisingly": 116131,
      "superior": 115677,
      "intuitively": 56672,
      "functioning": 46067,
      "notions": 80413,
      "reviews": 102594,
      "returned": 102481,
      "utilizes": 130550,
      "exceptional": 39929,
      "captured": 15424,
      "utility": 130475,
      "builds": 14903,
      "adopted": 3900,
      "wellmotivated": 132426,
      "captures": 15442,
      "items": 57107,
      "spontaneous": 112029,
      "demonstrated": 28908,
      "significantly": 108875,
      "wordbyword": 133701,
      "improvements": 52785,
      "historical": 50514,
      "dynamics": 34351,
      "random": 95472,
      "walk": 131995,
      "consider": 21187,
      "meanings": 67739,
      "reflecting": 97618,
      "life": 63978,
      "cycle": 25481,
      "verification": 131569,
      "russian": 103487,
      "age": 4278,
      "periods": 87042,
      "organizing": 82480,
      "encyclopedic": 36787,
      "question": 95012,
      "answering": 6598,
      "largescale": 61396,
      "describing": 29447,
      "organize": 82471,
      "targeting": 117777,
      "engineers": 37058,
      "examination": 39730,
      "studying": 114554,
      "split": 111939,
      "sampling": 103599,
      "favour": 43224,
      "reporting": 99078,
      "figures": 44017,
      "resampling": 100383,
      "statistically": 113179,
      "quantified": 94838,
      "statements": 112548,
      "settings": 107788,
      "devoted": 30474,
      "interested": 56064,
      "systematic": 116657,
      "survey": 116160,
      "understand": 126734,
      "relevance": 98501,
      "success": 115061,
      "esp": 38428,
      "availability": 10900,
      "focused": 44845,
      "np": 80797,
      "bit": 14236,
      "caching": 15011,
      "higherorder": 50217,
      "skipping": 109999,
      "kneserney": 57721,
      "explorations": 41507,
      "limits": 64311,
      "showing": 108395,
      "studied": 114150,
      "separately": 106868,
      "rarely": 95753,
      "significant": 108717,
      "interactions": 55978,
      "katz": 57497,
      "smoothed": 110266,
      "cutoffs": 25463,
      "reductions": 97470,
      "38": 978,
      "50": 1070,
      "depending": 29268,
      "89": 1352,
      "fair": 42991,
      "contains": 21813,
      "details": 29789,
      "proofs": 92379,
      "portability": 88175,
      "impact": 51854,
      "porting": 88182,
      "statistics": 113192,
      "atis": 9677,
      "microsoft": 70058,
      "manually": 67022,
      "upenn": 127811,
      "gathering": 46528,
      "modestly": 74439,
      "initialized": 55242,
      "outperforms": 82835,
      "initialization": 55233,
      "indomain": 53955,
      "achieving": 2924,
      "04": 23,
      "absolute": 1721,
      "58": 1130,
      "measured": 67831,
      "minimum": 70199,
      "achievable": 2404,
      "nbest": 76688,
      "lists": 64713,
      "worked": 134884,
      "12": 245,
      "antispam": 6758,
      "filtering": 44080,
      "unwanted": 127775,
      "electronic": 35245,
      "mail": 66389,
      "messages": 68505,
      "variants": 130872,
      "adaboost": 3165,
      "predictions": 89157,
      "singer": 109704,
      "99": 1450,
      "conclusions": 20726,
      "clearly": 17736,
      "outperform": 82685,
      "naive": 76009,
      "bayes": 12565,
      "f1": 42645,
      "highprecision": 50364,
      "misclassification": 70291,
      "costs": 24396,
      "preference": 89237,
      "selects": 104859,
      "papers": 84500,
      "extends": 41823,
      "learns": 63191,
      "classes": 17057,
      "motivation": 74876,
      "twofold": 126139,
      "subjectverb": 114710,
      "objectverb": 81152,
      "sensedisambiguated": 105711,
      "richer": 102803,
      "investigates": 56848,
      "enriching": 37565,
      "respectively": 101117,
      "906": 1375,
      "remedy": 98857,
      "weaknesses": 132205,
      "semantically": 105361,
      "solves": 110635,
      "neighborhood": 76989,
      "precise": 88780,
      "integrating": 55788,
      "learned": 62198,
      "untagged": 127762,
      "frequent": 45859,
      "inheritance": 55195,
      "disambiguated": 31945,
      "illustrating": 51751,
      "motivations": 74880,
      "founded": 45353,
      "feasible": 43240,
      "experimental": 40523,
      "allwords": 5201,
      "semcor": 105485,
      "analyzer": 6030,
      "prefixed": 89255,
      "did": 30909,
      "upperbound": 127824,
      "senseval": 105727,
      "comprised": 20332,
      "40": 1004,
      "nouns": 80438,
      "adjectives": 3860,
      "agreement": 4386,
      "75": 1248,
      "arbitration": 8595,
      "competing": 19615,
      "attained": 9723,
      "100": 157,
      "real": 96050,
      "wsd": 135168,
      "selection": 104762,
      "works": 134916,
      "embedded": 35352,
      "leaves": 63265,
      "room": 103258,
      "unsupervised": 127593,
      "discovery": 32133,
      "morphologically": 74758,
      "orthographic": 82592,
      "unannotated": 126583,
      "returns": 102483,
      "ranked": 95633,
      "pairs": 83479,
      "looking": 65223,
      "orthographically": 82601,
      "edit": 34529,
      "mutual": 75970,
      "rely": 98673,
      "morpheme": 74666,
      "concatenation": 20540,
      "distributional": 32692,
      "substrings": 114932,
      "affix": 4257,
      "frequency": 45836,
      "encouraging": 36776,
      "proportion": 92512,
      "qualitative": 94549,
      "discovered": 32122,
      "kanji": 57489,
      "lack": 58675,
      "delimiters": 28612,
      "laborintensive": 58665,
      "lexicosyntactic": 63939,
      "vulnerable": 131972,
      "unsegmented": 127558,
      "simplicity": 109565,
      "comparable": 19125,
      "surpassing": 116114,
      "stateoftheart": 112566,
      "analyzers": 6032,
      "chinese": 16736,
      "additionally": 3596,
      "twolevel": 126148,
      "granularities": 48742,
      "compatible": 19596,
      "bracket": 14547,
      "ensembles": 37625,
      "disambiguating": 31947,
      "samples": 103578,
      "ensemblebased": 37623,
      "senseval2": 105728,
      "unigrams": 127159,
      "bigrams": 14008,
      "cooccurrences": 23241,
      "classifier": 17515,
      "assessing": 9505,
      "instance": 55591,
      "difficulty": 31691,
      "pairwise": 83669,
      "agree": 4382,
      "instances": 55616,
      "included": 53243,
      "bigram": 14005,
      "graph": 48754,
      "infer": 54096,
      "identity": 51637,
      "exploring": 41658,
      "morphemes": 74670,
      "raw": 95847,
      "shares": 108124,
      "certain": 15929,
      "brown": 14718,
      "et": 38679,
      "al": 4588,
      "1992": 463,
      "grown": 49185,
      "analyze": 5950,
      "adjacent": 3848,
      "eigenvector": 35219,
      "decomposition": 28157,
      "nearestneighbor": 76727,
      "twodimensional": 126135,
      "rendering": 98888,
      "neighborhoods": 76991,
      "extending": 41814,
      "morphology": 74782,
      "ask": 9294,
      "function": 45995,
      "tion": 122214,
      "primarily": 90637,
      "marks": 67269,
      "plurals": 87768,
      "3rd": 995,
      "deep": 28194,
      "broad": 14666,
      "mrs": 74936,
      "play": 87686,
      "role": 103153,
      "pos": 88206,
      "tool": 122368,
      "requiring": 100333,
      "easily": 34427,
      "command": 18787,
      "detects": 30102,
      "datasets": 27287,
      "merging": 68481,
      "unix": 127357,
      "readily": 95963,
      "showed": 108367,
      "wellformed": 132410,
      "uninflected": 127167,
      "languageindependent": 60359,
      "icons": 51311,
      "answer": 6515,
      "needs": 76883,
      "clues": 17916,
      "intelligent": 55840,
      "reasonable": 96206,
      "purely": 94412,
      "chart": 16644,
      "stores": 113394,
      "intermediate": 56130,
      "prolog": 92247,
      "engine": 37025,
      "brings": 14641,
      "polynomial": 88019,
      "depends": 29279,
      "predicates": 88860,
      "turing": 125951,
      "questions": 95267,
      "reply": 98970,
      "french": 45799,
      "robert": 102989,
      "computer": 20476,
      "incapable": 53209,
      "passing": 85625,
      "probe": 90877,
      "perceptual": 85933,
      "associations": 9630,
      "humans": 51067,
      "live": 64845,
      "embodied": 36043,
      "appropriately": 8447,
      "contrary": 22676,
      "claim": 17007,
      "collection": 18454,
      "issuing": 57083,
      "indexes": 53809,
      "million": 70093,
      "shed": 108148,
      "light": 63985,
      "sub": 114643,
      "cognition": 18261,
      "mistaken": 70346,
      "compound": 20150,
      "splitting": 111952,
      "compounded": 20156,
      "challenge": 15986,
      "gold": 48428,
      "gains": 46379,
      "bleu": 14274,
      "germanenglish": 47930,
      "im": 51758,
      "afraid": 4270,
      "2001": 541,
      "brief": 14609,
      "generalaudience": 46735,
      "focusing": 44920,
      "datadriven": 26654,
      "strike": 113592,
      "factorization": 42871,
      "resolve": 100796,
      "discriminative": 32205,
      "factored": 42860,
      "estimate": 38627,
      "mainly": 66477,
      "factorizing": 42878,
      "sequential": 107214,
      "event": 39494,
      "resolved": 100809,
      "adopt": 3889,
      "reflect": 97599,
      "factorizations": 42874,
      "combine": 18614,
      "generalize": 46802,
      "traversing": 125547,
      "formulate": 45271,
      "categorization": 15761,
      "nearest": 76718,
      "firstly": 44562,
      "predication": 88865,
      "according": 1965,
      "neighbors": 76996,
      "speaking": 111338,
      "uneven": 127055,
      "sensitive": 105732,
      "fixed": 44594,
      "bias": 13789,
      "knn": 57725,
      "numbers": 80999,
      "deciding": 27930,
      "preliminary": 89261,
      "properly": 92429,
      "classify": 17647,
      "smaller": 110225,
      "crossvalidation": 25087,
      "allowed": 5103,
      "computation": 20343,
      "sort": 110682,
      "subsumes": 114935,
      "accessibility": 1929,
      "check": 16691,
      "characterize": 16547,
      "naturallanguage": 76632,
      "anusaaraka": 6761,
      "overcoming": 83306,
      "barrier": 11431,
      "india": 53812,
      "indian": 53815,
      "accessible": 1930,
      "load": 64895,
      "divided": 32908,
      "image": 51759,
      "equivalents": 38259,
      "special": 111348,
      "notation": 80382,
      "devised": 30465,
      "short": 108199,
      "weeks": 132315,
      "postedited": 88469,
      "grammatically": 48731,
      "style": 114564,
      "scenario": 103818,
      "assisted": 9582,
      "telugu": 120072,
      "kannada": 57490,
      "marathi": 67171,
      "bengali": 13031,
      "punjabi": 94399,
      "hindi": 50459,
      "near": 76708,
      "future": 46256,
      "everybody": 39624,
      "pitch": 87571,
      "build": 14739,
      "connecting": 21133,
      "servers": 107312,
      "follows": 44991,
      "strings": 113603,
      "implies": 52039,
      "preservation": 89915,
      "going": 48427,
      "narrow": 76187,
      "areas": 8888,
      "specialized": 111383,
      "putting": 94470,
      "remain": 98768,
      "create": 24601,
      "pertaining": 87186,
      "bank": 11418,
      "paninian": 83692,
      "organisational": 82461,
      "technologies": 120022,
      "overcome": 83272,
      "acquiring": 3042,
      "transforming": 124476,
      "storing": 113402,
      "organisations": 82462,
      "consisting": 21445,
      "supporting": 116025,
      "utilities": 130474,
      "indexing": 53810,
      "transcription": 123994,
      "downstream": 33985,
      "integrated": 55771,
      "workflow": 134892,
      "decisionmaking": 27951,
      "processes": 91603,
      "deployment": 29316,
      "designers": 29634,
      "field": 43931,
      "phonological": 87307,
      "contrasts": 22744,
      "unit": 127200,
      "phoneme": 87277,
      "fl": 44627,
      "family": 43055,
      "phonemic": 87284,
      "distinctive": 32557,
      "cantonese": 15161,
      "dutch": 34276,
      "mandarin": 66911,
      "compressed": 20296,
      "ukrainian": 126549,
      "verified": 131573,
      "genre": 47828,
      "nonrelated": 80286,
      "creates": 24690,
      "interrelated": 56296,
      "covers": 24553,
      "spheres": 111927,
      "activities": 3130,
      "verify": 131578,
      "statement": 112546,
      "invariant": 56680,
      "ideographic": 51644,
      "tabular": 117275,
      "tutorial": 126000,
      "nondeterministic": 80187,
      "pushdown": 94457,
      "automata": 10423,
      "earleys": 34373,
      "lr": 65578,
      "annotating": 6264,
      "predicateargument": 88853,
      "initiated": 55251,
      "multilayered": 75187,
      "dedicated": 28179,
      "aligning": 4933,
      "decisions": 27954,
      "realisation": 96098,
      "conclude": 20708,
      "remarks": 98855,
      "relates": 97935,
      "generalized": 46825,
      "smt": 110277,
      "begun": 12641,
      "treestructured": 125721,
      "easier": 34410,
      "implement": 51927,
      "modify": 74456,
      "service": 107319,
      "arrive": 9028,
      "abstract": 1760,
      "termination": 120251,
      "condition": 20751,
      "alignment": 4939,
      "generalizing": 46843,
      "recipe": 96787,
      "train": 122896,
      "sa": 103518,
      "rd": 95881,
      "regarding": 97655,
      "modern": 74380,
      "greek": 49076,
      "infrastructure": 55172,
      "faced": 42747,
      "steps": 113316,
      "subsequently": 114821,
      "insights": 55527,
      "spelling": 111898,
      "checker": 16694,
      "thesaurus": 121838,
      "outlining": 82639,
      "efficiencies": 35019,
      "inefficiencies": 54069,
      "directions": 31835,
      "chunk": 16952,
      "explored": 41619,
      "specifying": 111626,
      "contextual": 22440,
      "relying": 98754,
      "finegrained": 44331,
      "seen": 104529,
      "coarse": 18050,
      "bi": 13781,
      "aligned": 4908,
      "inbetween": 53205,
      "starting": 112407,
      "nonliteral": 80232,
      "difficult": 31605,
      "copes": 23277,
      "coding": 18243,
      "middle": 70068,
      "ground": 49094,
      "englishgerman": 37370,
      "fuse": 46222,
      "sentimental": 106820,
      "education": 34568,
      "sentiment": 106559,
      "subjectivity": 114704,
      "summarization": 115471,
      "cuts": 25464,
      "seeks": 104523,
      "identify": 51473,
      "viewpoints": 131721,
      "span": 111116,
      "classifying": 17668,
      "movie": 74892,
      "review": 102532,
      "thumbs": 121952,
      "polarity": 87903,
      "applies": 7149,
      "just": 57460,
      "portions": 88186,
      "greatly": 49042,
      "incorporation": 53573,
      "crosssentence": 25067,
      "recommendations": 97090,
      "visualization": 131823,
      "excellence": 39910,
      "european": 38766,
      "mining": 70220,
      "wg1": 132468,
      "carry": 15549,
      "comprehensive": 20244,
      "roadmap": 102988,
      "followup": 44995,
      "anticipated": 6755,
      "technological": 120018,
      "surge": 116085,
      "interests": 56096,
      "realworld": 96148,
      "childrens": 16729,
      "longitudinal": 65173,
      "series": 107267,
      "forms": 45230,
      "collocation": 18517,
      "connectivity": 21153,
      "hub": 50710,
      "authority": 10376,
      "children": 16722,
      "simply": 109619,
      "late": 61560,
      "follow": 44935,
      "faster": 43159,
      "dimension": 31741,
      "slower": 110103,
      "emerge": 36048,
      "presumptions": 89966,
      "reflects": 97624,
      "later": 61645,
      "shift": 108173,
      "authorities": 10375,
      "constantly": 21517,
      "adults": 3928,
      "rich": 102724,
      "area": 8865,
      "theories": 121808,
      "unification": 127098,
      "scientific": 103989,
      "terminological": 120252,
      "iso": 56969,
      "initiative": 55252,
      "portal": 88177,
      "markup": 67270,
      "associate": 9591,
      "structuring": 114121,
      "sharing": 108127,
      "initially": 55246,
      "shifts": 108190,
      "interdisciplinary": 56058,
      "ongoing": 81732,
      "denote": 29006,
      "departures": 29042,
      "correspondence": 24268,
      "deviations": 30453,
      "occurred": 81508,
      "resemble": 100726,
      "parliamentary": 84872,
      "proceedings": 91412,
      "proceeds": 91413,
      "monolingually": 74633,
      "accordingly": 2011,
      "extractors": 42574,
      "biomedical": 14164,
      "revealed": 102508,
      "decreasing": 28176,
      "impossibility": 52309,
      "limitation": 64159,
      "external": 41992,
      "terminologies": 120254,
      "exploits": 41489,
      "acquired": 3037,
      "reliability": 98604,
      "extractor": 42571,
      "challenging": 16221,
      "compositionality": 20140,
      "interpreting": 56285,
      "relaxed": 98424,
      "hypothesis": 51261,
      "face": 42728,
      "fixing": 44626,
      "compositional": 20105,
      "believe": 12694,
      "radical": 95438,
      "collections": 18501,
      "concerns": 20692,
      "emphasizes": 36142,
      "evolution": 39690,
      "accounted": 2047,
      "composition": 20090,
      "expresses": 41730,
      "criteria": 24790,
      "interpretations": 56278,
      "en": 36346,
      "la": 58376,
      "insufficient": 55719,
      "freely": 45778,
      "intends": 55875,
      "revised": 102644,
      "objective": 81057,
      "maintenance": 66528,
      "enrichment": 37571,
      "presentation": 89771,
      "organization": 82466,
      "archive": 8861,
      "conducted": 20909,
      "international": 56182,
      "standardization": 112343,
      "initiatives": 55254,
      "tei": 120058,
      "isotc": 56983,
      "37sc": 977,
      "note": 80384,
      "dimensional": 31748,
      "populated": 88162,
      "spaces": 111089,
      "topology": 122685,
      "unique": 127180,
      "fingerprints": 44528,
      "coherent": 18313,
      "1000": 173,
      "comprising": 20334,
      "13": 278,
      "reflection": 97621,
      "annotate": 6124,
      "outcome": 82615,
      "machinery": 66329,
      "intrinsic": 56350,
      "net": 77112,
      "projection": 92230,
      "opposed": 82147,
      "adult": 3926,
      "end": 36792,
      "ii": 51694,
      "phonotactics": 87321,
      "ilp": 51755,
      "phonotactic": 87320,
      "inductive": 54046,
      "discipline": 32016,
      "operators": 82078,
      "approaching": 8407,
      "direct": 31780,
      "bootstrapping": 14456,
      "courses": 24467,
      "ontological": 81826,
      "seed": 104498,
      "deployed": 29305,
      "strengths": 113571,
      "predicted": 88956,
      "fractal": 45362,
      "internet": 56189,
      "flows": 44687,
      "mean": 67594,
      "conformity": 21039,
      "perceived": 85911,
      "formally": 45203,
      "solving": 110638,
      "supplemented": 115946,
      "exists": 40344,
      "determination": 30117,
      "shall": 107907,
      "simplest": 109564,
      "opinion": 82080,
      "suppose": 116043,
      "logics": 65029,
      "recognising": 96799,
      "textual": 121672,
      "entailment": 37651,
      "handle": 49373,
      "rte": 103340,
      "firstorder": 44567,
      "contribution": 22767,
      "saturation": 103661,
      "subgraph": 114669,
      "graphes": 48927,
      "outils": 82623,
      "shifting": 108184,
      "reduces": 97378,
      "operation": 82051,
      "affects": 4242,
      "objectives": 81130,
      "realistic": 96100,
      "elaborated": 35225,
      "ladl": 58777,
      "transferred": 124227,
      "exploited": 41450,
      "determiners": 30140,
      "objects": 81146,
      "assessments": 9532,
      "validation": 130737,
      "recursive": 97281,
      "transition": 124479,
      "deeper": 28441,
      "treebanks": 125657,
      "86": 1328,
      "92": 1395,
      "strict": 113588,
      "romance": 103248,
      "portuguese": 88193,
      "fr": 45359,
      "essential": 38538,
      "head": 49574,
      "idioms": 51654,
      "guard": 49221,
      "excluded": 39954,
      "com": 18541,
      "tune": 125926,
      "instrument": 55715,
      "valence": 130707,
      "em": 35333,
      "matrices": 67468,
      "polish": 87979,
      "frame": 45380,
      "forests": 45054,
      "nonprobabilistic": 80275,
      "postprocessing": 88506,
      "incorrect": 53574,
      "detected": 29824,
      "phenomenon": 87250,
      "newly": 79262,
      "valid": 130708,
      "combined": 18646,
      "reaches": 95912,
      "45": 1032,
      "39": 986,
      "framebased": 45390,
      "contemporary": 21844,
      "practices": 88739,
      "domainindependent": 33713,
      "unrestricted": 127510,
      "refer": 97494,
      "correctness": 24184,
      "achievements": 2726,
      "evaluations": 39450,
      "finnish": 44554,
      "ccg": 15858,
      "pcfg": 85826,
      "link": 64632,
      "refers": 97580,
      "normal": 80325,
      "inputs": 55478,
      "concentrated": 20553,
      "misspelled": 70341,
      "counterparts": 24439,
      "overall": 83209,
      "decline": 27981,
      "declines": 27983,
      "tens": 120176,
      "percentage": 85920,
      "misspellings": 70344,
      "purposebuilt": 94444,
      "443": 1029,
      "return": 102479,
      "ungrammatical": 127081,
      "608": 1157,
      "340": 954,
      "149": 335,
      "prediction": 89024,
      "attempts": 9761,
      "latent": 61572,
      "lsa": 65585,
      "reliable": 98608,
      "longdistance": 65148,
      "lsabased": 65586,
      "cache": 15008,
      "reranking": 100365,
      "4gram": 1059,
      "highorder": 50356,
      "similarities": 109175,
      "exposure": 41692,
      "simulate": 109630,
      "paragraph": 84561,
      "association": 9618,
      "w1": 131978,
      "w2": 131980,
      "strongly": 113741,
      "increases": 53635,
      "decreases": 28174,
      "slightly": 110019,
      "operationalizing": 82065,
      "overestimates": 83314,
      "prime": 90658,
      "targets": 117789,
      "featurebased": 43332,
      "preserves": 89931,
      "toolkit": 122408,
      "conception": 20594,
      "ease": 34405,
      "spreading": 112053,
      "eventually": 39618,
      "filters": 44092,
      "adequate": 3830,
      "anaphoric": 6073,
      "compounds": 20158,
      "modular": 74463,
      "java": 57219,
      "custom": 25432,
      "theorem": 121784,
      "proving": 94146,
      "atoms": 9685,
      "formulas": 45270,
      "environment": 38183,
      "opensource": 82015,
      "pivot": 87578,
      "mildly": 70084,
      "contextsensitive": 22438,
      "formalisms": 45188,
      "treebased": 125667,
      "multicomponent": 75059,
      "tuples": 125948,
      "meroitic": 68486,
      "decipherment": 27933,
      "undeciphered": 126652,
      "ancient": 6089,
      "cognates": 18260,
      "borrowed": 14474,
      "paired": 83469,
      "une": 127049,
      "du": 34222,
      "pour": 88625,
      "automatique": 10845,
      "elements": 35276,
      "creole": 24741,
      "dialect": 30516,
      "inspiration": 55556,
      "1984": 456,
      "manual": 66971,
      "1997": 468,
      "nous": 80451,
      "dans": 25537,
      "les": 63392,
      "langue": 60999,
      "des": 29371,
      "sur": 116053,
      "anglais": 6111,
      "lexicales": 63845,
      "systeme": 116701,
      "ce": 15879,
      "layered": 61755,
      "kernel": 57531,
      "dialects": 30534,
      "instantiated": 55644,
      "multidialectal": 75064,
      "fuzzy": 46328,
      "collocations": 18520,
      "lexicography": 63870,
      "classical": 17084,
      "crisp": 24788,
      "grade": 48583,
      "membership": 68274,
      "express": 41695,
      "ttest": 125912,
      "utilized": 130538,
      "examined": 39761,
      "contained": 21772,
      "novels": 80785,
      "constituting": 21560,
      "gutenberg": 49286,
      "drawbacks": 34121,
      "decide": 27926,
      "251": 825,
      "whats": 132469,
      "framenet": 45393,
      "multidocument": 75073,
      "evolving": 39702,
      "proposing": 93621,
      "essentially": 38574,
      "composed": 20064,
      "stages": 112156,
      "trying": 125901,
      "representative": 100006,
      "clusters": 17963,
      "jargon": 57217,
      "evaluates": 39030,
      "variation": 130886,
      "mwes": 75995,
      "predicative": 88867,
      "clearcut": 17733,
      "involved": 56879,
      "largecoverage": 61330,
      "consequences": 21174,
      "active": 3102,
      "voice": 131922,
      "passive": 85628,
      "transitive": 124518,
      "patient": 85697,
      "receiving": 96381,
      "action": 3075,
      "purpose": 94422,
      "boolean": 14427,
      "subgroups": 114673,
      "permutation": 87052,
      "isomorphism": 56982,
      "property": 92495,
      "resulted": 101424,
      "evolve": 39697,
      "transformations": 124276,
      "families": 43052,
      "derivational": 29328,
      "readable": 95950,
      "regularities": 97736,
      "wordbased": 133695,
      "binary": 14116,
      "analogies": 5410,
      "hold": 50586,
      "doubt": 33972,
      "monotonicity": 74647,
      "assertion": 9464,
      "entails": 37685,
      "know": 57728,
      "epidemic": 38208,
      "spread": 112046,
      "quickly": 95388,
      "entail": 37649,
      "downward": 34064,
      "resourcelean": 100899,
      "recovering": 97143,
      "bayesian": 12577,
      "discovering": 32124,
      "typological": 126481,
      "implications": 51989,
      "typology": 126503,
      "universal": 127274,
      "implication": 51988,
      "facts": 42904,
      "come": 18758,
      "assisting": 9587,
      "wellknown": 132415,
      "deserve": 29508,
      "careful": 15507,
      "cope": 23270,
      "exploration": 41495,
      "tracking": 122746,
      "mentions": 68443,
      "ne": 76706,
      "patternbased": 85722,
      "originally": 82566,
      "termhood": 120249,
      "loglikelihood": 65043,
      "olac": 81638,
      "extension": 41834,
      "dravidian": 34107,
      "2000": 537,
      "character": 16416,
      "reviewed": 102589,
      "cognitive": 18262,
      "principles": 90684,
      "difficulties": 31675,
      "accomplished": 1954,
      "restructuring": 101353,
      "linkage": 64640,
      "mentioned": 68434,
      "discursive": 32230,
      "limit": 64152,
      "referents": 97567,
      "fresh": 45883,
      "xmlbased": 135215,
      "widely": 132533,
      "recognized": 97059,
      "proliferation": 92246,
      "runs": 103473,
      "counter": 24422,
      "standards": 112351,
      "mandatory": 66916,
      "morphosyntactic": 74813,
      "applicability": 6818,
      "analyse": 5430,
      "phrasestructure": 87471,
      "interaction": 55938,
      "syntactical": 116500,
      "polarities": 87900,
      "translated": 124550,
      "consequence": 21172,
      "usual": 130397,
      "sheds": 108160,
      "grouping": 49147,
      "synonyms": 116351,
      "lemma": 63326,
      "pair": 83418,
      "lexicographers": 63865,
      "judge": 57424,
      "interannotator": 56043,
      "ranges": 95616,
      "67": 1189,
      "annotator": 6484,
      "71": 1218,
      "mathematics": 67465,
      "recursion": 97278,
      "universals": 127343,
      "conflicting": 21035,
      "soft": 110483,
      "constructs": 21707,
      "reasoning": 96221,
      "partofspeech": 85523,
      "assamese": 9457,
      "signed": 108711,
      "generalpurpose": 46880,
      "simplification": 109575,
      "aids": 4439,
      "proteinprotein": 93665,
      "ppis": 88691,
      "accelerate": 1867,
      "informatics": 54350,
      "researchers": 100674,
      "collaborating": 18353,
      "advance": 3929,
      "stateofart": 112552,
      "ppi": 88689,
      "neglected": 76969,
      "characteristic": 16514,
      "substantial": 114849,
      "poses": 88259,
      "nontechnical": 80305,
      "abstracts": 1825,
      "pipeline": 87521,
      "ripple": 102861,
      "simplified": 109599,
      "ps": 94175,
      "17th": 410,
      "century": 15921,
      "genitive": 47824,
      "omission": 81650,
      "letters": 63407,
      "writers": 135082,
      "plural": 87766,
      "symbols": 116325,
      "abbreviations": 1466,
      "thanks": 121768,
      "isolation": 56979,
      "punctuation": 94392,
      "continent": 22592,
      "16th": 394,
      "19th": 472,
      "standardised": 112342,
      "problematic": 91290,
      "speakers": 111306,
      "errant": 38270,
      "complaint": 19723,
      "visiting": 131776,
      "websites": 132299,
      "misuses": 70355,
      "mistakes": 70348,
      "misuse": 70354,
      "noticed": 80401,
      "accompany": 1945,
      "highlight": 50254,
      "lexemes": 63722,
      "trace": 122717,
      "origin": 82499,
      "governing": 48541,
      "arabicfrench": 8572,
      "sport": 112037,
      "arabic": 8489,
      "doesnt": 33413,
      "reason": 96195,
      "ameliorate": 5304,
      "integrates": 55783,
      "transliteration": 125521,
      "nooj": 80320,
      "platform": 87651,
      "transducers": 124021,
      "muc": 75022,
      "conferences": 20981,
      "albanian": 4662,
      "inside": 55516,
      "linear": 64329,
      "concatenated": 20533,
      "declensions": 27980,
      "treat": 125549,
      "exceed": 39901,
      "adverbs": 4098,
      "critique": 24851,
      "findings": 44287,
      "tokens": 122296,
      "ratio": 95830,
      "course": 24464,
      "speed": 111861,
      "author": 10359,
      "hypothesizes": 51291,
      "englishrussian": 37425,
      "dimensionality": 31757,
      "visualize": 131835,
      "embedding": 35368,
      "2d": 874,
      "3d": 991,
      "plot": 87758,
      "draw": 34109,
      "expert": 41213,
      "chapter": 16409,
      "assume": 9644,
      "systematically": 116690,
      "spatial": 111261,
      "markers": 67230,
      "characterizing": 16556,
      "accounting": 2048,
      "highlighted": 50276,
      "commonsense": 18981,
      "geometry": 47868,
      "sketched": 109966,
      "pragmatic": 88747,
      "formalized": 45200,
      "adequacy": 3824,
      "entits": 37901,
      "tude": 125924,
      "descriptive": 29501,
      "analyzed": 6019,
      "static": 113060,
      "crosslinguistic": 25053,
      "validity": 130748,
      "underlies": 126672,
      "fulfilling": 45926,
      "generality": 46746,
      "dependences": 29060,
      "partwhole": 85597,
      "developmental": 30437,
      "languagespecific": 60988,
      "variability": 130830,
      "onwards": 81860,
      "bringing": 14640,
      "capacities": 15227,
      "determinants": 30116,
      "radically": 95440,
      "grounded": 49104,
      "paradigmatic": 84557,
      "analogy": 5416,
      "paradigms": 84558,
      "constitute": 21557,
      "implicit": 52002,
      "table": 117252,
      "script": 104251,
      "operations": 82066,
      "sake": 103540,
      "simplifications": 109598,
      "wikipedia": 132639,
      "histories": 50542,
      "deriving": 29370,
      "metadata": 68534,
      "edits": 34563,
      "prepared": 89298,
      "dont": 33944,
      "interesting": 56066,
      "negative": 76921,
      "romanian": 103250,
      "recorded": 97131,
      "motifs": 74830,
      "le": 61839,
      "semantical": 105359,
      "think": 121864,
      "piece": 87493,
      "today": 122235,
      "opinions": 82115,
      "excellent": 39914,
      "consumer": 21719,
      "numerous": 81012,
      "product": 92024,
      "forums": 45310,
      "blogs": 14372,
      "impossible": 52310,
      "informed": 55167,
      "purchase": 94407,
      "seller": 104966,
      "monitor": 74542,
      "positive": 88314,
      "neutral": 78753,
      "appear": 6794,
      "langues": 61000,
      "italian": 57084,
      "specialty": 111402,
      "pedagogical": 85841,
      "rapidly": 95725,
      "precede": 88771,
      "lifecycle": 63980,
      "progressively": 92195,
      "replaced": 98932,
      "par": 84516,
      "montrons": 74657,
      "depend": 29043,
      "specialise": 111376,
      "est": 38575,
      "plus": 87769,
      "forme": 45223,
      "ou": 82610,
      "editors": 34562,
      "cnl": 17982,
      "specifically": 111521,
      "cnls": 17983,
      "ace": 2396,
      "materials": 67444,
      "expansive": 40379,
      "bulgarian": 14948,
      "webbased": 132273,
      "placed": 87611,
      "formation": 45213,
      "primary": 90645,
      "principal": 90669,
      "keywords": 57642,
      "debate": 27887,
      "tagset": 117491,
      "facilitate": 42760,
      "standardize": 112344,
      "25": 810,
      "tagsets": 117493,
      "dataset": 26702,
      "partsofspeech": 85595,
      "competitive": 19631,
      "accuracies": 2057,
      "management": 66902,
      "englishlithuanianenglish": 37418,
      "deploying": 29314,
      "run": 103459,
      "windows": 132717,
      "linux": 64684,
      "mac": 65736,
      "virtual": 131739,
      "supported": 116021,
      "managing": 66910,
      "accessing": 1934,
      "notes": 80390,
      "continuation": 22602,
      "closely": 17848,
      "obviously": 81487,
      "answers": 6725,
      "fence": 43877,
      "modeldriven": 72354,
      "specification": 111610,
      "modelbased": 72349,
      "processors": 91862,
      "decouples": 28167,
      "unlike": 127425,
      "constrain": 21563,
      "ambiguities": 5275,
      "relatedness": 97915,
      "collocational": 18519,
      "wikipediabased": 132689,
      "mixed": 70393,
      "index": 53803,
      "spearman": 111342,
      "rho": 102717,
      "coefficient": 18247,
      "079": 71,
      "075": 65,
      "2007": 554,
      "087": 91,
      "078": 69,
      "2009": 558,
      "svm": 116230,
      "appendix": 6815,
      "esa": 38419,
      "2011": 562,
      "enhance": 37462,
      "section": 104484,
      "pronunciation": 92364,
      "persian": 87077,
      "synthesis": 116593,
      "popular": 88074,
      "researches": 100720,
      "possess": 88359,
      "exceptions": 39933,
      "vowels": 131958,
      "absent": 1720,
      "existence": 40035,
      "homograph": 50615,
      "phonetic": 87286,
      "pronunciations": 92369,
      "normalization": 80332,
      "pubmed": 94380,
      "affiliation": 4252,
      "normalizing": 80362,
      "canonical": 15151,
      "involves": 56888,
      "98": 1446,
      "associating": 9617,
      "authors": 10377,
      "geopolitical": 47871,
      "potentially": 88601,
      "social": 110323,
      "organizations": 82470,
      "topic": 122496,
      "weak": 132177,
      "augmenting": 10339,
      "nlms": 79556,
      "mars": 67278,
      "correcting": 24136,
      "ocr": 81525,
      "citations": 16992,
      "normalized": 80355,
      "country": 24450,
      "graphical": 48928,
      "download": 33978,
      "tuned": 125931,
      "readability": 95943,
      "shotgun": 108358,
      "versions": 131620,
      "optimized": 82229,
      "indexed": 53808,
      "downloaded": 33981,
      "xml": 135214,
      "widespread": 132613,
      "adapt": 3169,
      "tigerxml": 121965,
      "binding": 14142,
      "registry": 97688,
      "showcasing": 108365,
      "handling": 49412,
      "topological": 122676,
      "tandem": 117575,
      "inverse": 56696,
      "lambda": 58790,
      "lambdacalculus": 58792,
      "operator": 82077,
      "combinatorial": 18604,
      "categorial": 15718,
      "directed": 31811,
      "assign": 9538,
      "robot": 103008,
      "automatizing": 10847,
      "instructions": 55708,
      "composing": 20079,
      "querying": 95004,
      "ewc": 39707,
      "adhoc": 3845,
      "utilised": 130469,
      "ntcir": 80815,
      "interrater": 56294,
      "formality": 45189,
      "dimensions": 31763,
      "writing": 135085,
      "likert": 64146,
      "scale": 103702,
      "rating": 95818,
      "bottlenecks": 14500,
      "scoring": 104223,
      "sentencelevel": 106142,
      "requirement": 100230,
      "subproblem": 114801,
      "dale": 25521,
      "intentional": 55916,
      "influences": 54323,
      "1986": 458,
      "recency": 96384,
      "17": 398,
      "424": 1020,
      "performing": 86941,
      "60": 1146,
      "recencybased": 96386,
      "2006": 552,
      "undertaken": 127035,
      "tales": 117555,
      "ottoman": 82609,
      "mary": 67280,
      "essay": 38528,
      "proposal": 92518,
      "literary": 64733,
      "enabled": 36369,
      "singularities": 109863,
      "transcribing": 123991,
      "implied": 52037,
      "succession": 115197,
      "elementary": 35271,
      "alternating": 5218,
      "conditions": 20814,
      "discussing": 32315,
      "objectoriented": 81143,
      "consideration": 21280,
      "persons": 87153,
      "occupations": 81492,
      "actions": 3087,
      "machines": 66330,
      "animals": 6118,
      "thoughts": 121911,
      "causes": 15836,
      "ambiguous": 5289,
      "associativity": 9639,
      "precedence": 88773,
      "enforce": 37002,
      "industrial": 54056,
      "stable": 112111,
      "portfolio": 88181,
      "welldocumented": 132404,
      "interoperability": 56197,
      "biggest": 14002,
      "arises": 9002,
      "population": 88164,
      "adapted": 3280,
      "planner": 87634,
      "restaurant": 101317,
      "match": 67354,
      "presentations": 89772,
      "adapting": 3303,
      "individuals": 53946,
      "planners": 87635,
      "documented": 33127,
      "demonstration": 28985,
      "affecting": 4231,
      "responses": 101251,
      "oral": 82259,
      "sociolinguistic": 110477,
      "1968": 449,
      "handwritten": 49429,
      "transcribed": 123982,
      "transcriber": 123989,
      "audio": 10223,
      "stored": 113383,
      "consultation": 21713,
      "added": 3355,
      "transducer": 124017,
      "cascades": 15573,
      "cascade": 15567,
      "recognize": 97050,
      "speaker": 111280,
      "lexica": 63723,
      "converted": 23099,
      "campaign": 15087,
      "partir": 85506,
      "moderne": 74429,
      "definitional": 28531,
      "conversion": 23085,
      "plain": 87622,
      "2010": 561,
      "adverbial": 4097,
      "ending": 36851,
      "ment": 68407,
      "gross": 49093,
      "encoded": 36469,
      "lg": 63948,
      "havent": 49549,
      "deleting": 28594,
      "permuting": 87055,
      "paraphrasing": 84845,
      "lgpllr": 63950,
      "license": 63969,
      "twitter": 126092,
      "hashtags": 49525,
      "29": 869,
      "intuitive": 56667,
      "interactive": 56010,
      "myanmar": 75997,
      "augmented": 10320,
      "transformationbased": 124274,
      "extensions": 41844,
      "statisticsbased": 113212,
      "tagger": 117353,
      "critical": 24802,
      "expressing": 41732,
      "unable": 126567,
      "clarify": 17026,
      "measurement": 67843,
      "paragraphs": 84576,
      "iii": 51718,
      "framing": 45748,
      "agendasetting": 4293,
      "vector": 131293,
      "positioned": 88307,
      "positions": 88310,
      "communicate": 19016,
      "communicated": 19021,
      "codified": 18242,
      "bangla": 11415,
      "cfg": 15968,
      "accepted": 1896,
      "entry": 38177,
      "checking": 16696,
      "avoided": 11245,
      "factoring": 42868,
      "stratified": 113549,
      "taln": 117565,
      "2012": 563,
      "tfidf": 121758,
      "categorical": 15722,
      "ancestor": 6079,
      "ancestors": 6081,
      "adds": 3818,
      "extra": 42057,
      "survive": 116214,
      "wikinews": 132638,
      "18": 413,
      "hints": 50509,
      "laws": 61691,
      "psychology": 94217,
      "criminology": 24776,
      "business": 14974,
      "medicine": 68234,
      "horse": 50657,
      "pinpoint": 87516,
      "rational": 95832,
      "heuristics": 49881,
      "fast": 43125,
      "resembles": 100728,
      "favor": 43213,
      "conflicts": 21036,
      "explains": 41280,
      "misleading": 70300,
      "estimates": 38642,
      "minimized": 70196,
      "duration": 34274,
      "computations": 20461,
      "states": 113035,
      "fillerrole": 44030,
      "decompositions": 28162,
      "subsequent": 114811,
      "tensor": 120186,
      "paradigm": 84531,
      "race": 95421,
      "remote": 98867,
      "disambiguate": 31939,
      "crosslanguage": 24919,
      "divergences": 32786,
      "varieties": 130953,
      "situate": 109882,
      "cover": 24480,
      "influential": 54327,
      "transduction": 124024,
      "experimenting": 40749,
      "rogets": 103148,
      "computerized": 20510,
      "1987": 459,
      "wordnetbased": 133783,
      "benchmarks": 12882,
      "miller": 70092,
      "charles": 16640,
      "30": 899,
      "judges": 57435,
      "correlate": 24190,
      "traced": 122719,
      "65": 1179,
      "gets": 47950,
      "correlations": 24250,
      "885": 1348,
      "replicate": 98958,
      "toefl": 122240,
      "esl": 38425,
      "300": 909,
      "readers": 95956,
      "8200": 1305,
      "quantifies": 94841,
      "segmentations": 104658,
      "boundaries": 14513,
      "penalty": 85851,
      "scaling": 103771,
      "penalties": 85850,
      "coefficients": 18248,
      "configurable": 21002,
      "suit": 115387,
      "segmenters": 104665,
      "indus": 54054,
      "expanded": 40359,
      "began": 12629,
      "counting": 24445,
      "east": 34468,
      "syllabic": 116298,
      "sounds": 110709,
      "geometric": 47861,
      "shapes": 107941,
      "categorize": 15772,
      "hundreds": 51147,
      "production": 92050,
      "3500": 963,
      "rebus": 96332,
      "sounding": 110706,
      "lingua": 64412,
      "consistently": 21407,
      "linguists": 64628,
      "inscriptions": 55505,
      "continued": 22605,
      "evidenced": 39681,
      "brahmi": 14550,
      "6th": 1208,
      "legitimately": 63323,
      "decipher": 27931,
      "secure": 104495,
      "pictorial": 87487,
      "instructional": 55706,
      "reached": 95905,
      "considers": 21336,
      "enriches": 37563,
      "dealing": 27866,
      "straight": 113426,
      "forward": 45322,
      "consisted": 21352,
      "55": 1109,
      "42": 1016,
      "rhetorical": 102709,
      "representativeness": 100024,
      "conveying": 23126,
      "topical": 122589,
      "assigning": 9553,
      "scanners": 103785,
      "impose": 52303,
      "hard": 49448,
      "estimators": 38676,
      "cataphoric": 15708,
      "substitutes": 114926,
      "guaranteed": 49216,
      "sublinear": 114723,
      "token": 122244,
      "systme": 117244,
      "tongue": 122364,
      "organisation": 82460,
      "mental": 68408,
      "syllables": 116307,
      "observations": 81174,
      "weve": 132464,
      "try": 125891,
      "knows": 58336,
      "syntagmatic": 116526,
      "conclusion": 20724,
      "capable": 15188,
      "computers": 20511,
      "guided": 49252,
      "portion": 88183,
      "gaining": 46374,
      "suggestions": 115365,
      "unl": 127367,
      "predicate": 88846,
      "preserving": 89932,
      "networking": 77495,
      "synchronously": 116339,
      "facilitating": 42808,
      "opportunity": 82145,
      "addresses": 3792,
      "encodings": 36737,
      "calculus": 15038,
      "induces": 54012,
      "conditioned": 20797,
      "benchmark": 12727,
      "disjoint": 32365,
      "presence": 89346,
      "excluding": 39955,
      "unlikely": 127458,
      "indicator": 53878,
      "society": 110472,
      "analysed": 5436,
      "diachronic": 30486,
      "google": 48515,
      "books": 14421,
      "dated": 27833,
      "centuries": 15920,
      "growing": 49158,
      "20th": 754,
      "started": 112402,
      "period": 87037,
      "beginning": 12634,
      "contributed": 22760,
      "decrease": 28169,
      "tendencies": 120166,
      "contributes": 22761,
      "prominent": 92249,
      "answered": 6596,
      "gender": 46578,
      "media": 68077,
      "14000": 325,
      "treats": 125569,
      "nuanced": 80823,
      "styles": 114616,
      "strong": 113616,
      "orientations": 82485,
      "multifaceted": 75103,
      "gendered": 46621,
      "say": 103671,
      "fewer": 43883,
      "homophily": 50623,
      "correlated": 24205,
      "emerges": 36062,
      "position": 88281,
      "audiences": 10221,
      "mainstream": 66497,
      "norms": 80367,
      "transitionbased": 124498,
      "pluggable": 87762,
      "packages": 83404,
      "maltparser": 66889,
      "dropin": 34172,
      "conforming": 21038,
      "weka": 132388,
      "package": 83403,
      "wrapper": 135078,
      "timbl": 121975,
      "suggested": 115339,
      "lowresource": 65486,
      "scenarios": 103838,
      "regardless": 97663,
      "principled": 90679,
      "usability": 127852,
      "specifications": 111611,
      "ahead": 4416,
      "continuations": 22603,
      "unfinished": 127071,
      "lookahead": 65219,
      "onthefly": 81824,
      "wordalignment": 133693,
      "aligner": 4931,
      "giza": 48219,
      "parametric": 84798,
      "goldstandard": 48452,
      "450": 1035,
      "10000": 179,
      "fly": 44726,
      "neural": 77823,
      "nplm": 80800,
      "perspectives": 87173,
      "huge": 50713,
      "nonparametric": 80270,
      "500000": 1085,
      "evidence": 39632,
      "native": 76219,
      "corrected": 24131,
      "child": 16717,
      "proposals": 92522,
      "sufficient": 115269,
      "encoding": 36680,
      "grammaticality": 48727,
      "judgements": 57434,
      "mappings": 67159,
      "amenable": 5305,
      "rapid": 95707,
      "agent": 4294,
      "urgent": 127846,
      "wizard": 132759,
      "oz": 83396,
      "react": 95923,
      "failure": 42985,
      "non": 80145,
      "experts": 41237,
      "sees": 104557,
      "reordering": 98897,
      "treetostring": 125733,
      "submission": 114724,
      "decided": 27928,
      "publicly": 94287,
      "accomplish": 1951,
      "moses": 74823,
      "reorder": 98895,
      "englishitalian": 37412,
      "englishurdu": 37460,
      "hamming": 49307,
      "reproduce": 100079,
      "regulatory": 97780,
      "industry": 54062,
      "regulations": 97779,
      "ensure": 37635,
      "environmental": 38194,
      "guides": 49268,
      "issued": 57033,
      "validate": 130713,
      "formalizing": 45202,
      "sparql": 111216,
      "assist": 9567,
      "drafting": 34083,
      "acquire": 3030,
      "assessment": 9519,
      "aka": 4585,
      "comprises": 20333,
      "prototypical": 93690,
      "prerequisite": 89342,
      "ad": 3160,
      "hoc": 50585,
      "diagnose": 30499,
      "extend": 41779,
      "incorporates": 53501,
      "endtoend": 36867,
      "merge": 68476,
      "lmf": 64875,
      "cuebased": 25206,
      "cues": 25208,
      "apriori": 8478,
      "classbased": 17055,
      "approached": 8033,
      "descriptors": 29507,
      "dealt": 27880,
      "discriminate": 32192,
      "validating": 130736,
      "iterating": 57118,
      "distinctions": 32556,
      "files": 44022,
      "principally": 90675,
      "horizontal": 50655,
      "bounds": 14528,
      "treating": 125563,
      "traces": 122720,
      "configurations": 21006,
      "finetuned": 44418,
      "glosses": 48296,
      "nearly": 76730,
      "ontology": 81833,
      "outcomes": 82618,
      "immediate": 51843,
      "cooperation": 23254,
      "investigated": 56833,
      "rise": 102863,
      "convert": 23092,
      "procedural": 91385,
      "paraphrases": 84838,
      "engines": 37059,
      "analysing": 5475,
      "combinations": 18600,
      "everyday": 39625,
      "systematicity": 116699,
      "productivity": 92068,
      "regularly": 97774,
      "questioned": 95261,
      "science": 103973,
      "philosophy": 87267,
      "underspecified": 126733,
      "grades": 48594,
      "contextually": 22586,
      "composite": 20089,
      "quantum": 94904,
      "factorising": 42870,
      "factorisation": 42869,
      "meet": 68250,
      "modeled": 72355,
      "deemed": 28190,
      "lemmatizer": 63345,
      "expertise": 41235,
      "respective": 101113,
      "translators": 125512,
      "emerged": 36049,
      "translator": 125510,
      "stemming": 113236,
      "lemmatization": 63335,
      "generates": 47155,
      "removing": 98882,
      "affixes": 4258,
      "root": 103271,
      "interacts": 56042,
      "universities": 127346,
      "regional": 97677,
      "helps": 49806,
      "getting": 47955,
      "info": 54329,
      "competency": 19611,
      "foreign": 45040,
      "track": 122722,
      "students": 114147,
      "competencies": 19610,
      "mastered": 67350,
      "educators": 34575,
      "parents": 84860,
      "record": 97129,
      "accomplishments": 1958,
      "tracked": 122742,
      "enrollment": 37573,
      "curriculum": 25420,
      "emerging": 36064,
      "gujarati": 49273,
      "transliterated": 125518,
      "latentvariable": 61642,
      "triples": 125811,
      "documentlevel": 33132,
      "sparsity": 111249,
      "noting": 80402,
      "concentration": 20558,
      "physical": 87476,
      "assertions": 9465,
      "saved": 103665,
      "forming": 45229,
      "secondly": 104473,
      "thirdly": 121868,
      "solved": 110629,
      "graphtheoretic": 48980,
      "rest": 101313,
      "smallest": 110247,
      "half": 49295,
      "grounding": 49120,
      "samplingbased": 103618,
      "contributions": 22783,
      "investigating": 56853,
      "alignments": 5007,
      "enforcing": 37007,
      "adjustment": 3879,
      "grams": 48735,
      "nrccanada": 80807,
      "tweets": 126017,
      "sms": 110274,
      "messagelevel": 68500,
      "submissions": 114747,
      "stood": 113370,
      "obtaining": 81432,
      "termlevel": 120261,
      "surfaceform": 116082,
      "emoticons": 36093,
      "lexiconbased": 63916,
      "replicated": 98962,
      "computing": 20516,
      "knowing": 57736,
      "manuallycreated": 67094,
      "hot": 50665,
      "cold": 18347,
      "warm": 132021,
      "freezing": 45797,
      "contrasting": 22722,
      "begin": 12630,
      "crowdsourcing": 25115,
      "flesh": 44647,
      "obtains": 81451,
      "outperforming": 82796,
      "preparing": 89300,
      "korean": 58344,
      "spmrl": 111958,
      "2013": 568,
      "collected": 18395,
      "sejong": 104680,
      "nonprofit": 80276,
      "summarizing": 115622,
      "ian": 51298,
      "accomplishing": 1957,
      "delivers": 28621,
      "intervention": 56323,
      "parametrized": 84800,
      "explained": 41270,
      "automate": 10425,
      "90": 1364,
      "refine": 97586,
      "eurovoc": 38777,
      "multilabel": 75149,
      "categorisation": 15758,
      "hierarchically": 50019,
      "organised": 82464,
      "institutions": 55701,
      "member": 68272,
      "eu": 38756,
      "official": 81595,
      "viewing": 131719,
      "assignment": 9560,
      "retrain": 102368,
      "possibly": 88447,
      "descriptor": 29506,
      "lending": 63348,
      "crosslingual": 24923,
      "plagiarism": 87618,
      "ranking": 95656,
      "memory": 68284,
      "centre": 15910,
      "tm": 122231,
      "professionally": 92085,
      "twentytwo": 126079,
      "231": 795,
      "professionals": 92090,
      "terminology": 120255,
      "ner": 77016,
      "departments": 29039,
      "uptodate": 127833,
      "developments": 30438,
      "eye": 42631,
      "monitoring": 74544,
      "automated": 10427,
      "23": 785,
      "countries": 24447,
      "19": 432,
      "functionality": 46065,
      "multilinguality": 75409,
      "compiled": 19715,
      "activity": 3132,
      "daily": 25506,
      "improper": 52326,
      "lapse": 61003,
      "classifies": 17646,
      "opinionated": 82112,
      "classic": 17081,
      "topicbased": 122605,
      "boosted": 14439,
      "explanations": 41289,
      "electoral": 35240,
      "national": 76210,
      "election": 35236,
      "millions": 70114,
      "beneficial": 12945,
      "post": 88451,
      "emotion": 36094,
      "detecting": 29827,
      "mood": 74658,
      "policies": 87939,
      "predicting": 88974,
      "thousand": 121912,
      "3class": 990,
      "helpful": 49793,
      "cornell": 23388,
      "approximations": 8471,
      "consuming": 21726,
      "worse": 135062,
      "projected": 92226,
      "continuousvalued": 22657,
      "lowdimensional": 65408,
      "multilayer": 75167,
      "optimize": 82222,
      "europarl": 38763,
      "englishfrench": 37362,
      "phrasebased": 87379,
      "0710": 60,
      "formalization": 45193,
      "ontologybased": 81854,
      "dictionarybased": 30902,
      "turkish": 125959,
      "expressive": 41770,
      "agglutinative": 4336,
      "immature": 51841,
      "synsets": 116358,
      "hypernyms": 51218,
      "hyponyms": 51245,
      "selftraining": 104955,
      "alleviate": 5028,
      "tuning": 125934,
      "gather": 46525,
      "minimise": 70189,
      "guidelines": 49258,
      "extreme": 42590,
      "restrictive": 101349,
      "limiting": 64307,
      "feasibility": 43234,
      "100000": 181,
      "day": 27839,
      "touch": 122703,
      "cohesion": 18332,
      "chains": 15981,
      "denotation": 29003,
      "fusion": 46230,
      "indicators": 53880,
      "topicality": 122596,
      "let": 63404,
      "similarly": 109346,
      "illustration": 51752,
      "poem": 87778,
      "bitext": 14237,
      "bitexts": 14240,
      "onetoone": 81729,
      "tackle": 117282,
      "counterpart": 24434,
      "referred": 97570,
      "synonymous": 116350,
      "handtagged": 49427,
      "scalable": 103685,
      "log": 64977,
      "extractions": 42542,
      "resolver": 100810,
      "relational": 98087,
      "coreferential": 23383,
      "resolves": 100812,
      "78": 1268,
      "68": 1192,
      "35": 959,
      "resolvers": 100811,
      "polysemous": 88024,
      "97": 1438,
      "95": 1418,
      "trec": 125570,
      "devise": 30461,
      "capitalizes": 15246,
      "wordtoword": 134341,
      "sr": 112085,
      "gradually": 48616,
      "proceed": 91409,
      "texttotext": 121668,
      "paraphrase": 84803,
      "competes": 19613,
      "sentencealigned": 106134,
      "freeopensource": 45793,
      "apertium": 6774,
      "wordforword": 133715,
      "handcoded": 49332,
      "entirely": 37726,
      "integrative": 55818,
      "plays": 87719,
      "chain": 15970,
      "submodels": 114782,
      "subtask": 114938,
      "hopes": 50651,
      "desirable": 29658,
      "integrity": 55819,
      "practicality": 88726,
      "tackles": 117318,
      "leverage": 63572,
      "adaptive": 3328,
      "largest": 61524,
      "conll2008": 21099,
      "confirming": 21025,
      "pose": 88248,
      "idiosyncratic": 51658,
      "mwe": 75991,
      "suitability": 115389,
      "pointwise": 87888,
      "pmi": 87771,
      "llr": 64857,
      "productive": 92066,
      "asynchronous": 9675,
      "conversation": 22934,
      "blog": 14367,
      "graphbased": 48887,
      "beat": 12616,
      "huang": 50709,
      "linked": 64642,
      "geosemantic": 47876,
      "geographic": 47848,
      "recognised": 96798,
      "vast": 131271,
      "confused": 21050,
      "geo": 47846,
      "rated": 95812,
      "203": 745,
      "subjects": 114708,
      "approximates": 8466,
      "surpass": 116089,
      "mastering": 67351,
      "compelling": 19598,
      "succeeds": 115060,
      "inadequate": 53196,
      "imperfect": 51921,
      "possessing": 88361,
      "biases": 13863,
      "singleton": 109850,
      "2004": 545,
      "7year": 1285,
      "boy": 14534,
      "simon": 109352,
      "sign": 108676,
      "asl": 9312,
      "possesses": 88360,
      "specificity": 111613,
      "keyphrases": 57622,
      "louis": 65340,
      "hassan": 49526,
      "fail": 42945,
      "church": 16972,
      "gale": 46415,
      "idiomatic": 51650,
      "sheaf": 108147,
      "theoretic": 121787,
      "contextuality": 22536,
      "resolutions": 100795,
      "rank": 95621,
      "jim": 57237,
      "lambek": 58793,
      "compute": 20463,
      "corpuses": 24083,
      "reliably": 98625,
      "clinical": 17760,
      "tempeval": 120076,
      "preparation": 89294,
      "semeval2015": 105525,
      "exercise": 39984,
      "times": 122165,
      "discrete": 32160,
      "subtasks": 114962,
      "narrative": 76166,
      "container": 21778,
      "tamil": 117567,
      "prompt": 92324,
      "forecast": 45025,
      "expectation": 40381,
      "act": 3060,
      "englishhindi": 37400,
      "infancy": 54088,
      "extracts": 42576,
      "phonemes": 87282,
      "converts": 23105,
      "dun": 34265,
      "syntaxique": 116586,
      "syntacticsemantic": 116525,
      "changing": 16398,
      "englishswedish": 37432,
      "api": 6776,
      "berkeley": 13045,
      "swedish": 116262,
      "defines": 28508,
      "769": 1263,
      "749": 1247,
      "convolutional": 23147,
      "dubbed": 34250,
      "pooling": 88038,
      "varying": 131252,
      "capturing": 15459,
      "longrange": 65176,
      "multiclass": 75050,
      "distant": 32445,
      "supervision": 115863,
      "greater": 49035,
      "strongest": 113738,
      "skipped": 109998,
      "skip": 109978,
      "generalizes": 46834,
      "extensive": 41850,
      "31": 918,
      "127": 273,
      "strength": 113566,
      "kb": 57500,
      "257": 831,
      "kit": 57705,
      "orthography": 82604,
      "digital": 31714,
      "statistic": 113073,
      "atypical": 10215,
      "permitting": 87050,
      "resourcepoor": 100906,
      "nes": 77100,
      "alphabet": 5205,
      "leveraged": 63639,
      "secondary": 104468,
      "observe": 81186,
      "intensity": 55882,
      "emotional": 36116,
      "perceive": 85908,
      "lines": 64405,
      "shakespeare": 107906,
      "plots": 87759,
      "associative": 9638,
      "shape": 107938,
      "movements": 74887,
      "articulatory": 9235,
      "acoustic": 2998,
      "likewise": 64149,
      "differentiate": 31588,
      "selecting": 104743,
      "activate": 3096,
      "brain": 14551,
      "cosine": 24334,
      "adjacency": 3846,
      "divide": 32905,
      "mouth": 74883,
      "versus": 131626,
      "clustered": 17931,
      "differentiated": 31589,
      "arousal": 9022,
      "downwards": 34065,
      "formant": 45205,
      "coordinate": 23259,
      "mert": 68489,
      "converge": 22918,
      "optimum": 82245,
      "accord": 1959,
      "rss": 103335,
      "moving": 74918,
      "formed": 45224,
      "coast": 18062,
      "pilot": 87505,
      "update": 127789,
      "navigation": 76683,
      "encs": 36784,
      "der": 29326,
      "altered": 5213,
      "publishing": 94379,
      "lexicographical": 63869,
      "thesauri": 121837,
      "storage": 113377,
      "display": 32378,
      "targeted": 117762,
      "lapproche": 61002,
      "montague": 74650,
      "linguist": 64415,
      "montagovian": 74649,
      "1970": 450,
      "utilise": 130468,
      "proposons": 93634,
      "au": 10216,
      "sest": 107337,
      "thematically": 121781,
      "reinforced": 97787,
      "thematic": 121778,
      "calculate": 15015,
      "reinforce": 97786,
      "aggregating": 4354,
      "caused": 15827,
      "outofcontext": 82641,
      "mb": 67560,
      "newsgroups": 79404,
      "nonparallel": 80262,
      "semisupervised": 105590,
      "transductive": 124026,
      "irreplaceable": 56954,
      "expanding": 40361,
      "unlabeled": 127368,
      "highperformance": 50359,
      "economic": 34511,
      "asset": 9536,
      "vietnam": 131667,
      "malaysia": 66878,
      "computerize": 20509,
      "underresourced": 126725,
      "khmer": 57667,
      "explanatory": 41302,
      "combinatory": 18610,
      "coming": 18785,
      "completed": 19777,
      "fem": 43870,
      "noted": 80389,
      "ipa": 56926,
      "editing": 34542,
      "african": 4272,
      "bambara": 11409,
      "hausa": 49547,
      "doc": 32939,
      "unicode": 127087,
      "visibility": 131756,
      "arxiv": 9278,
      "multilanguage": 75165,
      "swiss": 116267,
      "avalanche": 11150,
      "bulletin": 14953,
      "twice": 126080,
      "catalogue": 15690,
      "predefined": 88820,
      "predetermined": 88840,
      "proofreading": 92377,
      "sublanguage": 114717,
      "offset": 81617,
      "winter": 132731,
      "seasons": 104375,
      "danger": 25531,
      "origins": 82579,
      "hardly": 49491,
      "ratings": 95822,
      "forecasters": 45026,
      "professional": 92076,
      "declared": 27974,
      "interestingly": 56093,
      "reproduces": 100085,
      "implements": 51984,
      "poor": 88052,
      "hmm": 50580,
      "icon": 51308,
      "fmeasures": 44728,
      "simulated": 109634,
      "soccer": 110322,
      "games": 46422,
      "stream": 113550,
      "comments": 18808,
      "establishes": 38607,
      "commentaries": 18803,
      "isizulu": 56962,
      "south": 110959,
      "africa": 4271,
      "speak": 111276,
      "home": 50609,
      "imperative": 51919,
      "options": 82249,
      "inform": 54333,
      "elaborate": 35224,
      "existential": 40040,
      "quantification": 94836,
      "conjunction": 21067,
      "singular": 109859,
      "library": 63961,
      "methodological": 69234,
      "arts": 9277,
      "tourism": 122706,
      "regarded": 97652,
      "response": 101187,
      "authoring": 10373,
      "documentation": 33119,
      "importantly": 52296,
      "nonlinguistic": 80229,
      "loose": 65230,
      "fullfledged": 45930,
      "noncompositionality": 80174,
      "expect": 40380,
      "host": 50660,
      "recognizes": 97065,
      "feedback": 43821,
      "crisis": 24782,
      "investigations": 56867,
      "2014": 573,
      "financial": 44252,
      "sciences": 103987,
      "facing": 42813,
      "communities": 19057,
      "spam": 111113,
      "routing": 103322,
      "notably": 80379,
      "behaviours": 12683,
      "continue": 22604,
      "checkers": 16695,
      "rgl": 102707,
      "lowlevel": 65472,
      "ideal": 51343,
      "introducing": 56625,
      "inflectional": 54291,
      "questionanswering": 95251,
      "suffers": 115248,
      "gazetteer": 46562,
      "capitalization": 15242,
      "apart": 6769,
      "substitute": 114924,
      "embeddings": 35538,
      "maps": 67161,
      "sphere": 111926,
      "cooccurring": 23246,
      "lie": 63973,
      "movies": 74912,
      "tv": 126004,
      "retraining": 102371,
      "adaptable": 3203,
      "title": 122219,
      "f1scores": 42722,
      "unbiased": 126608,
      "disease": 32332,
      "biological": 14160,
      "bionlp": 14217,
      "traditionally": 122886,
      "workers": 134886,
      "platforms": 87662,
      "amazons": 5272,
      "mechanical": 67935,
      "turk": 125954,
      "amt": 5405,
      "aggregate": 4344,
      "ncbi": 76698,
      "refining": 97598,
      "benchmarking": 12876,
      "protocol": 93673,
      "iterations": 57123,
      "arrived": 9029,
      "reproduced": 100084,
      "593": 1134,
      "max": 67494,
      "0984": 116,
      "merged": 68478,
      "145": 331,
      "week": 132314,
      "06": 42,
      "worker": 134885,
      "judged": 57426,
      "balance": 11395,
      "wellannotated": 132392,
      "simlex999": 109351,
      "genuine": 47840,
      "wordsim353": 134335,
      "men": 68405,
      "incentivizes": 53212,
      "arguably": 8906,
      "wider": 132606,
      "adjective": 3854,
      "concreteness": 20744,
      "consequently": 21178,
      "surpassed": 116099,
      "ceiling": 15880,
      "plenty": 87748,
      "quantify": 94844,
      "guiding": 49270,
      "keyword": 57633,
      "legal": 63312,
      "procurement": 91864,
      "noncontent": 80176,
      "heavily": 49648,
      "tensorbased": 120195,
      "additive": 3640,
      "multiplicative": 75762,
      "largerscale": 61395,
      "vectors": 131406,
      "resolving": 100814,
      "regression": 97689,
      "superiority": 115698,
      "modelindependent": 72369,
      "repair": 98904,
      "stir": 113351,
      "transcripts": 124010,
      "latency": 61563,
      "informationtheoretic": 55122,
      "disfluency": 32356,
      "overhead": 83324,
      "distances": 32438,
      "biology": 14163,
      "charniak": 16641,
      "rf": 102703,
      "evalb": 38779,
      "sekine": 104681,
      "zhang": 135536,
      "normalisation": 80330,
      "500": 1080,
      "globally": 48286,
      "minute": 70278,
      "140": 323,
      "tweet": 126008,
      "unintentionally": 127173,
      "encourages": 36767,
      "shorthand": 108308,
      "notations": 80383,
      "strip": 113611,
      "spellings": 111919,
      "bare": 11428,
      "conforms": 21041,
      "stanford": 112359,
      "te": 119682,
      "attachment": 9695,
      "correctly": 24170,
      "stanfords": 112385,
      "vital": 131847,
      "degraded": 28560,
      "crfbased": 24761,
      "crf": 24742,
      "gazetteers": 46563,
      "lowest": 65463,
      "69": 1197,
      "87": 1336,
      "79": 1271,
      "finer": 44397,
      "politeness": 87988,
      "compiler": 19719,
      "contrastive": 22725,
      "indicates": 53857,
      "merits": 68484,
      "remains": 98786,
      "intractable": 56335,
      "divergence": 32782,
      "noises": 80083,
      "guide": 49228,
      "partition": 85508,
      "mass": 67319,
      "topn": 122675,
      "sophisticated": 110665,
      "networkbased": 77480,
      "algebra": 4676,
      "lexicographic": 63866,
      "manipulation": 66929,
      "commands": 18792,
      "identifiers": 51466,
      "manipulations": 66933,
      "recurring": 97277,
      "trend": 125740,
      "poorly": 88064,
      "fit": 44575,
      "relaxing": 98426,
      "simplistic": 109618,
      "discussions": 32328,
      "gate": 46498,
      "accessed": 1928,
      "apis": 6781,
      "attributive": 10214,
      "precedes": 88774,
      "functionalities": 46064,
      "meeting": 68253,
      "portals": 88178,
      "sites": 109876,
      "companies": 19116,
      "attitude": 10135,
      "lengthy": 63389,
      "costly": 24382,
      "surveys": 116211,
      "forum": 45301,
      "posts": 88512,
      "participation": 85361,
      "evaluator": 39492,
      "option": 82246,
      "medical": 68194,
      "synthetic": 116611,
      "visual": 131783,
      "patients": 85704,
      "drugs": 34201,
      "acceptance": 1895,
      "prepare": 89297,
      "owl": 83392,
      "starshaped": 112395,
      "transforms": 124477,
      "croatian": 24853,
      "profile": 92098,
      "wordorder": 133785,
      "typedriven": 126238,
      "polymorphism": 88018,
      "slow": 110100,
      "lineartime": 64403,
      "syntaxdriven": 116583,
      "eliminates": 35298,
      "borrow": 14473,
      "subtype": 115011,
      "geoquery": 47873,
      "jobs": 57241,
      "plausibility": 87676,
      "conduct": 20823,
      "matrix": 67471,
      "logistic": 65030,
      "inquiry": 55502,
      "elected": 35235,
      "lexis": 63943,
      "th": 121765,
      "seminal": 105584,
      "female": 43871,
      "sex": 107888,
      "women": 132892,
      "promote": 92315,
      "wrote": 135165,
      "weaker": 132189,
      "looks": 65226,
      "rhetoric": 102708,
      "advancement": 3961,
      "comment": 18797,
      "liu": 64839,
      "physics": 87481,
      "volume": 131936,
      "11": 213,
      "december": 27917,
      "ling": 64409,
      "weld": 132389,
      "dramatic": 34087,
      "absence": 1714,
      "washington": 132038,
      "city": 17000,
      "government": 48542,
      "spurious": 112060,
      "fine": 44320,
      "submodule": 114785,
      "preferred": 89248,
      "34": 952,
      "accommodate": 1939,
      "unbalanced": 126605,
      "complicated": 19954,
      "svms": 116247,
      "surpasses": 116102,
      "inherited": 55196,
      "inter": 55929,
      "innate": 55272,
      "ug": 126543,
      "decades": 27907,
      "ago": 4381,
      "learnable": 62197,
      "roots": 103275,
      "unveil": 127773,
      "contradicting": 22666,
      "universally": 127341,
      "sole": 110528,
      "cosmic": 24343,
      "humanistic": 51037,
      "supertagging": 115719,
      "bangalore": 11414,
      "joshi": 57404,
      "scholars": 103961,
      "supertags": 115720,
      "inception": 53213,
      "chapters": 16413,
      "edited": 34540,
      "book": 14407,
      "blend": 14269,
      "fullscale": 45932,
      "lexicalsyntactic": 63864,
      "vocabularies": 131862,
      "degradation": 28548,
      "queried": 94910,
      "demo": 28638,
      "translationbased": 125441,
      "nonspecific": 80291,
      "indepth": 53792,
      "optimisation": 82180,
      "tricks": 125770,
      "lag": 58778,
      "stylistic": 114624,
      "drama": 34086,
      "stylistics": 114636,
      "distinguishing": 32588,
      "burrows": 14971,
      "delta": 28629,
      "concluded": 20721,
      "preserved": 89929,
      "distinguishable": 32585,
      "2002": 542,
      "adept": 3823,
      "imprint": 52325,
      "norwegian": 80373,
      "writings": 135110,
      "characterization": 16545,
      "distinctiveness": 32564,
      "characters": 16592,
      "navigating": 76682,
      "horizon": 50654,
      "shortcoming": 108274,
      "mode": 70489,
      "wordsense": 134328,
      "uncover": 126647,
      "glove": 48297,
      "skipgram": 109980,
      "etymological": 38755,
      "turkic": 125957,
      "exhibit": 39995,
      "leveraging": 63664,
      "cognate": 18256,
      "strives": 113613,
      "lms": 64879,
      "exhaustive": 39990,
      "tailored": 117500,
      "groundwork": 49131,
      "journals": 57413,
      "subproblems": 114803,
      "boundary": 14518,
      "enlarge": 37548,
      "carefully": 15514,
      "chinesetoenglish": 16867,
      "147": 333,
      "date": 27821,
      "decreased": 28172,
      "dominant": 33927,
      "trials": 125766,
      "stringbased": 113602,
      "standalone": 112180,
      "overviews": 83383,
      "anticipate": 6754,
      "suits": 115423,
      "breaking": 14578,
      "highdimensional": 50152,
      "maintain": 66500,
      "greedy": 49067,
      "variational": 130906,
      "branch": 14557,
      "727": 1229,
      "inflections": 54298,
      "postagger": 88457,
      "arceager": 8597,
      "9332": 1407,
      "fastest": 43199,
      "4000": 1009,
      "optimizing": 82242,
      "64": 1176,
      "beams": 12611,
      "cluster": 17923,
      "tunable": 125925,
      "released": 98474,
      "apache": 6765,
      "academic": 1853,
      "story": 113405,
      "harder": 49485,
      "bridging": 14602,
      "gap": 46439,
      "reverse": 102524,
      "crossword": 25092,
      "answerers": 6597,
      "handful": 49364,
      "freelyavailable": 45792,
      "taskspecific": 119616,
      "harvest": 49517,
      "matched": 67374,
      "prioritize": 90772,
      "queue": 95380,
      "crawling": 24599,
      "maximizes": 67515,
      "impacting": 51905,
      "ordering": 82436,
      "scales": 103767,
      "unrelated": 127501,
      "discarded": 32007,
      "benefit": 12957,
      "collapse": 18365,
      "collapsing": 18367,
      "collapsed": 18366,
      "noticeable": 80397,
      "hypothesize": 51287,
      "recognizers": 97064,
      "handcrafted": 49335,
      "characterlevel": 16557,
      "hyperparameters": 51234,
      "f1score": 42704,
      "72": 1223,
      "se": 104284,
      "mantic": 66969,
      "fulfills": 45928,
      "logicbased": 65024,
      "graded": 48589,
      "weighted": 132334,
      "corrects": 24189,
      "sick": 108657,
      "utilization": 130493,
      "numerical": 81007,
      "extrinsic": 42614,
      "completion": 19791,
      "emails": 35343,
      "legitimate": 63322,
      "indicated": 53856,
      "sequencetosequence": 107150,
      "graphemetophoneme": 48922,
      "targetlanguage": 117779,
      "captioning": 15255,
      "images": 51807,
      "caption": 15251,
      "qualitatively": 94578,
      "credit": 24740,
      "rival": 102882,
      "statoftheart": 113213,
      "bidirectional": 13898,
      "shortterm": 108312,
      "lstm": 65587,
      "spectral": 111629,
      "pcfgs": 85827,
      "decode": 27984,
      "baselines": 12343,
      "loss": 65245,
      "074": 64,
      "multitask": 75807,
      "layers": 61759,
      "jointly": 57334,
      "27": 848,
      "arabicenglish": 8570,
      "chineseenglish": 16846,
      "autoencoder": 10397,
      "longer": 65154,
      "recurrent": 97158,
      "longshort": 65186,
      "preserve": 89920,
      "reconstruct": 97107,
      "multisentence": 75779,
      "decodes": 28068,
      "reconstructed": 97111,
      "rouge": 103288,
      "grid": 49088,
      "encode": 36419,
      "visualizing": 131840,
      "vectorbased": 131401,
      "vision": 131758,
      "intensification": 55878,
      "salience": 103541,
      "backpropagation": 11319,
      "variance": 130862,
      "gates": 46524,
      "nets": 77115,
      "lstms": 65696,
      "informativeness": 55162,
      "implicature": 52000,
      "amazon": 5263,
      "judgments": 57440,
      "correlation": 24220,
      "inherent": 55186,
      "variables": 130852,
      "genres": 47835,
      "compatibility": 19595,
      "makeup": 66817,
      "connotation": 21155,
      "writer": 135081,
      "subtly": 115003,
      "sentiments": 106830,
      "projecting": 92228,
      "victim": 131651,
      "bad": 11353,
      "happened": 49437,
      "distressed": 32607,
      "typed": 126235,
      "connotative": 21157,
      "interplay": 56200,
      "confirm": 21012,
      "analytical": 5946,
      "subtle": 114996,
      "nonverbal": 80316,
      "abilities": 1477,
      "rasa": 95769,
      "elaboration": 35228,
      "supervising": 115862,
      "coordinates": 23262,
      "activation": 3098,
      "gating": 46533,
      "priori": 90769,
      "interacting": 55937,
      "textbased": 121423,
      "openended": 81985,
      "validated": 130732,
      "521": 1098,
      "crossplatform": 25066,
      "gpl": 48549,
      "v3": 130692,
      "plugin": 87765,
      "consist": 21346,
      "uninterpretable": 127176,
      "999": 1452,
      "booking": 14419,
      "ticket": 121956,
      "converses": 23083,
      "converse": 23080,
      "subtitles": 114991,
      "helpdesk": 49790,
      "opendomain": 81956,
      "transcript": 123992,
      "stylometry": 114641,
      "devising": 30466,
      "networked": 77494,
      "straightforwardly": 113435,
      "pivotal": 87585,
      "webquestions": 132287,
      "guarantees": 49219,
      "parallelizability": 84701,
      "freebase": 45773,
      "big": 13988,
      "analytics": 5948,
      "cooking": 23250,
      "recipes": 96791,
      "aggregated": 4351,
      "analogously": 5415,
      "server": 107311,
      "carrying": 15561,
      "novelty": 80788,
      "plane": 87632,
      "bounding": 14526,
      "box": 14531,
      "projections": 92238,
      "apparent": 6785,
      "sensible": 105731,
      "umbrella": 126564,
      "englishbased": 37355,
      "listing": 64711,
      "scattered": 103816,
      "cloud": 17888,
      "propositional": 93631,
      "username": 129081,
      "says": 103675,
      "usernames": 129082,
      "ubiquitous": 126508,
      "suggestive": 115368,
      "demographics": 28651,
      "decompose": 28152,
      "subunits": 115013,
      "138": 306,
      "2003": 543,
      "subcorpus": 114656,
      "rankings": 95697,
      "basically": 12539,
      "comma": 18786,
      "delimited": 28610,
      "lowercased": 65453,
      "wellbalanced": 132395,
      "systematize": 116700,
      "neuralnetworkbased": 78739,
      "initialize": 55240,
      "recommend": 97083,
      "sufficiently": 115291,
      "stopping": 113374,
      "manifolds": 66925,
      "arithmetic": 9005,
      "subspaces": 114845,
      "spanned": 111184,
      "ndimensional": 76705,
      "euclidean": 38759,
      "manifold": 66924,
      "quotient": 95418,
      "subgroup": 114672,
      "rotations": 103281,
      "kernels": 57535,
      "relationspecific": 98340,
      "subsystems": 114937,
      "shifted": 108182,
      "isolated": 56973,
      "transport": 125537,
      "trade": 122772,
      "depict": 29296,
      "subsystem": 114936,
      "precisely": 88784,
      "shuffled": 108646,
      "subword": 115014,
      "graphemes": 48921,
      "dependent": 29262,
      "corroborate": 24321,
      "separated": 106867,
      "uncommon": 126638,
      "judgment": 57438,
      "vsms": 131967,
      "predict": 88868,
      "ignoring": 51682,
      "collect": 18371,
      "crowdworkers": 25126,
      "fluent": 44702,
      "impacted": 51904,
      "affected": 4229,
      "differently": 31593,
      "setups": 107864,
      "vsm": 131966,
      "multilingualism": 75408,
      "compensate": 19599,
      "characterbased": 16484,
      "upto": 127832,
      "extents": 41991,
      "competitions": 19630,
      "bench": 12726,
      "accordance": 1960,
      "tdil": 119677,
      "govt": 48546,
      "item": 57104,
      "trigger": 125781,
      "render": 98885,
      "84": 1317,
      "residual": 100737,
      "comply": 19963,
      "affect": 4212,
      "hurdles": 51162,
      "attested": 10134,
      "syntaxbased": 116581,
      "overlapped": 83332,
      "repeatedly": 98911,
      "avoid": 11230,
      "unigram": 127152,
      "meteor": 68568,
      "wellsuited": 132440,
      "mcmc": 67584,
      "fluency": 44691,
      "widelyused": 132602,
      "adequately": 3836,
      "ngrambased": 79457,
      "catch": 15715,
      "indirectly": 53889,
      "wmt": 132764,
      "preferring": 89251,
      "biased": 13854,
      "overly": 83348,
      "equally": 38228,
      "ascertain": 9283,
      "evolutionary": 39694,
      "competent": 19612,
      "realizing": 96125,
      "completeness": 19787,
      "proficiency": 92094,
      "postulating": 88531,
      "elusive": 35331,
      "metaembeddings": 68535,
      "resort": 100823,
      "repeated": 98909,
      "pick": 87484,
      "adjusting": 3878,
      "regularization": 97743,
      "choose": 16922,
      "unpruned": 127495,
      "calibration": 15045,
      "exploratory": 41508,
      "uncertainty": 126613,
      "telling": 120070,
      "trust": 125870,
      "miscalibration": 70290,
      "intervals": 56321,
      "political": 87989,
      "sociolinguistics": 110478,
      "cl": 17005,
      "themes": 121783,
      "featuring": 43797,
      "synergy": 116343,
      "hope": 50625,
      "closer": 17871,
      "collaboration": 18354,
      "organized": 82474,
      "retrospectively": 102478,
      "analytic": 5945,
      "enron": 37574,
      "posits": 88358,
      "employees": 36311,
      "breeding": 14587,
      "grounds": 49126,
      "executives": 39976,
      "clausal": 17687,
      "signals": 108700,
      "revisiting": 102654,
      "revisit": 102651,
      "levins": 63714,
      "600": 1150,
      "multiperspective": 75485,
      "verbnet": 131546,
      "translationese": 125446,
      "ted": 120049,
      "talks": 117561,
      "commentary": 18804,
      "instrumental": 55716,
      "enjoys": 37546,
      "nerc": 77098,
      "multinomial": 75469,
      "nave": 76675,
      "mnb": 70459,
      "fitting": 44590,
      "5000": 1083,
      "f1measure": 42702,
      "83": 1308,
      "81": 1297,
      "compounding": 20157,
      "wordformation": 133712,
      "knowledgerich": 58281,
      "shop": 108196,
      "analogybased": 5429,
      "splitter": 111951,
      "frequencybased": 45852,
      "evident": 39683,
      "sexual": 107893,
      "wants": 132019,
      "analyzes": 6033,
      "seeing": 104512,
      "pave": 85803,
      "streams": 113559,
      "highlighting": 50277,
      "necessity": 76768,
      "segmenting": 104666,
      "topically": 122598,
      "attractive": 10168,
      "interviews": 56330,
      "asked": 9300,
      "iwslt": 57160,
      "nist": 79495,
      "ter": 120201,
      "cleaning": 17714,
      "logs": 65054,
      "store": 113378,
      "trails": 122895,
      "taxonomies": 119658,
      "hyponymy": 51246,
      "specialization": 111381,
      "primer": 90660,
      "yielding": 135374,
      "feedforward": 43842,
      "dispersion": 32375,
      "shaping": 107942,
      "pre": 88765,
      "consonantal": 21509,
      "licensing": 63972,
      "trac": 122716,
      "revealing": 102509,
      "helping": 49804,
      "compose": 20057,
      "replies": 98969,
      "lda": 61826,
      "exchanges": 39947,
      "silver": 109063,
      "labelling": 58565,
      "contact": 21729,
      "ap": 6764,
      "recommended": 97095,
      "correlational": 24248,
      "multimodal": 75413,
      "views": 131722,
      "1m": 479,
      "v1": 130686,
      "v2": 130690,
      "cross": 24856,
      "l1": 58369,
      "l2": 58373,
      "ubuntu": 126516,
      "utterance": 130588,
      "multiturn": 75909,
      "inhouse": 55201,
      "bilstms": 14111,
      "cnns": 18037,
      "averaging": 11226,
      "plans": 87646,
      "nowadays": 80795,
      "perfect": 85936,
      "opus": 82251,
      "taggers": 117363,
      "executed": 39970,
      "automatized": 10846,
      "true": 125844,
      "casing": 15674,
      "understandable": 126792,
      "ranged": 95615,
      "reallife": 96126,
      "implementations": 51954,
      "year": 135233,
      "nara": 76156,
      "institute": 55697,
      "2015": 581,
      "asian": 9287,
      "attentional": 10044,
      "stating": 113071,
      "reasons": 96326,
      "blstmrnn": 14381,
      "rsa": 103332,
      "listener": 64706,
      "literal": 64728,
      "broadly": 14703,
      "gricean": 49084,
      "criticized": 24850,
      "unrealistic": 127496,
      "preventing": 90369,
      "nonlinear": 80220,
      "opens": 82006,
      "approximating": 8468,
      "integer": 55733,
      "dependence": 29053,
      "moderate": 74372,
      "compressions": 20330,
      "deletion": 28596,
      "orders": 82444,
      "magnitude": 66377,
      "ilpbased": 51757,
      "degrade": 28557,
      "hotel": 50667,
      "cnn": 17984,
      "nb": 76686,
      "outstanding": 83202,
      "combinational": 18599,
      "death": 27884,
      "ple": 87746,
      "frozen": 45903,
      "similes": 109350,
      "british": 14657,
      "simile": 109349,
      "marker": 67229,
      "vehicle": 131504,
      "happy": 49445,
      "eventuality": 39616,
      "sleep": 110009,
      "predicts": 89213,
      "temporal": 120089,
      "serial": 107265,
      "spectacular": 111627,
      "misspecified": 70340,
      "nonsensical": 80288,
      "annual": 6504,
      "surprising": 116125,
      "suffer": 115206,
      "profound": 92107,
      "reviewers": 102592,
      "reading": 95970,
      "longterm": 65197,
      "retained": 102360,
      "recalled": 96346,
      "windowbased": 132714,
      "memories": 68278,
      "selfsupervision": 104952,
      "qa": 94492,
      "paraphrased": 84835,
      "summaries": 115429,
      "imitates": 51836,
      "usfd": 129302,
      "drift": 34141,
      "compensation": 19603,
      "wnut": 132879,
      "drawing": 34125,
      "attempting": 9758,
      "largercontext": 61394,
      "corpuslevel": 24085,
      "intrasentence": 56340,
      "intersentence": 56306,
      "imdb": 51831,
      "bbc": 12591,
      "ments": 68472,
      "unconditional": 126639,
      "theme": 121782,
      "lexicosemantic": 63938,
      "leverages": 63641,
      "composes": 20077,
      "generations": 47705,
      "alleviates": 5055,
      "hypothesizing": 51292,
      "supersenses": 115714,
      "dozen": 34070,
      "substitution": 114928,
      "prototypes": 93689,
      "polysemy": 88027,
      "adaptively": 3339,
      "multidomain": 75084,
      "ts": 125905,
      "subtopic": 115004,
      "summarisation": 115466,
      "regions": 97680,
      "segmented": 104662,
      "conceptualizations": 20675,
      "anymore": 6763,
      "agglomerative": 4332,
      "treelike": 125674,
      "conceptually": 20679,
      "granularity": 48745,
      "bytes": 15006,
      "lstmbased": 65678,
      "reads": 96043,
      "lengths": 63387,
      "operate": 82047,
      "stateof": 112550,
      "theart": 121775,
      "scratch": 104239,
      "tokenization": 122275,
      "employing": 36312,
      "pivotbased": 87587,
      "sparseness": 111246,
      "sourcetopivot": 110948,
      "pivottotarget": 87593,
      "induce": 54000,
      "ibm": 51301,
      "targetdependent": 117759,
      "mined": 70149,
      "msa": 74940,
      "bag": 11358,
      "alternatively": 5253,
      "uncovers": 126651,
      "wikipedias": 132691,
      "insignificant": 55549,
      "interpretability": 56217,
      "jadavpur": 57174,
      "performer": 86939,
      "agreementbased": 4409,
      "attentionbased": 10057,
      "intricate": 56348,
      "unidirectional": 127088,
      "sourcetotarget": 110952,
      "targettosource": 117804,
      "finetuning": 44441,
      "keeping": 57517,
      "unnormalized": 127480,
      "dearth": 27882,
      "criticism": 24848,
      "criticisms": 24849,
      "discriminatively": 32220,
      "rnn": 102899,
      "distortion": 32593,
      "considerably": 21265,
      "rnns": 102958,
      "multisource": 75782,
      "encoderdecoder": 36589,
      "48": 1049,
      "impressive": 52316,
      "positional": 88296,
      "conditioning": 20808,
      "fertility": 43879,
      "encoder": 36490,
      "slot": 110032,
      "priors": 90777,
      "minus": 70277,
      "norm": 80324,
      "prone": 92330,
      "overfit": 83317,
      "assumes": 9653,
      "width": 132620,
      "zero": 135470,
      "dropout": 34174,
      "injects": 55270,
      "optimally": 82179,
      "region": 97676,
      "nonstandard": 80292,
      "flemish": 44645,
      "belgian": 12686,
      "cmc": 17973,
      "convincing": 23131,
      "peak": 85832,
      "intriguing": 56349,
      "wordembeddings": 133711,
      "computationallyefficient": 20460,
      "align": 4897,
      "speeches": 111854,
      "party": 85598,
      "specify": 111625,
      "supportive": 116033,
      "opposing": 82150,
      "filter": 44069,
      "dirichlet": 31921,
      "allocation": 5069,
      "congressional": 21061,
      "debates": 27893,
      "formative": 45216,
      "ondemand": 81666,
      "undergraduate": 126668,
      "classrooms": 17685,
      "powered": 88660,
      "elastic": 35230,
      "labs": 58672,
      "instructors": 55714,
      "upload": 127816,
      "unlimited": 127459,
      "student": 114136,
      "browser": 14728,
      "raters": 95814,
      "teaching": 119709,
      "seek": 104513,
      "emphasize": 36139,
      "teach": 119690,
      "crosscutting": 24876,
      "disciplines": 32017,
      "notable": 80375,
      "customize": 25459,
      "scaled": 103764,
      "microblog": 70045,
      "feeds": 43856,
      "surveying": 116209,
      "ancillary": 6093,
      "days": 27845,
      "syntaxsemantics": 116589,
      "addressing": 3806,
      "tricky": 125771,
      "innovative": 55284,
      "geographical": 47854,
      "dialectal": 30526,
      "reproducing": 100090,
      "hilbert": 50443,
      "geotagged": 47878,
      "atlas": 9683,
      "editor": 34560,
      "growth": 49189,
      "exchanging": 39948,
      "facebook": 42736,
      "unstructured": 127568,
      "heterogeneous": 49849,
      "zipfs": 135548,
      "law": 61681,
      "letter": 63406,
      "homogeneous": 50613,
      "homogeneity": 50612,
      "lium": 64844,
      "eastern": 34470,
      "finland": 44553,
      "ivectors": 57156,
      "discriminant": 32189,
      "plda": 87745,
      "dnn": 32928,
      "swivel": 116297,
      "descent": 29372,
      "piecewise": 87501,
      "unobserved": 127481,
      "proportional": 92515,
      "vectorized": 131405,
      "multiplication": 75760,
      "thousands": 121919,
      "rows": 103329,
      "columns": 18540,
      "parallelize": 84704,
      "richness": 102814,
      "remained": 98777,
      "mimic": 70131,
      "naturalistic": 76628,
      "wealth": 132212,
      "fi": 43921,
      "er": 38262,
      "rankbased": 95632,
      "suffix": 115297,
      "modifies": 74455,
      "org": 82456,
      "highlights": 50282,
      "customization": 25458,
      "subparts": 114799,
      "music": 75963,
      "rescore": 100387,
      "ptb": 94222,
      "nnlm": 80014,
      "clstm": 17907,
      "sections": 104491,
      "abstractions": 1798,
      "snapshot": 110305,
      "21": 756,
      "ignore": 51663,
      "dissimilar": 32396,
      "dissimilarities": 32400,
      "decomposing": 28156,
      "calculates": 15024,
      "decomposed": 28154,
      "ultradense": 126560,
      "densifier": 29030,
      "subspace": 114843,
      "reach": 95886,
      "lost": 65313,
      "compactness": 19115,
      "quotations": 95414,
      "latin": 61663,
      "establishing": 38614,
      "quotation": 95413,
      "parallels": 84708,
      "identical": 51354,
      "rephrasing": 98924,
      "infrequent": 55173,
      "4600": 1042,
      "babylonian": 11282,
      "seconds": 104481,
      "bioinformatics": 14159,
      "lab": 58379,
      "scientists": 104030,
      "authorial": 10372,
      "intertextuality": 56318,
      "genomics": 47826,
      "echoes": 34501,
      "knowledgebased": 58248,
      "scarcity": 103801,
      "failing": 42976,
      "empowering": 36340,
      "convolution": 23136,
      "joins": 57255,
      "differentially": 31586,
      "penalizes": 85848,
      "subcomponents": 114653,
      "limiteddomain": 64302,
      "grows": 49186,
      "exponentially": 41682,
      "synthesised": 116602,
      "outofdomain": 82646,
      "proposition": 93630,
      "uniformly": 127145,
      "061": 44,
      "071": 59,
      "083": 82,
      "player": 87713,
      "072": 61,
      "observing": 81244,
      "trends": 125745,
      "private": 90784,
      "aware": 11256,
      "analysts": 5944,
      "seventy": 107877,
      "lets": 63405,
      "apps": 8474,
      "devices": 30456,
      "categorized": 15773,
      "multichannel": 75040,
      "pretrained": 89981,
      "multigranular": 75117,
      "pretraining": 90231,
      "namedentity": 76134,
      "ned": 76772,
      "knowledgebase": 58247,
      "batched": 12558,
      "speedup": 111884,
      "nli": 79533,
      "sentenceencoding": 106141,
      "running": 103469,
      "prepositional": 89308,
      "pp": 88682,
      "attachments": 9698,
      "dual": 34225,
      "enforces": 37006,
      "mstparser": 74953,
      "forest": 45047,
      "taxonomical": 119657,
      "2016": 601,
      "10fold": 205,
      "907": 1377,
      "572": 1127,
      "957": 1426,
      "918": 1393,
      "978": 1445,
      "claimed": 17016,
      "levy": 63715,
      "switched": 116291,
      "drastically": 34099,
      "934": 1408,
      "602": 1154,
      "923": 1398,
      "655": 1182,
      "973": 1442,
      "815": 1301,
      "rewrite": 102689,
      "highlevel": 50242,
      "gist": 47973,
      "pure": 94409,
      "libraries": 63960,
      "participant": 85307,
      "irrelevant": 56945,
      "discernible": 32013,
      "visualized": 131837,
      "crowdsourced": 25103,
      "asr": 9422,
      "arising": 9003,
      "stories": 113396,
      "storylines": 113423,
      "storyline": 113422,
      "engaging": 37017,
      "slidingwindow": 110014,
      "tackled": 117316,
      "sideproduct": 108659,
      "morph": 74665,
      "minimalistic": 70185,
      "obvious": 81486,
      "tibetan": 121954,
      "promotes": 92321,
      "unity": 127273,
      "stability": 112104,
      "realize": 96120,
      "body": 14394,
      "blstm": 14380,
      "sensitivity": 105752,
      "disfluencies": 32355,
      "ensuring": 37646,
      "exceeds": 39907,
      "satisfy": 103657,
      "margin": 67182,
      "decouple": 28165,
      "measurements": 67845,
      "auxiliary": 10873,
      "bilstm": 14089,
      "reliance": 98628,
      "byte": 14996,
      "sizes": 109957,
      "rare": 95736,
      "corruptions": 24331,
      "assumed": 9651,
      "dependencybased": 29255,
      "semeval": 105492,
      "dialogbased": 30617,
      "partner": 85518,
      "setup": 107849,
      "babi": 11277,
      "weston": 132457,
      "teachers": 119700,
      "essays": 38532,
      "subcorpora": 114655,
      "educational": 34570,
      "establishments": 38625,
      "sweden": 116261,
      "a1": 1454,
      "a2": 1456,
      "b1": 11272,
      "b2": 11273,
      "digitization": 31731,
      "registration": 97687,
      "waiting": 131989,
      "howto": 50689,
      "pay": 85809,
      "intensities": 55881,
      "comparatively": 19215,
      "lots": 65335,
      "conducts": 20966,
      "gated": 46501,
      "genetic": 47817,
      "bible": 13892,
      "organisms": 82465,
      "plants": 87649,
      "confirms": 21027,
      "115": 240,
      "expense": 40408,
      "patent": 85661,
      "claims": 17017,
      "owing": 83388,
      "irregular": 56941,
      "accumulating": 2054,
      "forced": 45014,
      "lessons": 63398,
      "campaigns": 15091,
      "garnered": 46494,
      "influencing": 54326,
      "exercises": 39989,
      "authenticity": 10358,
      "076": 66,
      "contextindependent": 22370,
      "dropped": 34180,
      "pronoun": 92336,
      "dps": 34078,
      "prodrop": 91865,
      "nontrivial": 80309,
      "zp": 135553,
      "zps": 135554,
      "explainer": 41271,
      "pictures": 87490,
      "batch": 12555,
      "picks": 87486,
      "englishtofrench": 37440,
      "massive": 67322,
      "unfolds": 127075,
      "comprehensiveness": 20291,
      "timesensitive": 122194,
      "summary": 115631,
      "438": 1025,
      "curricula": 25419,
      "scalar": 103701,
      "engineered": 37039,
      "characterizes": 16555,
      "identitybased": 51640,
      "stimulus": 113349,
      "discrimination": 32202,
      "rigorously": 102858,
      "stimuli": 113348,
      "unless": 127422,
      "resides": 100735,
      "brains": 14555,
      "interpersonal": 56198,
      "worlds": 135058,
      "formulated": 45285,
      "projective": 92240,
      "flag": 44628,
      "axes": 11268,
      "minds": 70148,
      "pipelines": 87566,
      "resourcerich": 100921,
      "divergent": 32787,
      "fullysupervised": 45988,
      "intelligibility": 55861,
      "fidelity": 43929,
      "comprehension": 20162,
      "postediting": 88470,
      "qe": 94529,
      "differs": 31601,
      "concise": 20699,
      "hopefully": 50649,
      "nmt": 79819,
      "pushing": 94465,
      "posteriors": 88492,
      "beamsearch": 12612,
      "singletask": 109843,
      "siamese": 108649,
      "convnets": 23135,
      "multirelational": 75768,
      "vectorspace": 131497,
      "transe": 124030,
      "51": 1091,
      "convergence": 22920,
      "tradeoff": 122773,
      "illustrates": 51749,
      "rising": 102873,
      "encoders": 36631,
      "coupled": 24457,
      "coupling": 24462,
      "contextualized": 22539,
      "efficacy": 35009,
      "m2": 65731,
      "gec": 46575,
      "scorer": 104145,
      "dense": 29010,
      "optimizers": 82238,
      "conll2014": 21104,
      "notice": 80395,
      "4637": 1044,
      "subtitle": 114989,
      "crawled": 24596,
      "speakerbased": 111303,
      "05": 31,
      "qualities": 94581,
      "100k": 184,
      "strengthens": 113570,
      "salient": 103546,
      "degrees": 28578,
      "antonyms": 6760,
      "encoderdecoders": 36629,
      "vanilla": 130810,
      "approximator": 8473,
      "sides": 108660,
      "equip": 38236,
      "reparameterization": 98906,
      "bound": 14507,
      "autoencoders": 10415,
      "sourcetarget": 110944,
      "subphrases": 114800,
      "bilinear": 14017,
      "maxmargin": 67556,
      "163": 384,
      "equal": 38220,
      "regards": 97670,
      "psychometrics": 94221,
      "testset": 120615,
      "collecting": 18443,
      "imply": 52041,
      "word2vec": 133666,
      "3k": 994,
      "semeval2016": 105526,
      "participating": 85354,
      "spans": 111195,
      "parent": 84854,
      "inferencebased": 54250,
      "canonicalization": 15158,
      "reinforcement": 97790,
      "promise": 92256,
      "reward": 102670,
      "chatbot": 16662,
      "simulates": 109645,
      "policy": 87943,
      "informativity": 55166,
      "manages": 66908,
      "foster": 45335,
      "sustained": 116224,
      "centering": 15894,
      "incoherent": 53412,
      "markovian": 67268,
      "pseudo": 94177,
      "organizers": 82478,
      "obstacle": 81247,
      "spend": 111921,
      "clozestyle": 17898,
      "twostep": 126167,
      "ontonotes": 81857,
      "adversarial": 4099,
      "copious": 23281,
      "enjoy": 37542,
      "abundance": 1830,
      "branches": 14561,
      "discriminator": 32222,
      "indicative": 53876,
      "scans": 103787,
      "wordgenerating": 133716,
      "345": 956,
      "lowfrequency": 65468,
      "simultaneous": 109653,
      "nonterminal": 80306,
      "nonterminals": 80307,
      "hebrew": 49670,
      "hungarian": 51159,
      "coarsetofine": 18059,
      "semi": 105563,
      "idiosyncrasies": 51656,
      "agnostic": 4379,
      "comparably": 19194,
      "edinburgh": 34526,
      "englishczech": 37358,
      "englishromanian": 37422,
      "bpe": 14537,
      "openvocabulary": 82041,
      "backtranslations": 11344,
      "pervasive": 87202,
      "tied": 121962,
      "redundant": 97477,
      "incorporated": 53496,
      "add": 3347,
      "wmt16": 132828,
      "chrf3": 16941,
      "crfs": 24766,
      "neuralbased": 78733,
      "lexically": 63855,
      "benefited": 13000,
      "enriched": 37561,
      "capitalize": 15244,
      "greatest": 49040,
      "governed": 48540,
      "outset": 83192,
      "progresses": 92186,
      "encourage": 36748,
      "asymmetric": 9672,
      "viterbi": 131850,
      "cbow": 15846,
      "suboptimal": 114789,
      "averaged": 11220,
      "albeit": 4664,
      "elegant": 35264,
      "truly": 125858,
      "smart": 110255,
      "suggestion": 115355,
      "tap": 117578,
      "mobile": 70464,
      "responsible": 101309,
      "throughput": 121949,
      "usergenerated": 129069,
      "modest": 74437,
      "cambridge": 15078,
      "synergies": 116341,
      "restriction": 101347,
      "couple": 24454,
      "uncovered": 126649,
      "behave": 12642,
      "black": 14244,
      "rigidity": 102847,
      "timeline": 122154,
      "insightful": 55526,
      "timelines": 122160,
      "dispute": 32383,
      "080": 76,
      "disagreement": 31935,
      "segmentlevel": 104669,
      "bootstrapped": 14455,
      "disparate": 32370,
      "067": 51,
      "058": 40,
      "056": 38,
      "egyptian": 35209,
      "competition": 19620,
      "informal": 54337,
      "chat": 16651,
      "standardizing": 112349,
      "cuni": 25238,
      "causal": 15802,
      "bindings": 14143,
      "handannotated": 49329,
      "battery": 12563,
      "micro": 70034,
      "43": 1024,
      "performers": 86940,
      "sievebased": 108662,
      "46": 1039,
      "chatbots": 16667,
      "summarizes": 115620,
      "synthesizes": 116609,
      "appearing": 6810,
      "window": 132709,
      "romanianenglish": 103253,
      "notoriously": 80416,
      "accompanied": 1943,
      "wellcalibrated": 132397,
      "risk": 102874,
      "workflows": 134893,
      "marries": 67275,
      "votes": 131949,
      "served": 107310,
      "mesoscopic": 68491,
      "hinges": 50497,
      "grasp": 48985,
      "overlooked": 83345,
      "multiscale": 75772,
      "traits": 123972,
      "myriad": 75998,
      "randomized": 95523,
      "stress": 113578,
      "indirect": 53887,
      "gathered": 46527,
      "topicrelated": 122611,
      "undesirable": 127040,
      "choices": 16913,
      "crafted": 24590,
      "abundant": 1832,
      "lesser": 63396,
      "predictionbased": 89156,
      "benchmarked": 12875,
      "tsne": 125911,
      "python": 94478,
      "plurality": 87767,
      "capital": 15238,
      "scikitlearn": 104032,
      "weakness": 132204,
      "contextualize": 22538,
      "bagofwords": 11368,
      "acronyms": 3054,
      "enfr": 37009,
      "ende": 36843,
      "contextualization": 22537,
      "lingual": 64413,
      "unreliable": 127506,
      "debiasing": 27898,
      "framed": 45391,
      "plentiful": 87747,
      "equals": 38232,
      "kinyarwanda": 57698,
      "normative": 80364,
      "obligation": 81153,
      "permission": 87045,
      "prohibition": 92199,
      "diagram": 30514,
      "verifying": 131596,
      "satisfies": 103655,
      "causality": 15815,
      "timing": 122208,
      "luong": 65724,
      "resorts": 100826,
      "unks": 127366,
      "embed": 35347,
      "schema": 103892,
      "kbs": 57509,
      "fundamentally": 46132,
      "falls": 43033,
      "manuallyproduced": 67095,
      "execution": 39974,
      "gained": 46356,
      "uas": 126507,
      "crossdocument": 24879,
      "entering": 37692,
      "doc2vec": 32940,
      "mikolov": 70078,
      "2013a": 571,
      "struggled": 114130,
      "hyperparameter": 51226,
      "tease": 119738,
      "14": 309,
      "premises": 89289,
      "encompassing": 36741,
      "subdomains": 114666,
      "void": 131932,
      "domaindependent": 33711,
      "modulation": 74475,
      "undergo": 126665,
      "subjected": 114688,
      "cfgs": 15969,
      "seq2seq": 106884,
      "authored": 10371,
      "nlu": 79803,
      "expansions": 40378,
      "yielded": 135365,
      "critically": 24843,
      "manager": 66906,
      "gameplay": 46421,
      "attributed": 10188,
      "offline": 81610,
      "codes": 18197,
      "iteratively": 57139,
      "gibbs": 47961,
      "guarantee": 49213,
      "49": 1054,
      "job": 57239,
      "resumes": 102354,
      "kept": 57528,
      "pace": 83401,
      "incredible": 53717,
      "inefficient": 54074,
      "plugged": 87763,
      "47": 1047,
      "company": 19121,
      "phylogenetic": 87473,
      "borrowing": 14475,
      "grained": 48620,
      "09": 97,
      "exploitable": 41448,
      "psycholinguistics": 94209,
      "expects": 40405,
      "centerembedded": 15893,
      "extensively": 41963,
      "examinations": 39731,
      "restrict": 101335,
      "syntaxoriented": 116587,
      "inspects": 55555,
      "normed": 80366,
      "swap": 116253,
      "swapped": 116254,
      "121": 266,
      "analogical": 5408,
      "contextualised": 22528,
      "element": 35267,
      "positioning": 88309,
      "replete": 98955,
      "opportunities": 82141,
      "outofvocabulary": 82672,
      "logbilinear": 64986,
      "softmaxbased": 110511,
      "unlabelled": 127415,
      "fairly": 42994,
      "colloquial": 18521,
      "nonword": 80317,
      "refines": 97597,
      "tokenlevel": 122286,
      "sharedtask": 108121,
      "facilitated": 42800,
      "dbpedia": 27848,
      "pathway": 85694,
      "communicating": 19024,
      "birds": 14227,
      "john": 57243,
      "hikes": 50442,
      "philosophically": 87266,
      "resisted": 100745,
      "generalizations": 46801,
      "conveyed": 23121,
      "truthconditional": 125887,
      "beliefs": 12693,
      "characterword": 16635,
      "brevity": 14588,
      "codeswitching": 18224,
      "violence": 131737,
      "inappropriate": 53203,
      "quantity": 94899,
      "narratives": 76184,
      "communitybased": 19103,
      "website": 132291,
      "appropriateness": 8448,
      "automating": 10842,
      "inappropriateness": 53204,
      "96": 1428,
      "explosive": 41671,
      "corenlp": 23385,
      "wiki": 132624,
      "disadvantages": 31931,
      "runtimes": 103482,
      "incurring": 53753,
      "kenlm": 57527,
      "imposes": 52306,
      "longstanding": 65191,
      "dominance": 33926,
      "pbmt": 85821,
      "respects": 101179,
      "postedits": 88478,
      "pointing": 87840,
      "slang": 110004,
      "shorttext": 108352,
      "tackling": 117321,
      "socioeconomic": 110475,
      "laborious": 58668,
      "faces": 42751,
      "drawback": 34119,
      "accumulation": 2056,
      "incorrectly": 53582,
      "topmost": 122674,
      "constitutes": 21558,
      "tying": 126182,
      "evolves": 39700,
      "regularizing": 97772,
      "harming": 49506,
      "24": 798,
      "metalanguage": 68540,
      "appraisal": 7281,
      "martin": 67279,
      "white": 132473,
      "2005": 550,
      "corpusassisted": 24073,
      "film": 44068,
      "arguing": 8933,
      "firstperson": 44572,
      "wonderful": 132898,
      "entertaining": 37695,
      "bodily": 14393,
      "reaction": 95924,
      "lasting": 61554,
      "groupings": 49148,
      "appreciation": 7285,
      "neighbourhood": 77001,
      "adjectivenoun": 3859,
      "verbobject": 131548,
      "compositions": 20148,
      "noncanonical": 80167,
      "soon": 110664,
      "drops": 34186,
      "prominently": 92255,
      "newswire": 79417,
      "communitys": 19104,
      "nonobvious": 80258,
      "hitherto": 50577,
      "sight": 108667,
      "embrace": 36047,
      "131": 295,
      "dollars": 33422,
      "texas": 120619,
      "numeric": 81004,
      "formula": 45268,
      "familiarity": 43051,
      "152": 363,
      "125": 272,
      "curated": 25242,
      "favorably": 43216,
      "englisharabic": 37354,
      "lengthbased": 63385,
      "nce": 76701,
      "inapplicable": 53202,
      "refute": 97645,
      "food": 44998,
      "gru": 49197,
      "skewed": 109970,
      "distinctively": 32563,
      "deteriorates": 30108,
      "mixeddomain": 70407,
      "spite": 111934,
      "domainrelated": 33720,
      "translationrelated": 125449,
      "fullyunsupervised": 45992,
      "fullyneural": 45986,
      "slavic": 110007,
      "fixedlength": 44615,
      "default": 28467,
      "phonology": 87318,
      "unordered": 127482,
      "phylogeny": 87475,
      "clarifies": 17025,
      "italy": 57103,
      "goes": 48424,
      "easytouse": 34493,
      "extendable": 41800,
      "ecosystem": 34516,
      "progressed": 92185,
      "tremendously": 125737,
      "minimally": 70186,
      "texttospeech": 121657,
      "tts": 125913,
      "enumerate": 38179,
      "saying": 103674,
      "datarich": 26697,
      "verbalizations": 131539,
      "languageuniversal": 60998,
      "developer": 30327,
      "merely": 68474,
      "hallucinated": 49302,
      "inverted": 56706,
      "resnets": 100748,
      "bypass": 14992,
      "intrinsically": 56373,
      "resnet": 100746,
      "ud": 126527,
      "v13": 130689,
      "amrtotext": 5402,
      "traveling": 125542,
      "amr": 5380,
      "solver": 110633,
      "relied": 98638,
      "manuallycrafted": 67093,
      "pivoting": 87589,
      "unsatisfactory": 127511,
      "latticebased": 61672,
      "tokenized": 122282,
      "confronted": 21046,
      "1best": 477,
      "tokenizations": 122281,
      "propagate": 92383,
      "wordlattice": 133723,
      "compactly": 19114,
      "arbitrarily": 8577,
      "preceding": 88775,
      "factorized": 42876,
      "tendency": 120167,
      "subjectverbobject": 114715,
      "speechtotranslation": 111859,
      "documenting": 33131,
      "endangered": 36838,
      "dyer": 34289,
      "als": 5210,
      "kmeans": 57713,
      "warping": 132034,
      "coordinated": 23261,
      "burst": 14973,
      "opened": 81982,
      "venues": 131509,
      "coarsegrained": 18054,
      "topiclevel": 122609,
      "streaming": 113552,
      "neverending": 78760,
      "tail": 117498,
      "guidance": 49225,
      "fixation": 44593,
      "motivates": 74869,
      "predictors": 89211,
      "surprisal": 116121,
      "uni": 127084,
      "super": 115667,
      "restricting": 101345,
      "embeddingbased": 35532,
      "cpus": 24577,
      "wmt15": 132825,
      "negligible": 76975,
      "neuropsychological": 78749,
      "diagnosis": 30503,
      "prevents": 90373,
      "impaired": 51908,
      "prosodic": 93639,
      "healthy": 49637,
      "elderly": 35234,
      "mild": 70083,
      "impairment": 51909,
      "070": 56,
      "055": 37,
      "053": 36,
      "ternary": 120414,
      "errorcorrecting": 38348,
      "populate": 88161,
      "asymptotic": 9674,
      "glass": 48221,
      "syllable": 116301,
      "variablelength": 130851,
      "consonantvowel": 21511,
      "abugida": 1829,
      "alphabetic": 5207,
      "alternation": 5219,
      "polymorphic": 88017,
      "subcategories": 114649,
      "centroid": 15918,
      "fillers": 44031,
      "edge": 34521,
      "centroids": 15919,
      "underpinned": 126719,
      "denoted": 29007,
      "texting": 121440,
      "abuse": 1840,
      "crime": 24773,
      "driving": 34161,
      "privacy": 90781,
      "ready": 96047,
      "united": 127225,
      "nations": 76217,
      "v10": 130687,
      "ratios": 95846,
      "github": 47975,
      "impressively": 52324,
      "unnecessary": 127478,
      "denoising": 28994,
      "strengthening": 113569,
      "concern": 20683,
      "clientserver": 17755,
      "dissertation": 32395,
      "device": 30454,
      "alternatives": 5254,
      "prevent": 90363,
      "staff": 112142,
      "hindienglish": 50478,
      "doctorpatient": 32944,
      "ranging": 95617,
      "synthetically": 116654,
      "drop": 34166,
      "attentive": 10105,
      "specially": 111398,
      "truefalse": 125857,
      "violation": 131735,
      "govern": 48538,
      "insults": 55730,
      "situated": 109883,
      "stacked": 112121,
      "cells": 15886,
      "internal": 56157,
      "cell": 15885,
      "pretty": 90354,
      "keystroke": 57627,
      "goaldriven": 48403,
      "flexibly": 44669,
      "productively": 92067,
      "cooperate": 23252,
      "neuralnetwork": 78737,
      "conjuncts": 21071,
      "conjunct": 21066,
      "solely": 110529,
      "genia": 47823,
      "enormously": 37554,
      "decade": 27906,
      "multicore": 75060,
      "reexamine": 97490,
      "replacement": 98940,
      "cores": 23387,
      "wordphrase": 133790,
      "conference": 20973,
      "correctincorrect": 24133,
      "rnnlm": 102952,
      "rnnlms": 102957,
      "advised": 4201,
      "tu": 125920,
      "2016b": 623,
      "2016a": 622,
      "2017": 624,
      "smtbased": 110298,
      "simplifying": 109616,
      "lisa": 64686,
      "moved": 74884,
      "laboratory": 58664,
      "announcing": 6503,
      "rollout": 103242,
      "32": 927,
      "ultimate": 126553,
      "contributing": 22765,
      "adoption": 3905,
      "delivery": 28628,
      "usecase": 128372,
      "lexstat": 63944,
      "transferring": 124237,
      "identifies": 51467,
      "pseudosyllables": 94196,
      "vowel": 131956,
      "compress": 20292,
      "generationbased": 47704,
      "humancomputer": 51004,
      "attracted": 10141,
      "retrievalbased": 102439,
      "synthesize": 116604,
      "meaningless": 67737,
      "fed": 43798,
      "rnnbased": 102940,
      "monotone": 74639,
      "postcorrection": 88464,
      "humanities": 51040,
      "deeplearning": 28455,
      "restaurants": 101327,
      "celebrities": 15882,
      "aforementioned": 4266,
      "28": 864,
      "inherently": 55192,
      "reaching": 95921,
      "macroaveraged": 66362,
      "frequencyinverse": 45854,
      "ignores": 51679,
      "recurrentneuralnetwork": 97276,
      "lstmlm": 65694,
      "baby": 11281,
      "artificially": 9270,
      "subwords": 115053,
      "scarce": 103788,
      "carries": 15548,
      "ordinal": 82448,
      "stated": 112542,
      "holding": 50592,
      "regard": 97648,
      "diversely": 32860,
      "adverse": 4187,
      "randomly": 95524,
      "vietnamese": 131668,
      "unusual": 127771,
      "las": 61550,
      "7353": 1233,
      "8066": 1295,
      "cogalexv": 18253,
      "meronymy": 68488,
      "solid": 110540,
      "autoencoderbased": 10414,
      "bits": 14241,
      "realvalued": 96143,
      "predictable": 88955,
      "locations": 64973,
      "deterministic": 30158,
      "ifthen": 51661,
      "computes": 20514,
      "twostage": 126157,
      "oneshot": 81713,
      "closing": 17884,
      "reconstruction": 97114,
      "remarkable": 98831,
      "mistakenly": 70347,
      "reconstructor": 97127,
      "summarises": 115469,
      "entitys": 38149,
      "biography": 14157,
      "pool": 88036,
      "entityrelated": 38145,
      "video": 131653,
      "youtube": 135461,
      "viewers": 131717,
      "producers": 91984,
      "china": 16735,
      "informational": 55115,
      "fiction": 43922,
      "records": 97135,
      "chemical": 16706,
      "protein": 93664,
      "digitized": 31732,
      "consortium": 21512,
      "aalto": 1460,
      "anomalous": 6507,
      "trajectory": 123979,
      "spatiotemporal": 111273,
      "motion": 74831,
      "confined": 21011,
      "road": 102987,
      "particle": 85371,
      "dataless": 26691,
      "contrasted": 22721,
      "heavy": 49664,
      "marriage": 67274,
      "undirected": 127045,
      "instability": 55588,
      "744": 1241,
      "fuses": 46227,
      "channels": 16408,
      "synthesized": 116606,
      "perceptron": 85930,
      "conducting": 20961,
      "pointer": 87820,
      "ptrnet": 94228,
      "195": 446,
      "240": 803,
      "billion": 14078,
      "librarys": 63965,
      "team": 119715,
      "helsinki": 49835,
      "concordances": 20737,
      "rates": 95815,
      "establish": 38576,
      "figure": 44016,
      "nns": 80016,
      "scala": 103680,
      "recreate": 97147,
      "atr": 9687,
      "clean": 17698,
      "hours": 50680,
      "legacy": 63311,
      "borndigital": 14471,
      "semiautomatically": 105573,
      "continuum": 22659,
      "tracing": 122721,
      "seamlessly": 104289,
      "literally": 64731,
      "lose": 65237,
      "reduplication": 97482,
      "hyphenation": 51241,
      "false": 43035,
      "positives": 88353,
      "checked": 16693,
      "transparent": 125535,
      "controls": 22863,
      "alleviating": 5062,
      "deteriorating": 30109,
      "allocate": 5067,
      "rejects": 97837,
      "intensive": 55886,
      "bulky": 14951,
      "consumption": 21728,
      "ownership": 83394,
      "energy": 36996,
      "10x": 212,
      "2x": 897,
      "quantization": 94902,
      "friendly": 45889,
      "partitions": 85513,
      "pe": 85831,
      "parallelism": 84700,
      "schedule": 103881,
      "hardware": 49497,
      "282": 867,
      "252": 826,
      "41": 1013,
      "3x": 1002,
      "titan": 122217,
      "gpu": 48574,
      "reproducible": 100088,
      "1990s": 461,
      "phoenix": 87271,
      "advances": 3973,
      "checks": 16703,
      "warning": 132032,
      "cer": 15927,
      "buy": 14984,
      "incompatible": 53417,
      "categorylevel": 15797,
      "1200": 263,
      "ecommerce": 34503,
      "conceptnet": 20596,
      "supplying": 115952,
      "connects": 21154,
      "edges": 34524,
      "narrower": 76195,
      "nn": 80013,
      "null": 80832,
      "yesno": 135325,
      "mouse": 74882,
      "pulearning": 94383,
      "historically": 50541,
      "97000": 1440,
      "bytepair": 15003,
      "discontiguous": 32025,
      "contours": 22661,
      "register": 97683,
      "timevarying": 122207,
      "f0": 42641,
      "accent": 1880,
      "contour": 22660,
      "prominence": 92248,
      "rhythmic": 102721,
      "nucleus": 80827,
      "localization": 64945,
      "partly": 85514,
      "median": 68188,
      "deviation": 30452,
      "labor": 58662,
      "resourceful": 100896,
      "quantities": 94894,
      "standardly": 112350,
      "prospects": 93656,
      "spellcheckers": 111897,
      "bantu": 11424,
      "reciprocal": 96795,
      "modifiers": 74453,
      "overcomes": 83301,
      "chained": 15980,
      "multitopic": 75905,
      "2015a": 600,
      "removal": 98869,
      "recommendation": 97084,
      "remove": 98872,
      "1303": 294,
      "095": 111,
      "task1": 118862,
      "090": 100,
      "malayalam": 66875,
      "task2": 118865,
      "081": 78,
      "erasure": 38267,
      "wordvector": 134345,
      "erase": 38266,
      "flip": 44672,
      "arranged": 9023,
      "regularise": 97735,
      "backpropagating": 11318,
      "capitalized": 15245,
      "predictor": 89210,
      "regimes": 97675,
      "shortcut": 108285,
      "shortcuts": 108286,
      "block": 14356,
      "discarding": 32008,
      "topologies": 122684,
      "flourishing": 44679,
      "frontier": 45900,
      "latest": 61651,
      "bulk": 14949,
      "drug": 34194,
      "lowering": 65455,
      "percent": 85917,
      "crimes": 24774,
      "mix": 70392,
      "behaviors": 12674,
      "impacts": 51906,
      "prevention": 90372,
      "accompanying": 1946,
      "violent": 131738,
      "situational": 109892,
      "car": 15498,
      "criminal": 24775,
      "threats": 121935,
      "225": 782,
      "casestudy": 15673,
      "persianenglish": 87089,
      "absorbed": 1755,
      "hyper": 51203,
      "187": 428,
      "personalized": 87140,
      "twophase": 126152,
      "responding": 101185,
      "aided": 4437,
      "judgement": 57433,
      "pagerank": 83409,
      "attached": 9690,
      "leaf": 61975,
      "delexicalized": 28602,
      "distinctly": 32565,
      "creative": 24725,
      "selfattention": 104872,
      "inject": 55255,
      "spectrum": 111632,
      "redditcom": 97304,
      "hacker": 49293,
      "overwhelming": 83385,
      "isnt": 56968,
      "timely": 122161,
      "prune": 94159,
      "marketing": 67243,
      "posted": 88466,
      "thread": 121930,
      "harvested": 49518,
      "omit": 81652,
      "tracks": 122767,
      "stacking": 112132,
      "startoftheart": 112415,
      "conventionally": 22916,
      "interleaved": 56116,
      "treatments": 125568,
      "sarcasm": 103638,
      "eyemovement": 42635,
      "37": 970,
      "93": 1403,
      "heldout": 49685,
      "empower": 36338,
      "5gram": 1139,
      "relating": 97936,
      "recommender": 97096,
      "spell": 111892,
      "presently": 89812,
      "editions": 34559,
      "reflected": 97614,
      "propagated": 92384,
      "originating": 82578,
      "snapshots": 110306,
      "era": 38263,
      "ml": 70431,
      "contradict": 22665,
      "intuition": 56663,
      "fading": 42944,
      "tensors": 120198,
      "clearer": 17734,
      "certainly": 15965,
      "revitalize": 102657,
      "tight": 121966,
      "interleaving": 56118,
      "highresource": 50420,
      "sourceside": 110934,
      "distillation": 32501,
      "stronger": 113732,
      "teacher": 119693,
      "speeds": 111883,
      "oracle": 82254,
      "revolutionized": 102664,
      "supposed": 116044,
      "switches": 116293,
      "knowledgegrounded": 58274,
      "factual": 42926,
      "versatile": 131599,
      "languageagnostic": 60353,
      "semitic": 105636,
      "harness": 49512,
      "syntacticosemantic": 116523,
      "standing": 112355,
      "amharic": 5314,
      "alongside": 5203,
      "300k": 913,
      "77": 1264,
      "loc": 64900,
      "misc": 70289,
      "tc": 119675,
      "chance": 16350,
      "prefer": 89235,
      "usingword": 130396,
      "complementarity": 19731,
      "8915": 1354,
      "advancing": 4028,
      "quantifiers": 94840,
      "bagofconcepts": 11366,
      "causing": 15842,
      "attentions": 10104,
      "appealing": 6790,
      "wordaligned": 133692,
      "meaningpreserving": 67738,
      "iv": 57151,
      "languageneutral": 60375,
      "mediums": 68247,
      "services": 107327,
      "abusive": 1842,
      "abbreviated": 1462,
      "alike": 5026,
      "taboo": 117274,
      "suspicious": 116221,
      "adaption": 3327,
      "benefiting": 13001,
      "057": 39,
      "clauselevel": 17695,
      "harmonic": 49507,
      "learnability": 62196,
      "ot": 82606,
      "expressivity": 41776,
      "affords": 4264,
      "upper": 127818,
      "postedit": 88468,
      "collective": 18508,
      "highvalue": 50437,
      "crowd": 25094,
      "visualisation": 131817,
      "characterisation": 16511,
      "boring": 14468,
      "emergence": 36055,
      "interact": 55930,
      "physically": 87479,
      "imitation": 51837,
      "onesentence": 81712,
      "biographies": 14156,
      "wikidata": 132629,
      "slotvalue": 110095,
      "hallucinating": 49303,
      "controlling": 22850,
      "interlocutors": 56128,
      "closest": 17883,
      "round": 103311,
      "manifest": 66918,
      "higherquality": 50218,
      "broader": 14697,
      "wishes": 132744,
      "subwordlevel": 115048,
      "morphemelevel": 74668,
      "displays": 32381,
      "consonants": 21510,
      "mimicking": 70137,
      "nonrecurrent": 80283,
      "alpha": 5204,
      "dnns": 32935,
      "injecting": 55262,
      "on3": 81662,
      "saves": 103666,
      "threads": 121932,
      "ic": 51304,
      "adopting": 3903,
      "rg": 102705,
      "homonyms": 50621,
      "neighboring": 76992,
      "southern": 110966,
      "really": 96131,
      "posing": 88278,
      "deeply": 28464,
      "researched": 100669,
      "demand": 28631,
      "thoroughly": 121899,
      "division": 32919,
      "serving": 107331,
      "located": 64957,
      "entirety": 37736,
      "workable": 134883,
      "eliminate": 35295,
      "opaque": 81877,
      "alternations": 5220,
      "canadian": 15095,
      "influenced": 54320,
      "encounters": 36747,
      "raised": 95458,
      "subgraphs": 114671,
      "lambada": 58789,
      "breakthroughs": 14585,
      "basics": 12540,
      "deceptive": 27924,
      "authorship": 10389,
      "attribution": 10209,
      "subsampling": 114806,
      "kldivergence": 57710,
      "sending": 105641,
      "premise": 89284,
      "approximated": 8462,
      "humanannotated": 50994,
      "meta": 68530,
      "parsingbased": 85283,
      "subsequence": 114809,
      "discovers": 32132,
      "contextaware": 22336,
      "geometrical": 47866,
      "lexeme": 63721,
      "corner": 23391,
      "request": 100103,
      "sum": 115424,
      "geometrically": 47867,
      "mu": 75021,
      "inner": 55274,
      "formatting": 45222,
      "gans": 46436,
      "discriminated": 32196,
      "golden": 48450,
      "machinegenerated": 66316,
      "game": 46417,
      "equilibrium": 38235,
      "transformer": 124285,
      "visually": 131844,
      "cast": 15676,
      "humanmachine": 51059,
      "chitchat": 16881,
      "120k": 265,
      "semeval2017": 105528,
      "competitively": 19700,
      "sts": 114132,
      "optimizes": 82239,
      "msc": 74944,
      "insertion": 55512,
      "nltk": 79802,
      "discoverability": 32121,
      "suite": 115413,
      "browsed": 14727,
      "humanreadable": 51064,
      "audiobased": 10246,
      "identities": 51636,
      "opensubtitle": 82034,
      "enhancing": 37534,
      "centres": 15916,
      "retailers": 102356,
      "yelp": 135318,
      "quora": 95410,
      "establishment": 38624,
      "brand": 14564,
      "builtin": 14947,
      "virtue": 131753,
      "depth": 29321,
      "provenance": 93739,
      "october": 81529,
      "revision": 102648,
      "revisions": 102650,
      "deleted": 28593,
      "forego": 45038,
      "hindered": 50450,
      "survival": 116213,
      "synthesizer": 116608,
      "musical": 75965,
      "imitate": 51835,
      "king": 57695,
      "james": 57177,
      "lepor": 63391,
      "reasonably": 96217,
      "tell": 120068,
      "weighting": 132361,
      "affective": 4234,
      "threedimensional": 121936,
      "assignments": 9563,
      "colors": 18535,
      "draws": 34138,
      "interprets": 56293,
      "color": 18529,
      "hardest": 49490,
      "newlycollected": 79279,
      "zeroshot": 135488,
      "citation": 16989,
      "sent2vec": 105760,
      "anthology": 6753,
      "10crossvalidation": 201,
      "wordalignmentbased": 133694,
      "embeddingsbased": 36041,
      "succinct": 115204,
      "seeking": 104520,
      "normally": 80363,
      "sports": 112038,
      "journalism": 57410,
      "detector": 30099,
      "heuristically": 49878,
      "twolayer": 126145,
      "unlexicalized": 127423,
      "pooled": 88037,
      "density": 29032,
      "stereotypical": 113342,
      "ucca": 126522,
      "crosslinguistically": 25057,
      "exhibits": 40011,
      "reentrancy": 97483,
      "dag": 25503,
      "discontinuous": 32026,
      "uncertain": 126611,
      "upstream": 127828,
      "segmenter": 104664,
      "errorprone": 38354,
      "treelstm": 125675,
      "tai": 117494,
      "latticelstm": 61673,
      "consume": 21717,
      "treelstms": 125678,
      "childsum": 16731,
      "forget": 45056,
      "microblogging": 70049,
      "experiencing": 40448,
      "renewed": 98892,
      "hashtag": 49524,
      "blackbox": 14250,
      "blstms": 14384,
      "skipthought": 110000,
      "fasttext": 43201,
      "correlating": 24219,
      "spanishenglish": 111178,
      "contextbased": 22350,
      "mtbased": 75010,
      "1st": 482,
      "4a": 1058,
      "iterative": 57124,
      "procrustes": 91863,
      "710": 1219,
      "radiology": 95449,
      "annotates": 6263,
      "radlex": 95453,
      "nonenglish": 80194,
      "shortly": 108311,
      "reddit": 97295,
      "graphstructured": 48978,
      "threaded": 121931,
      "controversial": 22864,
      "fostering": 45339,
      "engagement": 37014,
      "delivering": 28620,
      "listeners": 64707,
      "quantitatively": 94887,
      "presenters": 89805,
      "elicitation": 35291,
      "triggers": 125791,
      "audience": 10220,
      "trolling": 125836,
      "visitors": 131777,
      "leave": 63258,
      "exchange": 39940,
      "malicious": 66884,
      "disclosure": 32020,
      "excerpts": 39936,
      "trolls": 125837,
      "took": 122366,
      "reg": 97647,
      "speakerdependent": 111304,
      "personalised": 87126,
      "entitydriven": 38138,
      "shortage": 108269,
      "wrod": 135159,
      "perturbed": 87198,
      "humanlike": 51053,
      "typos": 126506,
      "adversary": 4186,
      "mismatched": 70310,
      "nonneural": 80253,
      "branching": 14562,
      "proportionally": 92516,
      "infrequently": 55177,
      "japanesetochinese": 57214,
      "englishtojapanese": 37456,
      "undertranslation": 127037,
      "nmts": 80012,
      "rerank": 100360,
      "japanesechinese": 57205,
      "08": 72,
      "inverting": 56708,
      "syntaxaware": 116573,
      "graphconvolutional": 48916,
      "gcns": 46569,
      "syntaxagnostic": 116572,
      "sparked": 111214,
      "snli": 110311,
      "recast": 96349,
      "mutually": 75987,
      "exclusive": 39957,
      "lowrank": 65483,
      "484": 1053,
      "ample": 5369,
      "phenomenal": 87249,
      "1988": 460,
      "interfering": 56113,
      "qlearning": 94534,
      "redundancy": 97476,
      "archives": 8863,
      "53": 1100,
      "handengineered": 49362,
      "precomputation": 88812,
      "typologically": 126491,
      "symmetric": 116326,
      "strategically": 113445,
      "humanhuman": 51025,
      "strategic": 113443,
      "wiktionary": 132701,
      "presumably": 89962,
      "wisdom": 132736,
      "exciting": 39951,
      "handdesigned": 49360,
      "exemplify": 39982,
      "onepass": 81677,
      "interval": 56320,
      "concluding": 20723,
      "modes": 74436,
      "riemannian": 102825,
      "sgns": 107903,
      "competitors": 19707,
      "svd": 116229,
      "disorder": 32368,
      "nongrammatical": 80211,
      "bow": 14529,
      "dementiabank": 28637,
      "cinderella": 16979,
      "duluth": 34258,
      "puns": 94404,
      "catalyze": 15706,
      "uncontrollable": 126644,
      "dilated": 31736,
      "iterated": 57115,
      "practitioners": 88744,
      "gpus": 48581,
      "augmentation": 10271,
      "backtranslation": 11330,
      "machinelearned": 66319,
      "messageresponse": 68504,
      "dotproduct": 33956,
      "fraction": 45363,
      "trialanderror": 125765,
      "convnet": 23134,
      "fullyconnected": 45983,
      "softmax": 110504,
      "initializations": 55239,
      "fourth": 45357,
      "task4": 118867,
      "674": 1190,
      "cltc": 17908,
      "adapts": 3341,
      "supervisory": 115935,
      "mismatch": 70303,
      "advantageous": 4065,
      "imagenet": 51806,
      "mnemonic": 70460,
      "memorize": 68280,
      "prevalent": 90357,
      "parallelized": 84705,
      "nonlinearities": 80227,
      "eases": 34409,
      "wmt14": 132807,
      "datatotext": 27809,
      "unclear": 126631,
      "exercised": 39988,
      "eliciting": 35294,
      "contentbased": 21982,
      "cheap": 16686,
      "94": 1412,
      "underfitting": 126664,
      "inability": 53189,
      "metaknowledge": 68539,
      "farther": 43110,
      "meetings": 68260,
      "footprints": 45009,
      "paris": 84864,
      "presidential": 89945,
      "elections": 35239,
      "squad": 112071,
      "binds": 14144,
      "recruited": 97149,
      "whword": 132482,
      "discrepancies": 32153,
      "exposed": 41687,
      "favoring": 43221,
      "wordembedding": 133708,
      "toolkits": 122423,
      "multi": 75025,
      "listen": 64705,
      "intelligently": 55860,
      "skills": 109974,
      "feedbacks": 43841,
      "collectively": 18509,
      "ballesteros": 11407,
      "persistent": 87093,
      "algebraic": 4679,
      "indicating": 53868,
      "pasa": 85601,
      "zar": 135468,
      "facets": 42756,
      "sourced": 110872,
      "deficits": 28480,
      "theano": 121774,
      "simplify": 109612,
      "topranked": 122691,
      "wolof": 132890,
      "phonetics": 87304,
      "observable": 81162,
      "subsaharan": 114805,
      "20k": 750,
      "lu": 65720,
      "atomic": 9684,
      "dstc": 34211,
      "thorough": 121876,
      "unwritten": 127777,
      "loanwords": 64899,
      "advent": 4091,
      "communications": 19046,
      "codeswitched": 18215,
      "moroccan": 74664,
      "darijaenglish": 25540,
      "jordan": 57403,
      "elman": 35307,
      "slu": 110108,
      "synergistic": 116342,
      "trim": 125803,
      "contiguous": 22590,
      "bt": 14731,
      "cited": 16994,
      "775": 1266,
      "mentioning": 68441,
      "macro": 66353,
      "articulation": 9234,
      "childdirected": 16719,
      "cds": 15878,
      "ads": 3925,
      "holds": 50594,
      "nonprojective": 80277,
      "erroneous": 38271,
      "conllx": 21111,
      "monotonic": 74642,
      "compliant": 19951,
      "worst": 135071,
      "54": 1104,
      "alzheimers": 5258,
      "id": 51314,
      "predications": 88866,
      "sid": 108658,
      "repeating": 98912,
      "diagnostic": 30506,
      "776": 1267,
      "723": 1226,
      "848": 1319,
      "minibatch": 70152,
      "pad": 83406,
      "shorter": 108291,
      "longest": 65168,
      "sorting": 110684,
      "padding": 83407,
      "shuffling": 108647,
      "floatingpoint": 44676,
      "footprint": 45008,
      "ensures": 37645,
      "bahasa": 11384,
      "indonesia": 53988,
      "totally": 122702,
      "22m": 784,
      "indonesian": 53989,
      "neu": 77822,
      "coder": 18195,
      "vinyals": 131728,
      "apple": 6816,
      "fruit": 45904,
      "supervisedlearning": 115860,
      "englishjapanese": 37413,
      "bahdanau": 11386,
      "reimplementation": 97784,
      "jointlytrained": 57402,
      "e2e": 34356,
      "bigger": 14000,
      "promises": 92261,
      "varied": 130942,
      "uniqueness": 127199,
      "naming": 76154,
      "segmental": 104569,
      "scaffold": 103677,
      "calls": 15073,
      "cheaper": 16688,
      "exponential": 41675,
      "protoroles": 93682,
      "advocate": 4202,
      "s2s": 103514,
      "fstbased": 45918,
      "idf": 51648,
      "inexpensive": 54085,
      "curation": 25249,
      "pathways": 85695,
      "tutor": 125999,
      "crosssituational": 25082,
      "gaze": 46560,
      "structurally": 113799,
      "goldberg": 48448,
      "mitigate": 70358,
      "conll": 21072,
      "macroaverage": 66361,
      "7th": 1279,
      "33": 942,
      "12th": 277,
      "suffixed": 115299,
      "11th": 243,
      "leipzig": 63325,
      "masking": 67311,
      "obfuscation": 81046,
      "posting": 88497,
      "anonymous": 6513,
      "logged": 64987,
      "anonymity": 6510,
      "protection": 93663,
      "hisher": 50511,
      "hide": 49919,
      "obfuscate": 81043,
      "pushed": 94459,
      "stylometric": 114640,
      "soundness": 110708,
      "turned": 125985,
      "classificationbased": 17505,
      "projectionbased": 92237,
      "hyponymhypernym": 51244,
      "reformulation": 97637,
      "assistants": 9575,
      "siri": 109871,
      "cortana": 24333,
      "execute": 39969,
      "chatting": 16685,
      "hinders": 50453,
      "hindrance": 50494,
      "dissatisfaction": 32388,
      "ia": 51296,
      "leaveoneout": 63264,
      "slight": 110015,
      "overestimate": 83312,
      "predictability": 88954,
      "hedging": 49676,
      "wmt17": 132837,
      "englishturkish": 37459,
      "bpebased": 14543,
      "ensembling": 37629,
      "crossgenre": 24916,
      "episode": 38211,
      "oov": 81863,
      "englishchinese": 37357,
      "oneself": 81711,
      "clouds": 17891,
      "glance": 48220,
      "publications": 94281,
      "trending": 125742,
      "york": 135452,
      "snippets": 110308,
      "wordcontext": 133706,
      "listwise": 64723,
      "neglecting": 76973,
      "hinder": 50448,
      "losses": 65308,
      "intentionally": 55917,
      "pun": 94385,
      "073": 62,
      "videos": 131663,
      "paying": 85812,
      "operating": 82050,
      "strictly": 113590,
      "sublexical": 114721,
      "placing": 87617,
      "lowerdimensional": 65454,
      "compositionally": 20147,
      "xgboost": 135190,
      "culture": 25233,
      "9065": 1376,
      "feeling": 43862,
      "trump": 125866,
      "forecasting": 45035,
      "winners": 132720,
      "upcoming": 127787,
      "certainty": 15966,
      "actor": 3136,
      "wins": 132730,
      "veridicality": 131567,
      "crowds": 25100,
      "surprise": 116122,
      "directionality": 31834,
      "hypernymhyponym": 51217,
      "encouraged": 36766,
      "spanishlanguage": 111182,
      "identifiable": 51358,
      "conjugation": 21065,
      "cap": 15162,
      "temporally": 120133,
      "fixedsize": 44619,
      "embodies": 36044,
      "classspecific": 17686,
      "gradientbased": 48609,
      "spirit": 111931,
      "testtime": 120618,
      "unconstrained": 126643,
      "break": 14576,
      "dialecte": 30530,
      "danalyse": 25530,
      "smantique": 110254,
      "algerian": 4682,
      "stem": 113231,
      "prefixes": 89256,
      "deception": 27922,
      "bottleneck": 14493,
      "truthful": 125889,
      "advertising": 4198,
      "reviewer": 102591,
      "weakly": 132190,
      "orderings": 82442,
      "englishtoczech": 37439,
      "tremendous": 125735,
      "recruitment": 97150,
      "companys": 19123,
      "grading": 48613,
      "venue": 131508,
      "subtrack": 115006,
      "linearchain": 64389,
      "statetransition": 113059,
      "edinburghs": 34527,
      "latvian": 61675,
      "nematus": 77007,
      "backtranslated": 11322,
      "novelties": 80787,
      "ablative": 1587,
      "sutskever": 116225,
      "underexplored": 126660,
      "namedentities": 76133,
      "syllablebased": 116304,
      "unaware": 126603,
      "lmbased": 64874,
      "rewritten": 102702,
      "estimations": 38674,
      "memoryaugmented": 68398,
      "mnmt": 70463,
      "encountered": 36743,
      "absa": 1712,
      "designs": 29654,
      "blossomed": 14377,
      "multiparty": 75479,
      "attracting": 10163,
      "academia": 1850,
      "surrogate": 116152,
      "temporalbased": 120131,
      "hybrids": 51202,
      "nontextual": 80308,
      "ideograph": 51643,
      "radicals": 95443,
      "costeffective": 24380,
      "inventories": 56690,
      "connectives": 21152,
      "signaled": 108697,
      "newsela": 79401,
      "ppdb": 88684,
      "91": 1382,
      "automation": 10844,
      "retellings": 102364,
      "releasing": 98499,
      "catalan": 15688,
      "galician": 46416,
      "ivas": 57153,
      "telecommunications": 120063,
      "selections": 104852,
      "greetings": 49082,
      "gratitude": 48988,
      "emotions": 36126,
      "tripadvisor": 125805,
      "taskbased": 118872,
      "iva": 57152,
      "teacherstudent": 119702,
      "welltrained": 132441,
      "sourcedomain": 110874,
      "lieu": 63977,
      "targetdomain": 117761,
      "44": 1027,
      "disagree": 31934,
      "grus": 49203,
      "fusing": 46229,
      "multidialect": 75063,
      "bilstmcrf": 14107,
      "stems": 113239,
      "clitics": 17803,
      "ranker": 95653,
      "determines": 30141,
      "englishfinnish": 37361,
      "wellstructured": 132432,
      "dstc2": 34213,
      "felicitous": 43866,
      "questionanswer": 95244,
      "homographs": 50618,
      "alleviated": 5053,
      "feeding": 43855,
      "configuration": 21003,
      "upfront": 127813,
      "layout": 61818,
      "layouts": 61822,
      "enduser": 36992,
      "removes": 98880,
      "0887": 95,
      "0788": 70,
      "storytelling": 113424,
      "persuade": 87182,
      "told": 122356,
      "dialogic": 30618,
      "personality": 87128,
      "tedious": 120054,
      "altogether": 5257,
      "infeasible": 54090,
      "expands": 40368,
      "bridges": 14600,
      "narrator": 76186,
      "perceptions": 85929,
      "personalities": 87127,
      "nerd": 77099,
      "exporting": 41684,
      "desires": 29674,
      "versatility": 131600,
      "licenses": 63971,
      "tolerance": 122357,
      "enterprise": 37693,
      "standardized": 112345,
      "fighting": 44011,
      "incompleteness": 53426,
      "coping": 23280,
      "prover": 93740,
      "obtainable": 81338,
      "deductive": 28187,
      "fracas": 45361,
      "marketplaces": 67248,
      "bought": 14506,
      "sold": 110527,
      "crossdomain": 24884,
      "slotfilling": 110075,
      "sectors": 104494,
      "learningbased": 63177,
      "1938": 443,
      "fifth": 44008,
      "languagedependent": 60358,
      "spent": 111925,
      "occasionally": 81489,
      "aggravates": 4343,
      "burden": 14968,
      "emphneural": 36145,
      "succeeding": 115059,
      "positively": 88349,
      "imageability": 51802,
      "implying": 52043,
      "kobayashi": 58337,
      "copy": 23283,
      "gu": 49208,
      "anonymized": 6512,
      "injection": 55267,
      "playing": 87716,
      "picking": 87485,
      "thresholds": 121948,
      "responsive": 101312,
      "nearoptimal": 76738,
      "mortality": 74820,
      "accountability": 2045,
      "singlelayer": 109833,
      "mimiciii": 70135,
      "steering": 113230,
      "steered": 113229,
      "faithfulness": 43007,
      "degrading": 28565,
      "ambitious": 5303,
      "chit": 16879,
      "friends": 45891,
      "grow": 49157,
      "bots": 14488,
      "memoryenhanced": 68404,
      "tractability": 122768,
      "neighbouring": 77003,
      "wmt2017": 132871,
      "opennmt": 82004,
      "finely": 44396,
      "filteringbased": 44091,
      "whilst": 132472,
      "intersection": 56304,
      "sender": 105640,
      "observer": 81242,
      "trigrams": 125801,
      "laplace": 61001,
      "goaloriented": 48404,
      "turnbased": 125983,
      "rewards": 102681,
      "selfattentive": 104907,
      "timestep": 122202,
      "targetside": 117795,
      "nonsequential": 80289,
      "harvesting": 49519,
      "stylistically": 114635,
      "figurative": 44013,
      "vastly": 131282,
      "hyperbolic": 51206,
      "delexicalize": 28601,
      "convincingness": 23133,
      "naturalness": 76650,
      "plan": 87628,
      "industries": 54061,
      "actionable": 3083,
      "wikihow": 132636,
      "sector": 104493,
      "skipgrams": 109997,
      "phi": 87261,
      "lends": 63349,
      "viability": 131637,
      "exception": 39923,
      "diacritic": 30494,
      "restoration": 101331,
      "wae": 131982,
      "vulnerability": 131971,
      "noisecontrastive": 80078,
      "initialisation": 55231,
      "singlemodel": 109834,
      "sorts": 110686,
      "interruptions": 56302,
      "wholly": 132479,
      "generalise": 46741,
      "restarts": 101316,
      "ai": 4418,
      "impractical": 52311,
      "generalisation": 46740,
      "rl": 102886,
      "hypothesised": 51284,
      "combinatorially": 18607,
      "74": 1234,
      "mitigating": 70381,
      "dependencyparsed": 29260,
      "typified": 126469,
      "nonetheless": 80198,
      "judging": 57437,
      "flavor": 44639,
      "cuisines": 25225,
      "famous": 43059,
      "prosody": 93649,
      "regularized": 97767,
      "replicability": 98956,
      "evergrowing": 39620,
      "unjustified": 127358,
      "estate": 38626,
      "avoiding": 11246,
      "reality": 96110,
      "stepwise": 113339,
      "weigh": 132318,
      "corporation": 23626,
      "publish": 94340,
      "widecoverage": 132532,
      "unpredictable": 127493,
      "habits": 49292,
      "rolebased": 103225,
      "rolespecific": 103241,
      "behavioral": 12671,
      "messaging": 68522,
      "craft": 24587,
      "android": 6100,
      "male": 66879,
      "crafting": 24592,
      "deficiencies": 28476,
      "optimizer": 82237,
      "fix": 44591,
      "robotic": 103009,
      "scene": 103875,
      "competitor": 19706,
      "635": 1172,
      "495": 1056,
      "compiling": 19721,
      "brazil": 14568,
      "portugal": 88192,
      "computerassisted": 20503,
      "mdt": 67593,
      "rudimentary": 103343,
      "headed": 49584,
      "multiitem": 75148,
      "wordforms": 133714,
      "sourcelanguage": 110875,
      "reimplement": 97783,
      "multiway": 75939,
      "repository": 99094,
      "mmc": 70454,
      "said": 103538,
      "splits": 111950,
      "unavailable": 126596,
      "trustable": 125871,
      "profiles": 92101,
      "triggered": 125789,
      "clauseembedding": 17693,
      "factuality": 42939,
      "outer": 82622,
      "weakened": 132187,
      "animate": 6119,
      "inanimate": 53201,
      "infinitival": 54272,
      "spreads": 112055,
      "microblogs": 70052,
      "weblogs": 132282,
      "instant": 55642,
      "messengers": 68527,
      "whatsapp": 132471,
      "americans": 5313,
      "feel": 43861,
      "brexit": 14589,
      "webscale": 132289,
      "commoncrawl": 18954,
      "365": 969,
      "143": 328,
      "crawl": 24593,
      "counterfactual": 24425,
      "accept": 1885,
      "acceptability": 1887,
      "retrofitted": 102474,
      "initializing": 55245,
      "sentencebased": 106139,
      "correlates": 24213,
      "bpemb": 14546,
      "275": 859,
      "typing": 126470,
      "testbed": 120555,
      "bet": 13493,
      "agency": 4290,
      "peoples": 85907,
      "fueled": 45922,
      "channel": 16404,
      "businesses": 14983,
      "routinely": 103319,
      "reputations": 100101,
      "landscape": 58801,
      "january": 57179,
      "hillary": 50445,
      "171": 404,
      "minimizes": 70197,
      "trivial": 125828,
      "paperbased": 84495,
      "undertaking": 127036,
      "scanning": 103786,
      "optical": 82155,
      "scanned": 103781,
      "federal": 43811,
      "separation": 106882,
      "unveiling": 127774,
      "resorted": 100824,
      "pretrain": 89971,
      "usecases": 128373,
      "099": 117,
      "innovation": 55281,
      "dailydialog": 25518,
      "humanwritten": 51128,
      "englishesperanto": 37360,
      "pas": 85600,
      "struggles": 114131,
      "inspection": 55554,
      "specialising": 111378,
      "emphasise": 36138,
      "isa": 56958,
      "specialisation": 111375,
      "adjusts": 3881,
      "specialised": 111377,
      "ohiostate": 81634,
      "ijcnlp": 51725,
      "gave": 46556,
      "facebooks": 42745,
      "70": 1209,
      "onehot": 81674,
      "personabased": 87108,
      "speakerspecific": 111337,
      "propbank": 92404,
      "conceptualization": 20674,
      "thought": 121909,
      "auto": 10393,
      "medium": 68243,
      "assembling": 9461,
      "arcs": 8864,
      "tac": 117276,
      "unexpected": 127056,
      "onetomany": 81726,
      "pressing": 89955,
      "friend": 45888,
      "wording": 133721,
      "joy": 57415,
      "coffee": 18252,
      "drink": 34143,
      "underrepresented": 126722,
      "assembled": 9459,
      "assembly": 9462,
      "switzerland": 116296,
      "seldom": 104682,
      "60k": 1158,
      "36": 965,
      "neuralsymbolic": 78740,
      "susceptible": 116216,
      "machinelearningbased": 66325,
      "attacks": 9710,
      "dl": 32923,
      "tpr": 122715,
      "sacrifices": 103524,
      "showcase": 108360,
      "httpsgithubcom": 50701,
      "unaligned": 126576,
      "tiny": 122211,
      "compressing": 20302,
      "hash": 49520,
      "gumbelsoftmax": 49277,
      "trick": 125769,
      "skill": 109972,
      "skilled": 109973,
      "reserved": 100732,
      "minority": 70273,
      "pressure": 89959,
      "posed": 88258,
      "resourced": 100892,
      "nontaskoriented": 80303,
      "activations": 3101,
      "multigenre": 75108,
      "sota": 110687,
      "nguyen": 79471,
      "sequencelevel": 107106,
      "iwslt14": 57170,
      "gigaword": 47963,
      "abstractive": 1800,
      "simulating": 109646,
      "anticipating": 6756,
      "complements": 19758,
      "transformers": 124442,
      "executing": 39973,
      "unstated": 127567,
      "offering": 81577,
      "nlpcc": 79793,
      "characterenhanced": 16509,
      "bagofword": 11367,
      "sized": 109956,
      "egregious": 35208,
      "smooth": 110263,
      "crowdgenerated": 25099,
      "nombank": 80138,
      "organic": 82458,
      "chemistry": 16711,
      "syntheses": 116592,
      "inorganic": 55287,
      "analogous": 5413,
      "expertannotated": 41231,
      "subsentential": 114808,
      "runtime": 103477,
      "distantly": 32486,
      "schemas": 103908,
      "usages": 127874,
      "realizations": 96119,
      "blacklist": 14257,
      "idiom": 51649,
      "blacklisted": 14258,
      "mistranslated": 70351,
      "gaps": 46486,
      "volumes": 131943,
      "inclusive": 53410,
      "provision": 94147,
      "centers": 15895,
      "customized": 25460,
      "contextresponse": 22374,
      "bandit": 11411,
      "thompson": 121873,
      "recallk": 96348,
      "interactionbased": 55977,
      "interdomain": 56063,
      "intradomain": 56336,
      "deploy": 29300,
      "launch": 61678,
      "site": 109875,
      "ethical": 38746,
      "offensive": 81539,
      "violations": 131736,
      "safety": 103536,
      "reproducibility": 100086,
      "spur": 112058,
      "safe": 103533,
      "ethically": 38750,
      "differentiable": 31582,
      "forces": 45016,
      "orderpreserving": 82443,
      "comprehensively": 20285,
      "imaging": 51822,
      "beating": 12619,
      "markedly": 67226,
      "remember": 98863,
      "lightweight": 64008,
      "endowing": 36861,
      "blind": 14352,
      "engage": 37012,
      "promoting": 92322,
      "complaining": 19722,
      "diminishing": 31778,
      "azure": 11271,
      "illustrative": 51754,
      "productspecific": 92074,
      "attempted": 9756,
      "replicates": 98963,
      "monte": 74651,
      "carlo": 15532,
      "allocates": 5068,
      "financially": 44262,
      "domaingeneral": 33712,
      "frobenius": 45895,
      "reconciling": 97104,
      "demands": 28634,
      "speechbased": 111852,
      "oversampling": 83354,
      "drilling": 34142,
      "oil": 81637,
      "gas": 46496,
      "wells": 132431,
      "sensors": 105757,
      "equipment": 38237,
      "accident": 1935,
      "mitigation": 70389,
      "symptom": 116330,
      "modals": 70488,
      "utmost": 130585,
      "sellers": 104967,
      "manufacturers": 67097,
      "dan": 25529,
      "qas": 94527,
      "purchasing": 94408,
      "footnotethe": 45007,
      "san": 103621,
      "empowered": 36339,
      "fits": 44585,
      "lacking": 58770,
      "tress": 125758,
      "reform": 97631,
      "reformulate": 97633,
      "aligns": 5025,
      "urdu": 127839,
      "lowquality": 65482,
      "echoing": 34502,
      "entangled": 37688,
      "endtask": 36866,
      "rollouts": 103243,
      "selfplay": 104929,
      "diverging": 32792,
      "hamshahri": 49312,
      "inspirations": 55559,
      "designer": 29633,
      "alternate": 5215,
      "session": 107335,
      "consulting": 21715,
      "detrimental": 30164,
      "alexa": 4668,
      "prize": 90787,
      "worldwide": 135060,
      "coherently": 18331,
      "minutes": 70279,
      "abound": 1709,
      "blends": 14273,
      "shedding": 108158,
      "stereotyping": 113343,
      "belongs": 12724,
      "immune": 51852,
      "stereotypes": 113340,
      "intensely": 55877,
      "headline": 49592,
      "misses": 70317,
      "tokenwise": 122355,
      "infant": 54089,
      "ids": 51660,
      "discriminable": 32188,
      "opposite": 82151,
      "discriminability": 32187,
      "conciseness": 20707,
      "groundtruth": 49127,
      "brazilian": 14569,
      "15000": 360,
      "ran": 95471,
      "elicited": 35292,
      "proportions": 92517,
      "semanticsrelated": 105483,
      "workplace": 134915,
      "recipients": 96794,
      "recipient": 96793,
      "actionbased": 3086,
      "intents": 55920,
      "rainbow": 95455,
      "recurrently": 97275,
      "irc": 56934,
      "citizens": 16999,
      "nar": 76155,
      "rose": 103277,
      "females": 43874,
      "males": 66882,
      "lion": 64685,
      "republic": 100094,
      "researcher": 100673,
      "publically": 94277,
      "sanskrit": 103631,
      "sandhi": 103625,
      "broken": 14709,
      "795": 1274,
      "titles": 122223,
      "browse": 14726,
      "extractive": 42545,
      "poetry": 87789,
      "neurocognitive": 78741,
      "poetics": 87788,
      "3000": 910,
      "250": 821,
      "nonfiction": 80207,
      "130": 292,
      "dickens": 30851,
      "cleaned": 17706,
      "poetic": 87787,
      "joyce": 57417,
      "exemplary": 39980,
      "george": 47875,
      "eliots": 35303,
      "loved": 65342,
      "joyces": 57418,
      "chamber": 16349,
      "boltzmann": 14406,
      "neurally": 78736,
      "occurs": 81520,
      "orientation": 82482,
      "emergent": 36060,
      "gca": 46564,
      "mixedeffects": 70408,
      "intra": 56334,
      "attain": 9719,
      "pbsmt": 85822,
      "timesteps": 122203,
      "positionaware": 88305,
      "snippet": 110307,
      "elementwise": 35288,
      "mathematically": 67464,
      "pca": 85824,
      "experiences": 40446,
      "hindering": 50452,
      "319": 926,
      "omissions": 81651,
      "biographical": 14155,
      "strengthen": 113568,
      "justified": 57479,
      "interventions": 56325,
      "sourcing": 110958,
      "proxy": 94155,
      "coarser": 18058,
      "edition": 34556,
      "sheffield": 108165,
      "centrality": 15907,
      "predominant": 89227,
      "corpusspecific": 24087,
      "scopes": 104042,
      "colloquially": 18524,
      "bot": 14480,
      "reliant": 98633,
      "warmstarting": 132025,
      "transmit": 125530,
      "wordvectors": 134346,
      "rigid": 102845,
      "aggressiveness": 4377,
      "recovered": 97142,
      "dangerous": 25532,
      "extremes": 42613,
      "recovers": 97144,
      "inherit": 55194,
      "manipulated": 66927,
      "marketplace": 67247,
      "interactively": 56039,
      "buyer": 14985,
      "listings": 64712,
      "bytelevel": 15000,
      "autoencoding": 10420,
      "160": 381,
      "parameterized": 84733,
      "upsampling": 127827,
      "chen": 16712,
      "esim": 38423,
      "signatures": 108710,
      "grapheme": 48917,
      "decoders": 28063,
      "signature": 108709,
      "graphemebased": 48920,
      "categorizing": 15775,
      "replying": 98971,
      "centralized": 15909,
      "reused": 102489,
      "096": 113,
      "097": 114,
      "0778": 68,
      "mlp": 70449,
      "554": 1120,
      "bilm": 14088,
      "exposing": 41689,
      "semisupervision": 105635,
      "struggle": 114123,
      "oie": 81635,
      "calculating": 15026,
      "pearson": 85835,
      "instancebased": 55613,
      "crossdataset": 24877,
      "hashing": 49522,
      "infusing": 55181,
      "textitwithout": 121441,
      "softattention": 110500,
      "viewpoint": 131720,
      "scholarly": 103959,
      "eventspecific": 39613,
      "celebrity": 15883,
      "accelerates": 1873,
      "645": 1178,
      "cloze": 17892,
      "covariates": 24478,
      "demographic": 28646,
      "covariate": 24476,
      "differential": 31585,
      "rdf": 95882,
      "verbalizes": 131541,
      "questionnaire": 95265,
      "nonexperts": 80202,
      "presupposition": 89967,
      "fscores": 45916,
      "complexities": 19898,
      "gauge": 46539,
      "arab": 8485,
      "health": 49615,
      "ordered": 82432,
      "factorize": 42875,
      "reordered": 98896,
      "invention": 56687,
      "penalized": 85847,
      "semeval2018": 105530,
      "threeway": 121945,
      "speer": 111891,
      "msr": 74950,
      "evolved": 39699,
      "trouble": 125842,
      "copying": 23295,
      "opentype": 82040,
      "vaswani": 131286,
      "englishtogerman": 37445,
      "03": 15,
      "relationaware": 98102,
      "advancements": 3968,
      "save": 103663,
      "bengalienglish": 13040,
      "codemixed": 18166,
      "codemixing": 18192,
      "transliterations": 125528,
      "phone": 87272,
      "exhibited": 40007,
      "isnotes": 56966,
      "hou": 50675,
      "antecedent": 6751,
      "moderation": 74378,
      "marginalize": 67213,
      "intuitions": 56666,
      "multicharacter": 75045,
      "preprocess": 89317,
      "fairer": 42993,
      "cumbersome": 25235,
      "parity": 84869,
      "barriers": 11432,
      "microsofts": 70062,
      "wsi": 135175,
      "germanic": 47942,
      "accepts": 1899,
      "slope": 110030,
      "sampled": 103570,
      "submitting": 114781,
      "383": 982,
      "distributionally": 32750,
      "framester": 45402,
      "manuallyconstructed": 67092,
      "semanticaware": 105389,
      "singlesentence": 109838,
      "comprehensible": 20161,
      "esperanto": 38527,
      "pytorch": 94491,
      "batching": 12562,
      "extensible": 41832,
      "rajpurkar": 95465,
      "maintained": 66505,
      "convinced": 23130,
      "enhancer": 37528,
      "borne": 14472,
      "aspectlevel": 9365,
      "binarization": 14115,
      "calculations": 15037,
      "unsuitable": 127591,
      "128": 274,
      "256": 830,
      "topk": 122669,
      "differentiating": 31591,
      "aiding": 4438,
      "credibility": 24738,
      "circulation": 16982,
      "discerning": 32014,
      "credible": 24739,
      "metalearning": 68543,
      "casting": 15680,
      "metalearner": 68541,
      "matter": 67486,
      "pomme": 88035,
      "crosslingually": 25052,
      "crowdsource": 25101,
      "featureenriched": 43343,
      "marian": 67221,
      "selfcontained": 104915,
      "differentiation": 31592,
      "societies": 110469,
      "postagging": 88458,
      "twin": 126087,
      "joined": 57249,
      "yoruba": 135456,
      "west": 132453,
      "tonal": 122361,
      "diacritics": 30497,
      "adr": 3924,
      "undiacritized": 127044,
      "diacritization": 30498,
      "sourcecode": 110871,
      "searchable": 104365,
      "lowresourced": 65573,
      "fewshot": 43896,
      "loop": 65228,
      "indiscriminately": 53890,
      "humanintheloop": 51035,
      "owner": 83393,
      "ontosensenet": 81859,
      "senseannotated": 105707,
      "verbcentric": 131544,
      "8483": 1320,
      "253": 827,
      "typelogical": 126240,
      "meantime": 67773,
      "decompositional": 28159,
      "plug": 87760,
      "popularly": 88159,
      "recalling": 96347,
      "repetitiveness": 98920,
      "todays": 122237,
      "negativity": 76966,
      "vader": 130697,
      "came": 15081,
      "ended": 36850,
      "synchronized": 116337,
      "till": 121972,
      "shortest": 108299,
      "hosting": 50664,
      "floating": 44674,
      "nonnegligible": 80252,
      "frustratingly": 45907,
      "lately": 61562,
      "performant": 86906,
      "counterintuitive": 24429,
      "sketches": 109967,
      "concordance": 20734,
      "userfriendly": 129068,
      "reusing": 102492,
      "displayed": 32379,
      "assessed": 9498,
      "helped": 49791,
      "silence": 109061,
      "negatives": 76965,
      "humor": 51137,
      "advertisements": 4196,
      "cartoon": 15565,
      "parodies": 84875,
      "proverbs": 93741,
      "songs": 110661,
      "humorous": 51143,
      "wit": 132745,
      "reidentification": 97782,
      "interdependence": 56054,
      "multivariate": 75931,
      "eeg": 34580,
      "lastly": 61556,
      "heart": 49645,
      "movement": 74885,
      "personcentered": 87152,
      "postevaluation": 88493,
      "2018": 650,
      "tions": 122215,
      "7511": 1255,
      "neologisms": 77010,
      "loan": 64897,
      "russianspeaking": 103512,
      "lemmatized": 63344,
      "opencorpora": 81955,
      "entered": 37691,
      "168": 391,
      "unsurprisingly": 127761,
      "multimedia": 75412,
      "marginals": 67218,
      "killed": 57668,
      "police": 87938,
      "cotraining": 24408,
      "fails": 42980,
      "clickbait": 17748,
      "instructor": 55713,
      "heshe": 49847,
      "deliver": 28617,
      "managers": 66907,
      "tasked": 118875,
      "agreements": 4411,
      "bloom": 14376,
      "specialpurpose": 111400,
      "multiinstance": 75144,
      "dementia": 28636,
      "irreversible": 56956,
      "progressive": 92190,
      "saliency": 103544,
      "consolidate": 21504,
      "derivative": 29334,
      "highway": 50439,
      "entailments": 37684,
      "unbounded": 126609,
      "domainagnostic": 33707,
      "ntuaslp": 80818,
      "ironic": 56936,
      "competed": 19606,
      "irony": 56937,
      "550": 1115,
      "visualizations": 131833,
      "workings": 134908,
      "2nd": 886,
      "enhancements": 37527,
      "echo": 34500,
      "gapping": 46485,
      "reconstructing": 97113,
      "paul": 85799,
      "likes": 64148,
      "tea": 119689,
      "overt": 83358,
      "obviates": 81484,
      "summarizer": 115618,
      "cnndaily": 18032,
      "doomed": 33948,
      "disentangled": 32346,
      "affording": 4263,
      "dissimilarity": 32401,
      "triplets": 125825,
      "motivating": 74873,
      "disentangles": 32352,
      "populations": 88171,
      "trial": 125763,
      "clinically": 17797,
      "aspectspecific": 9420,
      "multiaspect": 75033,
      "stance": 112166,
      "agrees": 4412,
      "disagrees": 31938,
      "fake": 43011,
      "2017s": 649,
      "cqa": 24580,
      "pathologies": 85683,
      "exposes": 41688,
      "pathological": 85681,
      "finetune": 44401,
      "largesize": 61522,
      "largesized": 61523,
      "workloads": 134914,
      "layerwise": 61815,
      "corrupt": 24325,
      "stances": 112178,
      "rationales": 95841,
      "rationale": 95839,
      "interdependencies": 56055,
      "intrasentential": 56343,
      "118": 242,
      "everincreasing": 39622,
      "informing": 55170,
      "locating": 64960,
      "ipdas": 56928,
      "shortlisting": 108309,
      "1500": 359,
      "ipda": 56927,
      "perturbation": 87193,
      "perturb": 87192,
      "bernoulli": 13048,
      "closedvocabulary": 17847,
      "invoking": 56874,
      "multihop": 75135,
      "kong": 58341,
      "disagreements": 31936,
      "sacrificing": 103525,
      "distill": 32498,
      "warrants": 132037,
      "warrant": 132035,
      "doubles": 33968,
      "semanticrole": 105393,
      "newlyproposed": 79280,
      "evaluators": 39493,
      "remarkably": 98848,
      "aspectbased": 9352,
      "genrebalanced": 47834,
      "nps": 80802,
      "vps": 131959,
      "clarity": 17028,
      "underrecognized": 126721,
      "inconsistency": 53437,
      "settle": 107848,
      "usersupplied": 129189,
      "sacrebleu": 103521,
      "eliminating": 35300,
      "scheduled": 103883,
      "begins": 12639,
      "worldknowledge": 135057,
      "tuple": 125947,
      "unify": 127147,
      "entityrelation": 38146,
      "realigning": 96096,
      "terabytes": 120203,
      "webcrawled": 132278,
      "gomezrodriguez": 48456,
      "on6": 81664,
      "inspires": 55586,
      "chooses": 16929,
      "righttoleft": 102843,
      "sampler": 103577,
      "hmms": 50583,
      "yellow": 135317,
      "l1s": 58372,
      "epoch": 38216,
      "epochs": 38217,
      "1030": 188,
      "sdp": 104279,
      "delayed": 28589,
      "sgd": 107902,
      "linearized": 64399,
      "administration": 3884,
      "slaves": 110006,
      "peculiar": 85839,
      "standpoint": 112357,
      "racial": 95426,
      "wanted": 132018,
      "hear": 49641,
      "ultimately": 126558,
      "spoke": 111961,
      "slavery": 110005,
      "hotels": 50671,
      "tourist": 122707,
      "price": 90633,
      "webpages": 132286,
      "diagnosed": 30500,
      "probing": 90887,
      "uncovering": 126650,
      "targetoriented": 117786,
      "obstacles": 81250,
      "originated": 82576,
      "openmt": 82002,
      "5th": 1144,
      "lexicalizing": 63854,
      "monologue": 74636,
      "parties": 85504,
      "partys": 85599,
      "leftright": 63304,
      "posthoc": 88496,
      "commenting": 18805,
      "referencebased": 97545,
      "discourseaware": 32104,
      "repetitive": 98917,
      "crossentropy": 24911,
      "flat": 44635,
      "modularized": 74471,
      "inequality": 54079,
      "interpreter": 56283,
      "demanding": 28633,
      "ukb": 126548,
      "inadvertently": 53200,
      "outofthebox": 82670,
      "pitfalls": 87576,
      "resurgence": 102355,
      "chatbased": 16661,
      "markets": 67249,
      "inhibit": 55200,
      "traffic": 122894,
      "sharp": 108144,
      "nearby": 76717,
      "ablation": 1581,
      "wikitext2": 132699,
      "200": 532,
      "sharply": 108146,
      "rough": 103308,
      "grave": 48989,
      "2017b": 648,
      "cachebased": 15009,
      "replicable": 98957,
      "punished": 94398,
      "appeared": 6808,
      "455": 1038,
      "wild": 132702,
      "gcdc": 46565,
      "unpaired": 127484,
      "164": 386,
      "winning": 132721,
      "27k": 863,
      "prompts": 92328,
      "mediocre": 68241,
      "s4": 103517,
      "substitutions": 114931,
      "commitment": 18846,
      "organizational": 82469,
      "propositions": 93633,
      "discriminators": 32227,
      "autoregressive": 10857,
      "selfcontradictory": 104916,
      "grices": 49086,
      "maxims": 67527,
      "enhances": 37529,
      "navigational": 76685,
      "dsms": 34205,
      "068": 52,
      "diminish": 31776,
      "dis": 31927,
      "dozens": 34071,
      "errorfree": 38352,
      "943": 1416,
      "926": 1401,
      "extrapolation": 42588,
      "maximise": 67503,
      "decomposable": 28149,
      "comprehending": 20160,
      "propara": 92402,
      "datapoints": 26695,
      "spanbased": 111140,
      "2017a": 647,
      "cubic": 25201,
      "stack": 112116,
      "cube": 25200,
      "japanesevietnamese": 57216,
      "rareword": 95768,
      "persists": 87094,
      "wmt2016": 132870,
      "overlaps": 83340,
      "webnlg": 132283,
      "semanticsaware": 105482,
      "personalitybased": 87136,
      "disentangle": 32342,
      "personage": 87111,
      "felix": 43867,
      "rotten": 103285,
      "tomatoes": 122358,
      "assets": 9537,
      "implicated": 51987,
      "narrowly": 76198,
      "locally": 64950,
      "computable": 20342,
      "highperforming": 50361,
      "relief": 98640,
      "machinetranslated": 66347,
      "voicecontrolled": 131927,
      "couples": 24461,
      "therapy": 121832,
      "dyadic": 34287,
      "looked": 65222,
      "perception": 85927,
      "partners": 85519,
      "singlespeaker": 109840,
      "reuters": 102493,
      "complicates": 19959,
      "adam": 3166,
      "mickiewicz": 70032,
      "tilde": 121971,
      "unfaithful": 127065,
      "worthy": 135076,
      "prefix": 89253,
      "highfrequency": 50240,
      "filled": 44025,
      "091": 102,
      "032": 18,
      "aspec": 9313,
      "preordering": 89293,
      "succinctly": 115205,
      "discern": 32010,
      "termed": 120248,
      "dream": 34140,
      "interfere": 56111,
      "148": 334,
      "searchbased": 104366,
      "132": 297,
      "literatures": 64783,
      "approximative": 8472,
      "conll2003": 21095,
      "entrainment": 38153,
      "follower": 44956,
      "semanticallyinformed": 105385,
      "rewording": 102688,
      "players": 87714,
      "pragmatically": 88759,
      "2019": 692,
      "submit": 114762,
      "anderson": 6094,
      "unspoken": 127565,
      "deploys": 29319,
      "bounded": 14525,
      "acknowledged": 2991,
      "questionable": 95242,
      "office": 81594,
      "bug": 14738,
      "realm": 96134,
      "proprietary": 93635,
      "conjecture": 21063,
      "unnatural": 127475,
      "attenuating": 10132,
      "repetition": 98914,
      "conclusively": 20732,
      "emnlp": 36083,
      "dates": 27834,
      "accented": 1881,
      "accents": 1882,
      "kaldi": 57487,
      "tdnn": 119681,
      "745": 1242,
      "concert": 20694,
      "acl": 2993,
      "workshops": 135017,
      "retain": 102357,
      "languagepair": 60376,
      "saw": 103670,
      "prospect": 93654,
      "fascinating": 43113,
      "emoji": 36087,
      "syncretism": 116340,
      "probabilistically": 90838,
      "disambiguates": 31946,
      "smoothly": 110270,
      "languagemodel": 60372,
      "differing": 31600,
      "birnn": 14228,
      "augments": 10351,
      "flowing": 44686,
      "replication": 98967,
      "generalisability": 46738,
      "repeatability": 98908,
      "continuing": 22608,
      "comparability": 19124,
      "jupyter": 57458,
      "anonymised": 6509,
      "bags": 11382,
      "systemic": 116703,
      "legislative": 63321,
      "legislation": 63320,
      "governments": 48545,
      "bills": 14086,
      "politics": 88008,
      "cursive": 25428,
      "joining": 57254,
      "nonjoiner": 80217,
      "zwnj": 135556,
      "085": 87,
      "coldstart": 18350,
      "userproduct": 129083,
      "frequencyguided": 45853,
      "rmse": 102897,
      "interchange": 56050,
      "nonhumorous": 80214,
      "ncrf": 76704,
      "acceleration": 1879,
      "lstmcrf": 65691,
      "databased": 26648,
      "loudness": 65339,
      "chunkbased": 16954,
      "asking": 9305,
      "chunklevel": 16967,
      "261": 846,
      "justifications": 57478,
      "82": 1303,
      "desktop": 29675,
      "densely": 29024,
      "positionindependent": 88308,
      "sequentially": 107261,
      "concatenates": 20535,
      "selfmatching": 104925,
      "route": 103316,
      "dnnbased": 32934,
      "requested": 100104,
      "nongoal": 80208,
      "partitioned": 85511,
      "unadapted": 126574,
      "geography": 47857,
      "affinity": 4255,
      "culturally": 25232,
      "crosscultural": 24874,
      "turntaking": 125998,
      "fluid": 44725,
      "endpoints": 36863,
      "pauses": 85802,
      "ood": 81861,
      "detectors": 30100,
      "ind": 53755,
      "thresholding": 121947,
      "tracker": 122743,
      "dst": 34206,
      "accumulates": 2053,
      "fisher": 44574,
      "02": 9,
      "cleaner": 17712,
      "completes": 19788,
      "notebook": 80388,
      "schemaorg": 103907,
      "backend": 11296,
      "involvement": 56887,
      "slots": 110083,
      "dozat": 34067,
      "manning": 66966,
      "pushes": 94464,
      "introductory": 56658,
      "authentic": 10354,
      "declined": 27982,
      "cousins": 24473,
      "labour": 58670,
      "encapsulate": 36417,
      "afford": 4259,
      "parallelization": 84703,
      "srnns": 112090,
      "subsequences": 114810,
      "irregularity": 56943,
      "puts": 94469,
      "interactivepredictive": 56040,
      "protocols": 93677,
      "keras": 57530,
      "tensorflow": 120196,
      "modularity": 74469,
      "novelly": 80784,
      "prototyping": 93692,
      "distribute": 32608,
      "511": 1093,
      "382": 981,
      "domainadversarial": 33705,
      "regularizer": 97768,
      "zeroresource": 135483,
      "prealignment": 88769,
      "conceptbased": 20593,
      "conceptlevel": 20595,
      "fused": 46226,
      "eec": 34579,
      "manifestations": 66920,
      "804": 1294,
      "101": 185,
      "exchanged": 39946,
      "smallfootprint": 110248,
      "loading": 64896,
      "cloudbased": 17890,
      "su": 114642,
      "wordtovector": 134339,
      "sequencebased": 107103,
      "preprocessed": 89319,
      "translationnmt": 125448,
      "abovementioned": 1711,
      "ref": 97491,
      "failures": 42990,
      "simulations": 109650,
      "spontaneously": 112036,
      "scripted": 104263,
      "navigate": 76681,
      "20172018": 646,
      "densenet": 29027,
      "metricbased": 69915,
      "amortised": 5323,
      "gradients": 48612,
      "posteriori": 88489,
      "grikoitalian": 49091,
      "griko": 49090,
      "330": 944,
      "topicspecific": 122668,
      "propelled": 92409,
      "explosion": 41669,
      "plethora": 87751,
      "assortment": 9640,
      "limitedsize": 64305,
      "172": 405,
      "department": 29036,
      "uh": 126544,
      "universitat": 127345,
      "babelnet": 11276,
      "textbook": 121430,
      "breaks": 14581,
      "genderneutral": 46624,
      "wed": 132311,
      "mitigates": 70379,
      "exclusively": 39958,
      "549": 1108,
      "enes": 36999,
      "lowlatency": 65471,
      "losing": 65240,
      "corrections": 24169,
      "traditions": 122893,
      "rendered": 98887,
      "derives": 29369,
      "generaldomain": 46736,
      "pinyin": 87517,
      "mccnn": 67579,
      "retrained": 102370,
      "nation": 76209,
      "lexicalbased": 63844,
      "statisticalbased": 113178,
      "942": 1415,
      "corrupting": 24329,
      "amplified": 5373,
      "l2r": 58375,
      "kullbackleibler": 58359,
      "chats": 16684,
      "disregarded": 32385,
      "apt": 8480,
      "protest": 93670,
      "suites": 115422,
      "semimarkov": 105579,
      "tokenizer": 122283,
      "nahuatl": 76001,
      "speechtotext": 111857,
      "discretization": 32185,
      "manipulating": 66928,
      "primal": 90635,
      "deduced": 28185,
      "p1": 83399,
      "833": 1311,
      "cent": 15888,
      "905": 1374,
      "763": 1260,
      "enforced": 37004,
      "wellstudied": 132433,
      "scarcely": 103800,
      "shi": 108170,
      "sememe": 105486,
      "sememes": 105491,
      "fortunately": 45300,
      "hownet": 50688,
      "baidu": 11392,
      "beats": 12621,
      "amateur": 5261,
      "pursue": 94450,
      "ms": 74939,
      "coco": 18065,
      "userdefined": 129063,
      "fourteen": 45356,
      "node2vec": 80030,
      "unexpectedly": 127057,
      "visible": 131757,
      "lens": 63390,
      "nonnative": 80240,
      "jfleg": 57227,
      "normalizes": 80361,
      "localglobal": 64944,
      "downsampling": 33984,
      "mere": 68473,
      "parliament": 84871,
      "iran": 56932,
      "agencies": 4289,
      "iranian": 56933,
      "imbalanced": 51827,
      "sexist": 107892,
      "debiased": 27897,
      "reversing": 102531,
      "reconcile": 97101,
      "sennrich": 105643,
      "granular": 48740,
      "iteration": 57119,
      "unifying": 127150,
      "imposing": 52307,
      "alternates": 5217,
      "infusion": 55182,
      "representable": 99157,
      "inevitably": 54081,
      "curate": 25241,
      "wn18rr": 132875,
      "utterancelevel": 130620,
      "multireference": 75767,
      "pyramidal": 94477,
      "generalizability": 46749,
      "grouped": 49146,
      "1520": 364,
      "registers": 97686,
      "motivational": 74879,
      "categorise": 15759,
      "openclass": 81953,
      "maxent": 67498,
      "reranker": 100363,
      "acnn": 2997,
      "click": 17747,
      "hyperlink": 51212,
      "50000": 1084,
      "seqtoseq": 106910,
      "awareness": 11262,
      "enjoyed": 37543,
      "paucity": 85797,
      "humanlevel": 51049,
      "exhibiting": 40009,
      "recurrences": 97156,
      "domaininvariant": 33715,
      "inserted": 55509,
      "wmt18": 132843,
      "selfattentionbased": 104905,
      "rephrase": 98921,
      "rewrites": 102691,
      "ninety": 79489,
      "websplit": 132309,
      "narayan": 76160,
      "originates": 82577,
      "inaccurate": 53191,
      "older": 81645,
      "younger": 135459,
      "college": 18512,
      "economics": 34514,
      "lessresourced": 63401,
      "uyghur": 130683,
      "135": 302,
      "disaster": 32003,
      "emergency": 36059,
      "engaged": 37013,
      "expertcrafted": 41233,
      "proofofconcept": 92374,
      "quantized": 94903,
      "suspect": 116219,
      "softly": 110502,
      "attending": 9783,
      "modelingbased": 72591,
      "117": 241,
      "cider": 16977,
      "decoupling": 28168,
      "negotiation": 76979,
      "bike": 14014,
      "selling": 104968,
      "degenerate": 28545,
      "di": 30485,
      "degeneracy": 28544,
      "explainable": 41264,
      "walking": 131998,
      "emits": 36080,
      "explainability": 41263,
      "portugueseenglish": 88205,
      "trajectories": 123978,
      "fluctuations": 44690,
      "concludes": 20722,
      "outlook": 82640,
      "realistically": 96109,
      "wave": 132048,
      "quantuminspired": 94905,
      "practically": 88727,
      "cnnbased": 18029,
      "stringtostring": 113606,
      "alphabets": 5209,
      "xu": 135224,
      "marginalizing": 67215,
      "transformerbased": 124399,
      "scaffolds": 103679,
      "inputoutput": 55477,
      "structureaware": 113987,
      "freetext": 45794,
      "multiplechoice": 75750,
      "bespoke": 13293,
      "timescales": 122193,
      "mistake": 70345,
      "departs": 29040,
      "corrupted": 24326,
      "confounds": 21045,
      "dynamicprogramming": 34350,
      "tractable": 122769,
      "justifies": 57480,
      "threepart": 121941,
      "onedimensional": 81672,
      "cmu": 17976,
      "wmt2018": 132872,
      "penalize": 85846,
      "paracrawl": 84529,
      "investigative": 56868,
      "processor": 91861,
      "journalists": 57412,
      "originate": 82575,
      "disclosures": 32021,
      "freedom": 45774,
      "crossborder": 24870,
      "collaborations": 18355,
      "fulltext": 45936,
      "preferable": 89236,
      "brands": 14566,
      "questionnaires": 95266,
      "consumed": 21718,
      "gapfilling": 46484,
      "hint": 50503,
      "untranslated": 127767,
      "multispeaker": 75795,
      "initiate": 55250,
      "v7": 130693,
      "opensubtitles2016": 82038,
      "languagepairs": 60377,
      "mistranslations": 70352,
      "badly": 11357,
      "manipulate": 66926,
      "proxies": 94151,
      "flipping": 44673,
      "perturbations": 87194,
      "770": 1265,
      "explanation": 41281,
      "psychological": 94210,
      "prevailing": 90355,
      "climate": 17757,
      "0853": 89,
      "arora": 9018,
      "equivalence": 38245,
      "charcnn": 16636,
      "ing": 55183,
      "interpolate": 56201,
      "mcdonald": 67580,
      "timedecay": 122151,
      "timeaware": 122141,
      "decaying": 27914,
      "dstc4": 34215,
      "accelerated": 1872,
      "proactive": 90794,
      "personas": 87151,
      "700": 1211,
      "multitreebank": 75907,
      "uppsala": 127825,
      "partof": 85521,
      "mlas": 70440,
      "vae": 130698,
      "poverty": 88627,
      "disparity": 32373,
      "obscure": 81158,
      "vivid": 131852,
      "imagebased": 51803,
      "artifacts": 9240,
      "thai": 121766,
      "parseme": 84901,
      "vmwes": 131861,
      "optional": 82248,
      "johnson": 57244,
      "chartbased": 16648,
      "constituencybased": 21533,
      "forcing": 45018,
      "improperly": 52327,
      "unstable": 127566,
      "lemmas": 63331,
      "udpipe": 126537,
      "rc": 95868,
      "softalignment": 110499,
      "maximization": 67504,
      "multihead": 75121,
      "sharper": 108145,
      "dictionaryguided": 30908,
      "alignmentbased": 5006,
      "rationality": 95843,
      "awe": 11266,
      "directional": 31832,
      "premisehypothesis": 89288,
      "asymmetry": 9673,
      "scitail": 104033,
      "noteworthy": 80394,
      "scan": 103779,
      "querybased": 94997,
      "bioasq": 14147,
      "regressionbased": 97718,
      "50k": 1088,
      "consequentially": 21177,
      "dull": 34256,
      "impeding": 51918,
      "overwhelmed": 83384,
      "skeleton": 109962,
      "readmission": 96041,
      "discharge": 32015,
      "hospital": 50658,
      "psychiatric": 94200,
      "ehrs": 35218,
      "ehr": 35216,
      "sliding": 110012,
      "depart": 29034,
      "senseaware": 105710,
      "propagates": 92386,
      "hearst": 49643,
      "excel": 39909,
      "represen": 99096,
      "opensubtitles": 82035,
      "nicts": 79477,
      "consulted": 21714,
      "neuron": 78743,
      "neurons": 78746,
      "gumbel": 49275,
      "imaginet": 51821,
      "copied": 23278,
      "suitably": 115411,
      "conversing": 23084,
      "modifying": 74459,
      "90k": 1380,
      "crossview": 25089,
      "elmo": 35309,
      "teaches": 119708,
      "recognizable": 97048,
      "mos": 74822,
      "expressiveness": 41775,
      "unleash": 127420,
      "relieve": 98665,
      "germantoenglish": 47944,
      "hate": 49528,
      "racist": 95430,
      "multiwoz": 75951,
      "wizardofoz": 132762,
      "breakthrough": 14583,
      "blocked": 14362,
      "10k": 208,
      "opensourced": 82033,
      "denotations": 29005,
      "parameterfree": 84730,
      "customercare": 25449,
      "anew": 6104,
      "62": 1164,
      "winograd": 132726,
      "wsc": 135167,
      "send": 105639,
      "weighs": 132320,
      "021": 11,
      "copa": 23268,
      "nested": 77102,
      "designated": 29574,
      "stacklstm": 112136,
      "rhythm": 102720,
      "goodness": 48514,
      "exceeding": 39904,
      "essence": 38537,
      "basing": 12541,
      "cut": 25462,
      "undermine": 126710,
      "pipelinebased": 87560,
      "newer": 79261,
      "canonicalized": 15160,
      "accepting": 1898,
      "keyvalue": 57629,
      "self": 104868,
      "disfluent": 32359,
      "swda": 116257,
      "declension": 27979,
      "inflect": 54278,
      "manuscript": 67098,
      "manuscripts": 67099,
      "military": 70090,
      "secret": 104483,
      "cipher": 16980,
      "plaintext": 87627,
      "agglutination": 4335,
      "guideline": 49257,
      "wang": 132006,
      "reconstructionbased": 97124,
      "dp": 34072,
      "holy": 50608,
      "grail": 48618,
      "turkishenglish": 125966,
      "marrying": 67277,
      "unimorph": 127164,
      "schemata": 103909,
      "compromise": 20337,
      "curve": 25429,
      "freeform": 45775,
      "humanchatbot": 51003,
      "pseudoparallel": 94189,
      "negatively": 76961,
      "anticipation": 6757,
      "controllable": 22824,
      "finished": 44530,
      "fullsentence": 45933,
      "waitk": 131990,
      "concurrently": 20747,
      "zhen": 135542,
      "deen": 28192,
      "nus": 81032,
      "kan": 57488,
      "semicrf": 105578,
      "para": 84528,
      "7000": 1212,
      "marginalized": 67214,
      "speeding": 111882,
      "phraselevel": 87401,
      "armenian": 9016,
      "richresource": 102816,
      "wordpair": 133789,
      "paraphrastic": 84853,
      "bidaf": 13896,
      "multinli": 75467,
      "bcws": 12593,
      "separators": 106883,
      "multigraph": 75120,
      "idiosyncrasy": 51657,
      "regularity": 97742,
      "nonredundant": 80285,
      "mweaware": 75994,
      "mweannotated": 75993,
      "192": 440,
      "severely": 107883,
      "pays": 85814,
      "triple": 125808,
      "relax": 98421,
      "1525": 365,
      "collects": 18510,
      "browsing": 14729,
      "journalistic": 57411,
      "distraction": 32597,
      "fever": 43881,
      "disentanglement": 32351,
      "disentangling": 32353,
      "manuallyannotated": 67087,
      "mmd": 70455,
      "utilitarian": 130473,
      "happiness": 49442,
      "favorite": 43222,
      "respond": 101180,
      "interacted": 55936,
      "150": 357,
      "sessions": 107336,
      "client": 17754,
      "sophistication": 110677,
      "englishvietnamese": 37461,
      "supportiveness": 116034,
      "convolutions": 23219,
      "assistive": 9588,
      "minorities": 70272,
      "unskilled": 127560,
      "guaranteeing": 49218,
      "multiheaded": 75132,
      "attend": 9777,
      "43k": 1026,
      "5k": 1142,
      "weather": 132215,
      "alarm": 4659,
      "reminder": 98865,
      "disasters": 32005,
      "confusing": 21051,
      "assemble": 9458,
      "mixtures": 70428,
      "concepttotext": 20654,
      "greedily": 49065,
      "deterioration": 30110,
      "countbased": 24417,
      "clwes": 17972,
      "talking": 117560,
      "listening": 64708,
      "trait": 123969,
      "stuck": 114135,
      "unfamiliar": 127067,
      "consult": 21712,
      "undertake": 127033,
      "urban": 127838,
      "cohesive": 18334,
      "longform": 65169,
      "humangenerated": 51021,
      "critic": 24801,
      "witnessed": 132750,
      "undoubtedly": 127048,
      "deployments": 29318,
      "pronounced": 92346,
      "photo": 87323,
      "hu": 50706,
      "guo": 49281,
      "215": 766,
      "restricts": 101350,
      "gloss": 48293,
      "dataefficient": 26683,
      "userprovided": 129086,
      "optimise": 82181,
      "precision1": 88809,
      "interpolating": 56203,
      "executions": 39975,
      "scone": 104035,
      "dramatically": 34089,
      "anglocentric": 6115,
      "ported": 88179,
      "demonstrations": 28990,
      "memorized": 68281,
      "knowledgeable": 58242,
      "iarpa": 51299,
      "babel": 11275,
      "cis": 16988,
      "decay": 27910,
      "sentencepairs": 106186,
      "fda": 43230,
      "smallvocabulary": 110253,
      "975": 1443,
      "688": 1195,
      "884": 1347,
      "sustain": 116222,
      "manners": 66965,
      "irrespective": 56955,
      "harm": 49501,
      "headfinal": 49587,
      "indispensable": 53891,
      "decides": 27929,
      "attract": 10138,
      "concisely": 20706,
      "screen": 104245,
      "phones": 87285,
      "innovatively": 55286,
      "unsolved": 127562,
      "personalization": 87137,
      "rumor": 103456,
      "misinformation": 70296,
      "fabrics": 42727,
      "enormous": 37553,
      "targetbased": 117756,
      "intertask": 56317,
      "calibrate": 15042,
      "regularize": 97763,
      "poesio": 87781,
      "2013b": 572,
      "preposition": 89307,
      "mln": 70448,
      "impartial": 51913,
      "phrased": 87400,
      "inflammatory": 54275,
      "onesided": 81716,
      "phrasing": 87472,
      "intensifiers": 55879,
      "elsevier": 35329,
      "rights": 102842,
      "morphologicallyrich": 74775,
      "holistic": 50599,
      "enumerates": 38180,
      "vice": 131647,
      "versa": 131598,
      "communityspecific": 19105,
      "unnamed": 127474,
      "cornerstone": 23392,
      "locates": 64959,
      "validates": 130735,
      "stateofthearts": 113034,
      "cuewords": 25221,
      "postings": 88502,
      "9000": 1372,
      "handpicked": 49424,
      "039": 22,
      "026": 13,
      "029": 14,
      "023": 12,
      "059": 41,
      "physicians": 87480,
      "pseudolabels": 94188,
      "diversitypromoting": 32901,
      "diversities": 32864,
      "mle": 70442,
      "complicate": 19952,
      "756": 1257,
      "bleu1": 14347,
      "exemplified": 39981,
      "gnns": 48324,
      "revolution": 102659,
      "adaptations": 3279,
      "lrl": 65580,
      "bhojpuri": 13778,
      "synchronic": 116334,
      "63": 1167,
      "resourceconstrained": 100890,
      "embeds": 36042,
      "understudy": 127031,
      "implanting": 51925,
      "unambiguous": 126578,
      "unacceptable": 126572,
      "converging": 22932,
      "lava": 61680,
      "carryover": 15563,
      "delexicalization": 28600,
      "reranks": 100381,
      "dstc7": 34218,
      "ahre": 4417,
      "wat": 132041,
      "indic": 53827,
      "languagerelated": 60378,
      "6000": 1151,
      "alter": 5211,
      "companions": 19120,
      "milliseconds": 70129,
      "hour": 50679,
      "narrowed": 76194,
      "nonconversational": 80186,
      "repeat": 98907,
      "absorbing": 1756,
      "forgetting": 45058,
      "coattention": 18063,
      "parikh": 84861,
      "meaningfully": 67735,
      "spurred": 112065,
      "stay": 113218,
      "sentencewise": 106552,
      "maximumlikelihood": 67553,
      "documentary": 33117,
      "beginnings": 12638,
      "massively": 67340,
      "fronts": 45902,
      "app": 6783,
      "googles": 48529,
      "invisible": 56872,
      "illegible": 51729,
      "searches": 104369,
      "sql": 112068,
      "priorities": 90771,
      "localized": 64947,
      "hypothesized": 51290,
      "zones": 135552,
      "consequent": 21175,
      "preconditions": 88814,
      "met": 68528,
      "sigmoid": 108668,
      "selfdistillation": 104918,
      "sans": 103630,
      "recurrence": 97153,
      "compile": 19711,
      "additively": 3645,
      "structuresensitive": 114120,
      "naturallyoccurring": 76649,
      "seamless": 104287,
      "migrating": 70076,
      "iterate": 57114,
      "ship": 108194,
      "complaints": 19725,
      "densities": 29031,
      "grain": 48619,
      "crossmodel": 25065,
      "wrt": 135166,
      "ablating": 1580,
      "reevaluating": 97487,
      "nlm": 79555,
      "dalvi": 25522,
      "ablate": 1579,
      "dstc6": 34217,
      "humancreated": 51013,
      "298": 872,
      "computergenerated": 20508,
      "679": 1191,
      "bist": 14231,
      "kiperwasser": 57699,
      "bert": 13050,
      "noncontextual": 80178,
      "centred": 15915,
      "star": 112390,
      "reflective": 97623,
      "noetic": 80040,
      "push": 94454,
      "potentials": 88623,
      "thorny": 121874,
      "confounding": 21043,
      "cola": 18346,
      "devlin": 30467,
      "gpt": 48552,
      "radford": 95432,
      "ate": 9676,
      "nearhuman": 76728,
      "seventh": 107876,
      "aptitude": 8481,
      "majorityclass": 66613,
      "symbolically": 116324,
      "emulated": 36344,
      "striking": 113593,
      "languagelike": 60371,
      "instruction": 55704,
      "exclusivity": 39961,
      "iconic": 51310,
      "vehicles": 131505,
      "passenger": 85621,
      "av": 10899,
      "amie": 5318,
      "automatedvehicle": 10482,
      "incabin": 53206,
      "seemingly": 104525,
      "wasserstein": 132039,
      "xnli": 135216,
      "385": 984,
      "initializes": 55244,
      "featureengineered": 43341,
      "aes": 4205,
      "unavailability": 126594,
      "sensetagged": 105725,
      "favorable": 43214,
      "avenues": 11166,
      "seq2seqbased": 106909,
      "wordoverlap": 133787,
      "summarise": 115468,
      "transfertransfo": 124250,
      "highcapacity": 50150,
      "informationretrieval": 55116,
      "personachat": 87109,
      "hits1": 50578,
      "linearly": 64401,
      "quadratic": 94540,
      "unfortunate": 127076,
      "blex": 14350,
      "symptoms": 116333,
      "selfreports": 104931,
      "coherency": 18311,
      "wei": 132316,
      "li": 63952,
      "qnetwork": 94536,
      "doctors": 32945,
      "commandandcontrol": 18790,
      "uttered": 130679,
      "patience": 85696,
      "eda": 34520,
      "shallower": 107933,
      "episodic": 38213,
      "typelevel": 126239,
      "raises": 95460,
      "modulating": 74474,
      "rivals": 102885,
      "englishlanguage": 37417,
      "shortened": 108288,
      "shortenings": 108290,
      "inconsistencies": 53433,
      "distorted": 32591,
      "ed": 34517,
      "achievement": 2725,
      "crc": 24600,
      "consumers": 21724,
      "flores": 44678,
      "nepalienglish": 77014,
      "sinhalaenglish": 109866,
      "failed": 42975,
      "naively": 76025,
      "nonentity": 80197,
      "degrades": 28562,
      "entityspecific": 38150,
      "restore": 101332,
      "squared": 112082,
      "glyph": 48319,
      "glyphs": 48320,
      "ingredients": 55185,
      "spots": 112043,
      "diseases": 32339,
      "cities": 16997,
      "scrambling": 104236,
      "bollywood": 14404,
      "transformational": 124273,
      "tagalog": 117345,
      "philippines": 87263,
      "archipelago": 8598,
      "islands": 56965,
      "beautiful": 12626,
      "cater": 15799,
      "philippine": 87262,
      "spacy": 111110,
      "st": 112098,
      "justifying": 57482,
      "dualencoder": 34245,
      "bucc": 14732,
      "secondstage": 104482,
      "animal": 6117,
      "cultures": 25234,
      "predominantly": 89229,
      "western": 132456,
      "richly": 102811,
      "featural": 43246,
      "multitude": 75908,
      "thereof": 121835,
      "contradiction": 22667,
      "won": 132896,
      "lorehlt": 65234,
      "edl": 34564,
      "sf": 107895,
      "security": 104497,
      "hiding": 49920,
      "drastic": 34098,
      "desideratum": 29511,
      "fairseq": 42999,
      "suis": 115386,
      "manytoone": 67102,
      "linearity": 64396,
      "ivory": 57157,
      "admits": 3888,
      "complexvalued": 19949,
      "underperform": 126712,
      "sponsored": 112027,
      "advertisers": 4197,
      "unpopular": 127488,
      "impression": 52314,
      "matchingbased": 67442,
      "deteriorate": 30107,
      "impressions": 52315,
      "cov": 24474,
      "clickthrough": 17752,
      "inferior": 54259,
      "deficient": 28479,
      "multiresolution": 75769,
      "shortlong": 108310,
      "querydocument": 95001,
      "gcn": 46566,
      "pain": 83417,
      "brackets": 14549,
      "board": 14390,
      "hampered": 49309,
      "transfers": 124248,
      "fantasy": 43062,
      "lexvec": 63945,
      "polylingual": 88016,
      "princeton": 90667,
      "semeval2019": 105542,
      "healthcare": 49630,
      "embodying": 36046,
      "comprehend": 20159,
      "red": 97292,
      "triaging": 125762,
      "facial": 42757,
      "opensmile": 82014,
      "subevent": 114667,
      "chronological": 16946,
      "microf1": 70054,
      "convention": 22874,
      "ascribed": 9285,
      "deidentification": 28580,
      "doctor": 32943,
      "recordings": 97134,
      "redacted": 97293,
      "redaction": 97294,
      "glue": 48308,
      "comparemt": 19493,
      "histograms": 50512,
      "likelihoods": 64130,
      "crucially": 25181,
      "propensity": 92411,
      "scenes": 103880,
      "usefully": 128952,
      "ma": 65733,
      "formulating": 45291,
      "dialoguelevel": 30818,
      "outlier": 82626,
      "marking": 67250,
      "possession": 88362,
      "fatal": 43211,
      "federated": 43812,
      "keyboard": 57614,
      "smartphones": 110258,
      "noniid": 80215,
      "hosted": 50661,
      "introspection": 56660,
      "supplements": 115948,
      "visualizes": 131839,
      "slovak": 110097,
      "land": 58800,
      "anecdotal": 6101,
      "subtrees": 115010,
      "provably": 93693,
      "underlie": 126671,
      "selectively": 104856,
      "boosts": 14447,
      "interpreters": 56284,
      "cai": 15014,
      "aspectual": 9421,
      "arrival": 9027,
      "london": 65058,
      "jane": 57178,
      "distractors": 32601,
      "133": 299,
      "deletions": 28599,
      "fullyautomated": 45982,
      "generalisable": 46739,
      "optimising": 82184,
      "166": 390,
      "nary": 76200,
      "oncology": 81665,
      "consecutive": 21162,
      "widening": 132605,
      "peters": 87207,
      "existent": 40039,
      "1944": 445,
      "06850": 53,
      "10th": 211,
      "barack": 11426,
      "obama": 81042,
      "metoo": 69864,
      "girls": 47972,
      "outliers": 82628,
      "atop": 9686,
      "confident": 21000,
      "coreferences": 23380,
      "confidences": 20999,
      "enumeration": 38182,
      "parallelizable": 84702,
      "stringtotree": 113607,
      "dynet": 34355,
      "depicts": 29299,
      "uniquely": 127198,
      "constraining": 21576,
      "acting": 3074,
      "unchanged": 126630,
      "singapore": 109703,
      "ingredient": 55184,
      "exemplarbased": 39978,
      "retrieves": 102467,
      "exemplar": 39977,
      "clusterbased": 17930,
      "condense": 20748,
      "converges": 22930,
      "quicker": 95387,
      "posbased": 88247,
      "cxg": 25477,
      "infersent": 54268,
      "happening": 49438,
      "terrorist": 120415,
      "pyramid": 94475,
      "attainable": 9722,
      "cwi": 25472,
      "movers": 74888,
      "therapists": 121831,
      "empathy": 36135,
      "meets": 68261,
      "iit": 51724,
      "reinflection": 97785,
      "1of100": 480,
      "existed": 40034,
      "mimicry": 70140,
      "odds": 81530,
      "posit": 88279,
      "retrieveandedit": 102457,
      "synthesizing": 116610,
      "repositories": 99093,
      "sqa": 112067,
      "catastrophic": 15709,
      "homonymy": 50622,
      "discourses": 32111,
      "definitive": 28541,
      "programmatic": 92120,
      "loops": 65229,
      "prespecified": 89951,
      "placeholder": 87612,
      "hurt": 51164,
      "hurting": 51167,
      "pb": 85820,
      "dg": 30484,
      "2008": 556,
      "switching": 116294,
      "ernie": 38269,
      "entitylevel": 38140,
      "masks": 67318,
      "datahungry": 26687,
      "lego": 63324,
      "saving": 103667,
      "encapsulating": 36418,
      "empowers": 36341,
      "json": 57420,
      "wikiqa": 132692,
      "modelagnostic": 72345,
      "routingbyagreement": 103325,
      "capsule": 15247,
      "rnmt": 102898,
      "emoticon": 36092,
      "publication": 94279,
      "reconciliation": 97103,
      "speedups": 111890,
      "reactions": 95926,
      "insufficiency": 55718,
      "ineffective": 54066,
      "referenced": 97550,
      "blended": 14271,
      "routine": 103318,
      "ties": 121963,
      "crises": 24781,
      "sensemaking": 105713,
      "tie": 121961,
      "paragraphlevel": 84575,
      "870": 1337,
      "724": 1227,
      "892": 1355,
      "profession": 92075,
      "nurse": 81031,
      "professions": 92091,
      "actively": 3123,
      "toponym": 122688,
      "epidemiology": 38210,
      "infected": 54093,
      "105": 192,
      "underline": 126673,
      "whwords": 132483,
      "dominates": 33938,
      "typologies": 126502,
      "yago": 135227,
      "lorelei": 65235,
      "humanitarian": 51038,
      "stakeholders": 112164,
      "decisionmakers": 27950,
      "darpa": 25543,
      "incidents": 53216,
      "fielded": 43989,
      "molecular": 74534,
      "digestion": 31712,
      "influx": 54328,
      "092": 106,
      "doubled": 33967,
      "runnerup": 103468,
      "separates": 106880,
      "grande": 48736,
      "englishportuguese": 37421,
      "umls": 126565,
      "silent": 109062,
      "disorders": 32369,
      "ssr": 112094,
      "impeded": 51916,
      "karaka": 57494,
      "phd": 87217,
      "detailing": 29788,
      "elucidate": 35330,
      "referencing": 97560,
      "windowing": 132716,
      "moderately": 74376,
      "unilm": 127161,
      "coqa": 23303,
      "cnndailymail": 18035,
      "rougel": 103306,
      "204": 746,
      "086": 90,
      "825": 1306,
      "bleu4": 14348,
      "documentgrounded": 33130,
      "highrank": 50417,
      "illegal": 51728,
      "haven": 49548,
      "scalably": 103700,
      "drugrelated": 34200,
      "timestamped": 122200,
      "sure": 116055,
      "borrows": 14478,
      "duality": 34247,
      "transduce": 124016,
      "marginalization": 67212,
      "adjustments": 3880,
      "waikato": 131983,
      "rings": 102860,
      "400": 1008,
      "893": 1356,
      "950": 1422,
      "971": 1441,
      "785": 1270,
      "868": 1335,
      "767": 1262,
      "862": 1331,
      "812": 1299,
      "laptop": 61004,
      "kgs": 57663,
      "knowledgedriven": 58265,
      "masked": 67286,
      "moment": 74537,
      "inseparable": 55507,
      "weibo": 132317,
      "lay": 61693,
      "hellaswag": 49692,
      "finish": 44529,
      "sits": 109880,
      "keys": 57626,
      "proves": 93742,
      "af": 4210,
      "zone": 135550,
      "misclassified": 70293,
      "coevolve": 18250,
      "releases": 98498,
      "australian": 10353,
      "desire": 29662,
      "l0": 58368,
      "regularisation": 97734,
      "mixes": 70412,
      "tractably": 122770,
      "reacts": 95927,
      "context2vec": 22334,
      "gnn": 48322,
      "adopts": 3909,
      "entitycentric": 38135,
      "formalise": 45183,
      "instantiation": 55647,
      "eventcentric": 39557,
      "690": 1199,
      "kl": 57707,
      "hyperparameterfree": 51233,
      "mdl": 67588,
      "multiparagraph": 75477,
      "headings": 49590,
      "heading": 49589,
      "threelevel": 121939,
      "duplication": 34273,
      "175": 408,
      "wordpiece": 133792,
      "urge": 127844,
      "sixteen": 109905,
      "force": 45011,
      "therefrom": 121834,
      "precursory": 88816,
      "eurlex": 38762,
      "unions": 127179,
      "multidisciplinary": 75072,
      "labelwise": 58659,
      "harry": 49515,
      "potter": 88624,
      "spells": 111920,
      "interrupt": 56301,
      "turnbyturn": 125984,
      "tokenbytoken": 122273,
      "relabel": 97839,
      "relabeling": 97841,
      "constitution": 21561,
      "presumes": 89965,
      "invariably": 56678,
      "overlook": 83343,
      "abbreviation": 1464,
      "radial": 95437,
      "deviates": 30451,
      "calling": 15071,
      "substitutionbased": 114930,
      "geolocating": 47859,
      "8000": 1292,
      "geolocation": 47860,
      "casualties": 15684,
      "civil": 17002,
      "war": 132020,
      "modelfree": 72366,
      "inefficiency": 54070,
      "vanishing": 130825,
      "tightly": 121969,
      "intralanguage": 56337,
      "characterrich": 16591,
      "mlm": 70445,
      "960": 1429,
      "unprecedented": 127489,
      "toptier": 122694,
      "pointers": 87834,
      "letting": 63409,
      "foundational": 45351,
      "sidestep": 108661,
      "conventions": 22917,
      "246": 807,
      "tabletotext": 117272,
      "wikibio": 132627,
      "wiseman": 132739,
      "twostaged": 126166,
      "delay": 28588,
      "ondevice": 81667,
      "fc": 43227,
      "continues": 22607,
      "easyfirst": 34487,
      "blank": 14264,
      "welldesigned": 132401,
      "contextualizing": 22585,
      "aggregates": 4353,
      "wolf": 132886,
      "hit": 50576,
      "stancebearing": 112177,
      "nondialogic": 80191,
      "rcn": 95878,
      "lowdata": 65404,
      "regime": 97674,
      "pretrains": 90353,
      "finetunes": 44439,
      "banking": 11420,
      "selfattentional": 104904,
      "lossless": 65311,
      "retrievable": 102377,
      "subsentence": 114807,
      "consolidation": 21506,
      "underexploited": 126659,
      "bestperforming": 13474,
      "math": 67447,
      "966": 1436,
      "968": 1437,
      "2040": 747,
      "poincare": 87793,
      "hyponym": 51243,
      "attaching": 9694,
      "disconnected": 32024,
      "slovene": 110098,
      "theses": 121845,
      "auc": 10217,
      "singleword": 109857,
      "happens": 49439,
      "lmtc": 64892,
      "berts": 13284,
      "exceptionally": 39930,
      "necessitate": 76764,
      "electricity": 35244,
      "environmentally": 38197,
      "carbon": 15499,
      "fuel": 45921,
      "equity": 38244,
      "deepen": 28438,
      "conll03": 21089,
      "progression": 92189,
      "receptive": 96785,
      "fillintheblank": 44064,
      "proficient": 92097,
      "extrinsically": 42628,
      "angle": 6112,
      "wont": 132899,
      "suffice": 115266,
      "ideally": 51346,
      "substantiated": 114918,
      "pertinent": 87190,
      "diversify": 32863,
      "pools": 88051,
      "8k": 1359,
      "noisily": 80085,
      "failsafe": 42984,
      "imputation": 53186,
      "pearsons": 85838,
      "spearmans": 111346,
      "178": 409,
      "headdriven": 49582,
      "anchor": 6083,
      "lrls": 65582,
      "pivots": 87592,
      "hrl": 50695,
      "charrnn": 16643,
      "noisiness": 80086,
      "245": 805,
      "continual": 22597,
      "vqa": 131960,
      "linguisticallyinformed": 64599,
      "whquestions": 132481,
      "polar": 87898,
      "acquires": 3041,
      "receives": 96380,
      "cotterell": 24410,
      "speculated": 111636,
      "methodologically": 69238,
      "natively": 76237,
      "drive": 34145,
      "viz": 131853,
      "attends": 9785,
      "milk": 70091,
      "maintains": 66527,
      "latencyquality": 61570,
      "cued": 25207,
      "wmt19": 132852,
      "checkpoint": 16700,
      "ldc": 61835,
      "antonym": 6759,
      "distilled": 32524,
      "ddm": 27851,
      "favouring": 43226,
      "placement": 87614,
      "beaten": 12618,
      "misclassify": 70294,
      "paralinguistic": 84582,
      "costsensitive": 24406,
      "confusionaware": 21058,
      "sibling": 108656,
      "subtypes": 115012,
      "instancelevel": 55614,
      "tackbp": 117279,
      "noisier": 80084,
      "deciphering": 27932,
      "interconnected": 56051,
      "coreferent": 23381,
      "pronominal": 92335,
      "questioning": 95264,
      "factaware": 42842,
      "remembering": 98864,
      "feed": 43814,
      "cognitively": 18286,
      "pairlevel": 83478,
      "plateau": 87650,
      "intensional": 55880,
      "recognise": 96797,
      "attributions": 10213,
      "unintended": 127171,
      "toxic": 122709,
      "undesired": 127042,
      "oregon": 82453,
      "fren": 45798,
      "labelers": 58481,
      "labeler": 58480,
      "labelspecific": 58658,
      "wordspecific": 134337,
      "continuity": 22609,
      "hungry": 51160,
      "targetlevel": 117781,
      "interview": 56327,
      "weaklysupervised": 132203,
      "spanpair": 111193,
      "humancurated": 51014,
      "breadth": 14575,
      "danish": 25534,
      "copenhagen": 23276,
      "linguisticallymotivated": 64601,
      "dog": 33415,
      "magnetoencephalography": 66376,
      "recording": 97133,
      "mboshi": 67575,
      "mediated": 68190,
      "mi": 70030,
      "substance": 114848,
      "categorizes": 15774,
      "forecasts": 45036,
      "alert": 4666,
      "successes": 115145,
      "filipino": 44023,
      "ulmfit": 126552,
      "selfsupervised": 104933,
      "ssn": 112093,
      "maml": 66890,
      "modernizing": 74433,
      "modernization": 74430,
      "profits": 92106,
      "overload": 83341,
      "discourseannotated": 32103,
      "forefront": 45037,
      "backgrounds": 11311,
      "commonalities": 18951,
      "resonate": 100822,
      "divides": 32912,
      "tone": 122362,
      "tones": 122363,
      "speaks": 111341,
      "comedy": 18773,
      "advice": 4199,
      "topicoriented": 122610,
      "shades": 107905,
      "enterprises": 37694,
      "nugget": 80829,
      "multistack": 75797,
      "jsd": 57419,
      "temporarily": 120135,
      "replaces": 98944,
      "emojis": 36089,
      "politicians": 88007,
      "61": 1159,
      "reqa": 100102,
      "hypothesisonly": 51285,
      "indication": 53874,
      "datafiltering": 26686,
      "sourcebased": 110870,
      "naver": 76677,
      "europes": 38775,
      "mts": 75020,
      "frenchenglish": 45830,
      "march": 67175,
      "adapter": 3295,
      "reorganize": 98903,
      "simplequestion": 109549,
      "transitivity": 124520,
      "unites": 127229,
      "supplementary": 115940,
      "genuinely": 47845,
      "deliberation": 28607,
      "uttering": 130680,
      "unbabels": 126604,
      "xlm": 135195,
      "reflexive": 97628,
      "pretending": 89970,
      "mature": 67493,
      "monomodal": 74638,
      "moments": 74538,
      "lowresources": 65576,
      "transferable": 124219,
      "expose": 41686,
      "degeneration": 28546,
      "cone": 20972,
      "closeness": 17869,
      "xlnet": 135210,
      "2018b": 690,
      "153": 366,
      "germanczech": 47929,
      "armed": 9013,
      "readytouse": 96049,
      "regressor": 97719,
      "tensor2tensor": 120194,
      "enlarging": 37550,
      "p005": 83397,
      "confirmatory": 21022,
      "summarized": 115617,
      "clausebased": 17692,
      "mc": 67577,
      "quac": 94539,
      "doqa": 33954,
      "intending": 55874,
      "cancer": 15096,
      "caregivers": 15531,
      "daytoday": 27846,
      "wellbeing": 132396,
      "observational": 81172,
      "turnlevel": 125989,
      "gs": 49204,
      "challenged": 16121,
      "sheer": 108164,
      "protolanguage": 93679,
      "historic": 50513,
      "phonologically": 87317,
      "wellattested": 132394,
      "rises": 102872,
      "predefine": 88819,
      "cuttingedge": 25466,
      "syntaxguided": 116585,
      "ride": 102822,
      "believed": 12707,
      "subordinate": 114794,
      "attributable": 10176,
      "billions": 14083,
      "abductive": 1469,
      "jenny": 57224,
      "house": 50686,
      "mess": 68493,
      "broke": 14708,
      "hobbs": 50584,
      "200k": 559,
      "conceptualize": 20676,
      "689": 1196,
      "914": 1387,
      "toxicity": 122713,
      "nepali": 77013,
      "metalwoz": 68559,
      "incar": 53210,
      "scheduling": 103889,
      "muse": 75962,
      "conneau": 21112,
      "reuses": 102491,
      "backward": 11348,
      "redirects": 97310,
      "misaligned": 70287,
      "untrained": 127766,
      "encouragingly": 36783,
      "tale": 117554,
      "revisited": 102653,
      "pack": 83402,
      "flair": 44631,
      "winner": 132719,
      "controversy": 22867,
      "argumentation": 8965,
      "machinereadable": 66326,
      "pdf": 85829,
      "360": 967,
      "encoderonly": 36630,
      "decoderonly": 28062,
      "openai": 81949,
      "pointergenerator": 87830,
      "skew": 109969,
      "overgeneralize": 83322,
      "optima": 82159,
      "escape": 38420,
      "hierarchicallyrefined": 50022,
      "persona": 87106,
      "lowerresourced": 65460,
      "dataselection": 26700,
      "domainadapted": 33703,
      "rethinking": 102367,
      "facto": 42847,
      "relevancy": 98521,
      "inventing": 56685,
      "inventors": 56691,
      "envision": 38203,
      "autocomplete": 10394,
      "conceive": 20548,
      "inventions": 56689,
      "technically": 119758,
      "gpt2": 48558,
      "sentencebert": 106140,
      "roberta": 102990,
      "sentencepair": 106183,
      "triplet": 125822,
      "optimizationbased": 82219,
      "graphbank": 48885,
      "deficiency": 28478,
      "remedies": 98856,
      "sequencelabeling": 107104,
      "languagemodeling": 60373,
      "crosscontext": 24871,
      "multitoken": 75904,
      "probes": 90881,
      "elmos": 35328,
      "diagnosing": 30502,
      "constructive": 21706,
      "1979": 454,
      "journey": 57414,
      "volunteers": 131946,
      "humanderived": 51017,
      "unite": 127224,
      "supremacist": 116050,
      "breakpoints": 14580,
      "mitigated": 70378,
      "anger": 6105,
      "chang": 16353,
      "wordplay": 133795,
      "homographic": 50617,
      "cmlm": 17975,
      "governmental": 48544,
      "12000": 264,
      "ideology": 51646,
      "softlabel": 110501,
      "nondifferentiable": 80192,
      "multiplicity": 75763,
      "flickr30k": 44670,
      "fmri": 44729,
      "eyetracking": 42638,
      "toolbox": 122407,
      "keeps": 57522,
      "mbert": 67562,
      "queen": 94909,
      "approx": 8451,
      "rotation": 103279,
      "multiencoder": 75100,
      "autocompletion": 10396,
      "shorten": 108287,
      "mrl": 74932,
      "hypersphere": 51240,
      "accommodated": 1940,
      "isomorphic": 56981,
      "26000": 844,
      "copyright": 23302,
      "brat": 14567,
      "conllu": 21109,
      "narrows": 76199,
      "roundtrip": 103314,
      "overestimation": 83316,
      "180": 422,
      "aggressively": 4376,
      "onlstm": 81820,
      "equivalently": 38258,
      "bpc": 14536,
      "enwik8": 38205,
      "icdm": 51305,
      "uwa": 130682,
      "facility": 42812,
      "visualise": 131821,
      "betweenness": 13776,
      "goto": 48536,
      "headlines": 49599,
      "npi": 80798,
      "cats": 15801,
      "steady": 113224,
      "untested": 127764,
      "oblique": 81155,
      "kbqa": 57508,
      "csqa": 25188,
      "redundantly": 97481,
      "assessors": 9535,
      "120": 262,
      "p082": 83398,
      "zipfian": 135545,
      "sami": 103558,
      "categorically": 15725,
      "troublesome": 125843,
      "singleturn": 109852,
      "flaws": 44642,
      "disregard": 32384,
      "dependencyaware": 29254,
      "necessitates": 76765,
      "102": 186,
      "football": 45005,
      "playbyplay": 87709,
      "anecdotes": 6103,
      "evince": 39686,
      "nsp": 80811,
      "mrc": 74923,
      "hans": 49433,
      "accomplishes": 1956,
      "enumerating": 38181,
      "propagating": 92388,
      "matters": 67491,
      "nonlatin": 80219,
      "pvalue": 94473,
      "005": 3,
      "linker": 64650,
      "susceptibility": 116215,
      "sift": 108665,
      "propensities": 92410,
      "refutation": 97644,
      "transferability": 124213,
      "dfs": 30483,
      "perfectly": 85937,
      "inspecting": 55553,
      "multiprocessing": 75765,
      "bertscore": 13289,
      "multilabeled": 75164,
      "farsi": 43109,
      "living": 64855,
      "sdss": 104283,
      "distracted": 32596,
      "conducive": 20822,
      "retriever": 102465,
      "row": 103326,
      "column": 18539,
      "queryfocused": 95003,
      "oneclass": 81671,
      "slotspecific": 110094,
      "alternately": 5216,
      "snips": 110309,
      "prohibits": 92208,
      "xlu": 135212,
      "intralayer": 56338,
      "complimentary": 19962,
      "76": 1259,
      "bertlike": 13282,
      "wikitext103": 132697,
      "108": 199,
      "158": 370,
      "665": 1187,
      "632": 1170,
      "909": 1378,
      "transmitted": 125531,
      "graphlevel": 48937,
      "embedders": 35367,
      "mtdnn": 75011,
      "documentbased": 33125,
      "plagues": 87621,
      "vaes": 130703,
      "cvae": 25469,
      "mixing": 70413,
      "preneural": 89291,
      "coarsely": 18057,
      "philosophical": 87265,
      "originality": 82565,
      "professor": 92092,
      "centered": 15892,
      "months": 74656,
      "april": 8475,
      "june": 57457,
      "leaders": 61881,
      "followers": 44957,
      "inclination": 53217,
      "convergent": 22929,
      "bestworst": 13490,
      "rankingbased": 95696,
      "entityrisk": 38148,
      "chess": 16715,
      "boards": 14391,
      "moves": 74891,
      "inspiring": 55587,
      "imprecise": 52313,
      "concatenate": 20532,
      "microaveraged": 70040,
      "humanrated": 51063,
      "illdefined": 51727,
      "genes": 47816,
      "chemicals": 16710,
      "mgb3": 70029,
      "subwordinformed": 115045,
      "onebest": 81670,
      "overnight": 83349,
      "strikingly": 113594,
      "pertain": 87185,
      "corroborating": 24324,
      "agreed": 4384,
      "recurrencebased": 97155,
      "manipuri": 66935,
      "fsts": 45919,
      "persuasion": 87183,
      "charity": 16639,
      "coach": 18048,
      "negotiate": 76978,
      "recommends": 97100,
      "reject": 97834,
      "liang": 63955,
      "sari": 103645,
      "factchecking": 42845,
      "uds": 126542,
      "bundled": 14965,
      "unifies": 127135,
      "predpatt": 89231,
      "vulnerabilities": 131968,
      "alterations": 5212,
      "textclassification": 121432,
      "watson": 132045,
      "comparatives": 19218,
      "numeral": 81002,
      "quotes": 95417,
      "exacerbated": 39710,
      "podcast": 87777,
      "continuousspace": 22656,
      "sindhi": 109701,
      "designation": 29575,
      "936": 1410,
      "nonannotated": 80151,
      "backing": 11312,
      "comorbidity": 19108,
      "physiological": 87482,
      "nlpbased": 79792,
      "submodular": 114783,
      "justify": 57481,
      "assesses": 9504,
      "universality": 127338,
      "surveillance": 116159,
      "smoking": 110260,
      "adversaries": 4185,
      "750": 1252,
      "834": 1312,
      "fewrel": 43894,
      "han": 49313,
      "commonlyused": 18976,
      "https": 50700,
      "nlps": 79801,
      "wngt": 132876,
      "suffering": 115247,
      "fasten": 43158,
      "taskagnostic": 118870,
      "333": 947,
      "186": 427,
      "107": 194,
      "seeds": 104511,
      "onetime": 81725,
      "intrusion": 56662,
      "denoised": 28993,
      "nmtbased": 80011,
      "wellresourced": 132428,
      "2019a": 718,
      "spot": 112041,
      "dative": 27837,
      "elaborating": 35227,
      "overarching": 83271,
      "turku": 125967,
      "genetics": 47822,
      "255": 829,
      "544": 1106,
      "637": 1174,
      "ace04": 2398,
      "ace05": 2400,
      "msra": 74951,
      "trading": 122786,
      "distantlysupervised": 32495,
      "viggo": 131727,
      "uptake": 127831,
      "7k": 1278,
      "keywordbased": 57641,
      "heavytailed": 49668,
      "month": 74655,
      "mlb": 70441,
      "baseball": 11494,
      "league": 61977,
      "draft": 34081,
      "simplifies": 109611,
      "strive": 113612,
      "openaccess": 81948,
      "nontarget": 80301,
      "foursquare": 45355,
      "memorization": 68279,
      "1579": 369,
      "datastore": 27808,
      "inserting": 55510,
      "mnli": 70461,
      "inferencing": 54256,
      "bertbased": 13266,
      "requisite": 100359,
      "fairness": 42998,
      "enzh": 38206,
      "crosssentential": 25081,
      "nonautoregressive": 80153,
      "nat": 76206,
      "errorcorrected": 38347,
      "mustc": 75966,
      "194": 444,
      "diverges": 32790,
      "hallucinations": 49305,
      "snmt": 110318,
      "simultaneity": 109652,
      "balancing": 11406,
      "srl": 112087,
      "laymen": 61817,
      "mr": 74922,
      "verifiability": 131568,
      "mislead": 70299,
      "verifies": 131576,
      "ebert": 34496,
      "entityenhanced": 38139,
      "el": 35223,
      "guessing": 49224,
      "affordable": 4260,
      "schutze": 103968,
      "singlesource": 109839,
      "357": 964,
      "rams": 95470,
      "9124": 1385,
      "139": 307,
      "taskindependent": 118878,
      "liked": 64116,
      "camembert": 15082,
      "omitting": 81659,
      "hallucination": 49304,
      "marginbased": 67219,
      "327": 937,
      "112": 237,
      "extensibility": 41831,
      "metatraining": 68567,
      "ln": 64894,
      "grubased": 49202,
      "eighth": 35221,
      "taskcompletion": 118873,
      "sceneaware": 103878,
      "schemaguided": 103903,
      "distractor": 32599,
      "didnt": 30924,
      "subclass": 114652,
      "hampers": 49311,
      "fleiss": 44643,
      "kappa": 57492,
      "outbreak": 82612,
      "infectious": 54094,
      "brittle": 14662,
      "noised": 80081,
      "bland": 14262,
      "cvaebased": 25470,
      "mid": 70066,
      "90s": 1381,
      "fold": 44933,
      "038": 21,
      "ancestral": 6082,
      "implicatures": 52001,
      "scraping": 104238,
      "exampleguided": 39803,
      "disregarding": 32386,
      "stc": 113220,
      "unresolved": 127509,
      "unsuited": 127592,
      "connectionist": 21138,
      "wait": 131988,
      "timings": 122210,
      "ctc": 25192,
      "opensubtitles2018": 82039,
      "transliterating": 125520,
      "bokmaal": 14402,
      "nynorsk": 81038,
      "nominals": 80143,
      "103m": 189,
      "twolayered": 126146,
      "prepossessing": 89316,
      "sg": 107899,
      "multicascaded": 75039,
      "roman": 103244,
      "directives": 31856,
      "fulfill": 45924,
      "imbalance": 51824,
      "isolating": 56978,
      "964": 1433,
      "transformerxl": 124475,
      "bleached": 14267,
      "manuals": 67096,
      "endusers": 36994,
      "born": 14469,
      "topologically": 122683,
      "laser": 61552,
      "multipurpose": 75766,
      "emitting": 36082,
      "traversal": 125543,
      "linearizing": 64400,
      "interleaves": 56117,
      "experienced": 40444,
      "dominate": 33936,
      "superseded": 115712,
      "highresourced": 50434,
      "feelings": 43863,
      "sadness": 103532,
      "disgust": 32361,
      "fear": 43231,
      "dataaugmentation": 26636,
      "concatenations": 20546,
      "lifelong": 63981,
      "lowcost": 65403,
      "nonsynthetic": 80299,
      "bai": 11391,
      "hypocoristic": 51242,
      "publiclyavailable": 94335,
      "finergrained": 44399,
      "concatenating": 20537,
      "automates": 10486,
      "documentcontext": 33126,
      "cyclic": 25485,
      "stacks": 112141,
      "triviaqa": 125834,
      "searchqa": 104374,
      "figer": 44009,
      "57": 1125,
      "utter": 130587,
      "autonomously": 10854,
      "distractions": 32598,
      "steer": 113228,
      "bart": 11433,
      "t5": 117248,
      "rouge1": 103302,
      "137": 305,
      "customizing": 25461,
      "makings": 66873,
      "oblivious": 81156,
      "textcnn": 121434,
      "bigrulstmcnn": 14013,
      "layoutlm": 61821,
      "textlevel": 121442,
      "transparency": 125533,
      "orthographies": 82603,
      "transcribe": 123981,
      "ann": 6122,
      "wikimedia": 132637,
      "phonemetographeme": 87283,
      "numerals": 81003,
      "appearances": 6807,
      "numeracy": 81001,
      "differentiates": 31590,
      "utilising": 130472,
      "speculation": 111637,
      "integral": 55739,
      "bioscope": 14219,
      "sfu": 107896,
      "427": 1023,
      "stumbling": 114563,
      "antecedents": 6752,
      "demonstrative": 28991,
      "koehn": 58340,
      "qualified": 94548,
      "encyclopedias": 36786,
      "typenet": 126242,
      "conllaida": 21108,
      "898": 1358,
      "alias": 4893,
      "trivia": 125826,
      "berlin": 13047,
      "prophetnet": 92511,
      "nstream": 80812,
      "onestepahead": 81722,
      "160gb": 382,
      "subfield": 114668,
      "wordsentence": 134331,
      "kurdish": 58362,
      "oversimplified": 83356,
      "dialectology": 30533,
      "reflexes": 97627,
      "invaluable": 56677,
      "protect": 93657,
      "confidentiality": 21001,
      "conform": 21037,
      "personally": 87150,
      "domainknowledge": 33718,
      "headwise": 49613,
      "precomputed": 88813,
      "qqp": 94537,
      "shot": 108355,
      "underperforms": 126715,
      "reformulates": 97635,
      "governance": 48539,
      "shareholders": 108123,
      "russia": 103486,
      "border": 14466,
      "diagrams": 30515,
      "dfa": 30482,
      "recordbreaking": 97130,
      "variances": 130864,
      "insensitive": 55506,
      "fitted": 44587,
      "sampleefficient": 103574,
      "continents": 22593,
      "smoother": 110267,
      "inversely": 56704,
      "fool": 45001,
      "genealogical": 46629,
      "110": 231,
      "colours": 18538,
      "levenshtein": 63570,
      "singlecharacter": 109821,
      "insertions": 55514,
      "permitted": 87049,
      "bigru": 14010,
      "transformersbert": 124473,
      "fra": 45360,
      "studio": 114301,
      "dal": 25520,
      "si": 108648,
      "proactively": 90795,
      "smoothness": 110271,
      "attains": 9728,
      "sina": 109699,
      "dot": 33955,
      "pandemic": 83690,
      "outbreaks": 82614,
      "prohibit": 92196,
      "globe": 48292,
      "epidemiological": 38209,
      "twentieth": 126076,
      "mtn": 75019,
      "universe": 127344,
      "shareable": 107963,
      "subnetworks": 114788,
      "questionandanswer": 95243,
      "helper": 49792,
      "manhattan": 66917,
      "mmi": 70456,
      "novices": 80794,
      "inexperienced": 54087,
      "iwslt15": 57172,
      "propaganda": 92380,
      "neutrality": 78759,
      "bertlarge": 13280,
      "56": 1121,
      "budget": 14733,
      "boxes": 14533,
      "sroie": 112091,
      "cord": 23306,
      "leaderboard": 61879,
      "decoded": 27988,
      "underspecification": 126732,
      "curating": 25248,
      "beings": 12684,
      "sends": 105642,
      "conveys": 23127,
      "ace2005": 2402,
      "knowledgeenriched": 58271,
      "injected": 55261,
      "asks": 9309,
      "clicking": 17750,
      "money": 74541,
      "raise": 95456,
      "informs": 55171,
      "retains": 102363,
      "ban": 11410,
      "caseless": 15636,
      "cohyponymy": 18338,
      "margins": 67220,
      "6k": 1205,
      "attraction": 10167,
      "taxi": 119654,
      "simulator": 109651,
      "timedependent": 122152,
      "tribute": 125768,
      "funeral": 46137,
      "press": 89953,
      "bilstmcrfs": 14110,
      "arabert": 8488,
      "pursuit": 94452,
      "hoping": 50652,
      "hardness": 49495,
      "depths": 29325,
      "accuracyspeed": 2322,
      "queryresponse": 95008,
      "impede": 51915,
      "committee": 18850,
      "acousticprosodic": 3027,
      "accelerating": 1877,
      "bertbase": 13264,
      "customise": 25455,
      "doubleannotation": 33966,
      "dstc8": 34219,
      "attaining": 9726,
      "hyperdimensional": 51210,
      "618": 1163,
      "462": 1043,
      "384": 983,
      "ca": 15007,
      "dissecting": 32390,
      "headers": 49586,
      "textonly": 121444,
      "tesseract": 120416,
      "humanlabeled": 51046,
      "morfessor": 74662,
      "phonetically": 87302,
      "irregularities": 56942,
      "offsets": 81622,
      "separable": 106843,
      "externally": 42052,
      "stepbystep": 113315,
      "inspire": 55560,
      "factchecked": 42844,
      "stanza": 112386,
      "pidgin": 87491,
      "nsurl2019": 80813,
      "184": 425,
      "142": 327,
      "mentionlevel": 68442,
      "competitiveness": 19705,
      "humanconstructed": 51008,
      "eacl": 34367,
      "2021": 740,
      "editbased": 34538,
      "predecessors": 88818,
      "tldr": 122229,
      "reweighting": 102687,
      "learnings": 63188,
      "stsb": 114134,
      "sst2": 112096,
      "2020": 721,
      "godard": 48421,
      "adherence": 3842,
      "dominated": 33937,
      "fragmented": 45376,
      "verticals": 131634,
      "fragmentation": 45375,
      "noncommercial": 80169,
      "competence": 19609,
      "temperature": 120075,
      "entityaware": 38133,
      "estonian": 38677,
      "serbian": 107264,
      "commercially": 18843,
      "love": 65341,
      "igbo": 51662,
      "furiously": 46139,
      "green": 49079,
      "compresses": 20301,
      "correlationbased": 24249,
      "giant": 47959,
      "embody": 36045,
      "logically": 65023,
      "usercentered": 129061,
      "tencent": 120137,
      "ft": 45920,
      "diminished": 31777,
      "styleaware": 114613,
      "manifested": 66922,
      "endofsequence": 36858,
      "eos": 38207,
      "endings": 36857,
      "domainbased": 33709,
      "impersonation": 51924,
      "stipulate": 113350,
      "overused": 83360,
      "operated": 82048,
      "longtail": 65194,
      "1318": 296,
      "learningtorank": 63189,
      "traintest": 123968,
      "mines": 70151,
      "indistribution": 53895,
      "anglicisms": 6114,
      "adventure": 4094,
      "refactoring": 97493,
      "needing": 76882,
      "mediation": 68192,
      "bertstyle": 13291,
      "convex": 23110,
      "inquisitive": 55503,
      "auditing": 10248,
      "billing": 14076,
      "inferencetime": 54255,
      "selfinduced": 104922,
      "circle": 16981,
      "school": 103964,
      "mobility": 70471,
      "routes": 103317,
      "streets": 113565,
      "accidents": 1937,
      "kornli": 58348,
      "korsts": 58350,
      "600k": 1153,
      "writes": 135084,
      "gpt3": 48572,
      "retranslation": 102376,
      "appending": 6814,
      "upgrade": 127814,
      "backtranslate": 11321,
      "1994": 465,
      "kudo": 58356,
      "contextlevel": 22372,
      "parameterefficient": 84728,
      "fare": 43105,
      "yes": 135324,
      "referencefree": 97551,
      "finance": 44249,
      "selftrained": 104954,
      "advertisement": 4194,
      "todbert": 122239,
      "amrs": 5400,
      "palm": 83687,
      "marco": 67177,
      "interdocument": 56062,
      "singlebest": 109819,
      "1320": 298,
      "icelandic": 51306,
      "insert": 55508,
      "dialoguebased": 30817,
      "dialogre": 30621,
      "speakeraware": 111302,
      "krishna": 58353,
      "verses": 131602,
      "prose": 93638,
      "empathetic": 36132,
      "encompass": 36738,
      "mesh": 68490,
      "multitasking": 75900,
      "aligners": 4932,
      "kg": 57653,
      "formulates": 45289,
      "knowledgeaware": 58243,
      "refinements": 97596,
      "convince": 23129,
      "threestage": 121942,
      "mask": 67282,
      "exclusion": 39956,
      "unmt": 127467,
      "thirteen": 121870,
      "g2p": 46330,
      "schwa": 103969,
      "multiobjective": 75476,
      "holtzman": 50604,
      "977": 1444,
      "nlprelated": 79800,
      "lecture": 63271,
      "slide": 110010,
      "ultra": 126559,
      "spectra": 111628,
      "nontargeted": 80302,
      "expertgenerated": 41234,
      "unequal": 127051,
      "delimit": 28609,
      "englishthai": 37434,
      "replicating": 98964,
      "delivered": 28619,
      "sparsification": 111248,
      "compromising": 20339,
      "endows": 36862,
      "165x": 389,
      "vaebased": 130702,
      "exemplars": 39979,
      "unexplored": 127060,
      "decoupled": 28166,
      "fetching": 43880,
      "pypi": 94474,
      "pes": 87205,
      "bilingually": 14075,
      "bracketing": 14548,
      "rid": 102820,
      "7x": 1284,
      "104": 190,
      "graphenhanced": 48926,
      "owns": 83395,
      "fulfil": 45923,
      "believes": 12712,
      "negating": 76909,
      "comedic": 18772,
      "genders": 46626,
      "translational": 125439,
      "sentencespecific": 106551,
      "humanunderstandable": 51125,
      "unimportant": 127166,
      "humanbased": 51000,
      "profiling": 92103,
      "wellchosen": 132398,
      "unfiltered": 127070,
      "metaclassifier": 68533,
      "interference": 56112,
      "restrained": 101334,
      "adapters": 3300,
      "v20": 130691,
      "867": 1334,
      "enja": 37541,
      "xlmroberta": 135207,
      "zeroshort": 135486,
      "paraphraser": 84836,
      "paraphrasers": 84837,
      "highfidelity": 50239,
      "surfacelevel": 116083,
      "diversified": 32862,
      "purportedly": 94421,
      "repurposes": 100097,
      "rephrases": 98922,
      "complexsimple": 19946,
      "knowledgetotext": 58282,
      "graphtosequence": 48982,
      "124": 269,
      "alarmed": 4660,
      "warming": 132023,
      "mutlitask": 75968,
      "overwhelmingly": 83387,
      "multiattribute": 75036,
      "ensured": 37644,
      "supersense": 115713,
      "streusle": 113586,
      "dimsum": 31779,
      "superficial": 115670,
      "toy": 122714,
      "unlearn": 127419,
      "hurts": 51169,
      "imparting": 51914,
      "uninformative": 127168,
      "attractors": 10173,
      "underperformed": 126713,
      "keyphrase": 57617,
      "genqa": 47827,
      "covid19": 24561,
      "spain": 111112,
      "france": 45753,
      "uk": 126547,
      "suffered": 115246,
      "virus": 131755,
      "keen": 57516,
      "arranging": 9025,
      "threestep": 121944,
      "overgeneration": 83323,
      "understudied": 127028,
      "prunes": 94162,
      "uninteresting": 127174,
      "hallucinate": 49301,
      "confounders": 21042,
      "regex": 97671,
      "stackoverflow": 112139,
      "regexes": 97673,
      "depictions": 29298,
      "plms": 87754,
      "derivationally": 29330,
      "plm": 87753,
      "unmodified": 127466,
      "generatively": 47762,
      "attracts": 10174,
      "aggressive": 4375,
      "conservative": 21186,
      "englishtochinese": 37436,
      "interpolations": 56205,
      "polyglot": 88009,
      "prohibitive": 92200,
      "critiques": 24852,
      "identically": 51355,
      "lowbias": 65402,
      "favored": 43220,
      "supplementing": 115947,
      "punish": 94397,
      "realignment": 96097,
      "contradictory": 22670,
      "evidencebased": 39679,
      "gleu": 48224,
      "hewitt": 49888,
      "unattested": 126593,
      "immediately": 51846,
      "111": 235,
      "begs": 12640,
      "nel": 77005,
      "catalog": 15689,
      "distilbert": 32496,
      "prohibited": 92198,
      "indications": 53875,
      "acoustically": 3026,
      "prevalence": 90356,
      "prefers": 89252,
      "perplexities": 87059,
      "burgeoning": 14970,
      "underperforming": 126714,
      "sst": 112095,
      "forwards": 45331,
      "backwards": 11352,
      "consequential": 21176,
      "axis": 11270,
      "tax": 119653,
      "legaldomain": 63318,
      "svo": 116249,
      "distills": 32529,
      "handlabeled": 49367,
      "noninformative": 80216,
      "cbr": 15850,
      "converters": 23103,
      "semeval2020": 105547,
      "memes": 68277,
      "harnessing": 49514,
      "persist": 87092,
      "youth": 135460,
      "resilience": 100743,
      "ehealth": 35215,
      "young": 135457,
      "undergraduates": 126670,
      "polarized": 87935,
      "contentrelated": 21984,
      "postlevel": 88503,
      "handcrafting": 49358,
      "occupational": 81491,
      "stratification": 113548,
      "prestige": 89961,
      "affirmative": 4256,
      "reinforces": 97831,
      "pointed": 87819,
      "18th": 430,
      "commons": 18979,
      "genderspecific": 46628,
      "henderson": 49837,
      "sds": 104282,
      "staying": 113219,
      "selfpaced": 104927,
      "150k": 362,
      "mds": 67591,
      "newsfeeds": 79402,
      "tls": 122230,
      "heritage": 49841,
      "871": 1339,
      "medieval": 68240,
      "intellectual": 55820,
      "postcorrections": 88465,
      "fusional": 46255,
      "sacrifice": 103522,
      "contextagnostic": 22335,
      "positiveunlabeled": 88354,
      "363": 968,
      "496": 1057,
      "eat": 34494,
      "visited": 131775,
      "medications": 68233,
      "structurebased": 113988,
      "cora": 23305,
      "iwpt": 57158,
      "4th": 1061,
      "elas": 35229,
      "largescaled": 61521,
      "inadequacy": 53195,
      "concentrates": 20556,
      "inherits": 55197,
      "parsbert": 84877,
      "207": 748,
      "transmission": 125529,
      "societal": 110466,
      "covid": 24560,
      "faq": 43066,
      "reputable": 100099,
      "tableqa": 117263,
      "whitebox": 132477,
      "renders": 98890,
      "revise": 102643,
      "biaffine": 13785,
      "convenience": 22868,
      "charged": 16638,
      "preannotation": 88770,
      "coronavirus": 23394,
      "viewer": 131716,
      "untruncated": 127770,
      "thirty": 121872,
      "groningen": 49092,
      "bender": 12943,
      "banks": 11422,
      "411": 1014,
      "prpn": 94158,
      "copies": 23279,
      "shen": 108166,
      "2018a": 689,
      "concurrent": 20746,
      "entitytype": 38151,
      "openly": 81998,
      "plugandplay": 87761,
      "leaderboards": 61880,
      "optimised": 82182,
      "mlms": 70447,
      "organizes": 82479,
      "500k": 1087,
      "latex": 61661,
      "chomsky": 16921,
      "cmv": 17979,
      "nextword": 79422,
      "logprobability": 65053,
      "psychometric": 94220,
      "extralinguistic": 42584,
      "tiktok": 121970,
      "instagram": 55590,
      "recasting": 96351,
      "quantifier": 94839,
      "vague": 130705,
      "donkey": 33942,
      "crosstask": 25084,
      "ht": 50696,
      "typographical": 126480,
      "aibased": 4430,
      "affairs": 4211,
      "surveyed": 116208,
      "midresource": 70073,
      "amortized": 5326,
      "lowestresource": 65467,
      "deviate": 30450,
      "groundbreaking": 49103,
      "earnings": 34401,
      "hapaxes": 49435,
      "presidents": 89950,
      "zipfmandelbrot": 135547,
      "sociopolitical": 110481,
      "evidenceaware": 39678,
      "event2mind": 39556,
      "customised": 25456,
      "valued": 130789,
      "discourseconfigurational": 32108,
      "380": 980,
      "definiteness": 28517,
      "guilt": 49272,
      "spanlevel": 111183,
      "felt": 43869,
      "ns": 80809,
      "recruit": 97148,
      "da": 25501,
      "openworld": 82045,
      "formalizations": 45194,
      "majorly": 66615,
      "closelyrelated": 17867,
      "pioneered": 87520,
      "garner": 46491,
      "reception": 96784,
      "steep": 113225,
      "chronic": 16945,
      "predicament": 88845,
      "acclaimed": 1938,
      "longtailed": 65196,
      "2305": 794,
      "1559": 367,
      "eventualities": 39615,
      "103": 187,
      "vicinity": 131650,
      "taxonomic": 119655,
      "annealing": 6123,
      "955": 1423,
      "399": 989,
      "russe": 103485,
      "fluently": 44724,
      "lesk": 63394,
      "9407": 1414,
      "rooted": 103274,
      "factbased": 42843,
      "buffer": 14737,
      "mono": 74545,
      "caveats": 15845,
      "xquad": 135217,
      "mlqa": 70451,
      "transferlearning": 124225,
      "socialnlp": 110465,
      "combat": 18542,
      "calibrated": 15043,
      "cord19": 23307,
      "32k": 941,
      "buying": 14986,
      "workload": 134913,
      "multigranularity": 75118,
      "cpwd": 24578,
      "metaphor": 68562,
      "peculiarities": 85840,
      "mrr": 74935,
      "preceded": 88772,
      "relu": 98672,
      "mixedinitiative": 70410,
      "sdrt": 104281,
      "attach": 9689,
      "commitments": 18847,
      "clarification": 17022,
      "886": 1350,
      "repeats": 98913,
      "labov": 58671,
      "partisan": 85507,
      "07": 55,
      "drinks": 34144,
      "eating": 34495,
      "entailed": 37650,
      "sentimentanalysis": 106823,
      "stock": 113363,
      "impedes": 51917,
      "huggingface": 50730,
      "xlmr": 135199,
      "expertwritten": 41253,
      "summarizationbased": 115594,
      "hin": 50447,
      "sentimentbased": 106826,
      "sentimentaware": 106825,
      "indoeuropean": 53952,
      "treetotree": 125734,
      "inquiries": 55501,
      "twostream": 126176,
      "macrof1": 66364,
      "quest": 95010,
      "vmwe": 131860,
      "nnlms": 80015,
      "vein": 131506,
      "emphasized": 36141,
      "participate": 85330,
      "occupy": 81494,
      "lyrics": 65730,
      "romanised": 103254,
      "song": 110659,
      "romanized": 103256,
      "lyric": 65729,
      "rhyming": 102719,
      "offenseval": 81537,
      "ubiquity": 126515,
      "immense": 51848,
      "officially": 81609,
      "nineteenth": 79488,
      "instructing": 55703,
      "robots": 103012,
      "deem": 28189,
      "academics": 1865,
      "awarded": 11254,
      "hypothesise": 51283,
      "nonacademic": 80148,
      "referendum": 97561,
      "selflearning": 104923,
      "labelpreserving": 58575,
      "formatted": 45221,
      "rouge2": 103304,
      "mrls": 74933,
      "coincide": 18341,
      "bar": 11425,
      "comprise": 20331,
      "supervisor": 115934,
      "sgcn": 107901,
      "tacred": 117329,
      "counterfactuals": 24427,
      "subtask1": 114960,
      "subtask2": 114961,
      "relativity": 98420,
      "textbooks": 121431,
      "distributive": 32778,
      "treestructure": 125720,
      "insideoutside": 55518,
      "batchify": 12560,
      "httpsgithubcomyzhangcscrfpar": 50705,
      "appearance": 6806,
      "missed": 70316,
      "tokenizers": 122284,
      "subcharacter": 114650,
      "accompanies": 1944,
      "corporate": 23625,
      "malay": 66874,
      "sockeye": 110482,
      "modernized": 74432,
      "complies": 19961,
      "lit": 64725,
      "terminals": 120250,
      "mltc": 70452,
      "labelaware": 58422,
      "lmpretrained": 64878,
      "infuse": 55178,
      "cointeractive": 18343,
      "correlative": 24264,
      "crossimpact": 24917,
      "mastodon": 67352,
      "lowers": 65462,
      "albert": 4665,
      "portrayals": 88190,
      "victims": 131652,
      "ages": 4331,
      "top3": 122484,
      "21k": 768,
      "singletoken": 109849,
      "infuses": 55180,
      "contingent": 22595,
      "rolefiller": 103226,
      "illsuited": 51734,
      "muc4": 75023,
      "targetopinion": 117782,
      "vacation": 130695,
      "spotlight": 112042,
      "vagueness": 130706,
      "truthfulness": 125890,
      "cner": 17980,
      "ccks2017": 15863,
      "workhorse": 134894,
      "denoise": 28992,
      "shannon": 107935,
      "deft": 28542,
      "defteval": 28543,
      "harmed": 49503,
      "checkthat": 16704,
      "dissemination": 32393,
      "amplify": 5376,
      "midas": 70067,
      "15th": 373,
      "upb": 127782,
      "meme": 68276,
      "funny": 46138,
      "sought": 110698,
      "circumvent": 16984,
      "yupik": 135465,
      "polysynthetic": 88031,
      "alaska": 4661,
      "30k": 917,
      "electra": 35241,
      "derivatives": 29335,
      "882": 1346,
      "prepend": 89306,
      "interspeaker": 56315,
      "equations": 38234,
      "treernn": 125679,
      "equation": 38233,
      "incoherence": 53411,
      "nigeria": 79481,
      "maithili": 66529,
      "magahi": 66371,
      "adversely": 4192,
      "purvanchal": 94453,
      "contextrelevant": 22373,
      "upvotes": 127834,
      "kld": 57709,
      "datas": 26698,
      "ere": 38268,
      "graphbanks": 48886,
      "moverscore": 74890,
      "notorious": 80415,
      "canada": 15094,
      "km": 57712,
      "gurevych": 49285,
      "6621": 1186,
      "4x": 1069,
      "perplexing": 87058,
      "invariance": 56679,
      "penalizing": 85849,
      "overconfidence": 83310,
      "endeavors": 36849,
      "cws": 25475,
      "isolates": 56977,
      "congruent": 21062,
      "incongruent": 53431,
      "birth": 14230,
      "multistep": 75801,
      "dilemma": 31740,
      "modellevel": 72594,
      "absolutely": 1753,
      "ls": 65584,
      "852": 1324,
      "actorcritic": 3137,
      "summarising": 115470,
      "abstracted": 1785,
      "pomdp": 88034,
      "drl": 34165,
      "reservation": 100730,
      "perceptrons": 85932,
      "mlps": 70450,
      "sumbt": 115428,
      "stably": 112115,
      "sums": 115663,
      "commonplace": 18978,
      "outofdistribution": 82642,
      "populous": 88172,
      "shortening": 108289,
      "deletionbased": 28598,
      "adem": 3822,
      "adversarially": 4183,
      "multipass": 75484,
      "unidirectionally": 127097,
      "nonpretrained": 80274,
      "cheaply": 16690,
      "posttraining": 88529,
      "controllability": 22823,
      "facetoface": 42755,
      "functionally": 46066,
      "admission": 3887,
      "savings": 103669,
      "heres": 49840,
      "onesizefitsall": 81717,
      "downloads": 33982,
      "stars": 112394,
      "forks": 45066,
      "topicaware": 122602,
      "topicagnostic": 122588,
      "dualattention": 34243,
      "crossattention": 24868,
      "conceived": 20551,
      "stimulate": 113344,
      "stylized": 114637,
      "imagination": 51819,
      "infers": 54267,
      "knowledgeenhanced": 58268,
      "grants": 48739,
      "favourable": 43225,
      "pin": 87515,
      "leader": 61878,
      "prefixing": 89258,
      "minimalist": 70182,
      "distort": 32590,
      "gtm": 49207,
      "das": 25546,
      "1975": 451,
      "complexquestions": 19945,
      "al2016": 4658,
      "rotational": 103280,
      "disclose": 32019,
      "1001": 183,
      "afforded": 4262,
      "errorful": 38353,
      "cat": 15685,
      "cmudog": 17977,
      "citing": 16998,
      "edr": 34565,
      "fills": 44066,
      "permutations": 87054,
      "roc": 103138,
      "930": 1404,
      "915": 1389,
      "hardtounderstand": 49496,
      "firstpass": 44571,
      "perpetually": 87056,
      "userspecific": 129186,
      "equips": 38243,
      "onestep": 81720,
      "ropes": 103276,
      "scenariobased": 103837,
      "contextspecific": 22439,
      "economical": 34513,
      "040": 26,
      "knowledgedistillation": 58264,
      "webpage": 132285,
      "urls": 127851,
      "klein": 57711,
      "commensurate": 18796,
      "voter": 131948,
      "committed": 18849,
      "adviceseeking": 4200,
      "dictate": 30852,
      "clts": 17909,
      "problemspecific": 91384,
      "routines": 103321,
      "staple": 112389,
      "reviewing": 102593,
      "repo": 98972,
      "overparameterization": 83351,
      "underutilization": 127038,
      "rejuvenated": 97838,
      "aspectcategory": 9360,
      "acsa": 3055,
      "categoryspecific": 15798,
      "neuronlevel": 78744,
      "objectively": 81129,
      "behavioural": 12682,
      "energybased": 36998,
      "gimpel": 47968,
      "seeded": 104510,
      "invent": 56683,
      "algo": 4684,
      "bf": 13777,
      "kd": 57514,
      "layerlevel": 61756,
      "12layer": 276,
      "hinton": 50505,
      "subpar": 114797,
      "qualityspeed": 94833,
      "openie": 81991,
      "123": 268,
      "pts": 94229,
      "incompetent": 53418,
      "deliberately": 28606,
      "offenses": 81536,
      "species": 111403,
      "datasetspecific": 27805,
      "ntcir15": 80816,
      "bed": 12627,
      "7485": 1246,
      "incongruence": 53430,
      "lengthiness": 63386,
      "hotpotqa": 50672,
      "fullwiki": 45938,
      "played": 87710,
      "summation": 115655,
      "queryaware": 94995,
      "lample": 58795,
      "persuasive": 87184,
      "fastgrowing": 43200,
      "nas": 76204,
      "microlevel": 70056,
      "mrpc": 74934,
      "treebanked": 125655,
      "filler": 44027,
      "bleurt": 14349,
      "iobes": 56923,
      "programmatically": 92121,
      "ethics": 38751,
      "autonomy": 10855,
      "consent": 21171,
      "firm": 44561,
      "investment": 56871,
      "employment": 36325,
      "excess": 39937,
      "congress": 21060,
      "751": 1254,
      "facilities": 42811,
      "trip": 125804,
      "lacked": 58769,
      "viceversa": 131649,
      "mismatching": 70312,
      "144": 330,
      "highrecall": 50419,
      "illness": 51733,
      "clinicians": 17798,
      "seriously": 107286,
      "ill": 51726,
      "amidst": 5317,
      "foodborne": 45000,
      "meter": 68576,
      "takeaway": 117503,
      "wmt2020": 132874,
      "wmt20": 132864,
      "medline": 68248,
      "englishtospanish": 37458,
      "secondbest": 104472,
      "56000": 1122,
      "nonexpert": 80201,
      "elicit": 35290,
      "specializing": 111397,
      "humanelicited": 51019,
      "hurdle": 51161,
      "nearperfect": 76739,
      "factually": 42940,
      "localize": 64946,
      "nonfactual": 80206,
      "decomposes": 28155,
      "arc": 8596,
      "localizing": 64949,
      "lays": 61824,
      "underpinnings": 126720,
      "arose": 9021,
      "longerterm": 65167,
      "reinforcing": 97832,
      "inspect": 55550,
      "cvc": 25471,
      "multibranch": 75037,
      "backbones": 11293,
      "multichoice": 75046,
      "spanextraction": 111145,
      "rectifying": 97152,
      "2022": 744,
      "elephant": 35289,
      "conveniently": 22873,
      "coincidence": 18342,
      "perceiving": 85916,
      "sorani": 110678,
      "frequentlyoccurring": 45882,
      "obscured": 81159,
      "veracity": 131510,
      "loses": 65239,
      "supervise": 115721,
      "tatoeba": 119650,
      "ng": 79424,
      "tl": 122227,
      "theres": 121836,
      "awdlstm": 11265,
      "na": 76000,
      "papel": 83693,
      "mag": 66370,
      "harassment": 49447,
      "exemplifying": 39983,
      "discounted": 32033,
      "feminist": 43876,
      "racialethnic": 95428,
      "justice": 57476,
      "xlwic": 135213,
      "wordincontext": 133717,
      "wic": 132485,
      "reformulating": 97636,
      "southeast": 110965,
      "asia": 9286,
      "america": 5306,
      "tunisian": 125945,
      "tokenizing": 122285,
      "rescale": 100384,
      "rescaled": 100385,
      "clip": 17800,
      "innovations": 55282,
      "ablations": 1586,
      "averages": 11225,
      "combinators": 18609,
      "biencoders": 13987,
      "crossencoders": 24909,
      "crossencoder": 24908,
      "biencoder": 13986,
      "negated": 76908,
      "speculative": 111638,
      "professors": 92093,
      "wake": 131992,
      "timesup": 122204,
      "coda": 18067,
      "momentum": 74539,
      "robertalarge": 103006,
      "humantohuman": 51122,
      "050": 34,
      "ape": 6770,
      "documentgraph": 33128,
      "outdomain": 82620,
      "negra": 76980,
      "tiger": 121964,
      "subjecting": 114689,
      "lgbt": 63949,
      "connotations": 21156,
      "surfacing": 116084,
      "signs": 109058,
      "lrp": 65583,
      "conservation": 21185,
      "scholarship": 103963,
      "intermingled": 56154,
      "adaptability": 3201,
      "timestamp": 122199,
      "jointlearning": 57333,
      "clusteringbased": 17961,
      "flawed": 44641,
      "lstmattention": 65677,
      "court": 24468,
      "disadvantage": 31930,
      "emulate": 36343,
      "rewarding": 102680,
      "i2b22010": 51295,
      "tydi": 126179,
      "40k": 1012,
      "informationseeking": 55118,
      "mt5": 75009,
      "checkpoints": 16702,
      "storm": 113404,
      "countless": 24446,
      "warmup": 132030,
      "warmstart": 132024,
      "inthe": 56331,
      "semiautoregressive": 105577,
      "autoregressively": 10872,
      "height": 49677,
      "logarithmic": 64984,
      "spider": 111929,
      "711": 1220,
      "695": 1200,
      "finishing": 44531,
      "opposition": 82154,
      "debated": 27890,
      "abortion": 1708,
      "tam": 117566,
      "understandability": 126791,
      "postpretraining": 88504,
      "gppl": 48550,
      "verbalize": 131540,
      "lama": 58788,
      "outputting": 83191,
      "unreasonably": 127499,
      "lince": 64322,
      "dialogpt": 30620,
      "neurosymbolic": 78751,
      "durations": 34275,
      "matres": 67467,
      "socially": 110462,
      "originals": 82574,
      "codeswitch": 18214,
      "agentcentric": 4316,
      "usercentric": 129062,
      "factoid": 42849,
      "lessstudied": 63403,
      "rationalize": 95844,
      "rationalizing": 95845,
      "receivers": 96379,
      "constitutional": 21562,
      "multitype": 75929,
      "reconciles": 97102,
      "parentchild": 84857,
      "topperforming": 122689,
      "discretized": 32186,
      "arrangement": 9024,
      "homonymous": 50620,
      "bat": 12554,
      "superglue": 115673,
      "usd": 127875,
      "sustainable": 116223,
      "eae": 34372,
      "lighter": 64003,
      "ctb": 25190,
      "wnut2020": 132884,
      "wet": 132461,
      "pixel": 87596,
      "intend": 55864,
      "systemlevel": 116704,
      "estimator": 38675,
      "whos": 132480,
      "uploading": 127817,
      "multiloss": 75411,
      "attenuation": 10133,
      "rumors": 103458,
      "economy": 34515,
      "bilateral": 14015,
      "seo": 106842,
      "intelligible": 55862,
      "smrt": 110273,
      "introspect": 56659,
      "conditionally": 20795,
      "cam": 15077,
      "localness": 64953,
      "neighbour": 77000,
      "corroborates": 24323,
      "modelspecific": 74371,
      "welldeveloped": 132403,
      "deeplearningbased": 28461,
      "expository": 41691,
      "multiview": 75933,
      "circumventing": 16987,
      "quartet": 94907,
      "nascent": 76205,
      "sublayer": 114719,
      "dictates": 30854,
      "077": 67,
      "lastlayer": 61555,
      "glu": 48307,
      "contexthypothesis": 22368,
      "passagelevel": 85613,
      "educated": 34567,
      "trustworthiness": 125875,
      "weighing": 132319,
      "agnews": 4378,
      "gan": 46428,
      "heterogeneity": 49848,
      "8379": 1315,
      "lowerlevel": 65456,
      "contributors": 22799,
      "4800": 1052,
      "severity": 107887,
      "chatroom": 16683,
      "scaffolding": 103678,
      "135k": 303,
      "133k": 301,
      "res": 100382,
      "bli": 14351,
      "angles": 6113,
      "tended": 120165,
      "quoted": 95416,
      "unedited": 127050,
      "consecutively": 21167,
      "antagonistic": 6750,
      "dominating": 33939,
      "hibert": 49890,
      "intertwined": 56319,
      "retrospective": 102477,
      "formulations": 45297,
      "4050": 1011,
      "161": 383,
      "redesigning": 97307,
      "multivocab": 75937,
      "citedevlin2018bert": 16996,
      "char": 16414,
      "lights": 64007,
      "solidly": 110543,
      "inuktitut": 56674,
      "uninformed": 127170,
      "neighbourhoods": 77002,
      "semeval2010": 105521,
      "multimodality": 75465,
      "alibaba": 4895,
      "effortless": 35185,
      "providers": 93994,
      "9135": 1386,
      "traversals": 125545,
      "elimination": 35302,
      "obscures": 81160,
      "doubts": 33975,
      "paperwe": 84515,
      "tutoring": 126002,
      "literacy": 64727,
      "ebooks": 34498,
      "moderator": 74379,
      "tks": 122226,
      "lagging": 58782,
      "cf": 15967,
      "selector": 104857,
      "editable": 34537,
      "quantifiable": 94835,
      "technologists": 120029,
      "unanswered": 126591,
      "repurpose": 100095,
      "373": 974,
      "nuggets": 80831,
      "classroom": 17684,
      "truncated": 125867,
      "sizeable": 109955,
      "stone": 113369,
      "intercultural": 56052,
      "wordtrigger": 134343,
      "6300": 1168,
      "cui": 25222,
      "drafts": 34084,
      "precludes": 88810,
      "harnesses": 49513,
      "sanity": 103628,
      "vernacular": 131597,
      "scribal": 104249,
      "camps": 15092,
      "cafiero": 15013,
      "412": 1015,
      "resourceintensive": 100898,
      "7645": 1261,
      "operationalized": 82064,
      "passes": 85623,
      "prioritized": 90773,
      "recycle": 97291,
      "minimises": 70190,
      "multisense": 75775,
      "617": 1162,
      "464": 1045,
      "documentaligned": 33116,
      "sl": 110003,
      "dwe": 34286,
      "enlightened": 37551,
      "mediaeval": 68186,
      "spreaders": 112052,
      "606": 1156,
      "666": 1188,
      "indoaryan": 53949,
      "2class": 873,
      "religious": 98669,
      "watching": 132044,
      "locationspecific": 64976,
      "cs": 25182,
      "contract": 22662,
      "blanks": 14266,
      "precompiled": 88811,
      "impair": 51907,
      "9090": 1379,
      "alerts": 4667,
      "perceives": 85915,
      "pedagogy": 85842,
      "perceptually": 85935,
      "misalignment": 70288,
      "confirmation": 21021,
      "clicks": 17751,
      "instantly": 55650,
      "noticeably": 80400,
      "1a": 475,
      "1b": 476,
      "qr": 94538,
      "friction": 45886,
      "casts": 15681,
      "locales": 64943,
      "nonhuman": 80213,
      "modelgenerated": 72367,
      "865": 1332,
      "939": 1411,
      "21000": 765,
      "phases": 87214,
      "tokenisation": 122274,
      "lemmatisation": 63334,
      "8x": 1363,
      "sublayers": 114720,
      "newstest2014": 79416,
      "embarrassingly": 35344,
      "cls": 17904,
      "trustworthy": 125877,
      "2layer": 885,
      "reranked": 100362,
      "akin": 4587,
      "mkqa": 70430,
      "colleagues": 18369,
      "metaphors": 68566,
      "complemented": 19756,
      "cotrained": 24407,
      "278": 861,
      "338": 951,
      "exiting": 40351,
      "emitted": 36081,
      "cascading": 15574,
      "4times": 1067,
      "programmer": 92124,
      "inconvenience": 53446,
      "singledomain": 109825,
      "flavors": 44640,
      "devanagari": 30168,
      "poems": 87780,
      "awadhi": 11251,
      "braj": 14556,
      "collated": 18368,
      "odqa": 81533,
      "voiceassistants": 131926,
      "erroneously": 38274,
      "caution": 15843,
      "ptlms": 94226,
      "ptlm": 94225,
      "onpar": 81821,
      "realizer": 96123,
      "javascript": 57221,
      "specifics": 111619,
      "noiseinvariant": 80082,
      "augmentations": 10319,
      "fabricate": 42725,
      "araelectra": 8575,
      "replacements": 98943,
      "146": 332,
      "mega": 68262,
      "on2": 81661,
      "linkages": 64641,
      "trie": 125772,
      "attended": 9782,
      "xsum": 135220,
      "524": 1099,
      "639": 1175,
      "opus100": 82253,
      "conflating": 21031,
      "multilingually": 75410,
      "longdocument": 65153,
      "quadratically": 94543,
      "truncating": 125868,
      "incur": 53751,
      "documentaware": 33124,
      "agendas": 4292,
      "composer": 20076,
      "infilling": 54270,
      "bartbased": 11441,
      "temporality": 120132,
      "delete": 28592,
      "abend": 1474,
      "rappoport": 95734,
      "exposition": 41690,
      "driver": 34157,
      "harmful": 49504,
      "165": 388,
      "prospective": 93655,
      "geared": 46571,
      "multitext": 75903,
      "substructure": 114933,
      "substituting": 114927,
      "marbert": 67174,
      "gamification": 46425,
      "twoway": 126178,
      "falsely": 43045,
      "fabricated": 42726,
      "lowend": 65418,
      "utilises": 130471,
      "nwp": 81037,
      "sorted": 110683,
      "recentlyintroduced": 96782,
      "causally": 15817,
      "substantively": 114923,
      "constraintaaai2021": 21588,
      "acute": 3156,
      "urgently": 127848,
      "094": 110,
      "incarnation": 53211,
      "killing": 57670,
      "incontext": 53445,
      "semanticallysimilar": 105388,
      "taskrelated": 118908,
      "nq": 80805,
      "transformersbased": 124472,
      "fullsize": 45935,
      "substituted": 114925,
      "marry": 67276,
      "awesome": 11267,
      "1d": 478,
      "dstc9": 34220,
      "autogenerated": 10422,
      "counteract": 24423,
      "dangers": 25533,
      "bbpe": 12592,
      "englishspeaking": 37431,
      "uralic": 127835,
      "emissions": 36078,
      "inconclusive": 53429,
      "ote": 82608,
      "t2": 117247,
      "1100": 233,
      "backtranslating": 11328,
      "animacy": 6116,
      "focal": 44732,
      "mps": 74921,
      "ba": 11274,
      "fillergap": 44028,
      "5grams": 1141,
      "lags": 58785,
      "965": 1435,
      "rural": 103483,
      "gat": 46497,
      "holdout": 50593,
      "nonsymbolic": 80298,
      "nmn": 79818,
      "genbert": 46577,
      "showcases": 108364,
      "enter": 37690,
      "humanity": 51044,
      "infodemic": 54332,
      "thirdparty": 121869,
      "docker": 32941,
      "adapterbased": 3298,
      "cdcr": 15875,
      "supplied": 115949,
      "peer": 85843,
      "systemgenerated": 116702,
      "917": 1391,
      "925": 1400,
      "holders": 50590,
      "holder": 50589,
      "bio": 14146,
      "hampering": 49310,
      "california": 15048,
      "1925": 441,
      "digitally": 31730,
      "header": 49585,
      "kaggle": 57485,
      "leakage": 61979,
      "protests": 93671,
      "semiautomated": 105566,
      "civic": 17001,
      "threefold": 121937,
      "494": 1055,
      "328": 938,
      "returning": 102482,
      "pseudotext": 94197,
      "waveform": 132049,
      "wav2vec": 132046,
      "hubert": 50711,
      "taskdependent": 118874,
      "320000": 934,
      "balances": 11405,
      "bipartite": 14223,
      "frontiers": 45901,
      "contrastively": 22743,
      "041": 27,
      "bestreported": 13483,
      "track1": 122741,
      "cardinal": 15501,
      "cv": 25468,
      "guess": 49223,
      "193": 442,
      "582": 1131,
      "bear": 12614,
      "resemblance": 100725,
      "probed": 90880,
      "traction": 122771,
      "fixeddimensional": 44612,
      "languagemodels": 60374,
      "delineate": 28614,
      "eschewing": 38421,
      "ninth": 79490,
      "r1": 95420,
      "cqr": 24583,
      "bodies": 14392,
      "selfexplanatory": 104920,
      "skeletons": 109963,
      "catchy": 15717,
      "excessive": 39938,
      "socialmedia": 110464,
      "hinglish": 50498,
      "093": 108,
      "codebase": 18163,
      "wordstructure": 134338,
      "as2": 9279,
      "mixedlanguage": 70411,
      "rcnn": 95879,
      "wordrepresentations": 133796,
      "assurance": 9670,
      "protected": 93659,
      "specialties": 111401,
      "diktnet": 31735,
      "frisiandutch": 45894,
      "fame": 43048,
      "frisian": 45893,
      "scraped": 104237,
      "textrich": 121447,
      "683": 1193,
      "avenue": 11164,
      "cbt": 15851,
      "indicbert": 53883,
      "englishmalayalam": 37420,
      "lstmcnn": 65690,
      "domainoriented": 33719,
      "699": 1204,
      "9001": 1373,
      "subjectarea": 114687,
      "mcauley": 67578,
      "114": 239,
      "typetoken": 126391,
      "swear": 116259,
      "laughter": 61677,
      "risks": 102881,
      "shapley": 107945,
      "aaai2021": 1458,
      "0938": 109,
      "tse": 125908,
      "colorless": 18533,
      "syntaxrelated": 116588,
      "opponent": 82137,
      "manifests": 66923,
      "hateful": 49544,
      "join": 57247,
      "joe": 57242,
      "democratic": 28645,
      "donald": 33940,
      "somebody": 110656,
      "humanrobot": 51065,
      "breaches": 14574,
      "cooperating": 23253,
      "400k": 1010,
      "201": 560,
      "continually": 22601,
      "activemisleading": 3125,
      "httpsgithubcomdeeplearnxmupssattention": 50702,
      "companion": 19119,
      "subtree": 115008,
      "rarer": 95767,
      "disconnect": 32023,
      "humancentric": 51002,
      "wordlike": 133760,
      "170": 403,
      "finedtuned": 44330,
      "3way": 1001,
      "098": 115,
      "1976": 452,
      "infill": 54269,
      "distinguishability": 32584,
      "deserves": 29509,
      "obviate": 81482,
      "ami": 5315,
      "proceeding": 91411,
      "gps": 48551,
      "eschews": 38422,
      "jumps": 57456,
      "wastes": 132040,
      "causeeffect": 15834,
      "semmeddb": 105638,
      "texttosql": 121664,
      "anaphors": 6077,
      "anaphor": 6060,
      "admissible": 3886,
      "singlestage": 109841,
      "noising": 80087,
      "domainaware": 33708,
      "conscientious": 21160,
      "delineating": 28616,
      "speechtospeech": 111856,
      "xtreme": 135221,
      "mtl": 75012,
      "converter": 23102,
      "overstatement": 83357,
      "semeval2021": 105555,
      "paramount": 84801,
      "triggering": 125790,
      "shrinking": 108645,
      "unsatisfying": 127514,
      "argmax": 8905,
      "shrink": 108643,
      "entmax": 38152,
      "1369": 304,
      "observes": 81243,
      "shaped": 107940,
      "listed": 64704,
      "unesco": 127053,
      "nonavailability": 80164,
      "impairs": 51910,
      "bundle": 14964,
      "hull": 50731,
      "blue": 14385,
      "morphologies": 74781,
      "demarcations": 28635,
      "satirical": 103647,
      "timeframe": 122153,
      "sarcastic": 103644,
      "unfavorable": 127068,
      "willing": 132708,
      "apology": 6782,
      "340k": 955,
      "reasoners": 96220,
      "skolt": 110001,
      "wellaligned": 132391,
      "discoursebased": 32107,
      "edus": 34576,
      "rivaling": 102884,
      "neuroscience": 78750,
      "peaks": 85834,
      "evoked": 39688,
      "onset": 81822,
      "beta": 13495,
      "bands": 11413,
      "threat": 121934,
      "attacking": 9709,
      "attackers": 9708,
      "poisoning": 87894,
      "sacrificed": 103523,
      "dubious": 34251,
      "throw": 121950,
      "somali": 110655,
      "swahili": 116252,
      "normalizer": 80360,
      "hyperlinks": 51213,
      "wellformedness": 132413,
      "excitement": 39950,
      "conceal": 20547,
      "anonymization": 6511,
      "ip": 56925,
      "episodes": 38212,
      "lift": 63983,
      "25x": 836,
      "recall10": 96345,
      "725": 1228,
      "rote": 103282,
      "consonant": 21508,
      "refuted": 97646,
      "nonneutral": 80257,
      "multiunit": 75930,
      "borders": 14467,
      "scibert": 103971,
      "shrinkage": 108644,
      "sexism": 107889,
      "referent": 97562,
      "shopping": 108197,
      "cart": 15564,
      "incident": 53214,
      "069": 54,
      "065": 49,
      "ats": 9688,
      "commandline": 18791,
      "contracts": 22664,
      "biasing": 13891,
      "spawned": 111275,
      "topscoring": 122693,
      "aggregators": 4372,
      "8th": 1360,
      "prlms": 90791,
      "fun": 45994,
      "multiplayer": 75486,
      "corruption": 24330,
      "anli": 6121,
      "artefacts": 9105,
      "coarsegrain": 18053,
      "subjectively": 114703,
      "isotropic": 56985,
      "safeguard": 103535,
      "20newsgroups": 753,
      "contradicts": 22673,
      "resourcescarce": 101070,
      "biomedicine": 14216,
      "inaccessible": 53190,
      "pimentel": 87512,
      "quarter": 94906,
      "mclwic": 67583,
      "enen": 36995,
      "927": 1402,
      "pmibased": 87773,
      "cutting": 25465,
      "remainder": 98776,
      "931": 1405,
      "failings": 42979,
      "affiliations": 4253,
      "lucene": 65722,
      "kendalls": 57525,
      "tau": 119651,
      "peerreview": 85845,
      "iclr": 51307,
      "recommending": 97098,
      "bottlenecked": 14499,
      "imputing": 53187,
      "topv2": 122695,
      "indigenous": 53885,
      "cnndm": 18036,
      "samsum": 103620,
      "logicalform": 65022,
      "referenceless": 97553,
      "questeval": 95011,
      "swapping": 116255,
      "delineated": 28615,
      "pause": 85800,
      "sun": 115666,
      "lightly": 64005,
      "pku": 87597,
      "weave": 132217,
      "acknowledge": 2990,
      "trades": 122785,
      "spoken dialog": 111975,
      "order logic": 82354,
      "coherence relation": 18308,
      "trees represent": 125708,
      "assumptions provide": 9668,
      "provide basis": 93765,
      "speech acts": 111642,
      "coherence relations": 18309,
      "prove useful": 93706,
      "useful determining": 128875,
      "finally sketch": 44234,
      "cue phrases": 25204,
      "dialog model": 30572,
      "lexicalized tree": 63852,
      "tree adjoining": 125575,
      "adjoining grammar": 3866,
      "grammar english": 48631,
      "english document": 37117,
      "document describes": 32986,
      "english written": 37353,
      "technical report": 119751,
      "english grammar": 37161,
      "feature structures": 43319,
      "range syntactic": 95606,
      "syntactic phenomena": 116449,
      "clause constructions": 17689,
      "relative clauses": 98345,
      "addition new": 3453,
      "new analyses": 78770,
      "old ones": 81644,
      "web page": 132241,
      "dependency constituency": 29134,
      "constituency tree": 21531,
      "rewriting paper": 102694,
      "paper present": 84075,
      "present new": 89578,
      "tags using": 117488,
      "linguistic dependency": 64463,
      "gives better": 48183,
      "representation compared": 99191,
      "compared previous": 19412,
      "previous approaches": 90380,
      "approaches use": 8390,
      "incremental parser": 53737,
      "adjoining grammars": 3868,
      "grammars paper": 48679,
      "paper describes": 83808,
      "incremental generation": 53728,
      "parsing tree": 85269,
      "adjoining languages": 3870,
      "algorithm presented": 4786,
      "parser generated": 84960,
      "far paper": 43094,
      "created need": 24676,
      "parsing incremental": 85130,
      "built far": 14923,
      "unknown words": 127365,
      "lexical acquisition": 63727,
      "presented paper": 89791,
      "linguistic properties": 64531,
      "words inferred": 134001,
      "surrounding context": 116155,
      "context parsing": 22215,
      "parsing sentences": 85229,
      "hpsg grammar": 50693,
      "grammar german": 48636,
      "lexical entries": 63761,
      "entries unknown": 38158,
      "takes place": 117537,
      "parsing requires": 85218,
      "pieces information": 87498,
      "access paths": 1920,
      "context actually": 21998,
      "provides new": 94056,
      "new information": 78956,
      "information type": 55063,
      "paper argues": 83740,
      "representation explicitly": 99235,
      "explicitly represent": 41391,
      "meaning situation": 67694,
      "analysis generation": 5632,
      "translation furthermore": 124833,
      "furthermore representation": 46211,
      "representation meaning": 99324,
      "meaning words": 67707,
      "words especially": 133941,
      "words convey": 133887,
      "choosing word": 16936,
      "context using": 22316,
      "using lexical": 129808,
      "network paper": 77380,
      "paper presents": 84148,
      "problem lexical": 91106,
      "lexical choice": 63737,
      "context apply": 22008,
      "apply new": 7200,
      "new statistical": 79183,
      "statistical approach": 113080,
      "approach representing": 7858,
      "representing context": 100043,
      "context word": 22327,
      "cooccurrence networks": 23230,
      "trained evaluated": 123140,
      "evaluated large": 38984,
      "large corpus": 61061,
      "corpus results": 23977,
      "results inclusion": 101849,
      "secondorder cooccurrence": 104478,
      "improves performance": 53006,
      "performance implemented": 86437,
      "syntactic structure": 116481,
      "structure language": 113890,
      "language modeling": 59423,
      "modeling paper": 72501,
      "language model": 59283,
      "meaningful information": 67720,
      "information word": 55095,
      "word history": 133309,
      "use long": 128130,
      "long distance": 65068,
      "distance dependencies": 32410,
      "dependencies model": 29096,
      "model assigns": 70693,
      "joint sequence": 57322,
      "lefttoright manner": 63309,
      "automatic speech": 10655,
      "speech recognition": 111734,
      "recognition model": 96909,
      "set experiments": 107440,
      "predictive power": 89207,
      "improvement standard": 52762,
      "model paper": 71690,
      "sequence words": 107100,
      "annotation model": 6339,
      "shallow approach": 107913,
      "approach text": 7971,
      "text generation": 120976,
      "generation order": 47523,
      "order support": 82414,
      "support efficient": 115966,
      "efficient development": 35068,
      "generation systems": 47652,
      "methods currently": 69416,
      "linguistically motivated": 64592,
      "surface realization": 116072,
      "techniques paper": 119945,
      "paper argue": 83739,
      "order improve": 82339,
      "suggest evaluate": 115311,
      "generation methods": 47473,
      "associated increased": 9596,
      "supports quick": 116037,
      "quick adaptation": 95384,
      "adaptation new": 3241,
      "new tasks": 79208,
      "tasks domains": 119074,
      "resources method": 101012,
      "method especially": 68809,
      "especially designed": 38448,
      "designed generating": 29595,
      "generating reports": 47255,
      "limited linguistic": 64251,
      "explaining word": 41279,
      "word meaning": 133359,
      "lexicon word": 63914,
      "semantic rules": 105262,
      "novel features": 80569,
      "artificial intelligence": 9246,
      "assigned words": 9552,
      "words occur": 134089,
      "occur close": 81496,
      "systems using": 117223,
      "using rules": 130132,
      "computational systems": 20436,
      "line argument": 64325,
      "grammar induction": 48638,
      "induction using": 54039,
      "using training": 130303,
      "training data": 123417,
      "data limited": 26090,
      "constituent information": 21538,
      "data learn": 26082,
      "learn structure": 62158,
      "cost building": 24347,
      "building large": 14855,
      "large annotated": 61029,
      "annotated corpora": 6143,
      "prohibitively expensive": 92202,
      "expensive work": 40438,
      "work aims": 134366,
      "aims improve": 4540,
      "labels training": 58650,
      "parse trees": 84890,
      "complex noun": 19851,
      "noun phrases": 80430,
      "inducing grammars": 54016,
      "labeled training": 58473,
      "constituent labels": 21539,
      "labels propose": 58632,
      "propose adaptation": 92527,
      "adaptation strategy": 3270,
      "fully labeled": 45964,
      "labeled corpora": 58427,
      "corpora results": 23578,
      "results suggest": 102227,
      "human annotators": 50750,
      "able automatically": 1609,
      "automatically extract": 10767,
      "syntactic processing": 116452,
      "spoken language": 111988,
      "japanese language": 57193,
      "arguments adjuncts": 8976,
      "subject object": 114680,
      "verbal arguments": 131532,
      "make analysis": 66621,
      "analysis japanese": 5665,
      "japanese sentences": 57199,
      "based empirical": 11672,
      "empirical investigation": 36172,
      "set type": 107620,
      "type hierarchy": 126203,
      "markov models": 67265,
      "models paper": 73696,
      "presents new": 89872,
      "new approach": 78784,
      "structures approach": 114058,
      "approach based": 7377,
      "based markov": 11828,
      "models layer": 73473,
      "structure represented": 113944,
      "markov model": 67259,
      "model output": 71682,
      "lower layer": 65435,
      "empirical evaluation": 36158,
      "evaluation method": 39266,
      "method yields": 69232,
      "good results": 48499,
      "newspaper texts": 79407,
      "texts evaluation": 121510,
      "evaluation nlp": 39310,
      "nlp components": 79590,
      "spoken dialogue": 111978,
      "programme language": 92123,
      "language speech": 60115,
      "speech technology": 111823,
      "development spoken": 30427,
      "language information": 59205,
      "information systems": 55025,
      "natural language": 76263,
      "language processing": 59832,
      "processing nlp": 91725,
      "nlp modules": 79647,
      "order compare": 82297,
      "formal evaluation": 45158,
      "evaluation carried": 39142,
      "evaluation procedure": 39334,
      "evaluation results": 39365,
      "performs better": 86984,
      "learning transformation": 63124,
      "transformation rules": 124272,
      "grammatical relations": 48716,
      "grammatical relationships": 48717,
      "important level": 52178,
      "processing present": 91774,
      "errordriven learning": 38351,
      "learning approach": 62337,
      "training test": 123912,
      "test set": 120492,
      "procedure achieves": 91389,
      "hierarchies using": 50024,
      "labeling paper": 58519,
      "paper explores": 83929,
      "automatic construction": 10503,
      "lexical knowledge": 63776,
      "knowledge base": 57762,
      "lexical resources": 63800,
      "resources present": 101028,
      "new robust": 79113,
      "robust approach": 103016,
      "approach linking": 7690,
      "linking existing": 64658,
      "existing lexicalsemantic": 40160,
      "lexicalsemantic hierarchies": 63862,
      "constraint satisfaction": 21583,
      "satisfaction algorithm": 103649,
      "algorithm relaxation": 4791,
      "select candidate": 104693,
      "candidate translations": 15129,
      "proposed bilingual": 93238,
      "bilingual dictionary": 14032,
      "english wordnet": 37348,
      "wordnet synset": 133780,
      "automatically derived": 10751,
      "spanish monolingual": 111163,
      "average 15": 11173,
      "method achieves": 68590,
      "achieves accuracy": 2733,
      "accuracy 80": 2073,
      "finally propose": 44221,
      "propose ways": 93159,
      "technique applied": 119768,
      "improve existing": 52382,
      "existing lexical": 40159,
      "lexical databases": 63755,
      "approach machine": 7698,
      "machine translation": 65879,
      "translation present": 125129,
      "present approach": 89371,
      "combines ideas": 18686,
      "approach implemented": 7616,
      "multilingual machine": 75274,
      "annotation graphs": 6320,
      "linguistic data": 64458,
      "data analysis": 25596,
      "analysis recent": 5795,
      "recent work": 96566,
      "work presented": 134690,
      "formal framework": 45159,
      "framework linguistic": 45602,
      "linguistic annotation": 64422,
      "annotation based": 6281,
      "based labeled": 11790,
      "offer simple": 81572,
      "simple powerful": 109493,
      "powerful method": 88671,
      "method representing": 69104,
      "representing complex": 100042,
      "complex annotation": 19798,
      "illustrate approach": 51737,
      "approach using": 8005,
      "text speech": 121322,
      "speech data": 111667,
      "data drawn": 25869,
      "annotation schemes": 6371,
      "multilevel annotation": 75189,
      "speech corpus": 111665,
      "corpus includes": 23835,
      "segment word": 104567,
      "coreference named": 23353,
      "named entity": 76041,
      "diverse set": 32843,
      "file formats": 44020,
      "approach facilitates": 7568,
      "comparison multiple": 19557,
      "multiple annotations": 75490,
      "signal based": 108682,
      "based different": 11646,
      "different theoretical": 31494,
      "theoretical models": 121797,
      "models discussion": 73091,
      "open door": 81902,
      "representing text": 100065,
      "text chunks": 120705,
      "chunks words": 16971,
      "words useful": 134293,
      "preprocessing step": 89334,
      "parsing information": 85132,
      "information extraction": 54563,
      "extraction information": 42357,
      "information retrieval": 54929,
      "data representation": 26339,
      "tagging task": 117453,
      "task paper": 118499,
      "paper examine": 83900,
      "seven different": 107870,
      "different data": 31073,
      "data representations": 26340,
      "problem recognizing": 91200,
      "noun phrase": 80427,
      "influence chunking": 54303,
      "suitable data": 115394,
      "memorybased learning": 68403,
      "able improve": 1652,
      "improve best": 52343,
      "best published": 13423,
      "chunking results": 16965,
      "results standard": 102200,
      "standard data": 112219,
      "data set": 26397,
      "learning machine": 62699,
      "translation paper": 125085,
      "introduce techniques": 56552,
      "techniques use": 120006,
      "increase coverage": 53592,
      "maintaining high": 66517,
      "high level": 50082,
      "space time": 111071,
      "time efficiency": 122019,
      "efficiency performance": 35037,
      "performance results": 86679,
      "results indicate": 101854,
      "indicate approach": 53832,
      "approach effective": 7513,
      "parsing propose": 85201,
      "combines advantages": 18677,
      "control parser": 22817,
      "allows user": 5196,
      "state art": 112423,
      "processing techniques": 91834,
      "techniques used": 120007,
      "used deal": 128473,
      "discuss various": 32292,
      "various aspects": 131044,
      "development corpus": 30375,
      "corpus annotation": 23662,
      "annotation parser": 6347,
      "parser evaluation": 84951,
      "recently developed": 96679,
      "developed corpus": 30263,
      "annotation scheme": 6368,
      "shortcomings current": 108281,
      "current methods": 25298,
      "corpus naturally": 23894,
      "naturally occurring": 76646,
      "english text": 37307,
      "text corpus": 120834,
      "corpus used": 24056,
      "used evaluate": 128523,
      "evaluate accuracy": 38783,
      "accuracy robust": 2266,
      "approach automatically": 7374,
      "automatically generating": 10781,
      "lexical transfer": 63834,
      "transfer rules": 124183,
      "method automatically": 68652,
      "using transfer": 130309,
      "transfer rule": 124182,
      "rule templates": 103355,
      "words new": 134077,
      "words provided": 134150,
      "provided words": 93993,
      "lexical categories": 63734,
      "categories required": 15750,
      "methods creating": 69412,
      "templates using": 120088,
      "generate new": 46975,
      "method consists": 68732,
      "finite set": 44537,
      "set templates": 107603,
      "directly using": 31919,
      "generation process": 47564,
      "process method": 91526,
      "higher level": 50186,
      "defined terms": 28504,
      "sets lexical": 107684,
      "set lexical": 107479,
      "parsing input": 85133,
      "input words": 55474,
      "phrasal level": 87325,
      "level using": 63518,
      "using corresponding": 129510,
      "corresponding lexical": 24291,
      "enables automatic": 36377,
      "automatic creation": 10506,
      "conversational systems": 23038,
      "describes speech": 29432,
      "speech speech": 111807,
      "speech translation": 111831,
      "achieve higher": 2479,
      "higher degree": 50174,
      "degree robustness": 28574,
      "robustness scalability": 103130,
      "data flow": 25951,
      "ability process": 1562,
      "process topdown": 91578,
      "considering signal": 21333,
      "engineering approach": 37044,
      "recognition techniques": 97028,
      "generalization concept": 46772,
      "pass beam": 85604,
      "beam search": 12598,
      "robust parsing": 103059,
      "extraction natural": 42404,
      "language queries": 60014,
      "queries paper": 94931,
      "parsing techniques": 85262,
      "applied building": 7048,
      "query generation": 94965,
      "generation module": 47490,
      "nlp architecture": 79577,
      "architecture aimed": 8614,
      "process natural": 91533,
      "restricted domain": 101339,
      "key issue": 57584,
      "issue nlp": 57013,
      "nlp systems": 79696,
      "utterances various": 130676,
      "various factors": 131096,
      "noisy environments": 80101,
      "low quality": 65380,
      "quality speech": 94794,
      "extracting meaningful": 42220,
      "contextfree languages": 22363,
      "languages methods": 60726,
      "methods discussed": 69454,
      "given contextfree": 48003,
      "contextfree grammar": 22360,
      "grammar including": 48637,
      "methods lead": 69583,
      "language methods": 59280,
      "respect existing": 101077,
      "existing literature": 40163,
      "experiments different": 40906,
      "different methods": 31256,
      "speech recognizer": 111790,
      "sentence using": 106121,
      "using surface": 130250,
      "surface expressions": 116057,
      "expressions examples": 41753,
      "purposes language": 94447,
      "language understanding": 60201,
      "understanding machine": 126882,
      "translation conversational": 124713,
      "processing paper": 91764,
      "verbs using": 131564,
      "using information": 129754,
      "information obtained": 54811,
      "obtained recall": 81400,
      "recall rate": 96342,
      "precision rate": 88797,
      "test sentences": 120491,
      "tense aspect": 120180,
      "developed new": 30291,
      "new method": 79006,
      "method method": 68960,
      "similarity input": 109244,
      "input example": 55334,
      "example sentences": 39794,
      "degree semantic": 28575,
      "semantic matching": 105105,
      "sentences method": 106396,
      "method uses": 69209,
      "method order": 69000,
      "effects noise": 34998,
      "tagged data": 117348,
      "data bilingual": 25705,
      "bilingual corpora": 14027,
      "corpora experiments": 23483,
      "experiments method": 40997,
      "method translate": 69199,
      "tenses aspects": 120184,
      "currently available": 25396,
      "does require": 33389,
      "require handcraft": 100140,
      "representations text": 99931,
      "text passage": 121171,
      "passage retrieval": 85610,
      "meaning natural": 67650,
      "language documents": 58981,
      "documents based": 33191,
      "based standard": 12066,
      "search strategy": 104353,
      "different levels": 31227,
      "levels abstraction": 63526,
      "knowledge representation": 58144,
      "search strategies": 104352,
      "outside nlp": 83199,
      "inference natural": 54171,
      "representation based": 99178,
      "represent meaning": 99120,
      "detailed description": 29764,
      "linguistic aspects": 64428,
      "aspects given": 9386,
      "exploiting syntactic": 41481,
      "structure natural": 113911,
      "presents attempt": 89820,
      "using syntactic": 130255,
      "language improved": 59155,
      "improved language": 52611,
      "language models": 59493,
      "models speech": 74081,
      "recognition structured": 97014,
      "model merges": 71511,
      "techniques automatic": 119840,
      "parsing language": 85138,
      "modeling using": 72585,
      "using original": 129994,
      "shiftreduce parser": 108187,
      "maximum likelihood": 67536,
      "belonging class": 12720,
      "algorithms employed": 4846,
      "training model": 123710,
      "model experiments": 71131,
      "wall street": 132001,
      "street journal": 113562,
      "broadcast news": 14687,
      "news corpora": 79323,
      "improvement perplexity": 52741,
      "perplexity word": 87076,
      "word error": 133267,
      "error rate": 38323,
      "rate word": 95810,
      "word lattice": 133332,
      "lattice rescoring": 61670,
      "approach language": 7661,
      "uses hierarchical": 129232,
      "hierarchical syntactic": 50009,
      "improve current": 52368,
      "modeling techniques": 72560,
      "techniques large": 119915,
      "large vocabulary": 61324,
      "vocabulary speech": 131906,
      "model new": 71578,
      "new language": 78974,
      "model speech": 72076,
      "linguistic analysis": 64421,
      "analysis presented": 5762,
      "hierarchical structure": 50005,
      "incrementally uses": 53749,
      "use extended": 128041,
      "currently used": 25418,
      "ngram markov": 79442,
      "models model": 73574,
      "algorithm model": 4772,
      "model parameters": 71704,
      "parameters set": 84784,
      "recognition performance": 96963,
      "trigram models": 125800,
      "models structured": 74105,
      "twopass speech": 126151,
      "experiments switchboard": 41163,
      "switchboard corpus": 116283,
      "rate conventional": 95789,
      "modeling speech": 72547,
      "experiments wsj": 41212,
      "wsj corpus": 135178,
      "rate wer": 95807,
      "models requirements": 73937,
      "text processing": 121204,
      "processing lexicons": 91697,
      "processing systems": 91807,
      "systems expand": 116886,
      "require larger": 100169,
      "senses word": 105722,
      "word existing": 133277,
      "existing systems": 40307,
      "systems incorporate": 116951,
      "contain information": 21742,
      "information largely": 54715,
      "representing meaning": 100056,
      "developing semantic": 30360,
      "semantic parsers": 105137,
      "concepts requirements": 20641,
      "measure based": 67781,
      "relations new": 98221,
      "new word": 79249,
      "word usage": 133620,
      "measure proposed": 67815,
      "proposed based": 93235,
      "relations allows": 98109,
      "words degree": 133903,
      "problem space": 91239,
      "changes new": 16389,
      "new products": 79087,
      "describes new": 29418,
      "approach classification": 7419,
      "based shallow": 12039,
      "processing machine": 91704,
      "machine learning": 65765,
      "learning techniques": 63092,
      "techniques implemented": 119905,
      "used commercial": 128433,
      "commercial setting": 18839,
      "word rate": 133427,
      "document document": 32994,
      "document despite": 32988,
      "conventional ngram": 22893,
      "ngram language": 79435,
      "models usually": 74307,
      "derived using": 29365,
      "rate paper": 95799,
      "paper investigate": 84012,
      "investigate use": 56822,
      "poisson distribution": 87896,
      "approach estimating": 7538,
      "estimating relative": 38647,
      "relative frequencies": 98353,
      "frequencies words": 45835,
      "words ngrams": 134079,
      "prior information": 90704,
      "considered using": 21309,
      "news task": 79376,
      "task approach": 117886,
      "approach demonstrates": 7478,
      "reduction perplexity": 97455,
      "perplexity 10": 87061,
      "news paper": 79354,
      "paper discusses": 83876,
      "statistical models": 113134,
      "models extracting": 73223,
      "finite state": 44538,
      "models identifying": 73350,
      "proper names": 92417,
      "named entities": 76028,
      "broadcast speech": 14690,
      "speech models": 111709,
      "models presented": 73775,
      "word attribute": 132932,
      "ngram based": 79426,
      "based formulation": 11725,
      "used models": 128637,
      "models task": 74159,
      "task named": 118429,
      "entity identification": 37944,
      "sparse training": 111237,
      "issues related": 57073,
      "experiments reported": 41116,
      "north american": 80369,
      "phrase recognition": 87365,
      "recognition combination": 96841,
      "performance machine": 86511,
      "learning algorithms": 62322,
      "combining output": 18739,
      "output different": 83066,
      "different systems": 31472,
      "systems paper": 117042,
      "paper apply": 83737,
      "generate different": 46928,
      "different classifiers": 31042,
      "classifiers using": 17644,
      "using different": 129567,
      "different representations": 31384,
      "representations data": 99583,
      "data combining": 25745,
      "techniques described": 119866,
      "described van": 29382,
      "best reported": 13426,
      "data sets": 26417,
      "phrases arbitrary": 87407,
      "treebank parser": 125643,
      "techniques applied": 119837,
      "applied natural": 7096,
      "language parsing": 59803,
      "parsing experiments": 85110,
      "experiments using": 41192,
      "using techniques": 130275,
      "statistical parser": 113147,
      "large gain": 61095,
      "corpus size": 23999,
      "size error": 109923,
      "error analysis": 38280,
      "analysis result": 5809,
      "inconsistent annotations": 53441,
      "penn treebank": 85860,
      "semiautomatic method": 105570,
      "method finding": 68840,
      "treebank annotations": 125627,
      "exploiting diversity": 41464,
      "diversity natural": 32885,
      "applying machine": 7251,
      "learning methods": 62734,
      "methods computational": 69387,
      "computational linguistics": 20392,
      "systems problems": 117077,
      "solutions using": 110583,
      "using various": 130353,
      "various techniques": 131217,
      "solutions developed": 110574,
      "observed errors": 81222,
      "thesis concerned": 121848,
      "problem domain": 91017,
      "domain penn": 33609,
      "parsing machine": 85149,
      "learning community": 62445,
      "community provides": 19088,
      "parser output": 84978,
      "output structured": 83127,
      "novel strategies": 80736,
      "combining parsers": 18743,
      "used learning": 128609,
      "parsers developed": 85018,
      "developed intention": 30281,
      "ensemble techniques": 37622,
      "ensemble parsers": 37618,
      "created using": 24689,
      "using underlying": 130334,
      "parser induction": 84963,
      "method producing": 69076,
      "spoken natural": 112014,
      "language dialogue": 58959,
      "dialogue interfaces": 30693,
      "interfaces semiautonomous": 56108,
      "semiautonomous systems": 105576,
      "systems central": 116775,
      "central idea": 15900,
      "transform input": 124252,
      "input speech": 55443,
      "speech signal": 111802,
      "levels representation": 63560,
      "representation corresponding": 99199,
      "linguistic knowledge": 64499,
      "knowledge dialogue": 57859,
      "dialogue knowledge": 30697,
      "knowledge domain": 57878,
      "domain knowledge": 33555,
      "final representation": 44121,
      "executable program": 39963,
      "program simple": 92112,
      "simple scripting": 109513,
      "scripting language": 104265,
      "language equivalent": 59009,
      "translation process": 125145,
      "process input": 91512,
      "outputmetaoutput distinction": 83155,
      "distinction permits": 32555,
      "perspicuous treatment": 87181,
      "diverse topics": 32857,
      "topics including": 122631,
      "including resolution": 53361,
      "user misconceptions": 129012,
      "methods described": 69435,
      "speech interface": 111700,
      "specific goal": 111446,
      "goal mind": 48367,
      "combination main": 18566,
      "main types": 66474,
      "people working": 85904,
      "makes efforts": 66779,
      "efforts deal": 35194,
      "compare ones": 19266,
      "ones work": 81710,
      "work paper": 134665,
      "point view": 87818,
      "grammar does": 48628,
      "does fully": 33353,
      "models developed": 73060,
      "developed based": 30256,
      "pruned model": 94161,
      "model shown": 72010,
      "pruning single": 94170,
      "backoff models": 11314,
      "models relative": 73912,
      "training set": 123845,
      "perplexity leads": 87066,
      "pruning criterion": 94166,
      "original size": 82546,
      "recognition error": 96869,
      "compare approach": 19221,
      "approaches select": 8330,
      "better performance": 13650,
      "surface natural": 116068,
      "language generation": 59066,
      "generation present": 47548,
      "present systems": 89735,
      "corpora systems": 23595,
      "domainspecific semantic": 33915,
      "syntactic dependency": 116392,
      "dependency information": 29147,
      "systems attempt": 116740,
      "produce grammatical": 91897,
      "semantic representation": 105226,
      "baseline uses": 12333,
      "generate phrase": 46987,
      "maximum entropy": 67532,
      "probability models": 90866,
      "generate word": 47043,
      "word phrase": 133403,
      "determine word": 30135,
      "word choice": 132948,
      "choice word": 16912,
      "word order": 133374,
      "phrase present": 87363,
      "present experiments": 89483,
      "travel domain": 125541,
      "optimality theory": 82178,
      "previous work": 90517,
      "new finitestate": 78920,
      "constraints presented": 21609,
      "prince smolensky": 90665,
      "smolensky 1993": 110262,
      "annotations large": 6441,
      "large speech": 61278,
      "speech corpora": 111664,
      "case studies": 15610,
      "studies paper": 114259,
      "discusses challenges": 32306,
      "challenges arise": 16133,
      "annotations case": 6413,
      "telephone conversations": 120065,
      "news switchboard": 79375,
      "various types": 131230,
      "additional annotation": 3490,
      "annotation carried": 6284,
      "group cases": 49134,
      "issues arise": 57037,
      "having different": 49555,
      "different formats": 31151,
      "formats levels": 45219,
      "general framework": 46660,
      "issues addressed": 57036,
      "using structural": 130230,
      "structural information": 113770,
      "information present": 54858,
      "present robust": 89681,
      "select set": 104711,
      "set candidates": 107383,
      "candidates node": 15137,
      "node target": 80026,
      "target taxonomy": 117725,
      "taxonomy bests": 119661,
      "bests matches": 13486,
      "matches node": 67381,
      "node source": 80024,
      "source taxonomy": 110851,
      "particular use": 85462,
      "wordnet 15": 133763,
      "15 wordnet": 354,
      "wordnet 16": 133766,
      "high precision": 50104,
      "ambiguity language": 5284,
      "language identification": 59126,
      "formal language": 45164,
      "language techniques": 60153,
      "dynamical systems": 34337,
      "systems new": 117021,
      "new features": 78917,
      "features needed": 43621,
      "needed distinguish": 76873,
      "distinguish information": 32571,
      "information generated": 54636,
      "generated input": 47087,
      "modelling framework": 72605,
      "learning algorithm": 62318,
      "algorithm proposed": 4789,
      "detection practical": 30023,
      "practical applications": 88701,
      "techniques yield": 120017,
      "accurate results": 2360,
      "promising technique": 92308,
      "early detection": 34387,
      "constraintbased grammars": 21590,
      "generation algorithms": 47305,
      "major linguistic": 66573,
      "knowledge source": 58178,
      "source parsing": 110804,
      "parsing generation": 85117,
      "input semantics": 55421,
      "representations vary": 99972,
      "generation algorithm": 47304,
      "linguistically plausible": 64595,
      "plausible semantic": 87685,
      "semantic analyses": 104975,
      "approaches generation": 8174,
      "source examples": 110757,
      "examples propose": 39867,
      "propose new": 92831,
      "new general": 78933,
      "general approach": 46633,
      "approach explicitly": 7551,
      "algorithm adding": 4690,
      "generation comparing": 47335,
      "form important": 45094,
      "different sets": 31422,
      "useful different": 128878,
      "different purposes": 31365,
      "small training": 110216,
      "training corpus": 123401,
      "annotations small": 6465,
      "corpus compare": 23699,
      "compare systems": 19301,
      "systems use": 117218,
      "use different": 128001,
      "different learning": 31224,
      "effect larger": 34598,
      "larger factor": 61366,
      "english different": 37113,
      "length measure": 63368,
      "better suited": 13733,
      "finding simple": 44284,
      "data help": 25999,
      "learning accurate": 62303,
      "statistical significance": 113161,
      "differences statistical": 30975,
      "significance testing": 108716,
      "metrics like": 69979,
      "recall precision": 96341,
      "commonly used": 18966,
      "detect differences": 29799,
      "exist different": 40020,
      "different techniques": 31482,
      "independence assumption": 53758,
      "identification use": 51455,
      "algorithms task": 4883,
      "task identifying": 118264,
      "processed different": 91597,
      "different combination": 31044,
      "combination methods": 18568,
      "methods outperformed": 69659,
      "outperformed best": 82779,
      "best individual": 13340,
      "majority vote": 66610,
      "systems standard": 117155,
      "managed improve": 66901,
      "published result": 94369,
      "set aspects": 107365,
      "dataoriented parsing": 26694,
      "stateofthe art": 112558,
      "art parsing": 9075,
      "parsing accuracy": 85061,
      "syntactic structures": 116487,
      "structures relevant": 114101,
      "proper linguistic": 92415,
      "model tries": 72237,
      "accuracy enhanced": 2150,
      "probabilistic topdown": 90830,
      "parsing paper": 85180,
      "paper examines": 83904,
      "parsing dynamic": 85105,
      "dynamic programming": 34322,
      "topdown parsing": 122492,
      "partial parses": 85297,
      "fully connected": 45951,
      "trees spanning": 125716,
      "left context": 63298,
      "partial semantic": 85299,
      "semantic interpretation": 105085,
      "parsing approaches": 85073,
      "parsing viable": 85279,
      "nonlocal information": 80236,
      "information improves": 54673,
      "improves parser": 53003,
      "parser accuracy": 84909,
      "substantially improves": 114898,
      "selectional restrictions": 104850,
      "constraints imposed": 21600,
      "linguistic constructions": 64451,
      "constraints interpretation": 21602,
      "despite limitations": 29706,
      "proven useful": 93735,
      "useful natural": 128907,
      "language applications": 58833,
      "applications used": 7034,
      "word sense": 133516,
      "sense disambiguation": 105660,
      "anaphora resolution": 6065,
      "practical value": 88724,
      "explore methods": 41560,
      "methods incorporate": 69547,
      "method employs": 68800,
      "second method": 104428,
      "method particularly": 69041,
      "particularly useful": 85502,
      "development practical": 30411,
      "practical systems": 88716,
      "grammars using": 48686,
      "computational complexity": 20365,
      "reduced training": 97375,
      "subset training": 114840,
      "set results": 107568,
      "results using": 102301,
      "journal corpus": 57408,
      "obtain better": 81263,
      "better estimation": 13575,
      "sample training": 103568,
      "training available": 123368,
      "models gaussian": 73268,
      "gaussian prior": 46548,
      "reduce overfitting": 97347,
      "overlapping features": 83336,
      "makes minimal": 66793,
      "difference performance": 30944,
      "performance approach": 86145,
      "approach applicable": 7356,
      "applicable situations": 6828,
      "large number": 61174,
      "computationally expensive": 20451,
      "statistical parsing": 113149,
      "little attention": 64791,
      "attention paid": 9962,
      "high accuracy": 50032,
      "accuracy statistical": 2289,
      "statistical parsers": 113148,
      "parsers paper": 85037,
      "paper proposes": 84350,
      "general allow": 46632,
      "different parsing": 31319,
      "parsing architectures": 85075,
      "events considered": 39570,
      "calculated order": 15022,
      "discuss advantages": 32232,
      "demonstrate usefulness": 28897,
      "compare parsers": 19267,
      "fundamental ways": 46131,
      "interpretation using": 56277,
      "using examples": 129650,
      "advantages method": 4082,
      "form noun": 45112,
      "noun noun": 80425,
      "sentences using": 106538,
      "using new": 129961,
      "new corpus": 78845,
      "corpus experimented": 23791,
      "obtained precision": 81391,
      "using method": 129878,
      "identification using": 51456,
      "rules paper": 103424,
      "identification methods": 51397,
      "methods using": 69835,
      "using supervised": 130238,
      "supervised learning": 115763,
      "syntactic analysis": 116368,
      "analysis usually": 5931,
      "identification important": 51390,
      "analyzing japanese": 6042,
      "sentences experiments": 106305,
      "experiments comparing": 40836,
      "previously available": 90589,
      "machinelearning methods": 66323,
      "decision tree": 27945,
      "method examplebased": 68820,
      "approach decision": 7474,
      "list new": 64695,
      "new methods": 79016,
      "method using": 69210,
      "highest similarity": 50238,
      "performed best": 86912,
      "probabilistic information": 90811,
      "retrieval using": 102438,
      "category information": 15786,
      "model does": 71019,
      "does use": 33407,
      "framework using": 45733,
      "model submitted": 72099,
      "submitted systems": 114774,
      "systems based": 116751,
      "based framework": 11726,
      "22 systems": 777,
      "systems participated": 117050,
      "comparative experiments": 19206,
      "experiments various": 41198,
      "experiments confirmed": 40853,
      "effectiveness using": 34974,
      "information probabilistic": 54870,
      "probabilistic constraint": 90800,
      "constraint logic": 21581,
      "logic programming": 64996,
      "thesis present": 121856,
      "present approaches": 89382,
      "processing approach": 91622,
      "approach called": 7408,
      "sound complete": 110703,
      "approach combines": 7427,
      "formal semantics": 45177,
      "inference based": 54117,
      "systems second": 117128,
      "second approach": 104386,
      "probability distribution": 90858,
      "probability model": 90865,
      "model efficiently": 71050,
      "probabilistic modeling": 90821,
      "modeling arbitrary": 72378,
      "finding correct": 44270,
      "addition address": 3397,
      "address problem": 3727,
      "problem computational": 90975,
      "inference task": 54231,
      "task present": 118548,
      "present various": 89762,
      "solve task": 110623,
      "automatic extraction": 10543,
      "present novel": 89596,
      "novel machine": 80622,
      "information verbs": 55087,
      "compare different": 19236,
      "different statistical": 31450,
      "statistical techniques": 113171,
      "applied problem": 7107,
      "problem learning": 91102,
      "algorithm used": 4811,
      "previously unknown": 90628,
      "dependency treebank": 29238,
      "used label": 128601,
      "czech treebank": 25498,
      "able achieve": 1593,
      "parsed text": 84900,
      "shared task": 108015,
      "task dividing": 118112,
      "dividing text": 32916,
      "syntactically related": 116515,
      "groups words": 49156,
      "text chunking": 120704,
      "background information": 11300,
      "information data": 54464,
      "sets present": 107695,
      "present general": 89500,
      "general overview": 46690,
      "overview systems": 83377,
      "systems taken": 117181,
      "briefly discuss": 14620,
      "discuss performance": 32274,
      "recognition word": 97043,
      "confusion networks": 21056,
      "networks new": 77677,
      "new framework": 78924,
      "word lattices": 133333,
      "improve accuracy": 52331,
      "accuracy speech": 2280,
      "recognition obtain": 96948,
      "representation set": 99408,
      "set alternative": 107354,
      "map decoding": 67105,
      "decoding approach": 28076,
      "words corresponding": 133892,
      "highest posterior": 50230,
      "posterior probability": 88487,
      "probability given": 90862,
      "model given": 71249,
      "models map": 73552,
      "decoder does": 28004,
      "does necessarily": 33370,
      "performance metric": 86530,
      "metric word": 69911,
      "method explicitly": 68827,
      "posterior probabilities": 88484,
      "probabilities word": 90853,
      "problem formulation": 91062,
      "search large": 104324,
      "large set": 61268,
      "set sentence": 107575,
      "local search": 64935,
      "small set": 110197,
      "set word": 107637,
      "word candidates": 132940,
      "addition improving": 3437,
      "improving accuracy": 53069,
      "method produces": 69075,
      "produces new": 92005,
      "new representation": 79104,
      "set candidate": 107382,
      "study properties": 114485,
      "networks examine": 77589,
      "examine use": 39758,
      "use tasks": 128318,
      "recognition hypotheses": 96888,
      "hypotheses using": 51260,
      "knowledge sources": 58180,
      "sources using": 110927,
      "using existing": 129651,
      "small amounts": 110129,
      "amounts annotated": 5328,
      "annotated grammatical": 6201,
      "relations training": 98271,
      "data grammatical": 25986,
      "boost performance": 14435,
      "performance using": 86828,
      "using small": 130186,
      "use existing": 128036,
      "types annotations": 126248,
      "extraction semantic": 42477,
      "semantic relations": 105208,
      "dictionary using": 30899,
      "constraint grammar": 21580,
      "grammar paper": 48649,
      "paper deals": 83794,
      "knowledge bases": 57798,
      "goal research": 48386,
      "lexical database": 63754,
      "semantic information": 105072,
      "relations extracted": 98172,
      "focuses extraction": 44898,
      "relations best": 98115,
      "hypernymy relations": 51225,
      "nominal verbal": 80141,
      "verbal adjectival": 131529,
      "uses morphological": 129247,
      "morphological inflection": 74708,
      "relations inferred": 98190,
      "morphological analyser": 74676,
      "syntax parsing": 116551,
      "parsing based": 85077,
      "proven successful": 93732,
      "highly inflected": 50326,
      "inflected languages": 54284,
      "languages basque": 60423,
      "processing time": 91840,
      "extracted relation": 42174,
      "extremely low": 42602,
      "extracted relations": 42175,
      "world wide": 135054,
      "wide web": 132531,
      "term descriptions": 120207,
      "texts paper": 121567,
      "paper propose": 84214,
      "propose method": 92761,
      "method extract": 68833,
      "technical terms": 119755,
      "web pages": 132242,
      "order utilize": 82426,
      "use linguistic": 128127,
      "linguistic patterns": 64518,
      "extract text": 42128,
      "text fragments": 120965,
      "descriptions use": 29499,
      "use language": 128111,
      "clustering method": 17947,
      "method summarize": 69172,
      "effectiveness method": 34903,
      "method way": 69222,
      "way experiments": 132080,
      "method information": 68908,
      "retrieval information": 102397,
      "precision recall": 88798,
      "ir systems": 56931,
      "systems given": 116923,
      "number retrieval": 80958,
      "retrieval systems": 102430,
      "available public": 11084,
      "relevant documents": 98544,
      "documents retrieved": 33287,
      "view problem": 131708,
      "problem propose": 91174,
      "propose evaluation": 92662,
      "novel documents": 80548,
      "used method": 128630,
      "method evaluate": 68813,
      "evaluate systems": 38932,
      "aim finding": 4456,
      "minimal set": 70177,
      "parse accuracy": 84880,
      "data oriented": 26193,
      "oriented parsing": 82498,
      "experiments penn": 41054,
      "leading improved": 61894,
      "accuracy previous": 2244,
      "previous models": 90432,
      "models tested": 74180,
      "dependency relations": 29220,
      "relations previous": 98239,
      "models neglect": 73622,
      "contribute higher": 22753,
      "referential properties": 97564,
      "properties noun": 92470,
      "phrases japanese": 87432,
      "language articles": 58847,
      "resolution japanese": 100765,
      "phrases generally": 87427,
      "generally classified": 46852,
      "generic noun": 47805,
      "rules used": 103452,
      "clue words": 17915,
      "words rules": 134180,
      "score given": 104075,
      "cost high": 24357,
      "work automatically": 134392,
      "scores using": 104217,
      "developed systems": 30302,
      "retrieval results": 102428,
      "locational information": 64972,
      "tasks got": 119151,
      "results tasks": 102255,
      "free parameters": 45760,
      "tasks automatically": 118948,
      "automatically determined": 10756,
      "number documents": 80877,
      "original query": 82538,
      "relevant terms": 98593,
      "tasks used": 119587,
      "used document": 128498,
      "achieved good": 2628,
      "problems natural": 91345,
      "generation nlg": 47511,
      "referring expression": 97572,
      "expression generation": 41737,
      "domainspecific representations": 33912,
      "representations general": 99666,
      "general linguistic": 46674,
      "linguistic representations": 64544,
      "representations paper": 99795,
      "communicative intent": 19053,
      "representations results": 99865,
      "linguistic structure": 64557,
      "meaning structure": 67697,
      "information application": 54377,
      "application domain": 6847,
      "domain current": 33496,
      "discourse context": 32047,
      "context approach": 22009,
      "sentence planning": 105985,
      "uses lexicalized": 129241,
      "treeadjoining grammar": 125622,
      "grammar formalism": 48635,
      "structure meaning": 113905,
      "meaning context": 67628,
      "representations provide": 99834,
      "nlg using": 79532,
      "using representations": 130112,
      "representations communicative": 99549,
      "syntax semantics": 116559,
      "semantics pragmatics": 105444,
      "sentence simultaneously": 106080,
      "assess progress": 9486,
      "various problems": 131177,
      "achieve desired": 2453,
      "correction errors": 24148,
      "used machine": 128617,
      "translation using": 125416,
      "method performed": 69044,
      "performed corpus": 86916,
      "corpus machine": 23871,
      "method constructed": 68735,
      "corpus based": 23677,
      "based corpus": 11613,
      "methods corpus": 69410,
      "correction experiments": 24150,
      "method corpus": 68747,
      "man woman": 66893,
      "case study": 15612,
      "great deal": 48998,
      "demonstrating ability": 28969,
      "ability machine": 1539,
      "algorithms automatically": 4831,
      "knowledge annotated": 57752,
      "corpora little": 23520,
      "little work": 64835,
      "machine paper": 65844,
      "paper step": 84437,
      "step direction": 113260,
      "tagging parsing": 117434,
      "parsing models": 85162,
      "paper compares": 83767,
      "different ways": 31562,
      "ways estimating": 132155,
      "statistical language": 113098,
      "models statistical": 74098,
      "statistical nlp": 113146,
      "models estimated": 73170,
      "data applications": 25610,
      "applications require": 7007,
      "conditional probability": 20772,
      "probability distributions": 90859,
      "conditional likelihood": 20763,
      "likelihood training": 64129,
      "data somewhat": 26468,
      "somewhat surprisingly": 110658,
      "surprisingly models": 116144,
      "conditional models": 20767,
      "access information": 1911,
      "topdown parser": 122490,
      "problem language": 91098,
      "recognition paper": 96952,
      "paper introduces": 84006,
      "probabilistic parsing": 90826,
      "approaches using": 8394,
      "parser presented": 84986,
      "terms accuracy": 120264,
      "parsers new": 85035,
      "model utilizes": 72289,
      "empirical results": 36176,
      "results improves": 101846,
      "improves previous": 53029,
      "work test": 134850,
      "test corpus": 120432,
      "model yields": 72336,
      "relative improvement": 98357,
      "improvement observed": 52733,
      "observed models": 81225,
      "models demonstrating": 73036,
      "information captured": 54405,
      "parsing model": 85158,
      "model orthogonal": 71622,
      "model small": 72061,
      "recognition experiment": 96879,
      "utility model": 130484,
      "model robust": 71942,
      "single pass": 109780,
      "parsing approach": 85072,
      "approach adopted": 7339,
      "perspective model": 87165,
      "model captures": 70810,
      "lexical items": 63773,
      "process building": 91434,
      "results provided": 102091,
      "model presented": 71768,
      "lead improved": 61855,
      "improved performance": 52626,
      "performance new": 86571,
      "model uses": 72262,
      "rate reduction": 95802,
      "models trigram": 74250,
      "significantly data": 108902,
      "wordbyword basis": 133702,
      "additional improvements": 3524,
      "random walk": 95517,
      "consider word": 21238,
      "word meanings": 133361,
      "russian words": 103510,
      "words distribution": 133921,
      "encyclopedic knowledge": 36788,
      "knowledge based": 57795,
      "based web": 12158,
      "web application": 132221,
      "question answering": 95018,
      "answering propose": 6664,
      "method generate": 68853,
      "generate largescale": 46964,
      "valuable nlp": 130768,
      "nlp research": 79681,
      "research based": 100428,
      "web search": 132252,
      "search web": 104362,
      "question use": 95235,
      "based word": 12162,
      "word senses": 133536,
      "automatically generated": 10777,
      "using distribution": 129599,
      "distribution performance": 32673,
      "systems frequently": 116912,
      "evaluated compared": 38963,
      "compared basis": 19335,
      "split training": 111944,
      "test data": 120433,
      "data results": 26359,
      "results obtained": 102004,
      "obtained using": 81427,
      "using single": 130181,
      "noise paper": 80061,
      "argue favour": 8920,
      "performance figures": 86381,
      "figures obtained": 44018,
      "data single": 26457,
      "additional information": 3526,
      "used make": 128622,
      "make statistically": 66730,
      "parameter settings": 84721,
      "performance learning": 86493,
      "learning computational": 62451,
      "computational grammars": 20381,
      "paper reports": 84404,
      "devoted studying": 30481,
      "application machine": 6859,
      "systematic survey": 116688,
      "factors success": 42901,
      "success learning": 115090,
      "annotated data": 6156,
      "data kind": 26060,
      "dependencies data": 29075,
      "data availability": 25666,
      "phrase np": 87359,
      "progress language": 92154,
      "modeling past": 72505,
      "past years": 85657,
      "years number": 135284,
      "number different": 80868,
      "different language": 31190,
      "modeling improvements": 72445,
      "models including": 73377,
      "kneserney smoothing": 57724,
      "techniques including": 119909,
      "mixture models": 70424,
      "models potential": 73759,
      "rarely studied": 95765,
      "clustering techniques": 17957,
      "techniques compare": 119853,
      "combination techniques": 18588,
      "model count": 70935,
      "perplexity reductions": 87069,
      "depending training": 29276,
      "data size": 26458,
      "size word": 109951,
      "highest reported": 50233,
      "extended version": 41813,
      "version paper": 131615,
      "paper contains": 83787,
      "presents study": 89904,
      "statistical syntactic": 113170,
      "syntactic knowledge": 116427,
      "knowledge framework": 57941,
      "investigate impact": 56763,
      "journal wsj": 57409,
      "approach applying": 7360,
      "atis data": 9678,
      "data using": 26608,
      "small data": 110146,
      "data manually": 26118,
      "surprisingly despite": 116133,
      "despite fact": 29688,
      "initialization methods": 55236,
      "methods based": 69333,
      "based indomain": 11768,
      "data achieving": 25566,
      "achieving significant": 2974,
      "absolute relative": 1748,
      "relative reduction": 98369,
      "reduction word": 97466,
      "baseline word": 12336,
      "nbest lists": 76694,
      "12 boosting": 250,
      "filtering paper": 44086,
      "problem automatically": 90947,
      "main conclusions": 66407,
      "conclusions drawn": 20727,
      "clearly outperform": 17740,
      "outperform baseline": 82689,
      "baseline learning": 12240,
      "naive bayes": 76013,
      "decision trees": 27948,
      "corpus achieving": 23643,
      "achieving high": 2951,
      "high levels": 50083,
      "f1 measure": 42660,
      "increasing complexity": 53667,
      "allows obtain": 5176,
      "important issue": 52174,
      "selectional preferences": 104849,
      "preference learning": 89239,
      "methods usually": 69836,
      "extends previous": 41829,
      "previous statistical": 90485,
      "presents model": 89869,
      "model learns": 71439,
      "different senses": 31411,
      "model tested": 72160,
      "tested word": 120586,
      "disambiguation task": 31988,
      "task uses": 118829,
      "extracted small": 42181,
      "syntactic dependencies": 116391,
      "paper investigates": 84028,
      "investigates use": 56852,
      "use richer": 128253,
      "present simple": 89699,
      "simple methods": 109470,
      "dependencies syntactic": 29117,
      "syntactic parse": 116434,
      "trees used": 125718,
      "evaluate impact": 38841,
      "nbest rescoring": 76696,
      "new model": 79024,
      "model achieves": 70551,
      "achieves improvement": 2808,
      "baseline results": 12297,
      "results reported": 102127,
      "upenn treebank": 127812,
      "corpora respectively": 23576,
      "accuracy combining": 2117,
      "combining semantic": 18747,
      "semantic syntactic": 105314,
      "modeling structured": 72554,
      "recognition shown": 96998,
      "ngram models": 79446,
      "models current": 73007,
      "models limited": 73507,
      "dependencies contribute": 29070,
      "significantly improved": 108935,
      "improved word": 52656,
      "model trained": 72182,
      "semantically syntactically": 105381,
      "data exploit": 25930,
      "dependencies paper": 29103,
      "models using": 74297,
      "modality paper": 70487,
      "describes experiments": 29403,
      "experiments carried": 40812,
      "carried using": 15545,
      "using variety": 130352,
      "methods including": 69546,
      "method used": 69206,
      "used previous": 128696,
      "previous study": 90505,
      "machine method": 65842,
      "paper extends": 83941,
      "theoretical motivation": 121798,
      "frequent word": 45863,
      "allow easy": 5078,
      "easy integration": 34480,
      "small corpus": 110145,
      "corpus disambiguated": 23760,
      "wordnet senses": 133779,
      "showing approach": 108396,
      "approach feasible": 7570,
      "experimental results": 40548,
      "results word": 102341,
      "task provided": 118585,
      "decision lists": 27936,
      "paper systems": 84462,
      "systems developed": 116846,
      "developed english": 30272,
      "english lexical": 37192,
      "tasks supervised": 119539,
      "supervised systems": 115841,
      "lists used": 64720,
      "training english": 123610,
      "english allwords": 37065,
      "allwords task": 5202,
      "task defined": 118052,
      "different feature": 31145,
      "feature sets": 43312,
      "sets language": 107679,
      "order extract": 82323,
      "extract information": 42088,
      "information text": 55037,
      "text defined": 120864,
      "features used": 43773,
      "used literature": 128615,
      "literature using": 64780,
      "using morphological": 129899,
      "morphological analyzer": 74685,
      "analyzer implemented": 6031,
      "good features": 48476,
      "features able": 43347,
      "able obtain": 1665,
      "systems used": 117219,
      "used features": 128551,
      "selected features": 104725,
      "task did": 118095,
      "did systems": 30923,
      "systems perform": 117056,
      "task task": 118778,
      "examples taken": 39884,
      "attained results": 9724,
      "baseline best": 12197,
      "75 precision": 1250,
      "coverage paper": 24519,
      "paper includes": 83980,
      "procedure used": 91405,
      "systems particular": 117052,
      "wsd task": 135173,
      "task memorybased": 118384,
      "shallow parsing": 107928,
      "parsing present": 85196,
      "learning approaches": 62356,
      "approaches shallow": 8336,
      "parsing apply": 85071,
      "parsing parsing": 85187,
      "parsing use": 85275,
      "use feature": 128051,
      "feature selection": 43306,
      "selection techniques": 104840,
      "techniques combination": 119851,
      "methods improving": 69543,
      "improving performance": 53135,
      "approach evaluated": 7541,
      "evaluated standard": 39007,
      "sets results": 107707,
      "results compared": 101597,
      "compared systems": 19473,
      "systems reveals": 117124,
      "reveals approach": 102514,
      "approach works": 8021,
      "room improvement": 103262,
      "improvement unsupervised": 52781,
      "unsupervised discovery": 127623,
      "morphologically related": 74764,
      "related words": 97912,
      "words based": 133834,
      "based orthographic": 11917,
      "orthographic semantic": 82596,
      "semantic similarity": 105279,
      "similarity present": 109282,
      "present algorithm": 89361,
      "algorithm takes": 4804,
      "unannotated corpus": 126585,
      "ranked list": 95646,
      "pairs output": 83597,
      "tries discover": 125778,
      "orthographically semantically": 82602,
      "semantically similar": 105377,
      "orthographic similarity": 82597,
      "similarity measured": 109259,
      "measured terms": 67840,
      "edit distance": 34531,
      "distance semantic": 32432,
      "mutual information": 75976,
      "does rely": 33384,
      "model distributional": 71013,
      "distributional properties": 32714,
      "properties word": 92492,
      "german english": 47889,
      "encouraging results": 36782,
      "results terms": 102263,
      "terms precision": 120361,
      "pairs various": 83663,
      "list terms": 64701,
      "terms qualitative": 120367,
      "qualitative analysis": 94551,
      "analysis types": 5915,
      "morphological patterns": 74717,
      "statistical segmentation": 113160,
      "sequences given": 107124,
      "given lack": 48053,
      "word delimiters": 132983,
      "japanese word": 57202,
      "word segmentation": 133489,
      "generally considered": 46853,
      "considered crucial": 21289,
      "crucial step": 25168,
      "step processing": 113297,
      "segmentation algorithms": 104574,
      "algorithms rely": 4878,
      "unknown word": 127364,
      "word problem": 133420,
      "problem contrast": 90982,
      "introduce novel": 56509,
      "statistical method": 113125,
      "method utilizing": 69218,
      "data despite": 25833,
      "despite simplicity": 29730,
      "algorithm yields": 4819,
      "yields performance": 135425,
      "performance long": 86500,
      "surpassing stateoftheart": 116119,
      "stateoftheart morphological": 112757,
      "morphological analyzers": 74686,
      "algorithm outperforms": 4776,
      "previously proposed": 90609,
      "additionally present": 3619,
      "incorporate multiple": 53479,
      "multiple segmentation": 75679,
      "novel evaluation": 80563,
      "evaluation metrics": 39283,
      "metrics based": 69926,
      "based notion": 11901,
      "account multiple": 2032,
      "multiple granularities": 75571,
      "granularities simultaneously": 48743,
      "samples paper": 103588,
      "presents evaluation": 89848,
      "participated english": 85337,
      "english spanish": 37281,
      "lexical sample": 63808,
      "sample tasks": 103567,
      "unigrams bigrams": 127160,
      "single classifier": 109713,
      "data assessing": 25623,
      "senseval2 paper": 105729,
      "comparative evaluation": 19205,
      "evaluation systems": 39412,
      "spanish english": 111153,
      "pairwise comparisons": 83672,
      "assess degree": 9474,
      "test instances": 120463,
      "tasks using": 119589,
      "identity paper": 51638,
      "describes results": 29428,
      "results experiments": 101780,
      "statistical methods": 113126,
      "methods infer": 69550,
      "infer syntactic": 54106,
      "syntactic behavior": 116373,
      "words morphemes": 134061,
      "raw corpus": 95850,
      "corpus unsupervised": 24054,
      "unsupervised fashion": 127637,
      "et al": 38680,
      "techniques analyze": 119834,
      "behavior based": 12647,
      "based words": 12170,
      "given word": 48173,
      "word use": 133623,
      "words corpus": 133889,
      "corpus words": 24068,
      "tend form": 120149,
      "automatic learning": 10575,
      "learning morphology": 62804,
      "derived corpus": 29347,
      "unsupervised learning": 127652,
      "consistent syntactic": 21404,
      "syntactic function": 116409,
      "method works": 69231,
      "task efficient": 118131,
      "deep processing": 28396,
      "broad coverage": 14670,
      "hpsg formalism": 50692,
      "use real": 128223,
      "real world": 96088,
      "world applications": 135019,
      "robustness performance": 103125,
      "performance issues": 86472,
      "play important": 87695,
      "important role": 52236,
      "pos tagging": 88224,
      "tagging word": 117464,
      "multilingual context": 75218,
      "easily comparable": 34442,
      "comparable languages": 19147,
      "languages using": 60951,
      "paper shows": 84430,
      "differences different": 30954,
      "different datasets": 31077,
      "merging different": 68482,
      "readily available": 95966,
      "available easy": 10988,
      "easy use": 34485,
      "research natural": 100563,
      "algorithm efficient": 4722,
      "efficient semantic": 35111,
      "semantic analysis": 104976,
      "wellformed natural": 132411,
      "expected input": 40392,
      "input information": 55351,
      "information communication": 54426,
      "communication systems": 19042,
      "input sequences": 55432,
      "semantic units": 105345,
      "units like": 127253,
      "answer users": 6589,
      "users needs": 129149,
      "analysis performed": 5748,
      "performed based": 86909,
      "based lexical": 11801,
      "lexical semantic": 63810,
      "semantic knowledge": 105090,
      "dependency analysis": 29130,
      "analysis syntactic": 5881,
      "syntactic morphological": 116431,
      "morphological clues": 74692,
      "reasonable accuracy": 96207,
      "reasonable time": 96215,
      "time propose": 122079,
      "method allowing": 68631,
      "sequences semantic": 107139,
      "algorithm inspired": 4755,
      "inspired idea": 55571,
      "chart parsing": 16647,
      "parsing known": 85135,
      "known natural": 58314,
      "intermediate parsing": 56137,
      "parsing results": 85222,
      "results order": 102015,
      "calculation time": 15036,
      "comparison using": 19585,
      "method brings": 68679,
      "time polynomial": 122070,
      "polynomial time": 88020,
      "task languageindependent": 118336,
      "languageindependent named": 60365,
      "entity recognition": 37991,
      "recognition background": 96821,
      "sets evaluation": 107668,
      "method present": 69063,
      "taken task": 117516,
      "task discuss": 118104,
      "turing test": 125953,
      "passing turing": 85626,
      "questions designed": 95299,
      "world paper": 135041,
      "paper possible": 84073,
      "approach answering": 7355,
      "questions use": 95373,
      "use statistical": 128296,
      "statistical information": 113097,
      "information extracted": 54559,
      "extracted large": 42160,
      "large collection": 61049,
      "collection text": 18492,
      "text particular": 121170,
      "particular possible": 85441,
      "possible answer": 88378,
      "queries search": 94938,
      "search engine": 104310,
      "million web": 70111,
      "simple algorithm": 109357,
      "shed light": 108149,
      "light nature": 63996,
      "nature human": 76655,
      "empirical methods": 36173,
      "compound splitting": 20153,
      "words challenge": 133845,
      "challenge nlp": 16067,
      "nlp applications": 79566,
      "applications machine": 6961,
      "translation mt": 124977,
      "introduce methods": 56458,
      "methods learn": 69584,
      "monolingual parallel": 74603,
      "parallel corpora": 84593,
      "corpora evaluate": 23472,
      "gold standard": 48437,
      "measure impact": 67801,
      "impact performance": 51884,
      "performance statistical": 86750,
      "statistical mt": 113135,
      "mt systems": 74995,
      "systems results": 117123,
      "results accuracy": 101489,
      "performance gains": 86401,
      "translation task": 125328,
      "overview history": 83367,
      "history natural": 50562,
      "language analysis": 58828,
      "discriminative model": 32210,
      "factored models": 42862,
      "model combining": 70846,
      "provide better": 93770,
      "previous works": 90566,
      "works mainly": 134960,
      "mainly focus": 66486,
      "models sequential": 74011,
      "allows model": 5171,
      "model event": 71116,
      "propose general": 92695,
      "framework adopt": 45416,
      "models use": 74274,
      "mixture model": 70423,
      "model combine": 70842,
      "method integrate": 68916,
      "lattice based": 61668,
      "types model": 126321,
      "used natural": 128645,
      "algorithm text": 4808,
      "text categorization": 120697,
      "categorization based": 15762,
      "classification process": 17352,
      "test training": 120548,
      "nearest neighbors": 76723,
      "generally speaking": 46876,
      "class distribution": 17033,
      "distribution training": 32685,
      "performance sensitive": 86697,
      "sensitive choice": 105735,
      "bias large": 13812,
      "problems propose": 91359,
      "propose improved": 92719,
      "algorithm uses": 4812,
      "uses different": 129225,
      "different categories": 31031,
      "fixed number": 44602,
      "used deciding": 128474,
      "samples training": 103595,
      "set preliminary": 107538,
      "preliminary experiments": 89268,
      "experiments chinese": 40820,
      "chinese text": 16827,
      "categorization method": 15765,
      "documents belonging": 33192,
      "method promising": 69077,
      "order deal": 82306,
      "contexts present": 22418,
      "present paper": 89635,
      "datadriven method": 26668,
      "language barrier": 58855,
      "makes text": 66811,
      "indian language": 53818,
      "man computer": 66892,
      "machine interpretation": 65764,
      "interpretation text": 56275,
      "image source": 51796,
      "source text": 110852,
      "text language": 121077,
      "image constructions": 51777,
      "source language": 110771,
      "read understand": 95942,
      "indian languages": 53821,
      "languages close": 60446,
      "learning time": 63106,
      "time output": 122063,
      "output language": 83085,
      "grammatically correct": 48732,
      "target language": 117629,
      "language style": 60131,
      "function human": 46025,
      "assisted translation": 9586,
      "translation currently": 124730,
      "punjabi hindi": 94400,
      "near future": 76709,
      "build systems": 14811,
      "using free": 129687,
      "model language": 71403,
      "information based": 54393,
      "based approach": 11510,
      "translation makes": 124927,
      "language machine": 59263,
      "equivalents target": 38260,
      "pairs languages": 83570,
      "available use": 11141,
      "use email": 128024,
      "language target": 60146,
      "subject areas": 114675,
      "subject domain": 114677,
      "produce good": 91894,
      "good quality": 48495,
      "modules work": 74532,
      "collaborative effort": 18358,
      "effort creating": 35166,
      "creating lexical": 24701,
      "resources paper": 101025,
      "model lexical": 71458,
      "resources developed": 100963,
      "english indian": 37170,
      "tree bank": 125581,
      "dependency trees": 29243,
      "trees based": 125681,
      "based paninian": 11921,
      "model bilingual": 70778,
      "architecture spoken": 8752,
      "language resources": 60053,
      "overcome challenges": 83276,
      "challenges associated": 16135,
      "architecture consisting": 8640,
      "tools supporting": 122476,
      "integrated workflow": 55782,
      "decisionmaking processes": 27953,
      "systems development": 116848,
      "development methods": 30399,
      "frequency counts": 45838,
      "measure use": 67828,
      "language makes": 59270,
      "linguistic unit": 64571,
      "word important": 133314,
      "needed use": 76881,
      "recognition propose": 96973,
      "propose family": 92675,
      "family measures": 43056,
      "distinctive features": 32559,
      "phonological rules": 87313,
      "rules test": 103449,
      "test robustness": 120487,
      "finally provide": 44227,
      "provide examples": 93817,
      "dutch english": 34278,
      "english german": 37147,
      "historical linguistics": 50528,
      "linguistics language": 64613,
      "language acquisition": 58810,
      "recognition information": 96894,
      "lexical base": 63731,
      "model world": 72331,
      "ukrainian language": 126550,
      "language article": 58846,
      "list words": 64703,
      "words selected": 134186,
      "base language": 11462,
      "selected list": 104727,
      "human activities": 50737,
      "verify statement": 131594,
      "parsing tutorial": 85271,
      "parsing basis": 85078,
      "nondeterministic pushdown": 80188,
      "pushdown automata": 94458,
      "earleys algorithm": 34374,
      "lr parsing": 65579,
      "predicateargument structure": 88854,
      "aims building": 4507,
      "treebank english": 125635,
      "particular attention": 85383,
      "attention devoted": 9817,
      "layer used": 61751,
      "equivalent sentences": 38251,
      "sentences languages": 106370,
      "field statistical": 43984,
      "statistical machine": 113104,
      "translation smt": 125268,
      "smt systems": 110296,
      "translation models": 124951,
      "models systems": 74150,
      "systems involving": 116959,
      "models tend": 74172,
      "article aims": 9109,
      "aims reduce": 4568,
      "systems order": 117035,
      "order make": 82365,
      "make easier": 66661,
      "use study": 128302,
      "goal article": 48328,
      "parsing algorithm": 85067,
      "article shows": 9158,
      "shows common": 108563,
      "including hierarchical": 53298,
      "parameter estimation": 84714,
      "grammar logic": 48647,
      "using generalized": 129699,
      "aim paper": 4466,
      "modern greek": 74390,
      "tools developed": 122446,
      "finally discuss": 44169,
      "discuss improvement": 32259,
      "future directions": 46270,
      "chunk extraction": 16953,
      "french using": 45829,
      "using limited": 129811,
      "limited resources": 64273,
      "way extracting": 132085,
      "grammar rules": 48655,
      "structures used": 114115,
      "limited lexical": 64250,
      "lexical information": 63770,
      "linguistic information": 64488,
      "obtained results": 81403,
      "results effective": 101742,
      "corpus presented": 23936,
      "model finegrained": 71188,
      "finegrained alignment": 44332,
      "sentential level": 106557,
      "word alignment": 132906,
      "bi multilingual": 13784,
      "useful resource": 128926,
      "number examples": 80883,
      "tend make": 120155,
      "alignment model": 4972,
      "model based": 70722,
      "predicateargument structures": 88858,
      "middle ground": 70070,
      "sentence word": 106128,
      "model currently": 70946,
      "additional languages": 3537,
      "sentiment analysis": 106560,
      "analysis using": 5930,
      "summarization based": 115482,
      "based minimum": 11845,
      "text span": 121316,
      "example application": 39773,
      "movie review": 74900,
      "review thumbs": 102584,
      "thumbs thumbs": 121953,
      "sentiment polarity": 106770,
      "propose novel": 92892,
      "document extracting": 33007,
      "implemented using": 51980,
      "using efficient": 129614,
      "contextual constraints": 22449,
      "art evaluation": 9051,
      "regarding document": 97656,
      "document processing": 33057,
      "visualization techniques": 131829,
      "set framework": 107448,
      "field text": 43987,
      "text mining": 121118,
      "mining field": 70232,
      "key topics": 57612,
      "working group": 134897,
      "carry detailed": 15550,
      "associated text": 9610,
      "mining process": 70256,
      "process identify": 91503,
      "identify relevant": 51544,
      "relevant research": 98574,
      "research topics": 100646,
      "related research": 97893,
      "research areas": 100416,
      "areas document": 8889,
      "document present": 33053,
      "present results": 89667,
      "comprehensive survey": 20274,
      "includes description": 53252,
      "description current": 29457,
      "current stateoftheart": 25350,
      "research identified": 100531,
      "domain text": 33674,
      "analyzing language": 6043,
      "development network": 30404,
      "network approach": 77137,
      "approach paper": 7785,
      "new measures": 79003,
      "inspired recent": 55580,
      "recent surge": 96547,
      "studies realworld": 114271,
      "networks word": 77816,
      "word forms": 133289,
      "quantitative comparison": 94864,
      "different paths": 31324,
      "network size": 77437,
      "follow different": 44936,
      "space develop": 110991,
      "approach requires": 7859,
      "requires little": 100293,
      "preprocessing words": 89341,
      "sentence structures": 106093,
      "characteristics words": 16544,
      "words usage": 134289,
      "articles start": 9220,
      "provide new": 93879,
      "approach study": 7937,
      "study language": 114420,
      "time language": 122041,
      "network theories": 77455,
      "resources using": 101064,
      "conceptual model": 20668,
      "uses recent": 129267,
      "recent iso": 96466,
      "markup framework": 67271,
      "terms given": 120331,
      "given concept": 47996,
      "resource studying": 100880,
      "translation shifts": 125251,
      "interdisciplinary approach": 56059,
      "approach brings": 7402,
      "corpus linguistics": 23870,
      "translation studies": 125298,
      "presents ongoing": 89887,
      "ongoing work": 81739,
      "work creation": 134449,
      "creation corpus": 24714,
      "corpus resource": 23972,
      "resource translation": 100884,
      "annotated translation": 6250,
      "correspondence source": 24271,
      "source target": 110829,
      "target text": 117730,
      "process resource": 91562,
      "systematic way": 116689,
      "way make": 132106,
      "make possible": 66706,
      "possible study": 88435,
      "need addressed": 76778,
      "addressed machine": 3785,
      "translation output": 125081,
      "resemble human": 100727,
      "human translation": 50978,
      "translation resource": 125218,
      "described paper": 29379,
      "english source": 37279,
      "source texts": 110853,
      "predicates arguments": 88861,
      "arguments annotated": 8978,
      "corresponding english": 24283,
      "grammatical semantic": 48720,
      "term extraction": 120210,
      "studies different": 114213,
      "different term": 31484,
      "biomedical domain": 14174,
      "applied highly": 7077,
      "technical texts": 119756,
      "new domains": 78890,
      "domains additional": 33727,
      "limitation paper": 64162,
      "propose use": 93138,
      "use external": 128042,
      "data order": 26190,
      "extraction tool": 42517,
      "different steps": 31451,
      "steps process": 113331,
      "candidates experiments": 15134,
      "method term": 69181,
      "extraction process": 42449,
      "process involving": 91515,
      "implemented term": 51978,
      "principle compositionality": 90677,
      "interpreting natural": 56286,
      "language texts": 60166,
      "paper aims": 83725,
      "face problems": 42735,
      "framework believe": 45438,
      "radical change": 95439,
      "propose approach": 92547,
      "paper concerns": 83772,
      "evolution time": 39692,
      "time crucial": 122008,
      "composition does": 20093,
      "does vary": 33408,
      "vary time": 131249,
      "expresses different": 41731,
      "different criteria": 31069,
      "depending context": 29269,
      "resources evaluating": 100971,
      "evaluating linguistic": 39068,
      "linguistic models": 64511,
      "models algorithms": 72739,
      "quantitative qualitative": 94875,
      "freely accessible": 45779,
      "based fact": 11712,
      "constructed using": 21669,
      "using manually": 129857,
      "manually revised": 67074,
      "annotated linguistic": 6207,
      "syntax coreference": 116532,
      "make available": 66623,
      "available online": 11060,
      "online research": 81794,
      "research purposes": 100601,
      "focus using": 44839,
      "using standard": 130210,
      "standard annotation": 112184,
      "existing resources": 40270,
      "resources maintenance": 101010,
      "prior actual": 90690,
      "generic model": 47802,
      "linguistic resource": 64546,
      "isotc 37sc": 56984,
      "text high": 121035,
      "high dimensional": 50061,
      "sparsely populated": 111244,
      "characterized terms": 16554,
      "study extent": 114379,
      "texts different": 121498,
      "different domains": 31105,
      "hierarchical structures": 50008,
      "structures data": 114067,
      "data use": 26599,
      "13 million": 285,
      "million words": 70112,
      "criteria paper": 24795,
      "networks study": 77782,
      "study english": 114369,
      "english language": 37183,
      "use syntactic": 128308,
      "network properties": 77396,
      "properties global": 92454,
      "global patterns": 48258,
      "use global": 128072,
      "global view": 48279,
      "issues paper": 57064,
      "syntactic relations": 116455,
      "overcome difficulty": 83281,
      "grammar based": 48623,
      "corpora annotation": 23410,
      "enables perform": 36397,
      "perform standard": 86075,
      "provide detailed": 93801,
      "phonotactics using": 87322,
      "using ilp": 129747,
      "experiments learning": 40987,
      "rules using": 103453,
      "approaching problem": 8410,
      "compared related": 19435,
      "related work": 97913,
      "work task": 134845,
      "task results": 118656,
      "results direct": 101729,
      "background knowledge": 11304,
      "advantage prior": 4055,
      "prior domain": 90701,
      "knowledge available": 57761,
      "available research": 11091,
      "resources resources": 101039,
      "methods make": 69612,
      "make use": 66741,
      "use morphological": 128153,
      "morphological syntactic": 74736,
      "model word": 72318,
      "word similarity": 133552,
      "seed lexicon": 104503,
      "methods deployed": 69434,
      "learning lexical": 62691,
      "grammar shown": 48656,
      "strengths weaknesses": 113577,
      "different word": 31565,
      "word classes": 132952,
      "particular focus": 85414,
      "focus paper": 44802,
      "language resource": 60052,
      "resource types": 100885,
      "task information": 118296,
      "document given": 33017,
      "given retrieval": 48105,
      "retrieval users": 102437,
      "users information": 129133,
      "information query": 54891,
      "user certain": 128969,
      "certain order": 15950,
      "users query": 129162,
      "relevant set": 98580,
      "measure document": 67793,
      "query relevance": 94982,
      "proposed approach": 93179,
      "approach does": 7503,
      "measures using": 67902,
      "textual entailment": 121693,
      "entailment rte": 37672,
      "rte task": 103341,
      "task using": 118830,
      "propose representation": 93021,
      "representation natural": 99347,
      "language semantics": 60076,
      "inspired existing": 55567,
      "existing representations": 40268,
      "firstorder logic": 44569,
      "significant contribution": 108747,
      "novel inference": 80600,
      "inference tasks": 54233,
      "reduces number": 97390,
      "parsing errors": 85108,
      "syntactic parser": 116438,
      "grammar lexicalization": 48645,
      "carried experiments": 15537,
      "experiments used": 41191,
      "performed applying": 86908,
      "approach results": 7864,
      "syntactic parsing": 116444,
      "sentences evaluation": 106297,
      "existing syntactic": 40306,
      "natural languages": 76606,
      "far complete": 43074,
      "complex objects": 19853,
      "takes form": 117532,
      "syntactic information": 116417,
      "chunking information": 16959,
      "information available": 54387,
      "available treebanks": 11138,
      "evaluation comparison": 39152,
      "comparison corpus": 19535,
      "86 precision": 1329,
      "text tagged": 121356,
      "parts speech": 85586,
      "romance languages": 103249,
      "languages french": 60595,
      "french portuguese": 45821,
      "includes features": 53254,
      "complement limited": 19729,
      "noun case": 80422,
      "extraction using": 42531,
      "cooccurrence matrices": 23228,
      "discusses new": 32312,
      "raw texts": 95864,
      "polish language": 87981,
      "novel technique": 80747,
      "selection algorithm": 104763,
      "algorithm performs": 4784,
      "obtained applying": 81343,
      "text second": 121273,
      "second new": 104432,
      "filtering incorrect": 44083,
      "text motivated": 121131,
      "motivated observation": 74856,
      "newly introduced": 79274,
      "matrices using": 67470,
      "using cooccurrence": 129502,
      "resources natural": 101015,
      "language parser": 59802,
      "wide variety": 132518,
      "used automatic": 128401,
      "automatic syntactic": 10684,
      "parsing natural": 85169,
      "languages necessary": 60759,
      "different approaches": 30999,
      "approaches research": 8328,
      "parsers able": 85011,
      "text 100": 120621,
      "use term": 128323,
      "factors create": 42887,
      "methods resources": 69730,
      "resources used": 101062,
      "evaluate compare": 38809,
      "compare parsing": 19268,
      "parsing systems": 85255,
      "systems research": 117117,
      "theoretical analysis": 121789,
      "analysis current": 5553,
      "parsing parser": 85186,
      "evaluation framework": 39217,
      "framework called": 45450,
      "used carry": 128418,
      "parser evaluations": 84952,
      "set new": 107505,
      "new evaluation": 78905,
      "evaluation resources": 39364,
      "resources english": 100969,
      "english results": 37265,
      "developed evaluation": 30274,
      "english used": 37328,
      "used evaluating": 128525,
      "set selected": 107573,
      "robustness evaluation": 103096,
      "refers ability": 97581,
      "ability deal": 1503,
      "inputs paper": 55489,
      "evaluations stateoftheart": 39485,
      "stateoftheart parsers": 112800,
      "parse sentences": 84887,
      "sentences containing": 106260,
      "misspelled words": 70342,
      "words propose": 134145,
      "evaluation based": 39130,
      "parsers output": 85036,
      "input sentences": 55426,
      "sentences noisy": 106412,
      "counterparts paper": 24440,
      "paper use": 84485,
      "parsers present": 85040,
      "present analysis": 89367,
      "parser performance": 84984,
      "levels results": 63563,
      "performance typically": 86820,
      "texts containing": 121483,
      "containing misspellings": 21802,
      "sentences best": 106228,
      "parser experiment": 84953,
      "parser able": 84906,
      "parse tree": 84889,
      "grammatical ungrammatical": 48726,
      "ungrammatical sentences": 127082,
      "respectively methods": 101149,
      "methods integrate": 69557,
      "integrate language": 55757,
      "model semantic": 71963,
      "word prediction": 133413,
      "current word": 25389,
      "prediction systems": 89129,
      "systems make": 116993,
      "use ngram": 128173,
      "models lm": 73515,
      "estimate probability": 38632,
      "following word": 44989,
      "models syntactic": 74146,
      "syntactic semantic": 116463,
      "want explore": 132015,
      "explore predictive": 41577,
      "latent semantic": 61601,
      "analysis lsa": 5686,
      "method shown": 69130,
      "shown provide": 108518,
      "provide reliable": 93908,
      "semantic dependencies": 105028,
      "dependencies words": 29126,
      "words context": 133879,
      "context present": 22223,
      "present evaluate": 89476,
      "evaluate methods": 38858,
      "standard language": 112249,
      "different forms": 31153,
      "methods significant": 69748,
      "significant improvements": 108781,
      "improvements compared": 52826,
      "cooccurrences word": 23243,
      "word semantic": 133513,
      "semantic similarities": 105276,
      "computational model": 20407,
      "model construction": 70903,
      "construction word": 21701,
      "texts built": 121466,
      "order simulate": 82404,
      "simulate effects": 109631,
      "values word": 130808,
      "similarities paragraph": 109184,
      "paragraph paragraph": 84570,
      "similarity words": 109341,
      "words w1": 134310,
      "w1 w2": 131979,
      "grammar provide": 48652,
      "provide insight": 93857,
      "insight sentence": 55525,
      "sentence semantics": 106065,
      "systems define": 116836,
      "feature based": 43253,
      "based tree": 12136,
      "structures original": 114094,
      "lexicon paper": 63902,
      "software toolkit": 110525,
      "toolkit designed": 122410,
      "construction maintenance": 21691,
      "use generative": 128068,
      "lexicon order": 63901,
      "use tool": 128330,
      "tool built": 122376,
      "open source": 81933,
      "anaphoric reference": 6074,
      "linguistic structures": 64558,
      "structures using": 114116,
      "using domainspecific": 129607,
      "generating sentences": 47261,
      "underlying linguistic": 126685,
      "languages uses": 60950,
      "languages based": 60422,
      "functional grammar": 46061,
      "languages used": 60948,
      "used tool": 128815,
      "linguistic description": 64464,
      "theorem proving": 121786,
      "present original": 89633,
      "methods recognizing": 69709,
      "recognizing textual": 97078,
      "textual inference": 121711,
      "resolution method": 100768,
      "recent methods": 96471,
      "second based": 104393,
      "based semantic": 12008,
      "relations text": 98269,
      "text presented": 121196,
      "grammar engineering": 48629,
      "engineering paper": 37054,
      "present opensource": 89632,
      "linguistic parsing": 64516,
      "opening way": 81997,
      "mildly contextsensitive": 70085,
      "treeadjoining grammars": 125623,
      "allows computation": 5133,
      "structures corresponding": 114066,
      "corresponding semantic": 24301,
      "semantic representations": 105235,
      "representations used": 99960,
      "used development": 128490,
      "word similarities": 133551,
      "meroitic undeciphered": 68487,
      "language ancient": 58830,
      "years various": 135310,
      "bilingual text": 14061,
      "ancient languages": 6092,
      "surrounding areas": 116154,
      "techniques borrowed": 119847,
      "information theory": 55042,
      "similar words": 109172,
      "defined words": 28507,
      "words extract": 133955,
      "partial meaning": 85293,
      "pour la": 88626,
      "unification grammar": 127099,
      "lexical functional": 63765,
      "categories based": 15730,
      "based syntactic": 12101,
      "semantic properties": 105186,
      "used implement": 128579,
      "prototype text": 93688,
      "generation main": 47467,
      "applications use": 7033,
      "task learning": 118348,
      "second language": 104414,
      "la langue": 58377,
      "grammar model": 48648,
      "model using": 72272,
      "build common": 14753,
      "common syntactic": 18931,
      "syntactic kernel": 116426,
      "article describes": 9121,
      "describes design": 29394,
      "group related": 49141,
      "structure like": 113899,
      "consists single": 21494,
      "specific language": 111456,
      "language included": 59159,
      "attribute types": 10185,
      "selected subset": 104736,
      "model hybrid": 71292,
      "principle used": 90678,
      "used group": 128568,
      "fuzzy logic": 46329,
      "collocation extraction": 18518,
      "important tasks": 52274,
      "tasks natural": 119325,
      "processing information": 91680,
      "retrieval machine": 102403,
      "methods used": 69831,
      "extraction methods": 42388,
      "propose fuzzy": 92693,
      "logic rules": 64998,
      "existing methods": 40175,
      "inference resulting": 54210,
      "demonstrated utility": 28936,
      "word pairs": 133392,
      "data based": 25689,
      "corpus million": 23881,
      "project gutenberg": 92219,
      "proposed method": 93334,
      "methods overcoming": 69661,
      "provides better": 94004,
      "better result": 13701,
      "methods new": 69641,
      "confidence measures": 20989,
      "measures statistical": 67898,
      "translation confidence": 124706,
      "confidence measure": 20988,
      "translation problem": 125143,
      "measures machine": 67878,
      "translation based": 124664,
      "based mutual": 11875,
      "lexical features": 63762,
      "features language": 43582,
      "model evaluate": 71103,
      "using combination": 129450,
      "measures based": 67854,
      "information yields": 55113,
      "classification error": 17198,
      "rate low": 95796,
      "step larger": 113277,
      "series experiments": 107273,
      "inducing similar": 54017,
      "frame semantics": 45388,
      "previously used": 90632,
      "used context": 128450,
      "context multidocument": 22182,
      "multidocument summarization": 75078,
      "summarization evolving": 115503,
      "evolving events": 39703,
      "stages stage": 112163,
      "vocabulary words": 131920,
      "words vocabulary": 134309,
      "later used": 61649,
      "used second": 128739,
      "second stage": 104456,
      "apply various": 7225,
      "clustering approaches": 17938,
      "approaches order": 8269,
      "order identify": 82337,
      "semantic roles": 105260,
      "roles use": 103237,
      "semantics paper": 105443,
      "syntactic variation": 116497,
      "report experiments": 98998,
      "experiments syntactic": 41165,
      "syntactic variations": 116498,
      "special type": 111374,
      "type multiword": 126214,
      "multiword expressions": 75945,
      "expressions mwes": 41756,
      "large french": 61094,
      "french corpus": 45802,
      "set examples": 107436,
      "statistical results": 113158,
      "results data": 101653,
      "finitestate techniques": 44547,
      "techniques results": 119977,
      "results frequently": 101800,
      "occur support": 81502,
      "methods extracting": 69503,
      "mathematical model": 67458,
      "survey models": 116178,
      "models context": 72978,
      "meaning syntactic": 67698,
      "phrase structure": 87371,
      "structuring sentences": 114122,
      "transitive verb": 124519,
      "parsed sentences": 84899,
      "sentences corpus": 106264,
      "corpus generated": 23816,
      "constituents sentences": 21555,
      "phrases form": 87426,
      "property yields": 92510,
      "computer program": 20487,
      "software developers": 110518,
      "morphological families": 74699,
      "machine readable": 65847,
      "dictionary paper": 30888,
      "linguistic computational": 64445,
      "morphological structure": 74735,
      "formal semantic": 45176,
      "semantic regularities": 105193,
      "regularities words": 97741,
      "model wordbased": 72326,
      "consists binary": 21471,
      "binary relations": 14135,
      "relations connect": 98127,
      "words morphologically": 134064,
      "analogies hold": 5411,
      "hold words": 50588,
      "words model": 134056,
      "lexicon french": 63884,
      "determining given": 30148,
      "does entail": 33345,
      "learning linguistic": 62693,
      "systems currently": 116828,
      "bayesian model": 12582,
      "standard form": 112242,
      "analysis linguistic": 5682,
      "linguistic typology": 64569,
      "state facts": 112494,
      "verbs adjectives": 131551,
      "analysis small": 5858,
      "small sample": 110191,
      "languages propose": 60819,
      "propose computational": 92590,
      "process model": 91528,
      "model able": 70508,
      "able discover": 1631,
      "careful application": 15509,
      "analysis able": 5480,
      "languages independent": 60639,
      "global features": 48238,
      "features joint": 43575,
      "joint entity": 57272,
      "entity detection": 37921,
      "tracking model": 122759,
      "model entity": 71087,
      "identifying textual": 51631,
      "textual mentions": 121720,
      "detection coreference": 29909,
      "coreference resolution": 23360,
      "resolution task": 100789,
      "task considering": 118008,
      "mentions names": 68462,
      "descriptions like": 29486,
      "task separate": 118694,
      "mention detection": 68419,
      "detection aspect": 29885,
      "limited using": 64298,
      "using local": 129825,
      "local features": 64920,
      "features learning": 43591,
      "modeling aspects": 72379,
      "task simultaneously": 118719,
      "able learn": 1661,
      "learn using": 62184,
      "highly complex": 50303,
      "nonlocal features": 80235,
      "features develop": 43453,
      "develop new": 30219,
      "new joint": 78966,
      "model explore": 71144,
      "explore utility": 41612,
      "utility features": 130480,
      "demonstrating effectiveness": 28971,
      "effectiveness task": 34960,
      "task pattern": 118526,
      "pattern based": 85711,
      "based term": 12111,
      "extraction approach": 42264,
      "approach japanese": 7653,
      "originally developed": 82567,
      "candidates based": 15132,
      "based loglikelihood": 11811,
      "approach suitable": 7946,
      "terms compound": 120295,
      "compound nouns": 20151,
      "olac extension": 81639,
      "extension dravidian": 41835,
      "dravidian languages": 34108,
      "paper intends": 83987,
      "project proposes": 92224,
      "ontological structure": 81828,
      "effective natural": 34717,
      "reference resolution": 97528,
      "framework cognitive": 45455,
      "model reference": 71878,
      "overcome difficulties": 83280,
      "difficulties previous": 31688,
      "approaches based": 8075,
      "entities model": 37821,
      "model accounts": 70528,
      "entities explicitly": 37782,
      "explicitly mentioned": 41377,
      "set potential": 107534,
      "important feature": 52163,
      "diverse phenomena": 32833,
      "approach provides": 7829,
      "fresh perspective": 45884,
      "problem reference": 91201,
      "framework syntactic": 45707,
      "syntactic annotations": 116372,
      "annotations widely": 6480,
      "widely recognized": 132550,
      "answer need": 6547,
      "need developed": 76797,
      "developed framework": 30278,
      "framework comprised": 45462,
      "abstract model": 1777,
      "model variety": 72298,
      "variety different": 130969,
      "different annotation": 30992,
      "tagging syntactic": 117451,
      "annotation coreference": 6288,
      "coreference annotation": 23339,
      "instantiated different": 55645,
      "annotators approach": 6489,
      "goals paper": 48420,
      "paper provide": 84374,
      "provide overview": 93889,
      "overview framework": 83366,
      "framework demonstrate": 45485,
      "demonstrate applicability": 28662,
      "diverse syntactic": 32851,
      "article proposes": 9153,
      "proposes method": 93602,
      "extract dependency": 42077,
      "dependency structures": 29230,
      "interactions words": 56008,
      "words using": 134296,
      "grammars lexicalized": 48675,
      "level words": 63523,
      "words dependency": 133905,
      "parsing process": 85200,
      "usual dependency": 130398,
      "dependency tree": 29233,
      "sheds new": 108163,
      "new light": 78995,
      "dependency parsing": 29171,
      "definitions present": 28539,
      "present method": 89545,
      "according dictionary": 1973,
      "large machine": 61134,
      "evaluate proposed": 38896,
      "method manually": 68954,
      "manually constructed": 67047,
      "constructed gold": 21661,
      "standard word": 112337,
      "given set": 48122,
      "specified set": 111622,
      "interannotator agreement": 56044,
      "annotator pair": 6485,
      "procedure propose": 91396,
      "method compared": 68710,
      "work word": 134879,
      "lexicon acquisition": 63872,
      "human languages": 50904,
      "recursive processing": 97289,
      "article make": 9140,
      "available linguistic": 11032,
      "linguistic community": 64441,
      "discuss linguistic": 32263,
      "languages related": 60839,
      "mathematical reasoning": 67460,
      "model theory": 72169,
      "parsing partofspeech": 85189,
      "partofspeech tagged": 85542,
      "texts natural": 121558,
      "language language": 59235,
      "language spoken": 60118,
      "spoken written": 112025,
      "formal languages": 45166,
      "languages languages": 60672,
      "used study": 128788,
      "formal logic": 45170,
      "information processing": 54873,
      "processing using": 91850,
      "using natural": 129933,
      "language called": 58864,
      "called natural": 15061,
      "input sentence": 55422,
      "aim produce": 4470,
      "grammatical structures": 48724,
      "structures sentences": 114104,
      "sentences assamese": 106219,
      "sentences parsing": 106440,
      "grammatical errors": 48700,
      "sentence error": 105858,
      "sentence sentence": 106066,
      "sentence simplification": 106077,
      "proteinprotein interaction": 93666,
      "interaction extraction": 55944,
      "extraction accurate": 42248,
      "accurate systems": 2367,
      "extracting proteinprotein": 42230,
      "proteinprotein interactions": 93668,
      "interactions ppis": 55994,
      "biomedical articles": 14165,
      "biomedical research": 14205,
      "research biomedical": 100433,
      "ppi extraction": 88690,
      "extraction problem": 42448,
      "neglected current": 76970,
      "current natural": 25304,
      "biomedical literature": 14193,
      "literature paper": 64763,
      "paper report": 84399,
      "impact automatic": 51859,
      "performance stateofart": 86741,
      "substantial improvement": 114858,
      "improvement recall": 52753,
      "recall sentence": 96343,
      "method applied": 68639,
      "significant impact": 108764,
      "effective sentence": 34744,
      "simplification automatic": 109578,
      "automatic processing": 10629,
      "biomedical text": 14206,
      "text complexity": 120807,
      "poses challenge": 88260,
      "challenge natural": 16059,
      "parsers typically": 85051,
      "typically trained": 126463,
      "trained largescale": 123185,
      "largescale corpora": 61412,
      "text propose": 121218,
      "text simplification": 121298,
      "simplification process": 109589,
      "reduce complexity": 97317,
      "improve performance": 52455,
      "performance syntactic": 86778,
      "syntactic parsers": 116442,
      "sentences syntactic": 106513,
      "steps text": 113336,
      "mining pipeline": 70255,
      "improvement performance": 52739,
      "processing steps": 91803,
      "evaluated method": 38988,
      "using corpus": 129506,
      "sentences annotated": 106208,
      "annotated syntactic": 6242,
      "results improvement": 101844,
      "simplified sentences": 109604,
      "sentences original": 106421,
      "original sentences": 82544,
      "17th century": 411,
      "century english": 15925,
      "english use": 37327,
      "contemporary english": 21848,
      "century 19th": 15922,
      "19th century": 473,
      "century use": 15926,
      "speakers english": 111315,
      "internet users": 56196,
      "users frequently": 129121,
      "common mistakes": 18892,
      "studying english": 114556,
      "different uses": 31548,
      "various models": 131135,
      "recognition translation": 97033,
      "entities case": 37754,
      "problem different": 91013,
      "nlp like": 79629,
      "automatic translation": 10704,
      "translation allows": 124635,
      "multilingual information": 75259,
      "information translation": 55057,
      "translation doesnt": 124767,
      "expected result": 40400,
      "ne context": 76707,
      "context propose": 22232,
      "method integrates": 68917,
      "translation transliteration": 125397,
      "used linguistic": 128614,
      "based local": 11808,
      "paper focus": 83946,
      "translation finally": 124823,
      "finally method": 44205,
      "method results": 69109,
      "results evaluation": 101771,
      "evaluation morphological": 39301,
      "electronic dictionaries": 35247,
      "dictionaries transducers": 30869,
      "albanian language": 4663,
      "analyze words": 6018,
      "segment text": 104566,
      "text study": 121335,
      "study relationship": 114501,
      "takes different": 117531,
      "identify words": 51578,
      "words created": 133894,
      "created simple": 24684,
      "simple concatenation": 109391,
      "morphological analysis": 74678,
      "event structure": 39543,
      "analysis example": 5607,
      "change word": 16375,
      "word types": 133612,
      "types word": 126386,
      "word tokens": 133605,
      "article provides": 9156,
      "lexical statistical": 63821,
      "statistical analysis": 113077,
      "speed word": 111881,
      "target texts": 117731,
      "dimensionality reduction": 31758,
      "text documents": 120883,
      "documents complex": 33205,
      "data important": 26019,
      "low dimensional": 65356,
      "dimensional embedding": 31749,
      "2d 3d": 875,
      "paper explore": 83912,
      "reduction methods": 97452,
      "knowledge order": 58091,
      "order achieve": 82266,
      "achieve better": 2424,
      "derived automatically": 29345,
      "corpus statistics": 24021,
      "linguistic resources": 64547,
      "dans la": 25538,
      "semantics language": 105428,
      "language provides": 60005,
      "provides means": 94049,
      "properties concepts": 92441,
      "conceptual representations": 20670,
      "representations space": 99900,
      "space propose": 111046,
      "propose formal": 92684,
      "analysis use": 5925,
      "use tools": 128331,
      "semantic content": 105018,
      "spatial relations": 111269,
      "presents semantic": 89896,
      "formal representations": 45175,
      "representations account": 99486,
      "account second": 2038,
      "formal set": 45178,
      "spatial concepts": 111262,
      "special attention": 111349,
      "les entits": 63393,
      "previous linguistic": 90419,
      "linguistic psycholinguistic": 64535,
      "psycholinguistic research": 94206,
      "reported paper": 99067,
      "focus language": 44778,
      "spatial entities": 111265,
      "experimental studies": 40732,
      "studies propose": 114265,
      "propose classification": 92581,
      "static dynamic": 113061,
      "cognitive processing": 18277,
      "formal computational": 45155,
      "computational analyses": 20357,
      "modelling categories": 72601,
      "research shows": 100625,
      "shows languagespecific": 108592,
      "space results": 111055,
      "question models": 95183,
      "models general": 73269,
      "general cognitive": 46642,
      "french paper": 45820,
      "describes details": 29396,
      "method morphological": 68971,
      "analysis research": 5808,
      "derivational morphology": 29329,
      "morphology morphological": 74793,
      "structure structure": 113963,
      "individual words": 53941,
      "morphological similarity": 74734,
      "properties morphological": 92467,
      "morphological paradigms": 74716,
      "directly used": 31916,
      "applications rely": 7005,
      "rely implicit": 98712,
      "implicit information": 52011,
      "information paper": 54821,
      "paper introduce": 83988,
      "based global": 11741,
      "information unique": 55067,
      "operations performed": 82072,
      "experiment conducted": 40456,
      "conducted generate": 20929,
      "report work": 99056,
      "work progress": 134708,
      "work focusing": 134539,
      "edit histories": 34534,
      "simple english": 109426,
      "english wikipedia": 37339,
      "task consider": 118005,
      "main approaches": 66397,
      "mixture different": 70418,
      "different operations": 31309,
      "using metadata": 129876,
      "simplification operations": 109588,
      "methods outperform": 69654,
      "reasonable baseline": 96208,
      "baseline yield": 12342,
      "manually prepared": 67073,
      "change way": 16374,
      "work proposed": 134750,
      "method learning": 68936,
      "learning english": 62547,
      "requires access": 100241,
      "access highquality": 1910,
      "negative polarity": 76941,
      "polarity items": 87913,
      "english languages": 37189,
      "approach applied": 7357,
      "applied languages": 7084,
      "study apply": 114323,
      "apply method": 7183,
      "analysis suggests": 5880,
      "suggests interesting": 115374,
      "rich syntactic": 102794,
      "language linguistic": 59255,
      "features explicitly": 43500,
      "lexicon features": 63883,
      "described literature": 29376,
      "properties make": 92464,
      "various natural": 131140,
      "describes method": 29413,
      "method build": 68680,
      "structure parsing": 113930,
      "obtain information": 81296,
      "information needed": 54793,
      "grammars used": 48685,
      "used parsing": 128681,
      "dependency relation": 29219,
      "patterns used": 85791,
      "used express": 128539,
      "set constraints": 107401,
      "opinion polarity": 82096,
      "people think": 85896,
      "piece information": 87494,
      "information various": 55085,
      "various decisionmaking": 131073,
      "people frequently": 85877,
      "frequently make": 45871,
      "opinions available": 82118,
      "available internet": 11019,
      "web resources": 132249,
      "opinions product": 82126,
      "product reviews": 92045,
      "reviews forums": 102609,
      "large information": 61117,
      "wide range": 132496,
      "read reviews": 95938,
      "informed decision": 55168,
      "customer opinions": 25440,
      "customer reviews": 25442,
      "opinion mining": 82089,
      "mining important": 70234,
      "extraction important": 42353,
      "important topics": 52283,
      "research area": 100410,
      "identification opinion": 51414,
      "review usually": 102587,
      "values positive": 130799,
      "positive negative": 88325,
      "negative neutral": 76935,
      "propose technique": 93111,
      "identifying polarity": 51609,
      "polarity reviews": 87923,
      "polarity adjectives": 87904,
      "evaluation shows": 39396,
      "shows technique": 108640,
      "naive bayesian": 76021,
      "bayesian classifiers": 12580,
      "dans les": 25539,
      "french italian": 45813,
      "italian corpora": 57085,
      "multiword term": 75949,
      "independent domain": 53765,
      "domain language": 33566,
      "language lexical": 59251,
      "depends domain": 29285,
      "rapidly evolving": 95727,
      "term study": 120245,
      "nous montrons": 80452,
      "controlled natural": 22840,
      "especially used": 38523,
      "paper new": 84052,
      "designed specifically": 29625,
      "different parsers": 31318,
      "parsers implemented": 85027,
      "large subset": 61285,
      "article presents": 9147,
      "languages main": 60714,
      "main features": 66420,
      "features new": 43627,
      "translation dictionaries": 124750,
      "comparison shown": 19573,
      "classification scheme": 17386,
      "proposed new": 93509,
      "new concepts": 78836,
      "new generation": 78936,
      "generation available": 47321,
      "universal partofspeech": 127317,
      "facilitate future": 42769,
      "future research": 46291,
      "research unsupervised": 100659,
      "unsupervised induction": 127643,
      "partofspeech categories": 85524,
      "categories addition": 15728,
      "different treebank": 31519,
      "universal set": 127335,
      "set result": 107567,
      "treebank data": 125632,
      "universal tagset": 127336,
      "dataset consisting": 26818,
      "different languages": 31195,
      "highlight use": 50275,
      "resource experiments": 100844,
      "experiments including": 40960,
      "competitive accuracies": 19632,
      "partofspeech tags": 85565,
      "multilingual lexicon": 75271,
      "database management": 26642,
      "mt paper": 74981,
      "presents design": 89838,
      "design development": 29530,
      "main requirements": 66458,
      "required mt": 100220,
      "programming language": 92129,
      "allows easily": 5145,
      "able run": 1685,
      "notes electronic": 80391,
      "topics covered": 122620,
      "electronic dictionary": 35248,
      "closely related": 17860,
      "related languages": 97874,
      "multiple answers": 75493,
      "efficient parser": 35101,
      "language specification": 60114,
      "applications implementation": 6934,
      "software development": 110519,
      "development data": 30378,
      "data integration": 26049,
      "mining natural": 70243,
      "unlike traditional": 127452,
      "kinds grammars": 57688,
      "needs general": 76892,
      "general parser": 46695,
      "able deal": 1628,
      "ambiguities paper": 5276,
      "lexical syntactic": 63826,
      "syntactic ambiguity": 116366,
      "enables use": 36406,
      "semantic relatedness": 105195,
      "relatedness measure": 97923,
      "knowledge new": 58086,
      "new semantic": 79122,
      "explicit semantic": 41344,
      "highest results": 50234,
      "using prediction": 130040,
      "svm classifier": 116231,
      "classifier trained": 17586,
      "wikipedia data": 132649,
      "data various": 26618,
      "word sentence": 133537,
      "level use": 63517,
      "use pronunciation": 128215,
      "pronunciation analogy": 92365,
      "analogy text": 5428,
      "persian language": 87079,
      "language text": 60161,
      "speech synthesis": 111810,
      "world text": 135051,
      "popular languages": 88096,
      "languages english": 60527,
      "spanish french": 111157,
      "languages similar": 60874,
      "similar importance": 109097,
      "importance research": 52076,
      "research persian": 100582,
      "increase complexity": 53590,
      "complexity text": 19943,
      "example short": 39795,
      "short vowels": 108267,
      "written text": 135151,
      "persian text": 87086,
      "text phonetic": 121178,
      "grammatical rules": 48719,
      "extraction normalization": 42415,
      "process involves": 91514,
      "rule matching": 103353,
      "matching multiple": 67419,
      "multiple dictionaries": 75539,
      "achieves 98": 2730,
      "clustering based": 17939,
      "local sequence": 64937,
      "sequence alignment": 106915,
      "alignment metrics": 4971,
      "learning based": 62383,
      "based finding": 11718,
      "paper authors": 83749,
      "location organization": 64966,
      "potentially help": 88613,
      "analyzing large": 6044,
      "large social": 61274,
      "social networks": 110443,
      "author disambiguation": 10361,
      "correcting errors": 24137,
      "errors ocr": 38395,
      "ocr output": 81527,
      "graphical user": 48934,
      "user interface": 129007,
      "interface available": 56098,
      "available download": 10987,
      "source sentence": 110810,
      "improve recall": 52515,
      "biomedical information": 14188,
      "use novel": 128178,
      "novel model": 80643,
      "model sentence": 71973,
      "automatic discourse": 10517,
      "discourse analysis": 32039,
      "analysis information": 5656,
      "simplification improving": 109583,
      "improving human": 53103,
      "approach produces": 7821,
      "versions original": 131624,
      "original sentence": 82542,
      "sentence combining": 105789,
      "constituent elements": 21537,
      "tool optimized": 122396,
      "scientific literature": 104011,
      "extraction improved": 42355,
      "improved fscore": 52607,
      "tool test": 122401,
      "object model": 81052,
      "model defined": 70961,
      "defined iso": 28500,
      "annotation framework": 6314,
      "best practices": 13411,
      "additional features": 3515,
      "features support": 43749,
      "support variety": 115997,
      "variety syntactic": 131020,
      "phenomena including": 87226,
      "constituent dependency": 21535,
      "different node": 31303,
      "finally case": 44150,
      "study german": 114394,
      "constituent structures": 21549,
      "translate english": 124526,
      "languages present": 60802,
      "translate natural": 124531,
      "language sentences": 60078,
      "sentences formulas": 106325,
      "representation language": 99273,
      "language uses": 60311,
      "operators using": 82079,
      "using input": 129759,
      "representation words": 99469,
      "words phrases": 134116,
      "phrases sentences": 87453,
      "languages including": 60634,
      "query languages": 94971,
      "answer set": 6580,
      "uses syntactic": 129287,
      "parser parse": 84981,
      "sentences construct": 106258,
      "semantic meaning": 105106,
      "meaning sentences": 67688,
      "sentences directed": 106279,
      "directed parsing": 31817,
      "parser used": 85004,
      "used addition": 128379,
      "addition inverse": 3443,
      "uses notion": 129257,
      "learn semantic": 62136,
      "words semantic": 134187,
      "existing statistical": 40297,
      "statistical learning": 113103,
      "approach assign": 7367,
      "assign weights": 9547,
      "deal multiple": 27863,
      "multiple meanings": 75613,
      "meanings words": 67752,
      "improved results": 52637,
      "standard corpora": 112216,
      "corpora natural": 23534,
      "language interfaces": 59219,
      "command control": 18788,
      "database queries": 26646,
      "queries language": 94924,
      "human level": 50906,
      "understand natural": 126761,
      "needs able": 76884,
      "text answer": 120647,
      "answer questions": 6563,
      "questions given": 95311,
      "given natural": 48065,
      "language respect": 60059,
      "respect text": 101107,
      "text needs": 121142,
      "able follow": 1640,
      "instructions given": 55711,
      "language achieve": 58808,
      "able process": 1673,
      "language able": 58805,
      "able capture": 1616,
      "capture knowledge": 15321,
      "knowledge text": 58206,
      "able translate": 1701,
      "text formal": 120961,
      "language discuss": 58972,
      "approach translation": 7985,
      "translation achieved": 124621,
      "words sentence": 134193,
      "approach uses": 8001,
      "method developed": 68774,
      "learn meaning": 62089,
      "words meaning": 134049,
      "initial lexicon": 55218,
      "lexicon present": 63903,
      "improved method": 52615,
      "method initial": 68909,
      "training sentence": 123835,
      "sentence meaning": 105945,
      "pairs evaluate": 83534,
      "methods compare": 69381,
      "compare existing": 19246,
      "query expansion": 94961,
      "selection using": 104845,
      "using ewc": 129647,
      "measure paper": 67809,
      "retrieval task": 102431,
      "task measure": 118382,
      "experiments open": 41048,
      "data proposed": 26294,
      "proposed technique": 93566,
      "data collection": 25740,
      "experiments demonstrated": 40896,
      "demonstrated promising": 28926,
      "promising results": 92290,
      "interrater agreement": 56295,
      "sentence formality": 105877,
      "important dimensions": 52144,
      "writing style": 135098,
      "style variation": 114609,
      "study conducted": 114338,
      "assessing sentence": 9518,
      "likert scale": 64147,
      "agreement results": 4405,
      "results different": 101725,
      "distributions different": 32757,
      "different sentence": 31413,
      "analysis identify": 5646,
      "main objective": 66446,
      "design automatic": 29518,
      "automatic scoring": 10646,
      "scoring mechanism": 104230,
      "study important": 114402,
      "important purpose": 52220,
      "learning content": 62455,
      "content selection": 21939,
      "selection rules": 104824,
      "rules generating": 103405,
      "object descriptions": 81049,
      "taskoriented dialogue": 118889,
      "ability generate": 1524,
      "task domain": 118115,
      "work large": 134602,
      "number models": 80914,
      "models proposed": 73822,
      "proposed paper": 93516,
      "corpus taskoriented": 24035,
      "sets based": 107650,
      "incremental model": 53733,
      "model use": 72253,
      "learning experiment": 62562,
      "automatically learn": 10797,
      "learn model": 62091,
      "model requires": 71912,
      "representation discourse": 99212,
      "discourse structure": 32090,
      "structure corpus": 113830,
      "corpus annotations": 23664,
      "annotations used": 6478,
      "used derive": 128483,
      "structure discourse": 113838,
      "representations discourse": 99599,
      "structure based": 113812,
      "based purely": 11961,
      "train test": 123037,
      "selection component": 104774,
      "corpus knowledge": 23850,
      "description generation": 29463,
      "generation dialogue": 47368,
      "separate content": 106847,
      "selection models": 104804,
      "models based": 72802,
      "based theoretical": 12119,
      "models independently": 73395,
      "achieve accuracies": 2412,
      "accuracies significantly": 2059,
      "significantly majority": 108977,
      "majority class": 66596,
      "class baseline": 17031,
      "unseen test": 127549,
      "performing significantly": 86969,
      "significantly better": 108882,
      "best performing": 13399,
      "performing models": 86959,
      "models combine": 72923,
      "sets achieving": 107644,
      "surprisingly simple": 116147,
      "structure knowledge": 113889,
      "knowledge empirical": 57889,
      "empirical comparison": 36155,
      "model discourse": 71003,
      "model generation": 71244,
      "generation task": 47656,
      "task automatic": 117908,
      "automatic transcription": 10702,
      "method evaluation": 68818,
      "nlp software": 79690,
      "corpus spanning": 24008,
      "literary texts": 64739,
      "various lexical": 131121,
      "morphological grammatical": 74707,
      "graphs built": 48940,
      "forms contemporary": 45236,
      "previous research": 90446,
      "textual analysis": 121673,
      "need focus": 76812,
      "tool analyze": 122371,
      "analyze possible": 5992,
      "work using": 134875,
      "using results": 130115,
      "results previous": 102052,
      "research proposing": 100597,
      "transcription method": 124000,
      "method words": 69229,
      "words sequences": 134206,
      "objectoriented semantics": 81144,
      "english natural": 37215,
      "approach problem": 7818,
      "problem natural": 91138,
      "understanding proposed": 126931,
      "proposed knowledge": 93318,
      "domain consideration": 33483,
      "consideration social": 21281,
      "social behavior": 110324,
      "behavior people": 12666,
      "english sentences": 37273,
      "sentences translated": 106528,
      "semantic database": 105027,
      "events messages": 39585,
      "machines things": 66343,
      "location time": 64968,
      "relations objects": 98223,
      "objects knowledge": 81148,
      "description semantics": 29474,
      "contextfree grammars": 22361,
      "traditional language": 122823,
      "processing tools": 91844,
      "tools need": 122460,
      "able parse": 1668,
      "following approach": 44961,
      "constraints paper": 21607,
      "chart parser": 16646,
      "data formats": 25957,
      "explore possibility": 41573,
      "research industrial": 100535,
      "commonly uses": 18975,
      "formats allow": 45218,
      "annotated spoken": 6239,
      "tools used": 122480,
      "used produce": 128702,
      "domain adaptation": 33428,
      "biggest challenges": 14004,
      "challenges development": 16148,
      "development deployment": 30383,
      "dialogue systems": 30762,
      "systems design": 116840,
      "challenge arises": 15996,
      "adapt features": 3178,
      "features dialogue": 43455,
      "dialogue domain": 30667,
      "domain user": 33691,
      "dialogue context": 30648,
      "promising approach": 92265,
      "generation uses": 47691,
      "knowledge automatically": 57760,
      "automatically adapted": 10713,
      "individual user": 53938,
      "user group": 128993,
      "sentence planner": 105984,
      "restaurant information": 101320,
      "complex information": 19820,
      "quality comparable": 94612,
      "tuned domain": 125932,
      "domain method": 33581,
      "method easily": 68788,
      "generally perform": 46872,
      "perform better": 85953,
      "better models": 13631,
      "models trained": 74192,
      "trained tested": 123307,
      "selection knowledge": 104790,
      "knowledge results": 58158,
      "results provide": 102086,
      "provide demonstration": 93800,
      "structure sentence": 113948,
      "sentence structure": 106092,
      "responses finally": 101267,
      "finally evaluate": 44179,
      "evaluate contribution": 38813,
      "ngram features": 79433,
      "features features": 43513,
      "features based": 43376,
      "higherlevel linguistic": 50215,
      "personal information": 87120,
      "annotation paper": 6345,
      "presents preliminary": 89890,
      "works online": 134965,
      "transcription corpus": 123996,
      "set metadata": 107492,
      "recognition named": 96912,
      "annotation tasks": 6385,
      "transducer cascades": 124018,
      "recognize named": 97055,
      "information speaker": 54998,
      "annotated corpus": 6145,
      "kind information": 57678,
      "paper evaluate": 83896,
      "evaluate various": 38943,
      "lexicon built": 63876,
      "new version": 79240,
      "obtained merging": 81383,
      "converted format": 23100,
      "used evaluation": 128527,
      "evaluation campaign": 39139,
      "features make": 43603,
      "make consistent": 66637,
      "including features": 53295,
      "plain text": 87625,
      "text format": 120962,
      "directly usable": 31914,
      "applications natural": 6970,
      "presents work": 89914,
      "work relies": 134773,
      "finegrained linguistic": 44363,
      "information provided": 54887,
      "provided existing": 93967,
      "various features": 131098,
      "features encoded": 43479,
      "hand simple": 49326,
      "simple compound": 109390,
      "different types": 31523,
      "manually evaluated": 67060,
      "freely available": 45780,
      "lgpllr license": 63951,
      "twitter messages": 126111,
      "using dataset": 129534,
      "dataset 29": 26713,
      "29 million": 870,
      "furthermore present": 46202,
      "present attempt": 89385,
      "attempt classify": 9735,
      "classes using": 17079,
      "approach overall": 7782,
      "sentences augmented": 106221,
      "transformationbased learning": 124275,
      "learning function": 62608,
      "tagging paper": 117431,
      "paper function": 83963,
      "tagging using": 117462,
      "using transformation": 130312,
      "transformation based": 124262,
      "based learning": 11799,
      "rules developed": 103394,
      "developed using": 30317,
      "achieving good": 2948,
      "results method": 101915,
      "lexical relations": 63795,
      "relations sentences": 98253,
      "sentences use": 106534,
      "use context": 127958,
      "context free": 22110,
      "free grammar": 45758,
      "language rely": 60034,
      "rely function": 98697,
      "visualization analysis": 131824,
      "analysis frames": 5628,
      "content analysis": 21855,
      "generate semantic": 47007,
      "semantic map": 105102,
      "statements using": 112549,
      "using freely": 129688,
      "relevant statistics": 98589,
      "discussed various": 32302,
      "social systems": 110459,
      "systems theory": 117196,
      "framing agendasetting": 45749,
      "communication information": 19036,
      "information network": 54797,
      "social network": 110438,
      "network analysis": 77134,
      "meaning vector": 67705,
      "vector space": 131369,
      "space vector": 111080,
      "network relations": 77407,
      "space words": 111087,
      "words related": 134164,
      "expected observed": 40393,
      "grammar using": 48664,
      "using predictive": 130041,
      "parser context": 84930,
      "bangla language": 11416,
      "language propose": 59997,
      "parser based": 84923,
      "grammar approach": 48622,
      "approach general": 7584,
      "language grammar": 59109,
      "proposed parser": 93518,
      "parser construct": 84929,
      "language successfully": 60134,
      "successfully parsed": 115191,
      "grammar checking": 48625,
      "language possible": 59815,
      "proposed scheme": 93541,
      "scheme based": 103914,
      "parsing method": 85156,
      "analysis translation": 5908,
      "analysis method": 5693,
      "using semantic": 130139,
      "spanning tree": 111191,
      "tree using": 125619,
      "wikipedia page": 132667,
      "given term": 48151,
      "sense minimal": 105692,
      "tree method": 125601,
      "method based": 68660,
      "tree evaluate": 125589,
      "evaluate method": 38852,
      "method text": 69187,
      "text classification": 120707,
      "increases precision": 53652,
      "provide hints": 93845,
      "reasoning natural": 96278,
      "understanding based": 126799,
      "based objectoriented": 11905,
      "semantics algorithms": 105396,
      "computer oriented": 20484,
      "oriented input": 82497,
      "semantic processing": 105184,
      "processing text": 91837,
      "text information": 121053,
      "information presented": 54863,
      "direct comparison": 31785,
      "objects question": 81149,
      "question database": 95144,
      "classes problems": 17072,
      "problems considered": 91309,
      "form answer": 45068,
      "reasoning used": 96323,
      "used knowledge": 128598,
      "reasoning based": 96229,
      "social psychology": 110449,
      "psychology proposed": 94218,
      "proposed algorithms": 93176,
      "computer systems": 20492,
      "systems closely": 116782,
      "closely connected": 17851,
      "connected text": 21129,
      "processing criminology": 91647,
      "criminology operation": 24777,
      "operation business": 82053,
      "business medicine": 14976,
      "medicine document": 68235,
      "document systems": 33091,
      "similarity recent": 109291,
      "systems approach": 116734,
      "approach argue": 7364,
      "parser combining": 84927,
      "effects natural": 34996,
      "natural sentence": 76619,
      "sentence processing": 106005,
      "estimates future": 38643,
      "space representations": 111051,
      "tensor product": 120191,
      "product representations": 92043,
      "representations paradigm": 99799,
      "representation space": 99415,
      "sentences paper": 106423,
      "based grammatical": 11745,
      "sentences combine": 106244,
      "challenge statistical": 16105,
      "sentences correct": 106265,
      "segmentation pos": 104618,
      "pos partofspeech": 88212,
      "partofspeech tagging": 85547,
      "tagging chunking": 117376,
      "information use": 55073,
      "use naive": 128159,
      "tags word": 117489,
      "tagged corpus": 117347,
      "corpus myanmar": 23892,
      "experiments analysis": 40765,
      "achieves good": 2790,
      "good result": 48498,
      "result simple": 101406,
      "simple sentences": 109517,
      "sentences complex": 106250,
      "complex sentences": 19876,
      "translation crosslanguage": 124722,
      "automata theory": 10424,
      "recently machine": 96707,
      "translation literature": 124912,
      "paper review": 84409,
      "rules machine": 103420,
      "translation systems": 125307,
      "present implementation": 89513,
      "translation rules": 125227,
      "rogets thesaurus": 103149,
      "thesaurus semantic": 121842,
      "measures semantic": 67895,
      "similarity using": 109328,
      "compare results": 19290,
      "results tests": 102268,
      "similarity measures": 109261,
      "noun pairs": 80426,
      "pairs human": 83559,
      "human judges": 50878,
      "using rogets": 130125,
      "rogets wordnet": 103152,
      "questions correct": 95291,
      "correct synonym": 24123,
      "group words": 49145,
      "agreement propose": 4401,
      "segmentation evaluation": 104585,
      "evaluation metric": 39276,
      "metric called": 69873,
      "quantifies similarity": 94843,
      "penalty function": 85852,
      "size propose": 109941,
      "improvement state": 52763,
      "propose using": 93148,
      "evaluate automatic": 38798,
      "terms human": 120335,
      "human performance": 50927,
      "indus script": 54055,
      "marathi language": 67172,
      "languages indian": 60642,
      "paper analyses": 83730,
      "writing systems": 135104,
      "script writing": 104262,
      "developed use": 30316,
      "spoken words": 112024,
      "similar sounding": 109147,
      "sounding words": 110707,
      "languages india": 60641,
      "corpora hundreds": 23495,
      "use early": 128021,
      "combined use": 18672,
      "text knowledge": 121070,
      "prototype prototype": 93686,
      "context knowledge": 22155,
      "limited data": 64227,
      "set called": 107381,
      "corpus corpus": 23728,
      "corpus collection": 23696,
      "text drawn": 120893,
      "sources used": 110926,
      "used test": 128807,
      "set evaluate": 107432,
      "systems available": 116750,
      "available corpus": 10967,
      "corpus domain": 23765,
      "representative corpus": 100008,
      "corpus evaluation": 23784,
      "major components": 66547,
      "components lexical": 20030,
      "knowledge model": 58063,
      "model evaluation": 71114,
      "grammar structures": 48658,
      "structures leads": 114084,
      "straight forward": 113427,
      "identification extraction": 51388,
      "framework semantic": 45676,
      "rhetorical analysis": 102710,
      "analysis corpus": 5550,
      "corpus prove": 23946,
      "corpus analyze": 23658,
      "address task": 3770,
      "task assigning": 117902,
      "tags context": 117468,
      "parsing task": 85257,
      "tagging use": 117461,
      "information propose": 54880,
      "parsing simple": 85239,
      "syntactic ambiguities": 116364,
      "ambiguities resolved": 5277,
      "furthermore traditional": 46217,
      "parsing tools": 85265,
      "mechanisms allow": 68053,
      "context account": 21997,
      "tool allows": 122370,
      "models arbitrary": 72767,
      "abstract syntax": 1782,
      "expression power": 41742,
      "based ngram": 11898,
      "use areas": 127898,
      "paraphrasing text": 84851,
      "translation word": 125430,
      "speech induction": 111696,
      "accurately identifying": 2384,
      "word large": 133331,
      "large scale": 61241,
      "experiments difficult": 40910,
      "difficult paper": 31651,
      "introduce new": 56471,
      "search algorithm": 104294,
      "sentence based": 105776,
      "vocabulary size": 131902,
      "dataset 100": 26704,
      "wsj section": 135179,
      "section penn": 104488,
      "treebank available": 125628,
      "sur le": 116054,
      "insights concerning": 55529,
      "number syllables": 80975,
      "relation words": 98086,
      "words paper": 134101,
      "describes tool": 29443,
      "tool based": 122374,
      "looking word": 65225,
      "access time": 1924,
      "target word": 117742,
      "linguistic databases": 64462,
      "automatically dictionary": 10759,
      "domains used": 33882,
      "combined database": 18651,
      "project explores": 92216,
      "nature language": 76657,
      "similar used": 109165,
      "existing natural": 40221,
      "processing methods": 91711,
      "methods limited": 69598,
      "limited scope": 64275,
      "understanding aims": 126796,
      "gain understanding": 46355,
      "understanding language": 126873,
      "initial input": 55216,
      "morphology language": 74789,
      "language using": 60312,
      "use frequency": 128060,
      "frequency distributions": 45841,
      "english french": 37139,
      "texts analyzed": 121455,
      "analyzed determine": 6023,
      "texts considered": 121481,
      "spanish texts": 111175,
      "texts second": 121600,
      "syntax language": 116539,
      "using recursive": 130096,
      "program uses": 92116,
      "methods analyze": 69308,
      "given sentences": 48118,
      "sentences based": 106226,
      "based sentence": 12015,
      "sentence patterns": 105982,
      "surrounding words": 116158,
      "words methods": 134054,
      "able understand": 1702,
      "understand structure": 126778,
      "structure simple": 113956,
      "sentences learn": 106375,
      "learn new": 62101,
      "new words": 79253,
      "words addition": 133808,
      "future work": 46320,
      "natural text": 76623,
      "approach universal": 7992,
      "universal networking": 127316,
      "language used": 60303,
      "used represent": 128725,
      "represent semantic": 99135,
      "data extracted": 25938,
      "presents novel": 89880,
      "novel approach": 80468,
      "approach converting": 7461,
      "performs morphological": 87010,
      "semantic lexical": 105099,
      "lexical analysis": 63730,
      "analysis text": 5895,
      "analysis produces": 5774,
      "like structure": 64099,
      "represented using": 100036,
      "universal language": 127309,
      "language translation": 60185,
      "translation method": 124933,
      "learning map": 62707,
      "map sentences": 67113,
      "logical form": 65002,
      "form structured": 45129,
      "classification probabilistic": 17343,
      "categorial grammars": 15721,
      "paper addresses": 83712,
      "addresses problem": 3802,
      "problem mapping": 91122,
      "mapping natural": 67140,
      "takes input": 117534,
      "input training": 55459,
      "set sentences": 107576,
      "sentences labeled": 106365,
      "lambda calculus": 58791,
      "loglinear model": 65045,
      "model represents": 71910,
      "conditioned input": 20801,
      "method task": 69178,
      "learning natural": 62825,
      "interfaces databases": 56107,
      "databases learned": 26651,
      "outperform previous": 82724,
      "previous methods": 90425,
      "methods benchmark": 69340,
      "approach extracting": 7563,
      "study presents": 114477,
      "based clustering": 11577,
      "clustering technique": 17956,
      "study uses": 114542,
      "uses combination": 129205,
      "cover aspects": 24481,
      "given corpus": 48008,
      "corpus suggests": 24027,
      "groups according": 49150,
      "according probability": 1995,
      "meaningful way": 67733,
      "way search": 132128,
      "search space": 104349,
      "average word": 11217,
      "word length": 133337,
      "cultural changes": 25227,
      "average length": 11199,
      "length words": 63384,
      "russian english": 103490,
      "words belonging": 133836,
      "diachronic text": 30490,
      "corpus google": 23821,
      "google books": 48517,
      "books ngram": 14424,
      "growing rapidly": 49180,
      "20th century": 755,
      "words contributed": 133886,
      "content words": 21976,
      "functional words": 46063,
      "words contribute": 133885,
      "length word": 63383,
      "shown words": 108541,
      "words reflect": 134162,
      "personal pronouns": 87123,
      "evaluation computational": 39155,
      "computational grammar": 20380,
      "languages natural": 60752,
      "prominent research": 92253,
      "make process": 66711,
      "parser development": 84936,
      "indian context": 53816,
      "formalism used": 45187,
      "used question": 128711,
      "question needs": 95189,
      "focus problem": 44806,
      "analyze different": 5964,
      "lexical variation": 63839,
      "variation social": 130899,
      "social media": 110348,
      "media present": 68154,
      "present study": 89723,
      "relationship gender": 98298,
      "linguistic style": 64561,
      "style social": 114595,
      "networks using": 77811,
      "using novel": 129978,
      "novel corpus": 80517,
      "twitter users": 126130,
      "users prior": 129157,
      "social variable": 110461,
      "users natural": 129145,
      "dataset various": 27270,
      "various styles": 131209,
      "gendered language": 46622,
      "language styles": 60132,
      "patterns identify": 85749,
      "identify individuals": 51509,
      "train statistical": 123028,
      "statistical classifier": 113086,
      "dataset examining": 26901,
      "language does": 58982,
      "does match": 33368,
      "classifiers model": 17619,
      "model gender": 71224,
      "gender social": 46613,
      "networks include": 77623,
      "significantly fewer": 108913,
      "computational methods": 20405,
      "methods social": 69762,
      "offers new": 81589,
      "new perspective": 79074,
      "emerges individuals": 36063,
      "transitionbased dependency": 124502,
      "dependency parsers": 29164,
      "makes possible": 66798,
      "software engineering": 110521,
      "use classifier": 127938,
      "experiments multilingual": 41026,
      "parsing variety": 85278,
      "earlier work": 34381,
      "work suggested": 134831,
      "good choice": 48469,
      "lowresource parsing": 65554,
      "support hypothesis": 115973,
      "machines better": 66331,
      "parsing performance": 85191,
      "size training": 109947,
      "principled approach": 90680,
      "direct mapping": 31791,
      "proposed improve": 93312,
      "representation systems": 99431,
      "systems query": 117091,
      "popular approach": 88075,
      "approach solve": 7922,
      "solve problem": 110603,
      "easy read": 34483,
      "look ahead": 65211,
      "possible continuations": 88395,
      "unfinished sentence": 127072,
      "lookahead features": 65221,
      "features difficult": 43459,
      "way existing": 132077,
      "references methods": 97559,
      "algorithms presented": 4875,
      "presented new": 89790,
      "specifically designed": 111539,
      "parser implementation": 84962,
      "shows approach": 108549,
      "approach practical": 7804,
      "efficient using": 35128,
      "using external": 129657,
      "external sources": 42044,
      "bilingual information": 14040,
      "alignment paper": 4976,
      "new simple": 79137,
      "based use": 12149,
      "information machine": 54743,
      "trained small": 123281,
      "results comparable": 101590,
      "comparable obtained": 19152,
      "obtained stateoftheart": 81417,
      "alignment error": 4956,
      "pairs sentences": 83636,
      "sentences provides": 106462,
      "provides results": 94068,
      "trained indomain": 123160,
      "indomain corpus": 53957,
      "sentences furthermore": 106326,
      "furthermore results": 46212,
      "obtained indicate": 81380,
      "use trained": 128336,
      "pair sentences": 83452,
      "sentences natural": 106405,
      "paper define": 83799,
      "language semantic": 60075,
      "relations events": 98157,
      "based definition": 11633,
      "consider text": 21233,
      "text understanding": 121383,
      "process having": 91499,
      "events basic": 39566,
      "basic unit": 12537,
      "joint space": 57323,
      "space neural": 111032,
      "neural probabilistic": 78632,
      "probabilistic language": 90812,
      "model statistical": 72087,
      "translation neural": 125005,
      "model nplm": 71595,
      "better perplexity": 13668,
      "perplexity ngram": 87067,
      "application area": 6833,
      "nlp specifically": 79692,
      "monolingual resources": 74609,
      "resources introduce": 100991,
      "using nonparametric": 129976,
      "nonparametric bayesian": 80272,
      "order facilitate": 82324,
      "various tasks": 131213,
      "tasks propose": 119405,
      "propose joint": 92731,
      "space model": 111023,
      "model experiment": 71125,
      "bleu points": 14302,
      "trained relatively": 123263,
      "relatively small": 98414,
      "sentence pairs": 105967,
      "case long": 15593,
      "long training": 65143,
      "training time": 123922,
      "language learning": 59241,
      "native language": 76223,
      "apparently requiring": 6788,
      "mistakes corrected": 70349,
      "problems language": 91331,
      "review recent": 102567,
      "results showing": 102176,
      "sufficient data": 115277,
      "encoding linguistic": 36707,
      "linguistic input": 64492,
      "input results": 55418,
      "results include": 101848,
      "ability learn": 1536,
      "learn linguistic": 62083,
      "prediction grammaticality": 89067,
      "language production": 59990,
      "simplicity approach": 109566,
      "analyse ability": 5431,
      "learn specific": 62154,
      "specific linguistic": 111460,
      "framework describing": 45487,
      "human language": 50896,
      "rapid development": 95713,
      "development natural": 30401,
      "understanding module": 126892,
      "conversational agent": 22985,
      "urgent need": 127847,
      "order test": 82418,
      "application real": 6877,
      "real users": 96087,
      "wizard oz": 132760,
      "environment used": 38193,
      "used agent": 128383,
      "agent able": 4295,
      "capture interactions": 15320,
      "understand people": 126765,
      "people react": 85886,
      "learning paradigm": 62872,
      "understanding natural": 126896,
      "language classification": 58876,
      "classification problem": 17344,
      "answers questions": 6744,
      "domain approach": 33470,
      "approach used": 7996,
      "language interface": 59218,
      "hierarchical model": 49976,
      "describes submission": 29433,
      "using publicly": 130066,
      "publicly available": 94289,
      "available tools": 11129,
      "accomplish task": 1952,
      "provided training": 93988,
      "data built": 25715,
      "built translation": 14944,
      "translation model": 124937,
      "using moses": 129900,
      "moses toolkit": 74824,
      "chart decoder": 16645,
      "decoder implemented": 28022,
      "language pair": 59728,
      "believe approach": 12695,
      "approach work": 8020,
      "language pairs": 59736,
      "significant improvement": 108765,
      "improvement baseline": 52684,
      "metrics detailed": 69950,
      "reproduce results": 100082,
      "results possible": 102044,
      "possible directions": 88397,
      "directions improvements": 31844,
      "approach modeling": 7713,
      "documents building": 33196,
      "increasingly complex": 53695,
      "ensure quality": 37641,
      "quality building": 94608,
      "particularly interested": 85485,
      "used validate": 128837,
      "sparql language": 111217,
      "language second": 60070,
      "second model": 104430,
      "model processes": 71794,
      "technical documents": 119746,
      "documents order": 33262,
      "order acquire": 82269,
      "related events": 97861,
      "tend appear": 120142,
      "set related": 107560,
      "generation usually": 47695,
      "usually manually": 130437,
      "methods inducing": 69549,
      "proposed recently": 93532,
      "typically use": 126466,
      "ad hoc": 3161,
      "propose probabilistic": 93000,
      "probabilistic approach": 90797,
      "events participants": 39591,
      "latent topics": 61620,
      "best explain": 13331,
      "text number": 121152,
      "novel application": 80466,
      "parsing endtoend": 85106,
      "extracted facts": 42156,
      "produced stateoftheart": 91969,
      "stateoftheart results": 112890,
      "substantially reducing": 114912,
      "engineering effort": 37047,
      "fully automatic": 45947,
      "step forward": 113266,
      "reports results": 99089,
      "main goal": 66424,
      "previously applied": 90587,
      "work extend": 134524,
      "apply technique": 7222,
      "experiments showed": 41134,
      "general obtain": 46689,
      "obtain good": 81286,
      "different tasks": 31477,
      "tasks important": 119168,
      "important step": 52256,
      "resources fully": 100982,
      "automatic lexical": 10577,
      "semantic classification": 105009,
      "classification nouns": 17309,
      "nouns work": 80449,
      "work present": 134674,
      "classification english": 17197,
      "automatically acquire": 10710,
      "previously known": 90604,
      "lexical classes": 63739,
      "achieved using": 2720,
      "particular aspects": 85382,
      "aspects linguistic": 9392,
      "linguistic contexts": 64454,
      "identify specific": 51559,
      "specific lexical": 111459,
      "theoretical background": 121790,
      "complexity task": 19941,
      "results despite": 101716,
      "useful tool": 128940,
      "semantic classes": 105007,
      "automatic detection": 10511,
      "results experimental": 101778,
      "experimental work": 40734,
      "work development": 134473,
      "classbased lexica": 17056,
      "automatic means": 10587,
      "based information": 11770,
      "use classifiers": 127939,
      "results approach": 101513,
      "approach help": 7600,
      "reducing human": 97419,
      "human effort": 50797,
      "effort required": 35183,
      "development language": 30393,
      "information identify": 54661,
      "unsupervised clustering": 127612,
      "task acquiring": 117841,
      "acquiring lexical": 3044,
      "complex problem": 19860,
      "problem typically": 91276,
      "number contexts": 80860,
      "contexts contribute": 22385,
      "information classification": 54416,
      "classification issues": 17239,
      "issues address": 57035,
      "address domain": 3676,
      "contexts work": 22437,
      "proposes use": 93620,
      "use automatically": 127912,
      "automatically obtained": 10805,
      "semantic class": 105006,
      "class unsupervised": 17052,
      "possible discriminate": 88398,
      "different lexical": 31236,
      "role information": 103179,
      "method accurately": 68582,
      "finegrained distinctions": 44344,
      "involving ambiguous": 56910,
      "ambiguous expressions": 5291,
      "effects sparse": 35004,
      "sparse data": 111224,
      "data noise": 26175,
      "approach parsing": 7790,
      "engine language": 37030,
      "language specific": 60112,
      "grammar parsing": 48650,
      "language independent": 59163,
      "grammar language": 48644,
      "specific rules": 111487,
      "rules resources": 103436,
      "given text": 48155,
      "consisting list": 21454,
      "related features": 97863,
      "role argument": 103158,
      "argument structure": 8960,
      "categories derived": 15735,
      "generative grammar": 47728,
      "complex syntactic": 19884,
      "weak points": 132180,
      "important resource": 52234,
      "semantic network": 105121,
      "network used": 77467,
      "used disambiguation": 128495,
      "disambiguation tasks": 31989,
      "tasks parsing": 119370,
      "rulebased semantic": 103372,
      "semantic tagging": 105323,
      "presented article": 89778,
      "extract semantic": 42112,
      "purpose project": 94437,
      "generation semantic": 47610,
      "italian dictionary": 57086,
      "project involves": 92221,
      "parsing semantic": 85224,
      "disambiguation techniques": 31990,
      "semantic features": 105051,
      "translation syntactic": 125305,
      "applied text": 7126,
      "text types": 121381,
      "process semantic": 91567,
      "semantics multiple": 105437,
      "multiple paraphrases": 75634,
      "paraphrases given": 84842,
      "given expression": 48033,
      "techniques application": 119836,
      "interesting information": 56077,
      "search engines": 104312,
      "answering systems": 6707,
      "probabilistic framework": 90805,
      "combination performs": 18576,
      "fundamental role": 46113,
      "broad range": 14674,
      "compound phrases": 20152,
      "everyday language": 39628,
      "provides novel": 94058,
      "novel probabilistic": 80689,
      "framework assessing": 45430,
      "productivity language": 92069,
      "language provide": 60003,
      "provide strong": 93926,
      "cognitive science": 18279,
      "semantic compositionality": 105013,
      "strong weak": 113731,
      "appear literature": 6798,
      "framework presented": 45656,
      "methods determining": 69442,
      "compositional noncompositional": 20123,
      "semantics addition": 105395,
      "addition suggest": 3479,
      "present methods": 89553,
      "methods allow": 69304,
      "joint probability": 57314,
      "modeling combination": 72395,
      "individual concepts": 53903,
      "necessary sufficient": 76760,
      "sufficient condition": 115274,
      "implies underlying": 52040,
      "underlying concepts": 126678,
      "probability space": 90870,
      "formal analysis": 45153,
      "analysis methods": 5695,
      "demonstrated applying": 28909,
      "empirical study": 36198,
      "society order": 110473,
      "people different": 85872,
      "different parts": 31320,
      "respective languages": 101115,
      "languages learning": 60684,
      "learning languages": 62679,
      "task machine": 118364,
      "machine translators": 66312,
      "perform task": 86086,
      "task order": 118480,
      "order develop": 82310,
      "develop machine": 30209,
      "machine translator": 66311,
      "need develop": 76796,
      "develop different": 30191,
      "different rules": 31394,
      "analysis stemming": 5875,
      "stemming lemmatization": 113237,
      "analysis paper": 5735,
      "paper created": 83791,
      "lemmatizer generates": 63346,
      "creating proper": 24706,
      "root word": 103273,
      "brief review": 14615,
      "human natural": 50918,
      "language present": 59820,
      "regional language": 97678,
      "language paper": 59788,
      "paper discuss": 83875,
      "evaluation paper": 39317,
      "english second": 37268,
      "foreign language": 45041,
      "language learners": 59240,
      "management systems": 66905,
      "terms language": 120343,
      "common european": 18875,
      "european framework": 38767,
      "framework english": 45521,
      "learning data": 62482,
      "data available": 25670,
      "decision making": 27937,
      "making easier": 66836,
      "easier efficient": 34413,
      "improving quality": 53150,
      "translation partofspeech": 125101,
      "translation indian": 124866,
      "emerging research": 36070,
      "means mapping": 67766,
      "mapping source": 67147,
      "text target": 121359,
      "language simple": 60096,
      "overall translation": 83263,
      "translation propose": 125150,
      "use stemming": 128299,
      "translation improved": 124859,
      "tagging stemming": 117450,
      "translation hindi": 124849,
      "hindi language": 50466,
      "text unsupervised": 121389,
      "unsupervised latent": 127650,
      "latent variable": 61625,
      "variable model": 130842,
      "model develop": 70985,
      "develop probabilistic": 30226,
      "latentvariable model": 61643,
      "model discover": 71004,
      "discover semantic": 32118,
      "corpora present": 23558,
      "analyze model": 5986,
      "novel interesting": 80604,
      "document contains": 32977,
      "issues including": 57052,
      "parameter learning": 84716,
      "analysis natural": 5714,
      "approach create": 7465,
      "create natural": 24630,
      "physical objects": 87478,
      "categories semantic": 15752,
      "abstract concepts": 1761,
      "applied proposed": 7109,
      "sentences semantic": 106488,
      "processing algorithms": 91612,
      "effectiveness algorithms": 34865,
      "number nodes": 80926,
      "semantics words": 105481,
      "knowledge used": 58225,
      "lexicon words": 63915,
      "words dictionary": 133910,
      "analysis reveals": 5818,
      "core words": 23337,
      "alignment method": 4968,
      "phrase pairs": 87361,
      "pairs phrase": 83606,
      "translation tables": 125324,
      "describes approach": 29385,
      "increase number": 53606,
      "method approach": 68642,
      "approach consists": 7451,
      "number ngrams": 80924,
      "normal distribution": 80326,
      "distribution used": 32688,
      "time translation": 122128,
      "leads better": 61920,
      "better evaluation": 13577,
      "results statistical": 102211,
      "translation tasks": 125347,
      "tasks original": 119354,
      "alignment approach": 4945,
      "approach furthermore": 7582,
      "translation quality": 125162,
      "building stateoftheart": 14890,
      "stateoftheart sentiment": 112948,
      "analysis tweets": 5910,
      "tweets paper": 126044,
      "created stateoftheart": 24685,
      "svm classifiers": 116233,
      "detect sentiment": 29817,
      "messages tweets": 68518,
      "tweets sms": 126057,
      "task detect": 118074,
      "tweets obtaining": 126043,
      "variety surfaceform": 131019,
      "semantic sentiment": 105270,
      "sentiment features": 106741,
      "features provided": 43679,
      "gain fscore": 46342,
      "available resources": 11095,
      "computing lexical": 20518,
      "words widespread": 134314,
      "application natural": 6868,
      "processing including": 91679,
      "including machine": 53319,
      "translation information": 124870,
      "different kinds": 31188,
      "list word": 64702,
      "contrast meaning": 22697,
      "propose automatic": 92566,
      "automatic method": 10591,
      "method identify": 68878,
      "pairs based": 83493,
      "based hypothesis": 11762,
      "hypothesis pair": 51268,
      "pair words": 83467,
      "strongly related": 113749,
      "exists pair": 40348,
      "large crowdsourcing": 61067,
      "crowdsourcing experiment": 25117,
      "determine human": 30127,
      "key features": 57570,
      "features different": 43457,
      "present automatic": 89388,
      "proposed measure": 93330,
      "precision large": 88794,
      "outperforming existing": 82808,
      "data shared": 26440,
      "task parsing": 118515,
      "morphologically rich": 74766,
      "rich languages": 102757,
      "languages document": 60512,
      "document gives": 33018,
      "gives brief": 48189,
      "brief description": 14610,
      "2013 shared": 569,
      "tokens used": 122350,
      "used shared": 128758,
      "constituent trees": 21550,
      "trees using": 125719,
      "using heuristics": 129733,
      "labeling rules": 58535,
      "analysis provided": 5783,
      "automatic morphological": 10610,
      "provided shared": 93981,
      "task generated": 118238,
      "designed implemented": 29601,
      "applications application": 6895,
      "end users": 36836,
      "language tasks": 60151,
      "extracting information": 42214,
      "based application": 11509,
      "semantic networks": 105125,
      "application software": 6883,
      "generates natural": 47165,
      "focused using": 44884,
      "output natural": 83099,
      "language human": 59125,
      "human intervention": 50871,
      "language input": 59208,
      "syntactically semantically": 116516,
      "sentences involving": 106360,
      "exploiting similarities": 41479,
      "similarities languages": 109181,
      "languages machine": 60712,
      "phrase tables": 87375,
      "paper develops": 83869,
      "automate process": 10426,
      "process generating": 91495,
      "tables method": 117266,
      "missing word": 70337,
      "learning language": 62670,
      "language structures": 60128,
      "structures based": 114060,
      "based large": 11794,
      "large monolingual": 61156,
      "monolingual data": 74565,
      "data mapping": 26119,
      "mapping languages": 67136,
      "languages small": 60880,
      "small bilingual": 110138,
      "bilingual data": 14030,
      "data uses": 26607,
      "distributed representation": 32617,
      "linear mapping": 64350,
      "mapping vector": 67156,
      "vector spaces": 131391,
      "spaces languages": 111097,
      "languages despite": 60494,
      "simplicity method": 109571,
      "surprisingly effective": 116135,
      "achieve 90": 2408,
      "translation words": 125431,
      "words english": 133939,
      "method makes": 68952,
      "used extend": 128540,
      "dictionaries translation": 30870,
      "highly multilingual": 50332,
      "subject domains": 114678,
      "european union": 38772,
      "union eu": 127178,
      "multilabel classification": 75150,
      "manually labelled": 67071,
      "labelled data": 58558,
      "data automatically": 25663,
      "automatically assign": 10722,
      "trained classifiers": 123088,
      "languages parallel": 60787,
      "parallel training": 84692,
      "data languages": 26074,
      "allows users": 5197,
      "document collections": 32975,
      "users change": 129100,
      "document representation": 33063,
      "result linguistic": 101381,
      "linguistic preprocessing": 64523,
      "increase speed": 53617,
      "consistency human": 21360,
      "process used": 91584,
      "used fully": 128561,
      "feature vector": 43326,
      "input various": 55469,
      "various language": 131112,
      "language technology": 60155,
      "tasks including": 119175,
      "including crosslingual": 53280,
      "clustering classification": 17941,
      "classification crosslingual": 17165,
      "plagiarism detection": 87619,
      "detection sentence": 30045,
      "sentence selection": 106059,
      "available translation": 11137,
      "translation memory": 124931,
      "22 languages": 773,
      "languages european": 60555,
      "general translation": 46724,
      "making available": 66824,
      "available large": 11027,
      "large translation": 61308,
      "memory tm": 68390,
      "produced translations": 91975,
      "typically used": 126467,
      "used translation": 128831,
      "improve speed": 52544,
      "studies language": 114245,
      "technology applications": 120033,
      "applications including": 6938,
      "terminology extraction": 120258,
      "extraction named": 42400,
      "recognition ner": 96922,
      "classification clustering": 17146,
      "new resource": 79109,
      "resource provide": 100871,
      "regarding size": 97660,
      "applications large": 6953,
      "news media": 79350,
      "media monitoring": 68128,
      "work automated": 134389,
      "languages particularly": 60794,
      "particularly important": 85484,
      "languages order": 60771,
      "order capture": 82290,
      "capture complementary": 15276,
      "news content": 79321,
      "content published": 21928,
      "published different": 94353,
      "different countries": 31068,
      "able access": 1590,
      "content languages": 21897,
      "extracted information": 42158,
      "present publicly": 89654,
      "publicly accessible": 94288,
      "50 languages": 1075,
      "discuss implications": 32257,
      "fact cover": 42817,
      "languages discuss": 60509,
      "achieve high": 2472,
      "available highly": 11005,
      "multilingual named": 75299,
      "resource paper": 100868,
      "new freely": 78927,
      "largescale multilingual": 61474,
      "multilingual news": 75313,
      "analysis combined": 5531,
      "names plus": 76151,
      "spelling variants": 111914,
      "20 different": 500,
      "different scripts": 31401,
      "used number": 128668,
      "learning systems": 63082,
      "systems learn": 116976,
      "improve machine": 52411,
      "translation results": 125222,
      "statistics current": 113195,
      "address issue": 3686,
      "details regarding": 29791,
      "resource available": 100830,
      "available daily": 10969,
      "quality mt": 94722,
      "mt output": 74978,
      "output using": 83145,
      "novel entity": 80562,
      "entity translation": 38112,
      "translation scheme": 125230,
      "scheme paper": 103934,
      "translation entities": 124793,
      "quality machine": 94710,
      "machine translated": 65877,
      "translated output": 124555,
      "output work": 83153,
      "using statistical": 130220,
      "rule based": 103345,
      "transliteration entities": 125522,
      "entities english": 37773,
      "english punjabi": 37255,
      "types entities": 126277,
      "entities proper": 37843,
      "names location": 76145,
      "location names": 64965,
      "calculated using": 15023,
      "translation toolkit": 125381,
      "toolkit moses": 122411,
      "comparative study": 19211,
      "study linguistic": 114429,
      "linguistic feature": 64478,
      "polarity classification": 87907,
      "classification sentiment": 17392,
      "classification widely": 17500,
      "widely studied": 132556,
      "negative opinion": 76937,
      "opinion paper": 82094,
      "paper using": 84488,
      "using movie": 129902,
      "review dataset": 102542,
      "dataset perform": 27092,
      "perform comparative": 85963,
      "study different": 114357,
      "single kind": 109746,
      "linguistic features": 64479,
      "bayes support": 12575,
      "support vector": 115999,
      "vector machine": 131310,
      "perform sentiment": 86063,
      "combination different": 18559,
      "different linguistic": 31237,
      "features classification": 43397,
      "classification accuracy": 17100,
      "tweets tweets": 126064,
      "hundreds millions": 51151,
      "beneficial downstream": 12946,
      "downstream natural": 34005,
      "applications question": 6996,
      "answering summarization": 6705,
      "summarization paper": 115545,
      "new task": 79196,
      "sentiment emotion": 106735,
      "emotion detection": 36100,
      "detection significantly": 30053,
      "significantly different": 108905,
      "number applications": 80839,
      "identifying key": 51602,
      "key issues": 57585,
      "create large": 24619,
      "large dataset": 61072,
      "thousand tweets": 121917,
      "purpose develop": 94426,
      "develop automatically": 30177,
      "task accuracy": 117830,
      "baseline finally": 12225,
      "finally resources": 44232,
      "detection helpful": 29965,
      "semantic parsing": 105141,
      "parsing framework": 85115,
      "learning inference": 62653,
      "inference framework": 54145,
      "formal representation": 45173,
      "automatic ranking": 10638,
      "mt outputs": 74979,
      "outputs using": 83188,
      "research machine": 100550,
      "good translations": 48506,
      "mt engines": 74963,
      "time consuming": 121998,
      "consuming expensive": 21727,
      "better worse": 13772,
      "paper approach": 83738,
      "approach provide": 7828,
      "taken different": 117508,
      "different mt": 31271,
      "provide solution": 93922,
      "intervention required": 56324,
      "ranking systems": 95689,
      "systems evaluations": 116881,
      "evaluations results": 39484,
      "results human": 101832,
      "human ranking": 50939,
      "learning semantic": 62991,
      "model pair": 71688,
      "target phrases": 117689,
      "continuousvalued vector": 22658,
      "vector representations": 131347,
      "semantic space": 105302,
      "space translation": 111072,
      "translation score": 125231,
      "pair new": 83445,
      "multilayer neural": 75177,
      "neural network": 78233,
      "weights learned": 132378,
      "data learning": 26083,
      "directly optimize": 31891,
      "endtoend machine": 36914,
      "experimental evaluation": 40531,
      "evaluation performed": 39320,
      "results new": 101988,
      "model significantly": 72019,
      "significantly improves": 108939,
      "performance stateoftheart": 86742,
      "phrasebased statistical": 87391,
      "translation leading": 124898,
      "language discourse": 58970,
      "discourse representation": 32081,
      "representation structures": 99422,
      "controlled language": 22838,
      "language statements": 60121,
      "achieved pipeline": 2666,
      "language syntax": 60139,
      "syntax semantic": 116556,
      "semantic discourse": 105032,
      "set previously": 107543,
      "reduction rules": 97457,
      "description evaluation": 29461,
      "evaluation potential": 39326,
      "potential limitations": 88571,
      "dictionarybased method": 30904,
      "used extract": 128543,
      "extract expressive": 42081,
      "concepts documents": 20614,
      "studies concerning": 114197,
      "agglutinative language": 4337,
      "dictionary instead": 30882,
      "widely used": 132560,
      "used concept": 128445,
      "concept extraction": 20568,
      "rarely used": 95766,
      "used domain": 128499,
      "domain concept": 33481,
      "taking account": 117540,
      "success rate": 115119,
      "rate high": 95793,
      "concepts concept": 20608,
      "extraction method": 42387,
      "method implemented": 68881,
      "documents collected": 33203,
      "collected different": 18417,
      "different corpora": 31066,
      "survey methods": 116177,
      "multilingual text": 75382,
      "mining applications": 70225,
      "useful information": 128896,
      "information content": 54446,
      "content different": 21872,
      "opinions information": 82124,
      "extraction text": 42515,
      "developed languages": 30284,
      "languages text": 60918,
      "text analysis": 120640,
      "analysis tools": 5902,
      "tools applied": 122432,
      "applied small": 7117,
      "small sets": 110201,
      "sets languages": 107681,
      "languages development": 60497,
      "language large": 59237,
      "alleviate problem": 5043,
      "providing training": 94143,
      "tuning results": 125940,
      "results usually": 102305,
      "various multilingual": 131139,
      "effort developing": 35169,
      "developing natural": 30351,
      "processing applications": 91616,
      "applications languages": 6952,
      "effort develop": 35168,
      "complex text": 19890,
      "shown feasibility": 108473,
      "feasibility approach": 43235,
      "approach development": 7490,
      "tools process": 122466,
      "online news": 81785,
      "news articles": 79287,
      "articles day": 9181,
      "day languages": 27841,
      "kind language": 57680,
      "resources make": 101011,
      "easier develop": 34412,
      "develop highly": 30203,
      "resources freely": 100980,
      "available simple": 11103,
      "simple parallel": 109490,
      "software tools": 110526,
      "semantic text": 105327,
      "analysis texts": 5897,
      "lexical chains": 63736,
      "coreference chains": 23343,
      "represent similar": 99141,
      "similar linguistic": 109107,
      "linguistic tools": 64568,
      "gives possibility": 48203,
      "study text": 114533,
      "text example": 120926,
      "summarization text": 115581,
      "parallel texts": 84690,
      "usually stored": 130454,
      "work shown": 134800,
      "fact texts": 42834,
      "pairs parallel": 83602,
      "words high": 133982,
      "high probability": 50109,
      "intermediate representation": 56139,
      "compression process": 20318,
      "word alignments": 132912,
      "alignments tackle": 5021,
      "words introduce": 134013,
      "methods binary": 69350,
      "compare performance": 19269,
      "performance different": 86295,
      "schemes applied": 103946,
      "applied extraction": 7075,
      "efficient algorithm": 35054,
      "words text": 134264,
      "text segments": 121279,
      "segments texts": 104677,
      "translations text": 125500,
      "text application": 120651,
      "usually referred": 130447,
      "compression algorithm": 20305,
      "unsupervised methods": 127668,
      "highquality information": 50386,
      "extraction paper": 42421,
      "context unsupervised": 22310,
      "unsupervised information": 127644,
      "training examples": 123620,
      "examples domain": 39826,
      "available paper": 11068,
      "presents scalable": 89895,
      "time number": 122060,
      "number extractions": 80888,
      "relational model": 98096,
      "model predicting": 71759,
      "based similarity": 12044,
      "assertions extracted": 9466,
      "extracted web": 42194,
      "90 precision": 1370,
      "probabilistic model": 90819,
      "experiments demonstrate": 40875,
      "improve f1": 52384,
      "allows handle": 5162,
      "precision 95": 88790,
      "corpus text": 24039,
      "relatedness fragments": 97921,
      "text automated": 120675,
      "automated manner": 10452,
      "pairwise relations": 83682,
      "relations words": 98284,
      "measure relatedness": 67819,
      "account lexical": 2030,
      "relatedness words": 97934,
      "aspects text": 9410,
      "tasks text": 119554,
      "text retrieval": 121267,
      "clustering paper": 17950,
      "approach measuring": 7707,
      "measuring semantic": 67929,
      "implicit semantic": 52017,
      "approach exploits": 7554,
      "links words": 64683,
      "approach introduce": 7646,
      "new measure": 79002,
      "measure semantic": 67822,
      "validate method": 130724,
      "evaluate performance": 38881,
      "performance semantic": 86695,
      "similarity relatedness": 109294,
      "relatedness synonym": 97929,
      "identification word": 51459,
      "word analogy": 132916,
      "evaluating performance": 39085,
      "performance method": 86528,
      "method measuring": 68959,
      "relatedness tasks": 97931,
      "paraphrase recognition": 84833,
      "shows proposed": 108615,
      "method outperforms": 69007,
      "outperforms lexiconbased": 82915,
      "lexiconbased method": 63917,
      "method semantic": 69119,
      "used data": 128469,
      "hybrid approaches": 51175,
      "small parallel": 110180,
      "corpora paper": 23551,
      "method automatic": 68651,
      "automatic inference": 10566,
      "rules based": 103383,
      "used statistical": 128783,
      "sentencealigned parallel": 106137,
      "extended set": 41811,
      "rules experiments": 103400,
      "experiments conducted": 40837,
      "conducted using": 20957,
      "translation transfer": 125391,
      "resulting translation": 101476,
      "quality close": 94610,
      "close obtained": 17825,
      "present entirely": 89474,
      "entirely unsupervised": 37735,
      "benefits information": 13009,
      "rest modules": 101315,
      "rules applied": 103380,
      "semantic dependency": 105029,
      "efficient largescale": 35085,
      "selection semantic": 104825,
      "parsing automatic": 85076,
      "automatic derivation": 10510,
      "meaning representation": 67668,
      "sentence plays": 105988,
      "plays critical": 87722,
      "critical role": 24828,
      "role deep": 103164,
      "processing natural": 91720,
      "systems semantic": 117130,
      "rely pipeline": 98727,
      "pipeline framework": 87540,
      "real applications": 96052,
      "maintaining competitive": 66510,
      "competitive performance": 19658,
      "parsing word": 85281,
      "word pair": 133390,
      "pair classification": 83423,
      "problem using": 91282,
      "using maximum": 129864,
      "entropy classifier": 38162,
      "feature space": 43315,
      "space use": 111074,
      "use far": 128049,
      "achieves stateoftheart": 2879,
      "stateoftheart performance": 112803,
      "performance evaluation": 86347,
      "evaluation data": 39167,
      "task pipeline": 118533,
      "expressions using": 41768,
      "semantic clustering": 105010,
      "issues natural": 57061,
      "understanding generation": 126851,
      "appropriate processing": 8427,
      "idiosyncratic nature": 51659,
      "diversity lexical": 32882,
      "syntactical semantic": 116503,
      "especially resource": 38503,
      "languages like": 60689,
      "like bengali": 64020,
      "present semantic": 89686,
      "clustering approach": 17937,
      "approach contributes": 7460,
      "present document": 89456,
      "measure similarity": 67825,
      "constituent words": 21552,
      "candidate phrase": 15116,
      "using vector": 130360,
      "model judge": 71390,
      "apply semantic": 7215,
      "types mwes": 126324,
      "pointwise mutual": 87889,
      "information pmi": 54846,
      "log likelihood": 64979,
      "likelihood ratio": 64124,
      "ratio llr": 95831,
      "employed extract": 36298,
      "approach outperforms": 7764,
      "outperforms competing": 82866,
      "topic segmentation": 122572,
      "segmentation labeling": 104595,
      "analysis shown": 5848,
      "shown useful": 108536,
      "applications present": 6990,
      "new corpora": 78844,
      "corpora email": 23466,
      "conversations annotated": 23047,
      "topics evaluate": 122627,
      "labeling tasks": 58547,
      "conversations propose": 23066,
      "computational framework": 20379,
      "approach extends": 7559,
      "stateoftheart methods": 112702,
      "methods considering": 69396,
      "features applying": 43364,
      "graphbased methods": 48901,
      "methods nlp": 69642,
      "segmentation propose": 104625,
      "novel unsupervised": 80763,
      "unsupervised models": 127672,
      "models exploit": 73209,
      "exploit finegrained": 41416,
      "conversational structure": 23036,
      "structure novel": 113924,
      "supervised model": 115806,
      "model combines": 70844,
      "combines lexical": 18689,
      "conversational topic": 23042,
      "labeling propose": 58531,
      "models respectively": 73946,
      "capture conversation": 15283,
      "different sources": 31439,
      "labeling performed": 58521,
      "best models": 13377,
      "models beat": 72822,
      "highly correlated": 50309,
      "correlated human": 24206,
      "human annotations": 50746,
      "annotations event": 6430,
      "semantics based": 105397,
      "mandarin chinese": 66913,
      "chinese data": 16752,
      "al 2000": 4590,
      "respectively study": 101168,
      "study focus": 114384,
      "event structures": 39544,
      "relatedness similarity": 97927,
      "similarity geographic": 109237,
      "geographic information": 47851,
      "information science": 54956,
      "vast number": 131278,
      "number tasks": 80978,
      "tasks information": 119200,
      "information integration": 54695,
      "largely ignored": 61340,
      "specific semantic": 111489,
      "article discuss": 9124,
      "semantic fields": 105056,
      "fields compare": 43991,
      "geosemantic similarity": 47877,
      "similarity dataset": 109216,
      "new open": 79063,
      "open dataset": 81889,
      "dataset designed": 26865,
      "designed evaluate": 29589,
      "evaluate computational": 38811,
      "computational measures": 20401,
      "larger existing": 61364,
      "existing datasets": 40104,
      "datasets kind": 27533,
      "geographic terms": 47853,
      "terms combined": 120293,
      "term pairs": 120238,
      "human subjects": 50970,
      "online used": 81812,
      "evaluation baseline": 39131,
      "degree given": 28571,
      "inconsistent source": 53443,
      "new algorithm": 78767,
      "model investigate": 71375,
      "investigate learning": 56775,
      "learning process": 62915,
      "input data": 55315,
      "learning model": 62756,
      "model explains": 71137,
      "types patterns": 126336,
      "exist language": 40024,
      "language use": 60299,
      "use data": 127978,
      "data collected": 25739,
      "american sign": 5310,
      "sign language": 108677,
      "language asl": 58848,
      "common form": 18880,
      "form source": 45125,
      "language feature": 59044,
      "index terms": 53806,
      "al 2012": 4603,
      "ontology construction": 81842,
      "automatic summarization": 10678,
      "summarization documents": 115500,
      "term frequency": 120215,
      "frequency tfidf": 45848,
      "fail advantage": 42950,
      "advantage semantic": 4061,
      "semantic relationships": 105221,
      "relationships terms": 98334,
      "specific terms": 111500,
      "demonstrate use": 28895,
      "use relational": 128241,
      "relations terms": 98266,
      "technique useful": 119820,
      "useful identifying": 128891,
      "identifying relevant": 51619,
      "relevant words": 98602,
      "words terms": 134261,
      "processing tasks": 91811,
      "tasks semantic": 119481,
      "approach natural": 7727,
      "provides high": 94039,
      "level mathematical": 63474,
      "mathematical framework": 67456,
      "framework model": 45612,
      "model contextual": 70909,
      "nature natural": 76661,
      "used provide": 128706,
      "global semantics": 48269,
      "semantics discourse": 105409,
      "logical semantics": 65020,
      "semantics sentence": 105464,
      "sentence discourse": 105824,
      "discourse introduce": 32063,
      "setting formulate": 107751,
      "notion semantic": 80409,
      "discourse coherent": 32044,
      "examples used": 39895,
      "rank using": 95630,
      "using quantitative": 130072,
      "quantitative measures": 94872,
      "measuring global": 67912,
      "similarity texts": 109322,
      "texts propose": 121584,
      "similarity measure": 109256,
      "contrary current": 22679,
      "stateoftheart approaches": 112580,
      "texts compared": 121477,
      "conducted experiments": 20922,
      "experiments methods": 41007,
      "methods reliably": 69721,
      "reliably identify": 98627,
      "identify different": 51491,
      "types texts": 126368,
      "texts clinical": 121474,
      "task currently": 118039,
      "currently preparation": 25413,
      "evaluation exercise": 39207,
      "task involves": 118312,
      "involves identifying": 56896,
      "describing events": 29450,
      "relations clinical": 98122,
      "clinical text": 17789,
      "subtasks included": 114968,
      "times events": 122170,
      "mentions entity": 68452,
      "entity types": 38118,
      "relation event": 97977,
      "creation time": 24724,
      "time identifying": 122035,
      "dependency parser": 29159,
      "parser using": 85006,
      "using hybrid": 129746,
      "hybrid approach": 51172,
      "tamil language": 117568,
      "language natural": 59704,
      "tool language": 122392,
      "analysis aims": 5492,
      "structural relationship": 113789,
      "relationship words": 98314,
      "words given": 133974,
      "given sentence": 48116,
      "developed language": 30283,
      "language tools": 60174,
      "translation major": 124925,
      "major application": 66532,
      "area natural": 8874,
      "translation language": 124887,
      "language structure": 60127,
      "play key": 87701,
      "key role": 57595,
      "given words": 48175,
      "sentence existing": 105860,
      "using rule": 130127,
      "data machine": 26105,
      "suitable handle": 115400,
      "handle larger": 49388,
      "larger data": 61359,
      "data better": 25702,
      "better accuracy": 13499,
      "accuracy learning": 2196,
      "learning training": 63115,
      "training proposed": 123801,
      "approach takes": 7957,
      "sentence input": 105908,
      "input produce": 55405,
      "result dependency": 101369,
      "structure using": 113980,
      "approach proposed": 7826,
      "proposed tool": 93572,
      "improve quality": 52506,
      "quality existing": 94658,
      "existing approaches": 40048,
      "approaches hybrid": 8182,
      "entity transliteration": 38113,
      "mt research": 74990,
      "proper transliteration": 92425,
      "entities domain": 37771,
      "domain paper": 33604,
      "paper address": 83701,
      "used englishhindi": 128516,
      "englishhindi language": 37407,
      "pair experiments": 83431,
      "english words": 37349,
      "approach extracts": 7566,
      "words applied": 133820,
      "converts english": 23106,
      "english equivalent": 37126,
      "word approach": 132925,
      "accuracy evaluation": 2155,
      "evaluation ranking": 39350,
      "using precision": 130036,
      "metrics evaluation": 69959,
      "evaluation plays": 39324,
      "plays crucial": 87726,
      "crucial role": 25161,
      "role development": 103169,
      "judge quality": 57425,
      "output human": 83081,
      "quality various": 94826,
      "various automatic": 131048,
      "automatic metrics": 10601,
      "exist present": 40029,
      "implementation results": 51950,
      "different metrics": 31259,
      "metrics used": 70018,
      "language comparisons": 58897,
      "like hindi": 64045,
      "free word": 45771,
      "order language": 82351,
      "evaluation integration": 39245,
      "set labels": 107472,
      "performance french": 86392,
      "extracting bilingual": 42202,
      "bilingual semantic": 14055,
      "aim research": 4473,
      "research make": 100553,
      "multilingual natural": 75302,
      "applications common": 6904,
      "common semantic": 18919,
      "languages paper": 60776,
      "annotated sentences": 6230,
      "berkeley framenet": 13046,
      "shared set": 108011,
      "result provide": 101398,
      "provide unified": 93945,
      "method comparing": 68714,
      "convolutional neural": 23175,
      "sentences ability": 106191,
      "ability accurately": 1482,
      "accurately represent": 2392,
      "represent sentences": 99140,
      "central language": 15902,
      "understanding convolutional": 126821,
      "convolutional architecture": 23148,
      "semantic modelling": 105113,
      "sentences network": 106408,
      "network uses": 77468,
      "uses dynamic": 129226,
      "pooling operation": 88046,
      "varying length": 131261,
      "graph sentence": 48867,
      "capable explicitly": 15199,
      "explicitly capturing": 41363,
      "capturing short": 15494,
      "longrange relations": 65183,
      "network does": 77231,
      "easily applicable": 34436,
      "applicable language": 6824,
      "language test": 60159,
      "experiments small": 41146,
      "small scale": 110194,
      "binary multiclass": 14132,
      "sentiment prediction": 106782,
      "question classification": 95136,
      "classification twitter": 17474,
      "twitter sentiment": 126122,
      "distant supervision": 32460,
      "network achieves": 77123,
      "achieves excellent": 2783,
      "excellent performance": 39916,
      "performance tasks": 86787,
      "25 error": 813,
      "error reduction": 38339,
      "task respect": 118652,
      "respect strongest": 101105,
      "strongest baseline": 113739,
      "model combination": 70841,
      "approach building": 7404,
      "building language": 14853,
      "systematic recursive": 116682,
      "using modified": 129895,
      "models contains": 72977,
      "order models": 82373,
      "models special": 74073,
      "special case": 111350,
      "paper motivate": 84049,
      "approach extensive": 7560,
      "extensive empirical": 41868,
      "experiment english": 40465,
      "text corpora": 120824,
      "corpora demonstrate": 23450,
      "models lead": 73475,
      "lead substantial": 61873,
      "substantial reduction": 114875,
      "comparison traditional": 19584,
      "furthermore investigate": 46184,
      "investigate behaviour": 56723,
      "domain specific": 33643,
      "specific corpus": 111424,
      "consistent improvements": 21382,
      "approach lies": 7686,
      "lies ability": 63975,
      "ability cope": 1500,
      "kb text": 57503,
      "yield improvements": 135344,
      "parallel multilingual": 84672,
      "multilingual corpus": 75225,
      "corpus persian": 23930,
      "parallel corpus": 84613,
      "european languages": 38770,
      "languages article": 60411,
      "describes primary": 29425,
      "basic language": 12526,
      "specific resources": 111486,
      "article introduces": 9138,
      "morphosyntactic features": 74816,
      "environments proposed": 38202,
      "deep architecture": 28198,
      "successful approaches": 115152,
      "approaches semantic": 8331,
      "text make": 121098,
      "use distributional": 128010,
      "distributional representations": 32717,
      "models match": 73556,
      "novel deep": 80535,
      "deep learning": 28245,
      "learning architecture": 62368,
      "architecture provides": 8736,
      "provides semantic": 94072,
      "neural models": 78158,
      "models language": 73448,
      "allows generation": 5160,
      "queries natural": 94929,
      "questions need": 95334,
      "especially suitable": 38515,
      "development semantic": 30419,
      "resourcepoor languages": 100908,
      "languages translation": 60931,
      "change using": 16373,
      "using multilingual": 129914,
      "multilingual data": 75234,
      "data named": 26149,
      "orthographic changes": 82593,
      "languages share": 60866,
      "share common": 107949,
      "leveraged improve": 63640,
      "improve named": 52427,
      "ner using": 77096,
      "using unsupervised": 130344,
      "unsupervised word": 127750,
      "word clusters": 132957,
      "secondary languages": 104470,
      "languages features": 60583,
      "features stateoftheart": 43743,
      "stateoftheart discriminative": 112642,
      "ner systems": 77083,
      "observe significant": 81209,
      "significant increases": 108818,
      "increases performance": 53651,
      "person location": 87096,
      "close languages": 17822,
      "languages provide": 60823,
      "provide valuable": 93952,
      "valuable features": 130762,
      "distant languages": 32453,
      "languages latent": 60679,
      "latent semantics": 61606,
      "emotional content": 36117,
      "statistical patterns": 113152,
      "word cooccurrences": 132975,
      "nouns verbs": 80446,
      "google search": 48522,
      "search queries": 104342,
      "resulting latent": 101449,
      "semantics capture": 105400,
      "links form": 64681,
      "spatial dimensions": 111263,
      "associated phonetic": 9606,
      "study explores": 114378,
      "articulatory acoustic": 9236,
      "cosine similarities": 24337,
      "computed using": 20475,
      "using latent": 129797,
      "different large": 31217,
      "scale text": 103755,
      "applying hierarchical": 7249,
      "hierarchical clustering": 49944,
      "clustering identify": 17946,
      "identify common": 51487,
      "structures text": 114110,
      "small large": 110168,
      "large size": 61273,
      "sequences close": 107116,
      "acoustic characteristics": 2999,
      "training statistical": 123882,
      "used training": 128823,
      "training procedure": 123788,
      "translation general": 124834,
      "problem approach": 90937,
      "local optimum": 64933,
      "feature functions": 43285,
      "contrary previous": 22682,
      "dimension corresponds": 31743,
      "dimensions new": 31771,
      "basic idea": 12525,
      "quite simple": 95405,
      "critical training": 24840,
      "functions experiments": 46072,
      "results better": 101552,
      "better results": 13702,
      "obtained language": 81381,
      "language knowledge": 59230,
      "language french": 59055,
      "automatic generation": 10545,
      "details language": 29790,
      "major applications": 66533,
      "conclude future": 20712,
      "open problems": 81922,
      "recent years": 96601,
      "developments area": 30439,
      "data created": 25806,
      "created new": 24677,
      "new types": 79232,
      "expand range": 40357,
      "range possible": 95593,
      "possible uses": 88443,
      "uses lexical": 129240,
      "lexical data": 63753,
      "data support": 26522,
      "support users": 115996,
      "translation article": 124649,
      "data main": 26108,
      "main methods": 66441,
      "used build": 128415,
      "represent lexical": 99117,
      "data paper": 26206,
      "groups people": 49153,
      "knowledge paper": 58094,
      "written languages": 135134,
      "french german": 45808,
      "present natural": 89567,
      "method called": 68686,
      "logic type": 64999,
      "type theory": 126229,
      "theory lambdacalculus": 121819,
      "semantics mathematical": 105432,
      "analysis present": 5761,
      "present extended": 89486,
      "analysis esa": 5599,
      "thematic information": 121780,
      "information category": 54410,
      "structure wikipedia": 113983,
      "measures relevance": 67894,
      "terms categories": 120289,
      "using measure": 129872,
      "wikipedia corpus": 132648,
      "corpus considered": 23716,
      "directed graph": 31816,
      "provides unique": 94092,
      "related categories": 97845,
      "sensitive noise": 105740,
      "words apply": 133821,
      "method french": 68847,
      "corpus evaluate": 23780,
      "corpus 20": 23631,
      "compared standard": 19447,
      "nonparallel corpus": 80265,
      "transductive learning": 124029,
      "learning parallel": 62875,
      "role machine": 103201,
      "scale coverage": 103708,
      "corpus resources": 23973,
      "resources web": 101065,
      "translation natural": 125001,
      "tasks article": 118942,
      "proposes semisupervised": 93616,
      "learning method": 62717,
      "expanding training": 40365,
      "corpus statistical": 24018,
      "extracting parallel": 42226,
      "parallel sentences": 84680,
      "sentences nonparallel": 106413,
      "corpus method": 23880,
      "method requires": 69106,
      "requires small": 100322,
      "small labeled": 110166,
      "labeled corpus": 58428,
      "corpus large": 23856,
      "large unlabeled": 61312,
      "unlabeled corpus": 127377,
      "corpus build": 23686,
      "especially short": 38509,
      "corpus experimental": 23788,
      "results combining": 101587,
      "method effectively": 68791,
      "effectively use": 34858,
      "building multilingual": 14865,
      "related information": 97868,
      "developing countries": 30338,
      "ones like": 81693,
      "underresourced language": 126726,
      "goal project": 48383,
      "development multilingual": 30400,
      "sense language": 105690,
      "data coming": 25748,
      "word xml": 133665,
      "state transducer": 112533,
      "resulting resource": 101469,
      "dictionaries paper": 30862,
      "model languages": 71411,
      "languages processed": 60816,
      "considered underresourced": 21308,
      "underresourced languages": 126727,
      "presented specific": 89798,
      "platform used": 87660,
      "used project": 128704,
      "networks natural": 77668,
      "terms based": 120281,
      "based analysis": 11505,
      "texts corpora": 121486,
      "corpora technique": 23599,
      "technique building": 119774,
      "technique based": 119771,
      "based methodology": 11838,
      "graphs constructed": 48942,
      "investigated language": 56838,
      "topics information": 122632,
      "automatic multilanguage": 10613,
      "multilanguage translation": 75166,
      "swiss avalanche": 116268,
      "avalanche bulletin": 11151,
      "bulletin swiss": 14958,
      "bulletin produced": 14954,
      "produced twice": 91976,
      "twice day": 126082,
      "languages lack": 60665,
      "lack time": 58758,
      "time available": 121985,
      "available manual": 11041,
      "manual translation": 67016,
      "translation fully": 124829,
      "fully automated": 45942,
      "automated translation": 10476,
      "translation employed": 124782,
      "employed based": 36292,
      "based catalogue": 11562,
      "catalogue predefined": 15702,
      "predefined phrases": 88829,
      "phrases predetermined": 87443,
      "predetermined rules": 88841,
      "rules phrases": 103427,
      "phrases combined": 87412,
      "combined produce": 18664,
      "produce sentences": 91929,
      "automatically translate": 10831,
      "translate sentences": 124537,
      "sentences german": 106331,
      "german target": 47915,
      "target languages": 117664,
      "italian english": 57088,
      "english subsequent": 37294,
      "subsequent proofreading": 114817,
      "proofreading correction": 92378,
      "catalogue phrases": 15695,
      "phrases limited": 87436,
      "limited small": 64280,
      "small sublanguage": 110212,
      "reduction daily": 97443,
      "daily translation": 25514,
      "translation costs": 124717,
      "costs expected": 24398,
      "expected offset": 40394,
      "offset initial": 81618,
      "initial development": 55210,
      "development costs": 30376,
      "costs years": 24405,
      "operational winter": 82058,
      "winter seasons": 132732,
      "seasons assess": 104376,
      "assess quality": 9487,
      "quality produced": 94761,
      "produced texts": 91972,
      "texts based": 121463,
      "based evaluation": 11691,
      "origins catalogue": 82580,
      "phrases versus": 87464,
      "versus manually": 131627,
      "manually written": 67083,
      "written translated": 135153,
      "translated texts": 124571,
      "texts mean": 121548,
      "mean recognition": 67605,
      "recognition rate": 96978,
      "rate 55": 95778,
      "55 users": 1112,
      "users hardly": 129128,
      "hardly distinguish": 49493,
      "distinguish types": 32579,
      "texts similar": 121607,
      "similar ratings": 109131,
      "ratings respect": 95825,
      "respect language": 101082,
      "language quality": 60008,
      "quality overall": 94740,
      "overall output": 83241,
      "output catalogue": 83056,
      "catalogue considered": 15691,
      "considered virtually": 21310,
      "virtually equivalent": 131749,
      "equivalent text": 38253,
      "text written": 121417,
      "written avalanche": 135115,
      "avalanche forecasters": 11158,
      "forecasters manually": 45031,
      "manually translated": 67079,
      "translated professional": 124557,
      "professional translators": 92081,
      "translators furthermore": 125513,
      "furthermore forecasters": 46175,
      "forecasters declared": 45027,
      "declared relevant": 27975,
      "relevant situations": 98581,
      "situations captured": 109895,
      "captured sufficient": 15438,
      "sufficient accuracy": 115270,
      "limited time": 64290,
      "theory framework": 121817,
      "acquisition paper": 3048,
      "initially developed": 55247,
      "main components": 66404,
      "constraints used": 21615,
      "used finally": 128554,
      "better representation": 13696,
      "yield better": 135332,
      "hmm based": 50581,
      "reports work": 99091,
      "nlp tools": 79785,
      "tools contest": 122444,
      "submitted runs": 114767,
      "bengali english": 13032,
      "english hindi": 37167,
      "hindi marathi": 50468,
      "marathi punjabi": 67173,
      "punjabi tamil": 94402,
      "tamil telugu": 117572,
      "hidden markov": 49894,
      "based model": 11846,
      "model used": 72254,
      "tested nlp": 120580,
      "respectively training": 101173,
      "training multilingual": 123719,
      "learn language": 62076,
      "novel framework": 80573,
      "framework learning": 45597,
      "generate language": 46960,
      "demonstrate capabilities": 28683,
      "capabilities developing": 15168,
      "english korean": 37182,
      "prior knowledge": 90705,
      "knowledge training": 58210,
      "sequence events": 106940,
      "events extracted": 39580,
      "events building": 39568,
      "building translation": 14895,
      "model supports": 72119,
      "novel algorithm": 80457,
      "algorithm learning": 4765,
      "human evaluations": 50844,
      "evaluations generated": 39464,
      "limited domain": 64231,
      "languages south": 60882,
      "south africa": 110960,
      "10 million": 139,
      "million people": 70103,
      "computational resources": 20424,
      "resources exist": 100973,
      "tool development": 122383,
      "focus natural": 44796,
      "preferences particular": 89246,
      "representation languages": 99275,
      "overall structure": 83260,
      "generation multilingual": 47493,
      "contribution paper": 22773,
      "paper twofold": 84483,
      "methodological approach": 69235,
      "automatically generate": 10773,
      "corpora second": 23582,
      "proof concept": 92371,
      "use cases": 127934,
      "languages case": 60434,
      "response challenges": 101193,
      "translation multilingual": 124990,
      "quite different": 95399,
      "different approach": 30998,
      "approach taken": 7956,
      "technical documentation": 119745,
      "paper methods": 84045,
      "specifically look": 111567,
      "methods detect": 69438,
      "propose methods": 92776,
      "look specific": 65217,
      "specific use": 111513,
      "generalpurpose method": 46887,
      "method detect": 68769,
      "extract multilingual": 42099,
      "nominal compounds": 80140,
      "german evaluate": 47890,
      "performing qualitative": 86965,
      "analysis results": 5811,
      "impact method": 51878,
      "method machine": 68947,
      "programming languages": 92131,
      "embedded cnl": 35353,
      "language proper": 59995,
      "useful feedback": 128887,
      "instead just": 55666,
      "extended abstract": 41802,
      "main concepts": 66405,
      "preliminary investigations": 89275,
      "information texts": 55039,
      "texts provide": 121587,
      "provide different": 93807,
      "domain dynamic": 33514,
      "representation main": 99319,
      "strategy provides": 113536,
      "practical solution": 88715,
      "social sciences": 110454,
      "methods tools": 69806,
      "tools automatically": 122438,
      "information natural": 54787,
      "texts survey": 121629,
      "structure learning": 113896,
      "methods text": 69799,
      "classification text": 17466,
      "classification task": 17425,
      "automatic classification": 10501,
      "text predefined": 121188,
      "predefined categories": 88823,
      "categories problem": 15745,
      "problem text": 91262,
      "studied different": 114156,
      "different communities": 31047,
      "like natural": 64072,
      "processing data": 91654,
      "data mining": 26130,
      "mining information": 70237,
      "retrieval text": 102435,
      "classification important": 17226,
      "important constituent": 52132,
      "tasks like": 119255,
      "topic identification": 122529,
      "spam filtering": 111115,
      "genre classification": 47830,
      "readability assessment": 95944,
      "performance text": 86798,
      "classification improves": 17230,
      "capturing nonlocal": 15482,
      "improvement text": 52777,
      "structure extraction": 113863,
      "extraction step": 42494,
      "pattern identification": 85715,
      "detailed study": 29786,
      "enable future": 36354,
      "work nlp": 134656,
      "nlp tasks": 79704,
      "information phrase": 54841,
      "answering information": 6634,
      "extraction machine": 42377,
      "translation text": 125376,
      "classification paper": 17318,
      "paper provides": 84381,
      "provides different": 94019,
      "levels classification": 63528,
      "detailed comparison": 29762,
      "resource grammar": 100846,
      "present ongoing": 89628,
      "ongoing research": 81738,
      "research investigating": 100540,
      "development important": 30390,
      "important component": 52123,
      "lowlevel linguistic": 65480,
      "morphology syntax": 74806,
      "20 languages": 510,
      "multilingual applications": 75196,
      "ideal case": 51344,
      "language require": 60049,
      "lexicon translation": 63908,
      "highly restricted": 50342,
      "knowledge particular": 58099,
      "particular language": 85428,
      "language average": 58854,
      "inflectional paradigms": 54297,
      "straightforward task": 113434,
      "paper focusing": 83956,
      "propose extension": 92671,
      "application developers": 6843,
      "semantic level": 105096,
      "demonstrate approach": 28664,
      "languages named": 60749,
      "recognition important": 96891,
      "extraction questionanswering": 42459,
      "questionanswering machine": 95252,
      "translation document": 124762,
      "document summarization": 33086,
      "paper forward": 83961,
      "languages particular": 60792,
      "various rulebased": 131186,
      "rulebased machine": 103364,
      "approaches available": 8074,
      "available approaches": 10948,
      "approaches named": 8243,
      "discuss related": 32281,
      "research field": 100505,
      "like indian": 64050,
      "suffers lack": 115260,
      "lack appropriate": 58682,
      "appropriate resources": 8429,
      "recognition requires": 96988,
      "requires large": 100284,
      "large data": 61068,
      "useful feature": 128885,
      "english assamese": 37071,
      "word embeddings": 133046,
      "embeddings supervised": 35965,
      "tasks analyze": 118929,
      "word embedding": 133003,
      "embedding method": 35439,
      "supervised tasks": 115842,
      "maps words": 67170,
      "similar contexts": 109081,
      "contexts measured": 22408,
      "compared word": 19491,
      "embeddings including": 35754,
      "including recent": 53357,
      "representations named": 99763,
      "ner chunking": 77030,
      "chunking dependency": 16958,
      "framework multilingual": 45616,
      "results proposed": 102061,
      "good better": 48465,
      "embeddings tasks": 35973,
      "tasks investigate": 119224,
      "results multilingual": 101965,
      "embeddings languages": 35773,
      "languages available": 60419,
      "public use": 94276,
      "recognition social": 97000,
      "entities named": 37825,
      "movies books": 74913,
      "tv shows": 126007,
      "systems need": 117017,
      "need retrain": 76853,
      "models new": 73637,
      "new entities": 78902,
      "entities arrive": 37745,
      "require retraining": 100194,
      "preliminary study": 89281,
      "focus entity": 44759,
      "entity type": 38115,
      "movie title": 74911,
      "using data": 129529,
      "collected twitter": 18437,
      "evaluation sets": 39386,
      "sets including": 107676,
      "entities corresponding": 37762,
      "movies training": 74915,
      "entities final": 37786,
      "final model": 44109,
      "model shows": 72012,
      "gives strong": 48209,
      "strong evidence": 113672,
      "evidence approach": 39635,
      "set entities": 107429,
      "disease mention": 32337,
      "pubmed abstracts": 94381,
      "relationships biomedical": 98316,
      "enables knowledge": 36389,
      "knowledge applied": 57754,
      "language process": 59830,
      "process bionlp": 91433,
      "attempt address": 9731,
      "address challenge": 3650,
      "research depends": 100463,
      "corpora evaluating": 23474,
      "extraction systems": 42500,
      "systems training": 117205,
      "training machine": 123695,
      "learning models": 62768,
      "corpora created": 23446,
      "small numbers": 110178,
      "expert annotators": 41216,
      "extended periods": 41809,
      "periods time": 87043,
      "recent studies": 96521,
      "studies shown": 114280,
      "crowdsourcing platforms": 25121,
      "amazons mechanical": 5273,
      "mechanical turk": 67936,
      "generate highquality": 46952,
      "highquality annotations": 50370,
      "annotations biomedical": 6412,
      "text investigated": 121063,
      "investigated use": 56844,
      "disease mentions": 32338,
      "abstracts used": 1827,
      "ncbi disease": 76699,
      "disease corpus": 32335,
      "crowdsourcing protocol": 25122,
      "documents training": 33305,
      "set gold": 107454,
      "overall measure": 83238,
      "based simple": 12048,
      "simple voting": 109542,
      "voting method": 131954,
      "documents span": 33296,
      "quality annotations": 94594,
      "increases number": 53649,
      "cost quality": 24371,
      "quality results": 94781,
      "results demonstrate": 101670,
      "valuable tool": 130772,
      "wellannotated corpora": 132393,
      "evaluating semantic": 39094,
      "semantic models": 105114,
      "genuine similarity": 47843,
      "similarity estimation": 109229,
      "estimation present": 38661,
      "standard resource": 112294,
      "resource evaluating": 100843,
      "distributional semantic": 32718,
      "models improves": 73371,
      "improves existing": 52977,
      "important ways": 52292,
      "gold standards": 48443,
      "similarity association": 109200,
      "entities associated": 37746,
      "models different": 73067,
      "wider range": 132609,
      "range applications": 95541,
      "adjective noun": 3856,
      "noun verb": 80434,
      "pairs independent": 83564,
      "enables finegrained": 36386,
      "performance models": 86539,
      "models concepts": 72950,
      "concepts different": 20612,
      "unlike existing": 127433,
      "standard evaluations": 112239,
      "automatic approaches": 10494,
      "stateoftheart models": 112722,
      "models perform": 73720,
      "future improvements": 46279,
      "improvements distributional": 52842,
      "unsupervised keyword": 127645,
      "keyword extraction": 57635,
      "legal texts": 63317,
      "texts work": 121653,
      "recently proposed": 96738,
      "proposed unsupervised": 93579,
      "extraction algorithm": 42260,
      "corpus polish": 23931,
      "public procurement": 94269,
      "language domain": 58984,
      "domain independent": 33547,
      "noncontent words": 80177,
      "heavily depends": 49652,
      "depends choice": 29280,
      "algorithm automatic": 4699,
      "automatic approach": 10492,
      "based statistical": 12071,
      "statistical properties": 113154,
      "term distribution": 120209,
      "evaluating neural": 39083,
      "neural word": 78729,
      "word representations": 133441,
      "provide comparative": 93780,
      "study neural": 114453,
      "representations traditional": 99938,
      "spaces based": 111091,
      "cooccurrence counts": 23223,
      "compositional tasks": 20136,
      "tasks use": 119586,
      "different semantic": 31407,
      "semantic spaces": 105307,
      "compositional models": 20120,
      "models test": 74179,
      "approaches tasks": 8371,
      "tasks involving": 119228,
      "verb disambiguation": 131521,
      "sentence similarity": 106073,
      "additionally evaluate": 3605,
      "spaces using": 111108,
      "using simple": 130176,
      "simple compositional": 109389,
      "compositional methods": 20118,
      "constrained language": 21571,
      "paraphrase detection": 84811,
      "detection dialogue": 29922,
      "dialogue act": 30629,
      "act tagging": 3071,
      "cooccurrence vectors": 23239,
      "compositional method": 20117,
      "embeddings robust": 35922,
      "stable performance": 112114,
      "tasks resolving": 119469,
      "lexical ambiguity": 63729,
      "regression models": 97704,
      "models meaning": 73559,
      "meaning paper": 67656,
      "provides method": 94050,
      "method improving": 68894,
      "compositional distributional": 20111,
      "distributional models": 32712,
      "contrast previous": 22702,
      "relatively simple": 98412,
      "models work": 74358,
      "work use": 134870,
      "use robust": 128255,
      "robust model": 103047,
      "linear regression": 64365,
      "disambiguation method": 31970,
      "suggest effectiveness": 115310,
      "effectiveness approach": 34866,
      "detection present": 30024,
      "informationtheoretic measures": 55125,
      "classifiers detecting": 17608,
      "detecting different": 29830,
      "different stages": 31447,
      "results switchboard": 102246,
      "par stateoftheart": 84525,
      "stateoftheart incremental": 112675,
      "detection methods": 29990,
      "methods better": 69346,
      "accuracy faster": 2163,
      "computational overhead": 20415,
      "using incremental": 129749,
      "metrics propose": 69994,
      "processing evaluation": 91668,
      "evaluation empirical": 39200,
      "study compare": 114336,
      "compare various": 19313,
      "distance measures": 32424,
      "developed computational": 30260,
      "computational biology": 20362,
      "stateoftheart parser": 112799,
      "charniak 2000": 16642,
      "describes different": 29398,
      "different tree": 31517,
      "evaluation article": 39123,
      "article argue": 9111,
      "similar information": 109099,
      "tree edit": 125586,
      "al 2011": 4602,
      "provides empirical": 94023,
      "empirical evidence": 36163,
      "high correlations": 50052,
      "correlations different": 24253,
      "twitter data": 126103,
      "data twitter": 26585,
      "encourages users": 36775,
      "users use": 129179,
      "bare minimum": 11429,
      "analysis twitter": 5912,
      "generally designed": 46854,
      "grammatical structure": 48722,
      "used english": 128514,
      "language order": 59726,
      "make sense": 66724,
      "performed level": 86928,
      "level individual": 63458,
      "individual tokens": 53936,
      "tokens words": 122353,
      "data presents": 26268,
      "presents findings": 89851,
      "raw data": 95853,
      "performance stanford": 86738,
      "biomedical texts": 14209,
      "paper performance": 84069,
      "performance dependency": 86286,
      "reported performance": 99068,
      "biomedical concepts": 14168,
      "parsers fail": 85023,
      "connected concepts": 21123,
      "performance terms": 86792,
      "recall fscore": 96340,
      "attachment score": 9697,
      "dependency parse": 29156,
      "texts addition": 121451,
      "investigation reveals": 56865,
      "play vital": 87707,
      "vital role": 131849,
      "degraded performance": 28561,
      "performance crfbased": 86263,
      "crfbased named": 24763,
      "performance crf": 86262,
      "crf based": 24743,
      "based systems": 12106,
      "task considered": 118007,
      "set language": 107473,
      "independent features": 53770,
      "features languages": 43585,
      "specific feature": 111441,
      "hindi english": 50464,
      "built wikipedia": 14946,
      "test results": 120486,
      "results achieves": 101496,
      "achieves highest": 2802,
      "measure 69": 67777,
      "used ner": 128652,
      "accuracy measure": 2207,
      "present open": 89630,
      "source morphological": 110794,
      "morphological analyzing": 74687,
      "using open": 129988,
      "source code": 110714,
      "tool available": 122373,
      "features word": 43785,
      "alignment important": 4962,
      "important natural": 52186,
      "processing task": 91809,
      "correspondence natural": 24269,
      "languages recently": 60836,
      "loglinear models": 65046,
      "models word": 74347,
      "received considerable": 96360,
      "considerable attention": 21242,
      "generative discriminative": 47724,
      "discriminative approaches": 32207,
      "major challenge": 66543,
      "challenge remains": 16095,
      "contrastive approach": 22726,
      "approach aims": 7342,
      "observed training": 81236,
      "partition functions": 85510,
      "functions based": 46070,
      "based observation": 11908,
      "probability mass": 90864,
      "posterior distributions": 88482,
      "allows efficient": 5148,
      "efficient accurate": 35052,
      "features experiments": 43499,
      "experiments approach": 40777,
      "approach achieves": 7301,
      "achieves significant": 2852,
      "improvements stateoftheart": 52913,
      "stateoftheart unsupervised": 113019,
      "alignment methods": 4969,
      "methods learning": 69587,
      "learning distributed": 62511,
      "distributed word": 32634,
      "representations natural": 99767,
      "natural logic": 76611,
      "logic reasoning": 64997,
      "distributed semantic": 32623,
      "representations proven": 99832,
      "proven valuable": 93738,
      "sophisticated language": 110670,
      "tasks remains": 119454,
      "remains open": 98811,
      "open question": 81924,
      "question possible": 95201,
      "possible train": 88438,
      "distributed representations": 32619,
      "representations support": 99916,
      "rich diverse": 102740,
      "logical reasoning": 65014,
      "address question": 3756,
      "question using": 95237,
      "using neural": 129946,
      "neural networkbased": 78420,
      "networkbased models": 77488,
      "models learning": 73492,
      "learning embeddings": 62535,
      "neural networks": 78433,
      "networks neural": 77674,
      "neural tensor": 78702,
      "tensor networks": 120190,
      "networks experiments": 77594,
      "experiments evaluate": 40939,
      "evaluate models": 38869,
      "models ability": 72637,
      "simulated data": 109636,
      "wordnet noun": 133775,
      "positive results": 88340,
      "results promising": 102059,
      "promising future": 92275,
      "representations applied": 99501,
      "errors digital": 38364,
      "lexicographic resources": 63868,
      "using dictionary": 129566,
      "language paradigm": 59796,
      "manual automatic": 66981,
      "error correction": 38292,
      "correction noisy": 24161,
      "lexicographic data": 63867,
      "language dictionary": 58961,
      "unique identifiers": 127188,
      "performed using": 86937,
      "simple commands": 109386,
      "set text": 107608,
      "errors generated": 38370,
      "generated automatically": 47052,
      "automatically correct": 10747,
      "problems discuss": 91315,
      "paradigm task": 84550,
      "bilingual dictionaries": 14031,
      "dictionaries experiments": 30859,
      "experiments improve": 40958,
      "turkish tweets": 125965,
      "tweets social": 126058,
      "media texts": 68174,
      "texts significant": 121606,
      "significant information": 108819,
      "information sources": 54995,
      "application areas": 6834,
      "areas including": 8891,
      "unfortunately existing": 127078,
      "existing solutions": 40287,
      "tasks named": 119321,
      "recognition perform": 96962,
      "formal texts": 45180,
      "texts usually": 121647,
      "usually perform": 130441,
      "perform poorly": 86043,
      "poorly applied": 88065,
      "purpose improving": 94429,
      "improving named": 53119,
      "tweets using": 126069,
      "sets experiments": 107671,
      "resources better": 100953,
      "better fit": 13590,
      "tweets observe": 126042,
      "different settings": 31425,
      "settings provided": 107834,
      "results detecting": 101720,
      "agreement errors": 4391,
      "article approach": 9110,
      "approach automatic": 7372,
      "steps required": 113333,
      "required develop": 100215,
      "develop simple": 30231,
      "processing application": 91615,
      "purpose use": 94443,
      "framework capable": 45451,
      "capable analyzing": 15192,
      "analyzing texts": 6056,
      "texts bulgarian": 121467,
      "language embedded": 58994,
      "software applications": 110513,
      "regular expressions": 97726,
      "used starting": 128777,
      "starting point": 112408,
      "applications related": 7004,
      "related language": 97872,
      "detection annotation": 29876,
      "word groups": 133308,
      "specific set": 111491,
      "set criteria": 107408,
      "mining text": 70265,
      "documents written": 33318,
      "language ability": 58804,
      "ability extract": 1521,
      "public opinion": 94267,
      "review sites": 102579,
      "make decisions": 66656,
      "used determining": 128487,
      "determining polarity": 30152,
      "forum posts": 45308,
      "classified positive": 17511,
      "test different": 120445,
      "classifiers provide": 17630,
      "detailed evaluation": 29773,
      "improving overall": 53133,
      "overall performance": 83245,
      "manually generated": 67064,
      "generated dataset": 47069,
      "dataset achieving": 26723,
      "accuracy performance": 2234,
      "performance systems": 86781,
      "systems automated": 116744,
      "comparable human": 19145,
      "human evaluator": 50851,
      "viable option": 131644,
      "text data": 120848,
      "analysis finally": 5623,
      "finally present": 44216,
      "present statistics": 89719,
      "statistics derived": 113197,
      "posts using": 88526,
      "using graph": 129716,
      "algorithms generate": 4856,
      "generate natural": 46972,
      "medical concepts": 68197,
      "need communicate": 76786,
      "medical information": 68208,
      "concepts expressed": 20621,
      "visual information": 131796,
      "information patients": 54833,
      "current state": 25340,
      "order increase": 82344,
      "context paper": 22208,
      "method graph": 68865,
      "data consisting": 25779,
      "use natural": 128162,
      "generation component": 47338,
      "representing deep": 100046,
      "deep semantic": 28420,
      "semantic structure": 105309,
      "ontology concepts": 81841,
      "language expressions": 59037,
      "network motifs": 77361,
      "networks constructed": 77546,
      "different texts": 31493,
      "croatian language": 24855,
      "preparing data": 89301,
      "data network": 26162,
      "networks present": 77702,
      "datasets furthermore": 27491,
      "furthermore compare": 46153,
      "results existing": 101773,
      "existing results": 40271,
      "results linguistic": 101891,
      "language similar": 60093,
      "similar languages": 109106,
      "belong family": 12716,
      "parsing significant": 85238,
      "significant progress": 108850,
      "representation introduce": 99269,
      "new techniques": 79211,
      "techniques tackle": 119993,
      "tackle problems": 117309,
      "second parser": 104438,
      "eliminates need": 35299,
      "grammar ccg": 48624,
      "fully exploit": 45956,
      "exploit power": 41435,
      "language theory": 60168,
      "enrich type": 37559,
      "order better": 82286,
      "better guide": 13598,
      "guide parsing": 49243,
      "accurate parses": 2350,
      "domains using": 33883,
      "using sentence": 130145,
      "learn semantics": 62141,
      "functional approach": 46055,
      "distributional semantics": 32723,
      "linear maps": 64351,
      "distributional vectors": 32744,
      "vectors representing": 131472,
      "vector representing": 131363,
      "representing sentence": 100064,
      "sentence conduct": 105803,
      "initial investigation": 55217,
      "uses matrix": 129243,
      "logistic regression": 65031,
      "regression classifier": 97696,
      "plausibility task": 87677,
      "function compare": 46008,
      "compare method": 19254,
      "corpusbased method": 24078,
      "method constructing": 68737,
      "training effective": 123597,
      "effective disambiguation": 34663,
      "words uses": 134295,
      "used author": 128399,
      "prove efficient": 93698,
      "compiled corpus": 19717,
      "early work": 34398,
      "work based": 134396,
      "based limited": 11805,
      "limited corpus": 64224,
      "corpus english": 23773,
      "article deals": 9119,
      "considered english": 21290,
      "men women": 68406,
      "relationships text": 98335,
      "morphology words": 74812,
      "text use": 121391,
      "advancement natural": 3965,
      "topological analysis": 122677,
      "analysis complex": 5536,
      "complex networks": 19845,
      "approaching human": 8408,
      "language complex": 58898,
      "finegrained entity": 44349,
      "tagging entity": 117390,
      "category labels": 15787,
      "labels mention": 58618,
      "mention entity": 68420,
      "entity document": 37927,
      "standard systems": 112316,
      "systems focus": 116909,
      "focus small": 44819,
      "set types": 107621,
      "types recent": 126344,
      "ling weld": 64410,
      "weld 2012": 132390,
      "suggests using": 115383,
      "using large": 129785,
      "large finegrained": 61092,
      "finegrained label": 44361,
      "label set": 58413,
      "dramatic improvements": 34088,
      "improvements downstream": 52843,
      "downstream tasks": 34032,
      "absence labeled": 1717,
      "data existing": 25914,
      "existing finegrained": 40136,
      "tagging systems": 117452,
      "systems obtain": 117029,
      "examples automatically": 39810,
      "automatically using": 10840,
      "entities types": 37884,
      "depends context": 29281,
      "generalization propose": 46794,
      "propose task": 93107,
      "task contextdependent": 118019,
      "local context": 64905,
      "context sentence": 22259,
      "sentence document": 105825,
      "new resources": 79110,
      "mentions annotated": 68445,
      "fine types": 44329,
      "types provide": 126341,
      "provide baseline": 93759,
      "baseline experimental": 12217,
      "data practice": 26255,
      "synonym extraction": 116346,
      "extraction large": 42368,
      "important task": 52262,
      "task natural": 118433,
      "processing used": 91849,
      "expansion question": 40376,
      "applications automatic": 6898,
      "automatic synonym": 10682,
      "applications previous": 6991,
      "previous studies": 90488,
      "extraction limited": 42374,
      "scale datasets": 103712,
      "datasets paper": 27609,
      "paper build": 83755,
      "build large": 14778,
      "capture challenges": 15272,
      "world scenarios": 135047,
      "cost function": 24353,
      "learning problem": 62910,
      "problem feature": 91056,
      "feature learning": 43291,
      "based deep": 11626,
      "deep neural": 28347,
      "network model": 77325,
      "model complicated": 70868,
      "synonym pairs": 116347,
      "based svms": 12099,
      "novel feature": 80568,
      "based neural": 11885,
      "network outperforms": 77377,
      "outperforms methods": 82917,
      "features specifically": 43740,
      "best performance": 13390,
      "performance model": 86534,
      "model surpasses": 72123,
      "improvement learning": 52718,
      "learning word": 63164,
      "representations relational": 99851,
      "relational graphs": 98091,
      "words relations": 134167,
      "representation similarity": 99410,
      "analogy detection": 5423,
      "words share": 134210,
      "connected semantic": 21128,
      "relations hand": 98181,
      "motivated close": 74843,
      "attributes relations": 10202,
      "relational graph": 98090,
      "relations propose": 98241,
      "method learn": 68934,
      "learn latent": 62077,
      "latent representation": 61598,
      "representation individual": 99261,
      "words proposed": 134147,
      "method considers": 68727,
      "cooccurrences words": 23245,
      "words existing": 133945,
      "approaches word": 8404,
      "word representation": 133435,
      "representation learning": 99281,
      "words cooccur": 133888,
      "accuracy word": 2318,
      "representations learnt": 99736,
      "learnt using": 63255,
      "using proposed": 130059,
      "method use": 69205,
      "learnt word": 63256,
      "solve semantic": 110617,
      "semantic word": 105355,
      "problems experimental": 91320,
      "possible learn": 88411,
      "learn better": 61993,
      "better word": 13769,
      "representations using": 99965,
      "important research": 52226,
      "linguistics introduced": 64612,
      "set languages": 107474,
      "human brain": 50771,
      "empirical studies": 36195,
      "grammar research": 48654,
      "entirely different": 37733,
      "knowledge transfer": 58211,
      "purpose language": 94430,
      "learning application": 62334,
      "approach originally": 7760,
      "improve parsing": 52452,
      "used small": 128765,
      "training datasets": 123570,
      "techniques learn": 119919,
      "learn probability": 62117,
      "related different": 97852,
      "parsing recent": 85211,
      "blend learning": 14270,
      "embedding word": 35529,
      "similarity neural": 109276,
      "neural machine": 77964,
      "neural language": 77934,
      "models learn": 73478,
      "learn word": 62188,
      "representations embeddings": 99619,
      "embeddings capture": 35584,
      "capture rich": 15370,
      "rich linguistic": 102764,
      "linguistic conceptual": 64446,
      "conceptual information": 20663,
      "information investigate": 54702,
      "investigate embeddings": 56751,
      "embeddings learned": 35781,
      "learned neural": 62235,
      "class neural": 17046,
      "model embeddings": 71054,
      "embeddings translation": 35992,
      "models outperform": 73673,
      "outperform learned": 82714,
      "monolingual models": 74597,
      "models tasks": 74164,
      "tasks require": 119459,
      "require knowledge": 100151,
      "knowledge conceptual": 57836,
      "conceptual similarity": 20671,
      "translating english": 124588,
      "french english": 45807,
      "desirable properties": 29660,
      "properties translation": 92489,
      "languages finally": 60586,
      "finally apply": 44146,
      "method training": 69193,
      "training neural": 123731,
      "neural translation": 78720,
      "models large": 73460,
      "large vocabularies": 61323,
      "vocabulary expansion": 131875,
      "algorithm results": 4792,
      "minimal degradation": 70157,
      "embedding quality": 35478,
      "quality embedding": 94638,
      "embedding spaces": 35502,
      "online demo": 81762,
      "translationbased embeddings": 125444,
      "embeddings used": 36003,
      "used applications": 128391,
      "according similarity": 2002,
      "similarity andor": 109197,
      "lexical function": 63764,
      "monolingual embeddings": 74588,
      "embeddings better": 35576,
      "language modelling": 59486,
      "modelling machine": 72609,
      "presents indepth": 89854,
      "indepth investigation": 53800,
      "models translation": 74247,
      "scaling neural": 103776,
      "models difficult": 73075,
      "difficult task": 31662,
      "task crucial": 118037,
      "crucial realworld": 25160,
      "realworld applications": 96149,
      "applications paper": 6983,
      "paper evaluates": 83899,
      "mt quality": 74985,
      "quality new": 94733,
      "scaling techniques": 103778,
      "scalable training": 103698,
      "training algorithms": 123353,
      "investigate noise": 56790,
      "noise contrastive": 80045,
      "contrastive estimation": 22728,
      "speed improvements": 111867,
      "improvements explore": 52857,
      "backoff ngram": 11315,
      "models neural": 73624,
      "models make": 73547,
      "make strong": 66732,
      "memory constrained": 68306,
      "traditional models": 122844,
      "models raw": 73856,
      "set recommendations": 107558,
      "scalable neural": 103697,
      "model mt": 71533,
      "stylistic analysis": 114625,
      "translation research": 125217,
      "stylistic properties": 114631,
      "received attention": 96357,
      "english translations": 37317,
      "translations using": 125505,
      "burrows delta": 14972,
      "differences observed": 30965,
      "large degree": 61077,
      "translations study": 125495,
      "study translations": 114537,
      "results work": 102346,
      "work suggest": 134830,
      "works author": 134924,
      "extent work": 41990,
      "work examines": 134499,
      "attempt investigate": 9748,
      "using relative": 130106,
      "graphs paper": 48967,
      "paper concerned": 83771,
      "nearest neighbor": 76719,
      "neighbor search": 76987,
      "information structure": 55008,
      "semantic model": 105111,
      "model ranked": 71844,
      "conflate different": 21029,
      "space provides": 111048,
      "provides important": 94041,
      "important information": 52171,
      "information different": 54484,
      "used determine": 128486,
      "define set": 28492,
      "introduce relative": 56529,
      "method uncover": 69203,
      "models provide": 73834,
      "wellknown semantic": 132423,
      "model glove": 71256,
      "skipgram model": 109986,
      "model annotating": 70665,
      "turkic languages": 125958,
      "languages exhibit": 60563,
      "relationships lexical": 98323,
      "exploring automated": 41659,
      "translation lexicon": 124908,
      "lexicon induction": 63887,
      "information extent": 54556,
      "relationships words": 98339,
      "words clear": 133860,
      "present methodology": 89552,
      "languages method": 60724,
      "research effort": 100482,
      "research improving": 100533,
      "phrase based": 87334,
      "based language": 11791,
      "used word": 128850,
      "word level": 133339,
      "level models": 63477,
      "models similar": 74049,
      "mt use": 75004,
      "preliminary experimental": 89266,
      "approach outperform": 7762,
      "outperform word": 82772,
      "word based": 132933,
      "quality phrase": 94751,
      "translation empirical": 124781,
      "challenge machine": 16050,
      "mt widely": 75007,
      "used approach": 128392,
      "approach apply": 7359,
      "apply word": 7226,
      "model lm": 71467,
      "sentence words": 106131,
      "words speech": 134235,
      "lms outperform": 64886,
      "nbest list": 76690,
      "list reranking": 64698,
      "languages syntactic": 60909,
      "necessary task": 76762,
      "translation challenging": 124691,
      "challenging task": 16316,
      "task develop": 118082,
      "morphological rich": 74728,
      "rich agglutinative": 102725,
      "agglutinative languages": 4339,
      "analysis used": 5926,
      "used understand": 128832,
      "language sentence": 60077,
      "grammatical information": 48711,
      "help understand": 49784,
      "understand language": 126754,
      "way literature": 132105,
      "literature survey": 64776,
      "understand different": 126745,
      "languages various": 60963,
      "various approaches": 131039,
      "approaches used": 8393,
      "used develop": 128488,
      "develop tools": 30245,
      "provides survey": 94085,
      "survey research": 116190,
      "research papers": 100578,
      "model important": 71304,
      "models phrasebased": 73743,
      "phrasebased models": 87386,
      "models generate": 73284,
      "generate target": 47027,
      "phrases words": 87468,
      "model dependency": 70969,
      "dependency phrases": 29216,
      "phrases research": 87449,
      "research work": 100665,
      "solving problem": 110647,
      "problem paper": 91155,
      "paper tackle": 84464,
      "tackle problem": 117303,
      "problem designing": 91006,
      "designing novel": 29644,
      "novel phrasebased": 80684,
      "model attempts": 70697,
      "attempts solve": 9775,
      "phrase language": 87353,
      "model determine": 70982,
      "phrase boundary": 87340,
      "largescale monolingual": 61464,
      "alleviate data": 5030,
      "data sparsity": 26480,
      "sparsity problem": 111259,
      "extensive experiments": 41892,
      "experiments chinesetoenglish": 40830,
      "chinesetoenglish translation": 16872,
      "translation phrasebased": 125119,
      "significantly improve": 108925,
      "improve translation": 52568,
      "absolute bleu": 1723,
      "bleu score": 14323,
      "survey word": 116207,
      "computational models": 20410,
      "language phenomena": 59811,
      "important factor": 52160,
      "quality efficiency": 94637,
      "research published": 100599,
      "dominant language": 33930,
      "pairs instead": 83565,
      "choice optimal": 16894,
      "approach new": 7740,
      "new translation": 79229,
      "task driven": 118123,
      "driven empirical": 34151,
      "present comprehensive": 89411,
      "statistical modeling": 113133,
      "task including": 118285,
      "modeling question": 72520,
      "important understand": 52286,
      "occur given": 81499,
      "given language": 48054,
      "end conduct": 36798,
      "conduct qualitative": 20888,
      "analysis word": 5938,
      "collection linguistic": 18477,
      "characteristics language": 16527,
      "pair select": 83451,
      "smt framework": 110283,
      "framework best": 45441,
      "model powerful": 71754,
      "learning highdimensional": 62633,
      "highdimensional word": 50156,
      "representations capture": 99529,
      "rich semantic": 102781,
      "prior work": 90740,
      "work learning": 134608,
      "representations does": 99608,
      "does account": 33323,
      "account word": 2043,
      "word ambiguity": 132913,
      "single representation": 109787,
      "representation word": 99463,
      "overcome limitation": 83289,
      "number word": 80995,
      "using greedy": 129720,
      "heuristic approaches": 49871,
      "approaches paper": 8274,
      "propose adaptive": 92528,
      "model nonparametric": 71593,
      "capable automatically": 15193,
      "required number": 100221,
      "representations words": 99988,
      "derive efficient": 29338,
      "model empirically": 71059,
      "empirically demonstrate": 36223,
      "demonstrate efficiency": 28732,
      "induction task": 54036,
      "analysis important": 5649,
      "linguistics natural": 64616,
      "studies word": 114300,
      "word structure": 133590,
      "word language": 133330,
      "language current": 58931,
      "current scenario": 25331,
      "analysis techniques": 5891,
      "techniques popular": 119950,
      "day day": 27840,
      "processing language": 91689,
      "morphology word": 74810,
      "word analyzed": 132921,
      "language contains": 58916,
      "contains complex": 21818,
      "structure work": 113986,
      "work used": 134872,
      "language limited": 59253,
      "database paper": 26643,
      "database consists": 26639,
      "grammatical features": 48703,
      "features words": 43790,
      "wordlevel semantic": 133754,
      "data stored": 26506,
      "domain semantic": 33633,
      "features argument": 43367,
      "fast accurate": 43126,
      "accurate dependency": 2329,
      "parser dependency": 84934,
      "tools natural": 122457,
      "processing important": 91678,
      "important applications": 52101,
      "applications downstream": 6919,
      "translation knowledge": 124883,
      "knowledge acquisition": 57742,
      "wsj test": 135182,
      "set ranks": 107556,
      "parsers fastest": 85024,
      "sentences second": 106485,
      "accuracy using": 2314,
      "cluster features": 17927,
      "parser trained": 85000,
      "trained syntactic": 123304,
      "different options": 31311,
      "specific tasks": 111498,
      "20 license": 511,
      "learning understand": 63130,
      "learn rich": 62132,
      "representations success": 99912,
      "recent nlp": 96484,
      "research developing": 100466,
      "developing models": 30349,
      "learn useful": 62183,
      "representations phrases": 99805,
      "sentences proved": 106459,
      "bridging gap": 14608,
      "phrasal semantics": 87328,
      "semantics neural": 105440,
      "language embedding": 58995,
      "embedding models": 35452,
      "models effectively": 73117,
      "effectively trained": 34853,
      "trained map": 123194,
      "map dictionary": 67106,
      "dictionary definitions": 30877,
      "lexical representations": 63798,
      "present applications": 89370,
      "concept given": 20573,
      "definition description": 28521,
      "tasks neural": 119338,
      "better existing": 13580,
      "commercial systems": 18841,
      "systems rely": 117106,
      "engineering results": 37056,
      "results highlight": 101825,
      "highlight effectiveness": 50259,
      "effectiveness neural": 34920,
      "neural embedding": 77901,
      "embedding architectures": 35375,
      "models understand": 74268,
      "conversational speech": 23031,
      "speech language": 111702,
      "applications involving": 6949,
      "limiting factor": 64308,
      "factor building": 42854,
      "better language": 13614,
      "model propose": 71810,
      "propose simple": 93050,
      "large amounts": 61006,
      "amounts data": 5332,
      "smaller training": 110244,
      "set closely": 107391,
      "techniques lead": 119917,
      "lead significant": 61867,
      "significant reduction": 108854,
      "lowresource languages": 65507,
      "twitter text": 126127,
      "text useful": 121393,
      "useful learning": 128900,
      "indomain text": 53984,
      "use word": 128359,
      "reductions perplexity": 97471,
      "perplexity additionally": 87062,
      "additionally introduce": 3611,
      "introduce method": 56455,
      "using social": 130193,
      "textual information": 121712,
      "useful data": 128872,
      "vocabulary coverage": 131871,
      "used infer": 128589,
      "corpus specifically": 24011,
      "specifically proposed": 111587,
      "lexical patterns": 63790,
      "patterns automatically": 85732,
      "automatically identify": 10784,
      "followed filtering": 44946,
      "different similarity": 31427,
      "compared model": 19384,
      "paper evaluated": 83898,
      "evaluated using": 39018,
      "using current": 129525,
      "current standard": 25339,
      "standard novel": 112287,
      "evaluation set": 39385,
      "shown good": 108474,
      "ccg parsing": 15859,
      "integration information": 55809,
      "information multiword": 54782,
      "combinatory categorial": 18611,
      "categorial grammar": 15719,
      "shown benefit": 108445,
      "adding information": 3383,
      "similar pipeline": 109122,
      "corpus contains": 23722,
      "contains sentences": 21838,
      "algorithm deal": 4713,
      "data limitations": 26089,
      "limitations approach": 64170,
      "study effect": 114360,
      "data parsing": 26221,
      "parser decisions": 84932,
      "decisions training": 27968,
      "obtained training": 81424,
      "data improves": 26028,
      "improves results": 53038,
      "standard model": 112266,
      "significantly outperforms": 109002,
      "outperforms baseline": 82840,
      "baseline model": 12253,
      "model gold": 71258,
      "model performs": 71735,
      "performs significantly": 87022,
      "parsing data": 85093,
      "effect results": 34610,
      "standard benchmark": 112203,
      "benchmark fail": 12825,
      "does significantly": 33402,
      "noticeable improvements": 80398,
      "improvements using": 52935,
      "parsing discuss": 85102,
      "discuss ways": 32293,
      "results finally": 101793,
      "recognition pipeline": 96965,
      "obtain different": 81276,
      "different results": 31389,
      "recognition neural": 96947,
      "character embeddings": 16428,
      "embeddings stateoftheart": 35952,
      "stateoftheart named": 112763,
      "rely handcrafted": 98704,
      "handcrafted features": 49338,
      "tasks partofspeech": 119373,
      "partofspeech pos": 85531,
      "tagging text": 117457,
      "work propose": 134709,
      "ner uses": 77095,
      "automatically learned": 10798,
      "learned features": 62219,
      "features approach": 43365,
      "wordlevel characterlevel": 133731,
      "characterlevel representations": 16583,
      "embeddings perform": 35859,
      "classification perform": 17329,
      "perform extensive": 85998,
      "extensive number": 41947,
      "number experiments": 80886,
      "using annotated": 129335,
      "corpora different": 23456,
      "texts portuguese": 121580,
      "texts spanish": 121616,
      "demonstrate neural": 28807,
      "successfully applied": 115175,
      "achieve stateoftheart": 2556,
      "outperforms stateoftheart": 82995,
      "points f1score": 87858,
      "72 points": 1224,
      "points f1": 87856,
      "meaning combination": 67621,
      "models nlp": 73645,
      "information require": 54923,
      "se mantic": 104285,
      "logicbased representations": 65027,
      "structure capture": 113815,
      "similarity ratings": 109290,
      "capture sentence": 15388,
      "logicbased approaches": 65025,
      "probabilistic logic": 90816,
      "models parsing": 73716,
      "base construction": 11450,
      "inference rules": 54213,
      "distributional information": 32705,
      "existing paraphrase": 40250,
      "use evaluate": 128033,
      "evaluate distributional": 38825,
      "lexical entailment": 63760,
      "entailment approaches": 37654,
      "easily added": 34433,
      "logical rules": 65018,
      "rules learns": 103415,
      "probabilistic inference": 90810,
      "inference solve": 54220,
      "inference problems": 54199,
      "problems efficiently": 91316,
      "evaluate approach": 38788,
      "approach use": 7995,
      "task textual": 118791,
      "representations particular": 99803,
      "dataset achieve": 26718,
      "fine tuning": 44328,
      "knowledge learning": 58050,
      "learning vector": 63157,
      "vector representation": 131341,
      "words important": 133990,
      "benefit natural": 12982,
      "available models": 11049,
      "bias caused": 13795,
      "lack knowledge": 58720,
      "inspired deep": 55565,
      "authors propose": 10384,
      "propose supervised": 93096,
      "words provide": 134149,
      "provide additional": 93752,
      "additional supervised": 3575,
      "learning framework": 62593,
      "framework knowledge": 45593,
      "knowledge rich": 58160,
      "vectors word": 131486,
      "perform intrinsic": 86015,
      "intrinsic evaluation": 56352,
      "relational similarity": 98099,
      "similarity prediction": 109281,
      "extrinsic evaluations": 42618,
      "like sentence": 64088,
      "sentence completion": 105791,
      "analysis experiments": 5612,
      "experiments results": 41120,
      "10 datasets": 128,
      "datasets proposed": 27632,
      "framework significantly": 45682,
      "syntactic features": 116403,
      "detection paper": 30012,
      "research explore": 100499,
      "subjects objects": 114709,
      "objects verbs": 81151,
      "detection achieve": 29868,
      "achieve objective": 2506,
      "conducted series": 20946,
      "syntactic similarity": 116480,
      "similarity sentences": 109308,
      "comparison results": 19570,
      "work sequencetosequence": 134794,
      "sequencetosequence neural": 107187,
      "neural net": 78229,
      "net models": 77114,
      "graphemetophoneme conversion": 48923,
      "translation methods": 124934,
      "model recently": 71863,
      "recently shown": 96763,
      "shown promising": 108513,
      "tasks machine": 119276,
      "models conditioned": 72955,
      "conditioned source": 20806,
      "source words": 110868,
      "words used": 134291,
      "text image": 121041,
      "image captioning": 51762,
      "captioning models": 15263,
      "images used": 51814,
      "used generate": 128564,
      "caption text": 15254,
      "past work": 85653,
      "work approach": 134376,
      "quality terms": 94808,
      "terms bleu": 120283,
      "bleu paper": 14298,
      "applicability models": 6820,
      "qualitatively different": 94579,
      "task input": 118301,
      "input output": 55379,
      "output vocabularies": 83149,
      "given output": 48071,
      "generation approach": 47310,
      "approach able": 7288,
      "rival stateoftheart": 102883,
      "able significantly": 1686,
      "significantly advance": 108880,
      "bidirectional long": 13934,
      "long shortterm": 65099,
      "shortterm memory": 108315,
      "memory lstm": 68322,
      "lstm neural": 65654,
      "networks use": 77808,
      "alignment information": 4963,
      "information used": 55074,
      "used conventional": 128452,
      "conventional approaches": 22876,
      "set predictions": 107537,
      "latentvariable pcfgs": 61644,
      "works creating": 134934,
      "models noise": 73647,
      "features training": 43765,
      "estimation model": 38656,
      "multiple models": 75618,
      "models addition": 72718,
      "addition simple": 3475,
      "simple variant": 109539,
      "compact models": 19110,
      "models experiments": 73203,
      "experiments natural": 41033,
      "parsing english": 85107,
      "comparable state": 19175,
      "art english": 9049,
      "achieve f1": 2458,
      "f1 score": 42674,
      "based smt": 12054,
      "networks vector": 77812,
      "phrasebased machine": 87382,
      "data method": 26126,
      "based recently": 11974,
      "recently introduced": 96695,
      "introduced neural": 56587,
      "representation models": 99333,
      "models words": 74357,
      "words sentences": 134197,
      "sentences time": 106521,
      "time models": 122051,
      "models used": 74286,
      "used end": 128512,
      "end end": 36806,
      "scores obtained": 104190,
      "obtained method": 81384,
      "bleu loss": 14289,
      "phrase table": 87374,
      "features combined": 43407,
      "improve bleu": 52344,
      "approach improving": 7629,
      "improving statistical": 53170,
      "building recent": 14879,
      "recent success": 96529,
      "application neural": 6871,
      "networks model": 77659,
      "model various": 72299,
      "architecture neural": 8707,
      "capture important": 15314,
      "multitask learning": 75821,
      "network parameters": 77384,
      "jointly proposed": 57385,
      "proposed methods": 93410,
      "methods results": 69733,
      "results significant": 102179,
      "overall improvement": 83234,
      "improvement 27": 52674,
      "18 bleu": 414,
      "chineseenglish translation": 16859,
      "translation stateoftheart": 125291,
      "network features": 77255,
      "features hierarchical": 43538,
      "hierarchical neural": 49987,
      "paragraphs documents": 84577,
      "coherent long": 18319,
      "long texts": 65140,
      "texts like": 121541,
      "longer documents": 65157,
      "documents challenging": 33198,
      "challenging problem": 16295,
      "recurrent networks": 97185,
      "networks models": 77663,
      "task training": 118800,
      "longshort term": 65187,
      "term memory": 120227,
      "multisentence paragraphs": 75781,
      "lstm model": 65639,
      "hierarchically builds": 50020,
      "paragraph embeddings": 84566,
      "sentences words": 106547,
      "reconstruct original": 97109,
      "standard metrics": 112263,
      "like rouge": 64086,
      "showing neural": 108418,
      "models able": 72641,
      "able encode": 1635,
      "encode texts": 36461,
      "way preserve": 132119,
      "preserve syntactic": 89928,
      "discourse coherence": 32042,
      "step generating": 113268,
      "generating coherent": 47202,
      "coherent text": 18328,
      "text units": 121386,
      "units neural": 127261,
      "significantly impact": 108924,
      "models described": 73046,
      "understanding neural": 126902,
      "nlp neural": 79653,
      "networks successfully": 77785,
      "applied nlp": 7102,
      "tasks resulting": 119471,
      "difficult interpret": 31636,
      "meaning meanings": 67647,
      "phrases paper": 87442,
      "similar work": 109173,
      "work computer": 134424,
      "computer vision": 20494,
      "negation intensification": 76915,
      "introduce simple": 56535,
      "simple straightforward": 109521,
      "straightforward methods": 113432,
      "contributes final": 22763,
      "gradient backpropagation": 48597,
      "measure information": 67804,
      "information flow": 54619,
      "test methods": 120471,
      "methods sentiment": 69741,
      "sentiment using": 106816,
      "simple recurrent": 109500,
      "recurrent nets": 97183,
      "wide applications": 132488,
      "applications understanding": 7031,
      "properties deep": 92444,
      "deep networks": 28346,
      "lstms outperform": 65714,
      "outperform simple": 82743,
      "corpus sentencelevel": 23987,
      "introduce corpus": 56400,
      "rated human": 95813,
      "scale corpus": 103707,
      "corpus annotated": 23660,
      "annotated using": 6257,
      "using amazon": 129332,
      "amazon mechanical": 5269,
      "experiments correlation": 40859,
      "annotations sentence": 6461,
      "conducted controlled": 20916,
      "setting despite": 107741,
      "inherent difficulty": 55188,
      "difficulty annotation": 31693,
      "annotation task": 6384,
      "linguistic variables": 64579,
      "document terms": 33094,
      "terms style": 120387,
      "corpus largest": 23858,
      "corpus released": 23965,
      "implied sentiments": 52038,
      "mental state": 68413,
      "event introduce": 39523,
      "investigate feasibility": 56759,
      "labels crowdsourcing": 58586,
      "crowdsourcing experiments": 25118,
      "experiments present": 41063,
      "present models": 89560,
      "models predicting": 73765,
      "verb predicates": 131525,
      "based distributional": 11653,
      "distributional word": 32745,
      "relations empirical": 98144,
      "results confirm": 101619,
      "various data": 131069,
      "data sources": 26475,
      "people use": 85900,
      "results potential": 102045,
      "potential use": 88593,
      "subtle biases": 114997,
      "neural architecture": 77831,
      "architecture able": 8603,
      "communicate natural": 19018,
      "communicative interactions": 19054,
      "verbal nonverbal": 131537,
      "modeling human": 72442,
      "language abilities": 58803,
      "current knowledge": 25289,
      "information processed": 54872,
      "work presents": 134694,
      "based largescale": 11795,
      "largescale neural": 61484,
      "architecture developed": 8645,
      "knowledge involved": 58027,
      "working memory": 134902,
      "memory model": 68349,
      "network takes": 77449,
      "input neural": 55373,
      "yields output": 135424,
      "flow information": 44682,
      "information working": 55108,
      "memory components": 68302,
      "components neural": 20038,
      "gating mechanisms": 46538,
      "capable learning": 15205,
      "language starting": 60119,
      "priori knowledge": 90770,
      "knowledge structure": 58192,
      "words role": 134179,
      "role different": 103170,
      "different classes": 31038,
      "classes words": 17080,
      "incremental learning": 53730,
      "process able": 91415,
      "expressive language": 41772,
      "model validated": 72292,
      "output sentences": 83119,
      "sentences expressing": 106308,
      "range language": 95572,
      "language presented": 59823,
      "download use": 33979,
      "written entirely": 135126,
      "best knowledge": 13343,
      "knowledge time": 58208,
      "time writing": 122139,
      "tools specifically": 122475,
      "accuracy 95": 2086,
      "comparison manually": 19552,
      "manually annotated": 67024,
      "evaluation accuracy": 39113,
      "results analysis": 101507,
      "analysis accuracy": 5483,
      "developed software": 30301,
      "language lemmatization": 59247,
      "used define": 128479,
      "lemmatization accuracy": 63336,
      "word vector": 133627,
      "representations datadriven": 99584,
      "datadriven representation": 26677,
      "learning words": 63171,
      "words technique": 134259,
      "importance nlp": 52070,
      "useful source": 128932,
      "source features": 110760,
      "features downstream": 43467,
      "semantic theories": 105335,
      "best present": 13414,
      "constructing interpretable": 21677,
      "interpretable word": 56257,
      "word vectors": 133633,
      "handcrafted linguistic": 49351,
      "resources like": 100998,
      "like wordnet": 64114,
      "wordnet framenet": 133771,
      "analyze performance": 5990,
      "stateoftheart evaluation": 112658,
      "evaluation methods": 39274,
      "methods distributional": 69455,
      "standard distributional": 112232,
      "distributional approaches": 32694,
      "approaches neural": 8253,
      "neural conversational": 77877,
      "conversational model": 23016,
      "model conversational": 70918,
      "conversational modeling": 23018,
      "machine intelligence": 65760,
      "restricted specific": 101344,
      "specific domains": 111433,
      "require handcrafted": 100141,
      "handcrafted rules": 49354,
      "simple approach": 109360,
      "approach task": 7962,
      "uses recently": 129268,
      "proposed sequence": 93548,
      "sequence sequence": 107052,
      "sequence framework": 106943,
      "predicting sentence": 89010,
      "sentence given": 105891,
      "given previous": 48087,
      "previous sentence": 90458,
      "strength model": 113567,
      "trained endtoend": 123129,
      "endtoend requires": 36961,
      "requires fewer": 100271,
      "straightforward model": 113433,
      "model generate": 71233,
      "generate simple": 47016,
      "given large": 48057,
      "training dataset": 123568,
      "preliminary results": 89276,
      "suggest despite": 115306,
      "objective function": 81078,
      "able extract": 1638,
      "extract knowledge": 42094,
      "specific dataset": 111427,
      "dataset large": 26996,
      "general domain": 46651,
      "domain dataset": 33499,
      "dataset movie": 27039,
      "movie subtitles": 74906,
      "dataset model": 27027,
      "model solution": 72064,
      "model perform": 71718,
      "perform simple": 86073,
      "forms common": 45234,
      "common sense": 18921,
      "sense reasoning": 105694,
      "lack consistency": 58694,
      "failure mode": 42987,
      "mode model": 70491,
      "model complex": 70865,
      "complex network": 19843,
      "methods widely": 69845,
      "widely employed": 132543,
      "study fundamental": 114389,
      "properties language": 92456,
      "language recent": 60024,
      "years methods": 135272,
      "methods complex": 69385,
      "proved useful": 93716,
      "useful create": 128871,
      "models despite": 73052,
      "despite large": 29702,
      "large studies": 61281,
      "studies devoted": 114209,
      "devoted represent": 30476,
      "represent texts": 99148,
      "texts physical": 121577,
      "physical models": 87477,
      "limited number": 64256,
      "number studies": 80974,
      "properties underlying": 92490,
      "performance natural": 86556,
      "tasks paper": 119359,
      "networks methods": 77658,
      "methods able": 69272,
      "performance current": 86266,
      "current statistical": 25370,
      "extracted texts": 42187,
      "traditional textual": 122881,
      "textual description": 121689,
      "performance obtained": 86585,
      "used proposed": 128705,
      "proposed model": 93422,
      "model generic": 71248,
      "generic framework": 47799,
      "textual applications": 121674,
      "plays pivotal": 87738,
      "pivotal role": 87586,
      "simple fast": 109431,
      "approach semantic": 7875,
      "feature vectors": 43329,
      "query candidate": 94950,
      "candidate logical": 15111,
      "form train": 45136,
      "train classifier": 122909,
      "classifier using": 17592,
      "product vectors": 92049,
      "vectors using": 131484,
      "simple features": 109432,
      "features achieves": 43352,
      "achieves average": 2735,
      "average f1": 11187,
      "dataset comparable": 26804,
      "comparable complex": 19137,
      "complex systems": 19886,
      "extending word": 41822,
      "embeddings embeddings": 35679,
      "learn embeddings": 62040,
      "flexible word": 44668,
      "embeddings input": 35764,
      "input does": 55325,
      "does need": 33372,
      "need additional": 76777,
      "additional training": 3587,
      "embeddings obtained": 35839,
      "space word": 111084,
      "use wordnet": 128367,
      "lexical resource": 63799,
      "easily applied": 34437,
      "performance word": 86857,
      "similarity word": 109332,
      "analysis big": 5520,
      "big data": 13991,
      "30 million": 903,
      "analysis consider": 5543,
      "consider large": 21210,
      "number text": 80980,
      "cooking recipes": 23251,
      "properties data": 92442,
      "look various": 65218,
      "approaches allow": 8057,
      "information high": 54657,
      "high low": 50086,
      "space embedding": 110997,
      "term set": 120244,
      "look analysis": 65212,
      "singular plural": 109860,
      "addition use": 3483,
      "space mapping": 111020,
      "bounding box": 14527,
      "languages called": 60432,
      "language cnl": 58882,
      "given different": 48019,
      "different names": 31276,
      "variety languages": 130987,
      "designed applied": 29580,
      "applied improve": 7078,
      "communication humans": 19034,
      "provide natural": 93878,
      "apparent differences": 6786,
      "languages general": 60605,
      "presented comprehensive": 89780,
      "survey existing": 116171,
      "100 languages": 164,
      "classification languages": 17247,
      "form single": 45124,
      "conceptual space": 20672,
      "space natural": 111031,
      "english end": 37124,
      "end formal": 36808,
      "article provide": 9154,
      "provide common": 93778,
      "common terminology": 18935,
      "common model": 18893,
      "understanding general": 126849,
      "general nature": 46683,
      "point researchers": 87814,
      "researchers interested": 100695,
      "help developers": 49711,
      "developers make": 30329,
      "design decisions": 29529,
      "user demographics": 128975,
      "gender language": 46603,
      "making use": 66871,
      "use unsupervised": 128350,
      "unsupervised morphology": 127675,
      "tasks demonstrate": 119034,
      "demonstrate effectiveness": 28705,
      "effectiveness proposed": 34925,
      "morphological features": 74701,
      "features compared": 43412,
      "character ngram": 16452,
      "ngram baseline": 79428,
      "multiword unit": 75950,
      "unit extraction": 127203,
      "extraction turkish": 42522,
      "measures mathematical": 67879,
      "based occurrences": 11912,
      "cooccurrences text": 23242,
      "purpose paper": 94436,
      "paper test": 84472,
      "national corpus": 76211,
      "statistical comparison": 113088,
      "according linguistic": 1986,
      "provide focus": 93832,
      "focus study": 44826,
      "corpus data": 23740,
      "data applying": 25613,
      "measures evaluating": 67861,
      "measures linguistic": 67877,
      "findings include": 44303,
      "corpus turkish": 24051,
      "generate good": 46948,
      "good word": 48510,
      "analyze critical": 5958,
      "critical components": 24807,
      "embedding training": 35519,
      "model corpus": 70930,
      "corpus training": 24044,
      "training parameters": 123776,
      "embedding algorithms": 35370,
      "compare using": 19311,
      "evaluate word": 38944,
      "analyzing semantic": 6049,
      "properties using": 92491,
      "using feature": 129669,
      "networks provide": 77717,
      "provide simple": 93920,
      "training word": 123952,
      "embeddings discover": 35658,
      "domain important": 33545,
      "corpus suitable": 24028,
      "suitable domain": 115397,
      "desired task": 29673,
      "using larger": 129792,
      "larger corpus": 61357,
      "corpus yields": 24071,
      "yields better": 135399,
      "results second": 102144,
      "faster models": 43185,
      "provide sufficient": 93933,
      "performance cases": 86196,
      "complex models": 19834,
      "sufficiently large": 115294,
      "rely development": 98684,
      "development set": 30420,
      "validation loss": 130741,
      "loss training": 65303,
      "training embedding": 123602,
      "linguistic regularities": 64541,
      "embeddings using": 36008,
      "using matrix": 129863,
      "work explored": 134519,
      "learning continuous": 62461,
      "continuous vector": 22641,
      "representations reflecting": 99848,
      "underlying semantics": 126699,
      "words simple": 134223,
      "simple vector": 109540,
      "using cosine": 129511,
      "cosine distances": 24336,
      "shown capture": 108448,
      "certain types": 15962,
      "past tense": 85651,
      "approach capture": 7410,
      "continuous word": 22647,
      "representations based": 99512,
      "just individual": 57463,
      "individual word": 53940,
      "euclidean space": 38760,
      "space called": 110976,
      "dimensions based": 31765,
      "based mathematical": 11831,
      "cosine distance": 24335,
      "distances word": 32442,
      "analogy tasks": 5426,
      "tasks approach": 118933,
      "approach performs": 7796,
      "better previous": 13680,
      "approaches given": 8175,
      "given task": 48150,
      "multilayer network": 75175,
      "network language": 77294,
      "unified framework": 127108,
      "framework structural": 45701,
      "structural analysis": 113757,
      "linguistic subsystems": 64563,
      "focus complex": 44747,
      "networks research": 77737,
      "modeling multiple": 72481,
      "multiple phenomena": 75639,
      "multilayer networks": 75176,
      "systems propose": 117082,
      "networks language": 77637,
      "framework modeling": 45613,
      "modeling linguistic": 72462,
      "structural properties": 113788,
      "aspects natural": 9397,
      "language systems": 60140,
      "systems represented": 117111,
      "linguistic units": 64577,
      "links model": 64682,
      "model relations": 71889,
      "language defined": 58940,
      "subword level": 115025,
      "network layers": 77302,
      "original text": 82554,
      "modeled language": 72358,
      "language obtained": 59724,
      "substantial differences": 114855,
      "structures different": 114069,
      "regardless language": 97668,
      "croatian english": 24854,
      "language dependent": 58942,
      "networks analysis": 77505,
      "topological structure": 122682,
      "structure syntactic": 113965,
      "languages findings": 60589,
      "network framework": 77259,
      "systematic approach": 116659,
      "approach model": 7712,
      "model linguistic": 71464,
      "unified view": 127133,
      "space modeling": 111024,
      "common evaluation": 18878,
      "evaluation practice": 39327,
      "space models": 111025,
      "measure models": 67808,
      "ability predict": 1559,
      "predict human": 88890,
      "human judgments": 50882,
      "relations word": 98282,
      "pairs existing": 83538,
      "existing evaluation": 40131,
      "sets consist": 107657,
      "collected english": 18419,
      "english word": 37342,
      "potential impact": 88560,
      "language word": 60343,
      "pairs presented": 83608,
      "presented human": 89785,
      "human scores": 50959,
      "english italian": 37176,
      "italian german": 57093,
      "german russian": 47912,
      "fluent language": 44712,
      "necessarily best": 76745,
      "correlate human": 24194,
      "used model": 128635,
      "model training": 72218,
      "training suggesting": 123892,
      "models humans": 73345,
      "semantic judgments": 105088,
      "judgments finally": 57446,
      "results improved": 101842,
      "correlations human": 24257,
      "language effect": 58991,
      "transitionbased parsing": 124514,
      "parsing modeling": 85161,
      "characters instead": 16611,
      "instead words": 55695,
      "makes applicable": 66754,
      "transitionbased parser": 124510,
      "parser uses": 85005,
      "lstm recurrent": 65660,
      "recurrent neural": 97186,
      "networks learn": 77640,
      "learn representations": 62127,
      "representations parser": 99802,
      "parser state": 84996,
      "representations representations": 99855,
      "representations constructed": 99562,
      "forms similar": 45260,
      "experiments morphologically": 41023,
      "languages parsing": 60789,
      "model benefits": 70760,
      "benefits incorporating": 13008,
      "words word": 134315,
      "survey paper": 116182,
      "paper survey": 84455,
      "disambiguation wsd": 32000,
      "major languages": 66570,
      "languages world": 60981,
      "world research": 135046,
      "research wsd": 100667,
      "wsd conducted": 135169,
      "approaches adopted": 8053,
      "different research": 31386,
      "research works": 100666,
      "art performance": 9076,
      "performance domain": 86312,
      "domain recent": 33623,
      "recent works": 96597,
      "finally survey": 44237,
      "bengali language": 13033,
      "language survey": 60136,
      "survey different": 116170,
      "sense definitions": 105658,
      "wordnet based": 133767,
      "sense definition": 105657,
      "definition words": 28530,
      "words available": 133832,
      "available bengali": 10953,
      "sentences automatically": 106222,
      "different groups": 31172,
      "sentences collected": 106242,
      "bengali text": 13034,
      "corpus developed": 23752,
      "developed tdil": 30308,
      "tdil project": 119678,
      "project govt": 92217,
      "govt india": 48547,
      "lexical item": 63772,
      "collected bengali": 18404,
      "model useful": 72259,
      "sentences applied": 106212,
      "applied algorithm": 7044,
      "sentences contain": 106259,
      "sentences different": 106277,
      "different meanings": 31248,
      "experiment achieved": 40450,
      "sense classification": 105653,
      "classification total": 17470,
      "affect results": 4225,
      "information main": 54745,
      "classification sentences": 17391,
      "automatic text": 10692,
      "classification machine": 17257,
      "learning information": 62654,
      "extraction word": 42538,
      "automatic machine": 10581,
      "translation evaluation": 124799,
      "metric based": 69871,
      "based dependency": 11636,
      "hypothesis reference": 51273,
      "human express": 50858,
      "express information": 41701,
      "avoid problems": 11240,
      "novel automatic": 80503,
      "automatic evaluation": 10525,
      "model need": 71563,
      "tree generate": 125594,
      "model quality": 71835,
      "probability order": 90867,
      "order obtain": 82375,
      "similarity introduce": 109245,
      "fscore new": 45913,
      "new metric": 79019,
      "experiment results": 40489,
      "gets stateoftheart": 47954,
      "performance level": 86494,
      "sentence level": 105924,
      "translation orthographic": 125078,
      "strong signal": 113712,
      "especially closely": 38439,
      "models wellsuited": 74336,
      "similarities propose": 109186,
      "model latent": 71419,
      "latent variables": 61635,
      "similarity features": 109233,
      "training computationally": 123392,
      "model address": 70633,
      "challenge perform": 16081,
      "perform approximate": 85946,
      "approximate inference": 8454,
      "scales large": 103769,
      "outperforms existing": 82886,
      "existing generative": 40139,
      "models exploiting": 73210,
      "orthographic features": 82594,
      "features improve": 43553,
      "using entropy": 129640,
      "adequately reflect": 3840,
      "like bleu": 64026,
      "does consider": 33336,
      "matched words": 67376,
      "words method": 134053,
      "bleu meteor": 14292,
      "removing biases": 98883,
      "mt metrics": 74971,
      "metrics using": 70020,
      "stateoftheart mt": 112758,
      "systems outputs": 117041,
      "evaluating different": 39048,
      "quality systems": 94807,
      "tend produce": 120160,
      "humans usually": 51118,
      "does address": 33326,
      "biases present": 13880,
      "present data": 89426,
      "data does": 25863,
      "knowledge types": 58218,
      "types features": 126290,
      "used useful": 128836,
      "lists using": 64721,
      "initially trained": 55248,
      "trained standard": 123292,
      "standard human": 112247,
      "adaptation domain": 3216,
      "diverse nbest": 32827,
      "syntax need": 116544,
      "problem solving": 91237,
      "syntax results": 116555,
      "evolution language": 39691,
      "language evolution": 59019,
      "giving new": 48216,
      "relation language": 98047,
      "using ensembles": 129637,
      "sets word": 107726,
      "embeddings distributed": 35660,
      "words deep": 133901,
      "learning beneficial": 62404,
      "beneficial tasks": 12955,
      "different embedding": 31120,
      "vary greatly": 131244,
      "instead relying": 55679,
      "embedding learning": 35428,
      "learning paper": 62867,
      "ensemble approach": 37577,
      "approach combining": 7428,
      "combining different": 18722,
      "experiments word": 41210,
      "similarity analogy": 109193,
      "networks shown": 77765,
      "shown improve": 108484,
      "performance range": 86648,
      "tasks designing": 119052,
      "designing training": 29653,
      "optimal settings": 82170,
      "settings paper": 107828,
      "correct number": 24111,
      "number units": 80992,
      "hidden layers": 49893,
      "layers introduce": 61783,
      "automatically adjusting": 10714,
      "hidden units": 49917,
      "regularization apply": 97744,
      "method language": 68927,
      "demonstrate ability": 28653,
      "ability correctly": 1501,
      "number hidden": 80893,
      "models machine": 73539,
      "translation decoder": 124738,
      "models maintain": 73545,
      "exploratory analysis": 41510,
      "processing models": 91713,
      "models models": 73577,
      "models natural": 73606,
      "distributions linguistic": 32765,
      "argue quality": 8927,
      "quality model": 94720,
      "posterior distribution": 88481,
      "pipeline components": 87533,
      "exploratory data": 41511,
      "user trust": 129048,
      "method analyze": 68634,
      "apply compare": 7166,
      "models contribute": 72983,
      "confidence intervals": 20987,
      "event extraction": 39512,
      "extraction task": 42502,
      "task computational": 117996,
      "computational sociolinguistics": 20434,
      "survey language": 116175,
      "language social": 60100,
      "linguistics cl": 64606,
      "article present": 9145,
      "present survey": 89731,
      "field computational": 43935,
      "aim provide": 4472,
      "provide comprehensive": 93785,
      "comprehensive overview": 20266,
      "social identity": 110343,
      "use social": 128283,
      "social interaction": 110345,
      "demonstrate potential": 28819,
      "potential synergy": 88588,
      "research communities": 100444,
      "largescale datadriven": 61418,
      "datadriven methods": 26669,
      "complement existing": 19728,
      "challenge methods": 16054,
      "conclude discussion": 20711,
      "open challenges": 81883,
      "open information": 81904,
      "information linguistic": 54732,
      "process divided": 91461,
      "divided steps": 32910,
      "document event": 33002,
      "event detection": 39507,
      "detection event": 29945,
      "methods proposed": 69693,
      "analysis application": 5494,
      "information analysis": 54374,
      "enron corpus": 37575,
      "corpus paper": 23920,
      "study paper": 114463,
      "issue meaning": 57005,
      "meaning grounded": 67637,
      "infer semantic": 54103,
      "meaning components": 67623,
      "english translation": 37314,
      "independently constructed": 53786,
      "constructed semantic": 21668,
      "lexicons german": 63925,
      "perform detailed": 85979,
      "detailed analysis": 29753,
      "analysis evaluation": 5606,
      "set bilingual": 107380,
      "englishfrench englishgerman": 37363,
      "englishgerman parallel": 37382,
      "direction translation": 31830,
      "literary works": 64740,
      "ted talks": 120051,
      "human machine": 50910,
      "translation specifically": 125281,
      "used task": 128803,
      "identification research": 51432,
      "research direction": 100472,
      "validate quality": 130730,
      "previous results": 90455,
      "supervised unsupervised": 115849,
      "experiments additional": 40761,
      "additional datasets": 3508,
      "datasets languages": 27539,
      "languages kannada": 60661,
      "recognition classification": 96833,
      "based multinomial": 11867,
      "nave bayes": 76676,
      "bayes mnb": 12571,
      "classifier named": 17560,
      "process identification": 91502,
      "identification proper": 51424,
      "proper nouns": 92420,
      "date time": 27831,
      "essential challenging": 38543,
      "task aim": 117860,
      "aim work": 4475,
      "work develop": 134471,
      "develop novel": 30220,
      "multinomial naive": 75473,
      "adopted paper": 3901,
      "based feature": 11714,
      "feature extraction": 43275,
      "extraction training": 42521,
      "corpus using": 24060,
      "using term": 130280,
      "frequency inverse": 45844,
      "inverse document": 56697,
      "document frequency": 33011,
      "frequency fitting": 45842,
      "evaluation discussed": 39188,
      "conducted training": 20954,
      "tokens test": 122339,
      "corpus 5000": 23634,
      "model works": 72330,
      "wordformation process": 133713,
      "process languages": 91518,
      "languages problematic": 60814,
      "problematic natural": 91292,
      "form word": 45148,
      "embeddings enable": 35686,
      "methods present": 69678,
      "present unsupervised": 89754,
      "unsupervised approach": 127602,
      "regularities semantic": 97740,
      "semantic vector": 105351,
      "space based": 110974,
      "high quality": 50110,
      "quality subsequent": 94805,
      "splitting algorithm": 111953,
      "algorithm based": 4700,
      "based analyses": 11503,
      "highly effective": 50319,
      "particularly ambiguous": 85468,
      "english machine": 37203,
      "translation experiments": 124811,
      "experiments semantic": 41123,
      "better translations": 13753,
      "solution text": 110566,
      "studied problem": 114169,
      "considered solved": 21302,
      "received little": 96365,
      "attention despite": 9815,
      "classification needs": 17304,
      "needs know": 76894,
      "document using": 33105,
      "partial information": 85289,
      "information document": 54504,
      "document processed": 33056,
      "devise method": 30462,
      "method make": 68950,
      "make predictions": 66709,
      "problem evident": 91038,
      "domains like": 33804,
      "possible paper": 88417,
      "paper analyzes": 83735,
      "bayes classifier": 12566,
      "problem specifically": 91245,
      "assess performance": 9485,
      "classifying documents": 17669,
      "increasingly number": 53705,
      "simple modification": 109474,
      "allows make": 5169,
      "information best": 54397,
      "used purpose": 128709,
      "extensive experimental": 41882,
      "evaluation effectiveness": 39197,
      "classification simple": 17407,
      "simple solution": 109519,
      "compared state": 19451,
      "art methodologies": 9058,
      "pave way": 85804,
      "development effective": 30384,
      "classification techniques": 17465,
      "techniques based": 119842,
      "general purpose": 46698,
      "realtime sentiment": 96139,
      "art benchmarks": 9037,
      "consider fact": 21200,
      "analysis highlighting": 5638,
      "explicitly incorporating": 41374,
      "incorporating uncertainty": 53570,
      "dataset high": 26962,
      "quality evaluate": 94655,
      "new problem": 79085,
      "oral history": 82261,
      "oral histories": 82260,
      "topically coherent": 122599,
      "make judgments": 66692,
      "making judgments": 66848,
      "speed task": 111879,
      "task extracting": 118194,
      "different people": 31325,
      "begin end": 12632,
      "low agreement": 65345,
      "approaches promising": 8293,
      "performs slightly": 87026,
      "slightly better": 110020,
      "does perform": 33377,
      "perform significantly": 86068,
      "used suggest": 128791,
      "segmentation task": 104640,
      "task needs": 118447,
      "needs better": 76888,
      "polish english": 87980,
      "english speech": 37291,
      "speech statistical": 111808,
      "iwslt 2014": 57161,
      "research explores": 100501,
      "various training": 131226,
      "training settings": 123858,
      "systems spoken": 117152,
      "language various": 60327,
      "parallel text": 84687,
      "2014 evaluation": 577,
      "used basis": 128407,
      "basis training": 12548,
      "training language": 123672,
      "models development": 73062,
      "comparable corpora": 19138,
      "bleu nist": 14297,
      "meteor ter": 68574,
      "ter metrics": 120202,
      "evaluate effects": 38830,
      "effects data": 34979,
      "morphological information": 74712,
      "deep analysis": 28197,
      "work automatic": 134390,
      "automatic data": 10507,
      "data correction": 25798,
      "automatic taxonomy": 10688,
      "query logs": 94972,
      "additional sources": 3573,
      "sources information": 110897,
      "information search": 54958,
      "store detailed": 113379,
      "detailed information": 29777,
      "information web": 55093,
      "web users": 132270,
      "users interactions": 129138,
      "daily basis": 25507,
      "common knowledge": 18886,
      "research shown": 100624,
      "shown possible": 108508,
      "possible extract": 88403,
      "methods obtain": 69647,
      "obtain similar": 81324,
      "queries query": 94937,
      "propose mixture": 92783,
      "lines research": 64407,
      "related queries": 97890,
      "used improve": 128580,
      "improve search": 52535,
      "effectiveness efficiency": 34886,
      "developed method": 30286,
      "method combines": 68703,
      "supervised classification": 115736,
      "classification model": 17271,
      "model successfully": 72107,
      "successfully extract": 115187,
      "hyponymy relations": 51247,
      "information language": 54711,
      "network models": 77341,
      "years neural": 135277,
      "machinelearning models": 66324,
      "models yielding": 74369,
      "yielding stateoftheart": 135385,
      "image recognition": 51790,
      "recognition speech": 97005,
      "speech processing": 111731,
      "processing recently": 91782,
      "recently neural": 96715,
      "models started": 74092,
      "applied textual": 7129,
      "models perspective": 73742,
      "perspective natural": 87166,
      "processing research": 91786,
      "attempt bring": 9732,
      "speed neural": 111870,
      "neural techniques": 78700,
      "input encoding": 55330,
      "feedforward networks": 43846,
      "networks convolutional": 77552,
      "convolutional networks": 23168,
      "networks recurrent": 77726,
      "networks computation": 77542,
      "gradient computation": 48599,
      "acoustic cues": 3000,
      "role shaping": 103218,
      "phonological structure": 87315,
      "language means": 59278,
      "paper introduced": 84005,
      "procedure order": 91394,
      "different contexts": 31061,
      "results applying": 101512,
      "provide evidence": 93816,
      "proved effective": 93712,
      "email responses": 35340,
      "topic prediction": 122561,
      "prediction approach": 89029,
      "approach propose": 7825,
      "email replies": 35339,
      "enable use": 36367,
      "use lda": 128121,
      "extract latent": 42096,
      "topics collection": 122618,
      "use latent": 128120,
      "label data": 58387,
      "data obtaining": 26188,
      "silver standard": 109066,
      "standard topic": 112326,
      "topic labelling": 122535,
      "set train": 107611,
      "classifier predict": 17572,
      "topic distribution": 122509,
      "response based": 101190,
      "based features": 11716,
      "current sentence": 25333,
      "sentence experimental": 105862,
      "results large": 101880,
      "domain proposed": 33620,
      "predicting best": 88977,
      "80 cases": 1287,
      "topics possible": 122649,
      "shows potential": 108609,
      "potential method": 88576,
      "interactive setting": 56029,
      "list likely": 64694,
      "networks multilingual": 77665,
      "multilingual multimodal": 75297,
      "multimodal representation": 75450,
      "learning recently": 62946,
      "recently lot": 96706,
      "learning common": 62443,
      "representations multiple": 99760,
      "multiple views": 75744,
      "views data": 131723,
      "data typically": 26587,
      "typically common": 126417,
      "representations learned": 99724,
      "learned using": 62278,
      "using parallel": 130000,
      "work address": 134354,
      "address realworld": 3759,
      "realworld scenario": 96175,
      "direct parallel": 31797,
      "parallel data": 84630,
      "propose model": 92784,
      "model learning": 71437,
      "common representation": 18915,
      "downstream applications": 33987,
      "applications focus": 6931,
      "transfer learning": 124074,
      "using pivot": 130018,
      "pivot language": 87580,
      "language ii": 59153,
      "language l1": 59232,
      "language l2": 59233,
      "performance multilingual": 86546,
      "multilingual document": 75246,
      "document classification": 32961,
      "available multilingual": 11050,
      "ted corpus": 120050,
      "multimodal retrieval": 75455,
      "new dataset": 78856,
      "dataset created": 26836,
      "created released": 24683,
      "released work": 98497,
      "learning baselines": 62401,
      "presents results": 89894,
      "utterance ranking": 130612,
      "ubuntu dialog": 126518,
      "dialog corpus": 30557,
      "largest publicly": 61539,
      "multiturn dialog": 75914,
      "corpus use": 24055,
      "previously reported": 90619,
      "reported models": 99066,
      "models independent": 73394,
      "independent evaluation": 53769,
      "evaluation using": 39435,
      "data second": 26378,
      "evaluate performances": 38892,
      "performances various": 86903,
      "bilstms cnns": 14112,
      "dataset create": 26835,
      "create ensemble": 24612,
      "predictions multiple": 89182,
      "models ensemble": 73158,
      "stateoftheart result": 112889,
      "dataset finally": 26933,
      "discuss future": 32253,
      "future plans": 46289,
      "plans using": 87647,
      "case machine": 15596,
      "translation machine": 124920,
      "terms quality": 120368,
      "available web": 11142,
      "provide reasonable": 93904,
      "reasonable translations": 96216,
      "translations systems": 125496,
      "systems perfect": 117055,
      "domains paper": 33827,
      "effects different": 34981,
      "different training": 31505,
      "training methods": 123707,
      "translation used": 125413,
      "used medical": 128628,
      "medical data": 68200,
      "corpora related": 23573,
      "open subtitles": 81940,
      "used ground": 128566,
      "systems bleu": 116761,
      "used order": 128674,
      "order evaluate": 82319,
      "evaluate results": 38915,
      "results various": 102310,
      "various systems": 131211,
      "systems experiments": 116891,
      "models hierarchical": 73324,
      "hierarchical models": 49977,
      "analysis polish": 5753,
      "data processing": 26281,
      "casing punctuation": 15675,
      "normalization phase": 80342,
      "used compare": 128437,
      "translation engine": 124789,
      "quality scores": 94785,
      "scores 50": 104148,
      "average results": 11212,
      "relatively high": 98393,
      "lowest score": 65464,
      "english polish": 37246,
      "general medical": 46675,
      "neural reranking": 78652,
      "year nara": 135234,
      "nara institute": 76157,
      "institute science": 55698,
      "science technology": 103986,
      "2015 workshop": 599,
      "workshop asian": 134998,
      "asian translation": 9290,
      "translation addition": 124625,
      "reranking component": 100369,
      "component using": 20011,
      "neural attentional": 77848,
      "neural mt": 78210,
      "reranking provides": 100378,
      "provides large": 94046,
      "objective evaluation": 81076,
      "evaluation measures": 39264,
      "measures bleu": 67855,
      "time results": 122093,
      "results carry": 101562,
      "carry manual": 15555,
      "manual evaluation": 66992,
      "main contributions": 66412,
      "grammatical correctness": 48693,
      "improvements lexical": 52868,
      "words partofspeech": 134109,
      "memory recurrent": 68371,
      "network bidirectional": 77169,
      "network blstmrnn": 77175,
      "blstmrnn shown": 14382,
      "shown effective": 108462,
      "tagging sequential": 117446,
      "sequential data": 107217,
      "data speech": 26492,
      "speech utterances": 111844,
      "utterances handwritten": 130642,
      "handwritten documents": 49430,
      "documents word": 33315,
      "powerful representation": 88676,
      "properties natural": 92468,
      "language study": 60130,
      "study propose": 114486,
      "use blstmrnn": 127927,
      "task tested": 118784,
      "treebank wsj": 125654,
      "set stateoftheart": 107589,
      "tagging accuracy": 117372,
      "accuracy achieved": 2093,
      "approach achieve": 7295,
      "achieve good": 2468,
      "good performance": 48490,
      "performance comparable": 86219,
      "stanford pos": 112375,
      "pos tagger": 88219,
      "rational speech": 95833,
      "acts model": 3144,
      "acts rsa": 3146,
      "rsa model": 103334,
      "model treats": 72235,
      "use recursive": 128234,
      "recursive process": 97288,
      "speaker listener": 111292,
      "agents reason": 4327,
      "manual specification": 67014,
      "semantic lexicon": 105100,
      "preventing use": 90371,
      "learn lexical": 62080,
      "knowledge data": 57850,
      "data address": 25580,
      "address concerns": 3669,
      "classifier uses": 17591,
      "layers representation": 61805,
      "nonlinear activation": 80221,
      "activation function": 3099,
      "opens new": 82009,
      "new application": 78781,
      "application domains": 6848,
      "domains new": 33824,
      "new possibilities": 79078,
      "learning effectively": 62529,
      "validate model": 130726,
      "task showing": 118709,
      "performance achieved": 86114,
      "incorporating features": 53539,
      "sentence compression": 105797,
      "approach sentence": 7877,
      "formulate task": 45284,
      "task constrained": 118014,
      "constrained optimization": 21572,
      "optimization problem": 82203,
      "problem solve": 91235,
      "integer linear": 55734,
      "linear programming": 64360,
      "programming ilp": 92127,
      "prohibitively slow": 92206,
      "techniques proposed": 119958,
      "alternative solution": 5244,
      "algorithm generates": 4745,
      "relying local": 98762,
      "orders magnitude": 82445,
      "magnitude faster": 66381,
      "producing better": 92015,
      "extensive evaluation": 41876,
      "evaluation demonstrates": 39176,
      "demonstrates quality": 28959,
      "does degrade": 33340,
      "single best": 109708,
      "results empirical": 101749,
      "study sentiment": 114512,
      "sentiment classification": 106692,
      "classification chinese": 17143,
      "using word": 130377,
      "embedding article": 35376,
      "features chinese": 43396,
      "chinese sentiment": 16819,
      "opinion corpus": 82085,
      "corpus built": 23688,
      "review websites": 102588,
      "embeddings represent": 35909,
      "used input": 128593,
      "input different": 55321,
      "different machine": 31241,
      "classification including": 17231,
      "including svm": 53389,
      "svm logistic": 116240,
      "regression convolutional": 97697,
      "network cnn": 77187,
      "ensemble methods": 37600,
      "methods methods": 69619,
      "performance compared": 86228,
      "compared ngram": 19399,
      "using naive": 129927,
      "bayes nb": 12572,
      "outstanding performance": 83204,
      "recall f1": 96336,
      "methods construct": 69400,
      "model testing": 72161,
      "testing corpus": 120589,
      "study focused": 114385,
      "focused automatic": 44849,
      "automatic identification": 10565,
      "british french": 14658,
      "french novels": 45819,
      "century beginning": 15924,
      "extracted using": 42190,
      "using rulebased": 130130,
      "rulebased algorithm": 103357,
      "semantic distance": 105033,
      "lexical diversity": 63759,
      "does mean": 33369,
      "temporal data": 120099,
      "demonstrate important": 28759,
      "languages strong": 60898,
      "strong relationship": 113705,
      "primary language": 90652,
      "language country": 58924,
      "temporal aspect": 120092,
      "data presenting": 26267,
      "level lexical": 63467,
      "given fact": 48034,
      "recent past": 96488,
      "studies published": 114269,
      "present surprising": 89730,
      "cultural linguistic": 25231,
      "linguistic characteristics": 64435,
      "profound consequences": 92108,
      "demonstrate simple": 28867,
      "time series": 122100,
      "solve problems": 110612,
      "argue evaluation": 8916,
      "important context": 52134,
      "context hope": 22124,
      "hope paper": 50633,
      "help researchers": 49777,
      "important use": 52288,
      "use special": 128289,
      "models analysis": 72748,
      "analysis data": 5555,
      "data natural": 26153,
      "temporal ordering": 120114,
      "explicit memory": 41334,
      "memory representations": 68379,
      "representations introduce": 99704,
      "new test": 79216,
      "test language": 120465,
      "models capture": 72877,
      "capture meaning": 15343,
      "unlike standard": 127451,
      "modelling benchmarks": 72600,
      "task predicting": 118542,
      "predicting syntactic": 89014,
      "function words": 46053,
      "words predicting": 134133,
      "greater semantic": 49039,
      "compare range": 19288,
      "range stateoftheart": 95602,
      "different way": 31561,
      "longterm contexts": 65199,
      "outperform stateoftheart": 82749,
      "stateoftheart neural": 112773,
      "predicting semantic": 89009,
      "words interestingly": 134012,
      "encoded single": 36484,
      "memory representation": 68378,
      "single words": 109816,
      "sentences allows": 106205,
      "trained effectively": 123123,
      "qa benchmark": 94496,
      "summaries news": 115451,
      "articles achieve": 9166,
      "performance chinese": 86202,
      "expert using": 41230,
      "analyse performance": 5434,
      "tagging chinese": 117375,
      "use property": 128216,
      "property chinese": 92496,
      "linked data": 64643,
      "ner shared": 77078,
      "task goal": 118246,
      "goal correctly": 48346,
      "tweet dataset": 126011,
      "dataset using": 27262,
      "employ structured": 36282,
      "structured learning": 114010,
      "data unsupervised": 26597,
      "clustering features": 17944,
      "topic drift": 122512,
      "key challenge": 57542,
      "media text": 68168,
      "result competitive": 101366,
      "provide analysis": 93757,
      "analysis components": 5539,
      "target dataset": 117597,
      "context task": 22294,
      "novel method": 80628,
      "method incorporate": 68896,
      "discourse information": 32060,
      "model introduce": 71368,
      "late fusion": 61561,
      "fusion approach": 46231,
      "approach recurrent": 7844,
      "recurrent language": 97177,
      "based long": 11812,
      "memory units": 68392,
      "units lstm": 127256,
      "intersentence dependencies": 56307,
      "separate evaluation": 106851,
      "evaluation corpora": 39161,
      "model improves": 71315,
      "improves perplexity": 53024,
      "perplexity significantly": 87074,
      "varying number": 131266,
      "number context": 80858,
      "context sentences": 22262,
      "sentences observe": 106415,
      "observe proposed": 81208,
      "incorporating additional": 53516,
      "additional inputs": 3529,
      "trained larger": 123184,
      "larger context": 61355,
      "context language": 22157,
      "words including": 133994,
      "increasing number": 53679,
      "sentences analysis": 106207,
      "model capturing": 70813,
      "document better": 32955,
      "grammar lexicon": 48646,
      "based theory": 12120,
      "range natural": 95582,
      "present methodological": 89551,
      "approach extraction": 7564,
      "extraction generation": 42348,
      "lexicon approach": 63873,
      "approach leverages": 7682,
      "corpora automatically": 23416,
      "extract set": 42118,
      "set crosslingual": 107409,
      "patterns based": 85733,
      "based data": 11620,
      "frame semantic": 45386,
      "semantic abstraction": 104970,
      "application programming": 6875,
      "programming interface": 92128,
      "api provided": 6780,
      "resources available": 100950,
      "available open": 11065,
      "characterbased neural": 16495,
      "translation introduce": 124874,
      "introduce neural": 56466,
      "model views": 72308,
      "views input": 131724,
      "sequences characters": 107115,
      "characters words": 16631,
      "wordlevel information": 133736,
      "information provides": 54888,
      "crucial source": 25167,
      "input model": 55369,
      "composes representations": 20078,
      "representations character": 99539,
      "character sequences": 16465,
      "words determined": 133909,
      "translated using": 124578,
      "using joint": 129771,
      "model target": 72140,
      "sequence word": 107099,
      "word generated": 133296,
      "character time": 16472,
      "representation generation": 99247,
      "words performed": 134113,
      "character level": 16440,
      "level model": 63476,
      "model capable": 70802,
      "capable interpreting": 15204,
      "unseen word": 127555,
      "benefit approach": 12960,
      "approach alleviates": 7344,
      "languages model": 60729,
      "model achieve": 70533,
      "results par": 102028,
      "wordbased models": 133699,
      "represent words": 99154,
      "multilingual supervision": 75378,
      "present neural": 89570,
      "network architecture": 77139,
      "architecture based": 8624,
      "based bidirectional": 11549,
      "bidirectional lstms": 13959,
      "words sentential": 134200,
      "sentential contexts": 106556,
      "representations suitable": 99915,
      "variations meaning": 130937,
      "meaning learn": 67646,
      "parameters model": 84770,
      "use crosslingual": 127975,
      "crosslingual supervision": 25018,
      "good representation": 48496,
      "word context": 132967,
      "context sufficient": 22284,
      "selecting correct": 104748,
      "correct translation": 24125,
      "translation second": 125233,
      "language evaluate": 59015,
      "evaluate quality": 38904,
      "quality representations": 94773,
      "low resource": 65384,
      "translation lexical": 124907,
      "lexical substitution": 63824,
      "substitution task": 114929,
      "task obtain": 118470,
      "obtain stateoftheart": 81327,
      "results gaussian": 101806,
      "gaussian mixture": 46545,
      "multiple word": 75746,
      "recently word": 96779,
      "increasingly focused": 53697,
      "representing word": 100068,
      "word semantics": 133515,
      "semantics previous": 105450,
      "suffer problem": 115236,
      "problem polysemy": 91163,
      "problem previous": 91169,
      "models represent": 73927,
      "words multiple": 134066,
      "distributed vectors": 32633,
      "representing words": 100069,
      "embedded space": 35361,
      "space paper": 111037,
      "propose gaussian": 92694,
      "model learn": 71426,
      "embeddings words": 36038,
      "based skipgram": 12050,
      "skipgram framework": 109984,
      "framework word": 45740,
      "represents word": 100077,
      "number senses": 80962,
      "word word": 133658,
      "propose dynamic": 92636,
      "number words": 80996,
      "words training": 134276,
      "training experiments": 123629,
      "experiments benchmarks": 40805,
      "benchmarks effectiveness": 12899,
      "model good": 71259,
      "better best": 13519,
      "embedding context": 35389,
      "combine sentence": 18642,
      "sentence context": 105809,
      "structural features": 113769,
      "features extracted": 43505,
      "extracted dependency": 42149,
      "using variations": 130351,
      "network classifier": 77185,
      "performance used": 86827,
      "input features": 55337,
      "features supervised": 43747,
      "computational semantics": 20431,
      "disambiguation problem": 31982,
      "problem content": 90978,
      "words tend": 134260,
      "words crucial": 133896,
      "units allow": 127231,
      "approach demonstrated": 7477,
      "text segmentation": 121276,
      "segmentation using": 104650,
      "similarity text": 109320,
      "ts aims": 125906,
      "long text": 65137,
      "text coherent": 120796,
      "structure text": 113971,
      "retrieval ir": 102400,
      "document summarisation": 33085,
      "current approaches": 25259,
      "approaches text": 8374,
      "similar use": 109164,
      "metrics measure": 69985,
      "regions text": 97681,
      "text document": 120881,
      "lexical cohesion": 63741,
      "cohesion words": 18333,
      "words various": 134302,
      "various nlp": 131153,
      "semantic web": 105354,
      "systems capture": 116768,
      "associated user": 9613,
      "user needs": 129015,
      "contents text": 21993,
      "segmentation based": 104577,
      "words sufficient": 134248,
      "based ontological": 11913,
      "text blocks": 120688,
      "conceptual relations": 20669,
      "hierarchical agglomerative": 49923,
      "agglomerative clustering": 4333,
      "represent text": 99147,
      "rich structure": 102792,
      "segmentation text": 104642,
      "various levels": 131119,
      "levels granularity": 63536,
      "method evaluated": 68815,
      "evaluated wellknown": 39025,
      "wellknown dataset": 132418,
      "dataset results": 27167,
      "segmentation promising": 104624,
      "enhance proposed": 37484,
      "method combining": 68704,
      "similarity lexical": 109251,
      "lexical similarity": 63819,
      "segmentation quality": 104626,
      "quality multilingual": 94726,
      "multilingual language": 75266,
      "lstmbased model": 65681,
      "reads text": 96046,
      "span annotations": 111117,
      "words characters": 133856,
      "characters analyze": 16594,
      "languages single": 60878,
      "single model": 109758,
      "small vocabulary": 110223,
      "multilingual models": 75285,
      "produce results": 91924,
      "results similar": 102188,
      "similar better": 109074,
      "stateof theart": 112551,
      "tagging named": 117413,
      "recognition use": 97037,
      "use provided": 128219,
      "external data": 41998,
      "learning scratch": 62986,
      "raw text": 95858,
      "language employed": 59000,
      "solve data": 110594,
      "data sparseness": 26478,
      "sparseness problem": 111247,
      "problem machine": 91114,
      "translation especially": 124796,
      "especially data": 38446,
      "data particular": 26222,
      "does exist": 33346,
      "sourcetopivot pivottotarget": 110949,
      "pivottotarget translation": 87594,
      "language errors": 59010,
      "errors models": 38386,
      "combined model": 18659,
      "model suffer": 72109,
      "paper directly": 83874,
      "employ word": 36289,
      "lexical model": 63784,
      "model ibm": 71293,
      "models additional": 72719,
      "additional resource": 3566,
      "pivot phrase": 87582,
      "addition propose": 3464,
      "pruning method": 94168,
      "method takes": 69175,
      "takes account": 117522,
      "experimental result": 40544,
      "result shows": 101403,
      "method significantly": 69133,
      "outperforms conventional": 82872,
      "considers source": 21344,
      "coverage furthermore": 24508,
      "entries lexicon": 38157,
      "lexicon model": 63898,
      "model phrase": 71748,
      "achieved improved": 2648,
      "using english": 129627,
      "english pivot": 37241,
      "language effective": 58992,
      "remains challenge": 98787,
      "modeling semantic": 72538,
      "relatedness target": 97930,
      "target context": 117592,
      "context words": 22329,
      "sentence different": 105822,
      "different context": 31058,
      "words different": 133912,
      "determining sentiment": 30155,
      "polarity sentence": 87925,
      "sentence target": 106099,
      "paper develop": 83864,
      "lstm models": 65644,
      "models target": 74156,
      "target information": 117625,
      "information automatically": 54386,
      "taken account": 117505,
      "benchmark dataset": 12752,
      "dataset twitter": 27250,
      "results modeling": 101952,
      "modeling sentence": 72541,
      "sentence representation": 106028,
      "representation standard": 99419,
      "standard lstm": 112252,
      "lstm does": 65616,
      "incorporating target": 53567,
      "information lstm": 54742,
      "significantly boost": 108894,
      "boost classification": 14430,
      "targetdependent lstm": 117760,
      "models achieve": 72657,
      "stateoftheart performances": 112859,
      "performances using": 86902,
      "parser external": 84955,
      "external sentiment": 42042,
      "sentiment lexicons": 106757,
      "analysis new": 5719,
      "new concept": 78835,
      "concept space": 20589,
      "representation textual": 99439,
      "textual data": 121684,
      "data mined": 26129,
      "novel concept": 80513,
      "model employs": 71064,
      "learning generate": 62612,
      "terms phrases": 120359,
      "phrases documents": 87419,
      "concepts derived": 20611,
      "target corpus": 117595,
      "implicit relations": 52016,
      "relations concepts": 98126,
      "link graph": 64635,
      "graph evaluate": 48805,
      "performance benchmark": 86170,
      "benchmark datasets": 12764,
      "datasets measuring": 27560,
      "sentences empirical": 106291,
      "results competitive": 101611,
      "compared prior": 19424,
      "prior stateoftheart": 90732,
      "study examine": 114372,
      "significance results": 108715,
      "reported different": 99060,
      "methods study": 69772,
      "study shows": 114518,
      "results performing": 102035,
      "performing methods": 86957,
      "statistically insignificant": 113180,
      "methods measuring": 69615,
      "based entity": 11689,
      "entity extraction": 37934,
      "extraction social": 42489,
      "2015 paper": 591,
      "carried jadavpur": 15542,
      "jadavpur university": 57175,
      "university participation": 127354,
      "participation 2015": 85362,
      "2015 task": 597,
      "task entity": 118151,
      "tool developed": 122379,
      "developed task": 30303,
      "task based": 117921,
      "based trigram": 12138,
      "trigram hidden": 125796,
      "utilizes information": 130553,
      "information like": 54729,
      "pos tag": 88213,
      "tag word": 117343,
      "level features": 63445,
      "features enhance": 43484,
      "enhance observation": 37478,
      "observation probabilities": 81166,
      "probabilities known": 90844,
      "known tokens": 58324,
      "tokens unknown": 122346,
      "unknown tokens": 127360,
      "tokens submitted": 122335,
      "runs english": 103474,
      "tested datasets": 120567,
      "datasets released": 27669,
      "joint training": 57327,
      "training bidirectional": 123377,
      "attentionbased neural": 10084,
      "translation attentional": 124653,
      "attentional mechanism": 10047,
      "mechanism proven": 68032,
      "proven effective": 93719,
      "effective improving": 34691,
      "improving endtoend": 53094,
      "endtoend neural": 36938,
      "structural divergence": 113766,
      "unidirectional attentionbased": 127089,
      "attentionbased models": 10082,
      "capture partial": 15356,
      "partial aspects": 85285,
      "aspects attentional": 9373,
      "attentional regularities": 10053,
      "regularities propose": 97739,
      "attentionbased endtoend": 10074,
      "translation instead": 124872,
      "instead training": 55687,
      "training sourcetotarget": 123872,
      "sourcetotarget targettosource": 110953,
      "targettosource translation": 117807,
      "approach encourages": 7526,
      "complementary models": 19749,
      "data experiments": 25924,
      "experiments chineseenglish": 40822,
      "englishfrench translation": 37367,
      "training significantly": 123862,
      "alignment translation": 4997,
      "independent training": 53779,
      "lexicon generation": 63885,
      "generation using": 47692,
      "using graphbased": 129719,
      "graphbased semisupervised": 48912,
      "semisupervised learning": 105603,
      "provide information": 93852,
      "information morphological": 54773,
      "syntactic roles": 116460,
      "roles words": 103240,
      "words language": 134025,
      "available languages": 11025,
      "coverage limited": 24513,
      "limited present": 64264,
      "words automatically": 133831,
      "automatically construct": 10740,
      "wide coverage": 132492,
      "small seed": 110195,
      "100 times": 168,
      "times size": 122187,
      "size high": 109926,
      "11 languages": 219,
      "languages addition": 60390,
      "addition automatically": 3402,
      "automatically created": 10749,
      "created lexicons": 24673,
      "performance downstream": 86316,
      "tasks morphological": 119308,
      "morphological tagging": 74742,
      "tagging dependency": 117383,
      "attentionbased convolutional": 10061,
      "network modeling": 77340,
      "pairs model": 83584,
      "sentences critical": 106269,
      "critical issue": 24814,
      "tasks answer": 118930,
      "answer selection": 6573,
      "paraphrase identification": 84816,
      "entailment te": 37679,
      "work deals": 134454,
      "task finetuning": 118218,
      "finetuning specific": 44514,
      "models sentences": 74003,
      "impact sentence": 51888,
      "sentence iii": 105900,
      "fully manually": 45966,
      "manually designed": 67056,
      "features work": 43791,
      "presents general": 89852,
      "attention based": 9798,
      "based convolutional": 11608,
      "sentences make": 106393,
      "make contributions": 66639,
      "applied wide": 7144,
      "variety tasks": 131021,
      "propose attention": 92557,
      "representation sentence": 99404,
      "takes consideration": 117530,
      "sentence pair": 105964,
      "pair representations": 83450,
      "representations powerful": 99808,
      "sentence representations": 106033,
      "representations iii": 99681,
      "te tasks": 119688,
      "tasks learning": 119250,
      "learning document": 62516,
      "document embeddings": 32999,
      "movie reviews": 74902,
      "loss semantic": 65297,
      "based methods": 11839,
      "methods achieve": 69277,
      "tasks sentiment": 119489,
      "embeddings methods": 35807,
      "proposed capture": 93239,
      "capture semantics": 15387,
      "methods task": 69793,
      "paragraph vector": 84573,
      "document vectors": 33109,
      "predicting words": 89023,
      "features model": 43609,
      "semantics word": 105480,
      "expressive power": 41773,
      "learned vectors": 62283,
      "vectors experimental": 131434,
      "results imdb": 101838,
      "imdb movie": 51832,
      "dataset shows": 27200,
      "shows model": 108598,
      "model outperforms": 71635,
      "outperforms previous": 82932,
      "based models": 11852,
      "models advantages": 72729,
      "robust results": 103068,
      "model combined": 70843,
      "combined models": 18660,
      "models source": 74070,
      "code model": 18116,
      "model published": 71831,
      "published paper": 94365,
      "contrastive entropy": 22727,
      "used metric": 128632,
      "metric evaluating": 69879,
      "evaluating language": 39064,
      "lack correlation": 58696,
      "extrinsic metrics": 42621,
      "like word": 64110,
      "shared vocabulary": 108117,
      "vocabulary model": 131883,
      "model comparison": 70859,
      "new discriminative": 78884,
      "based intrinsic": 11779,
      "intrinsic metric": 56365,
      "traditional word": 122884,
      "models like": 73503,
      "models propose": 73817,
      "discriminatively trained": 32221,
      "trained sentence": 123268,
      "network based": 77154,
      "model rnn": 71935,
      "model demonstrate": 70964,
      "shows strong": 108633,
      "strong correlation": 113661,
      "distortion levels": 32594,
      "considerably outperforms": 21276,
      "outperforms traditional": 83037,
      "multisource neural": 75786,
      "translation build": 124681,
      "model train": 72178,
      "probability target": 90871,
      "neural encoderdecoder": 77904,
      "encoderdecoder framework": 36597,
      "48 bleu": 1051,
      "strong attentionbased": 113622,
      "role context": 103163,
      "context types": 22308,
      "embeddings provide": 35894,
      "provide extensive": 93823,
      "types context": 126263,
      "skipgram word": 109994,
      "affects performance": 4247,
      "performance wide": 86851,
      "intrinsic extrinsic": 56359,
      "extrinsic nlp": 42622,
      "tasks results": 119472,
      "intrinsic tasks": 56370,
      "clear preference": 17729,
      "particular types": 85461,
      "types contexts": 126264,
      "careful tuning": 15513,
      "extrinsic tasks": 42627,
      "tasks benefit": 118963,
      "increasing embedding": 53671,
      "embedding dimensionality": 35394,
      "concatenation word": 20545,
      "learned different": 62211,
      "yield performance": 135346,
      "gains additional": 46382,
      "additional contribution": 3503,
      "learns word": 63246,
      "embeddings weighted": 36020,
      "weighted contexts": 132341,
      "words incorporating": 133997,
      "incorporating structural": 53559,
      "attentional neural": 10050,
      "model neural": 71566,
      "encoderdecoder models": 36604,
      "achieved impressive": 2645,
      "impressive results": 52322,
      "inductive biases": 54048,
      "paper extend": 83936,
      "model include": 71325,
      "biases word": 13889,
      "alignment models": 4973,
      "including positional": 53352,
      "positional bias": 88297,
      "translation directions": 124755,
      "improvements baseline": 52809,
      "attentional model": 10048,
      "model standard": 72080,
      "phrasebased model": 87385,
      "evaluating difficult": 39049,
      "difficult languages": 31640,
      "languages low": 60705,
      "resource setting": 100878,
      "leveraging sentencelevel": 63704,
      "sentencelevel information": 106163,
      "information encoder": 54525,
      "semantic slot": 105299,
      "slot filling": 110037,
      "network rnn": 77412,
      "architectures long": 8817,
      "lstm widely": 65674,
      "used sequence": 128751,
      "sequence labeling": 106966,
      "paper enhance": 83895,
      "labeling explicitly": 58499,
      "explicitly model": 41378,
      "model label": 71399,
      "label dependencies": 58389,
      "dependencies propose": 29108,
      "incorporate global": 53472,
      "global information": 48242,
      "input sequence": 55427,
      "sequence proposed": 107040,
      "lstm encodes": 65618,
      "encodes input": 36674,
      "fixed length": 44598,
      "length vector": 63381,
      "uses encoded": 129227,
      "encoded vector": 36486,
      "initial state": 55225,
      "state lstm": 112504,
      "lstm sequence": 65665,
      "combining methods": 18735,
      "methods predict": 69675,
      "label sequence": 58411,
      "sequence considering": 106925,
      "dependencies information": 29090,
      "information input": 54689,
      "filling task": 44058,
      "task essential": 118153,
      "essential component": 38546,
      "component natural": 19989,
      "understanding using": 127008,
      "standard atis": 112192,
      "corpus achieved": 23641,
      "achieved stateoftheart": 2703,
      "crosslingual transfer": 25027,
      "learning sequence": 63010,
      "sequence model": 107008,
      "algorithms typically": 4885,
      "language training": 60179,
      "data consists": 25781,
      "sequences sentences": 107140,
      "translations languages": 125473,
      "languages labeled": 60663,
      "labeled data": 58429,
      "data word": 26628,
      "alignments training": 5022,
      "data noisy": 26176,
      "instead propose": 55674,
      "use source": 128285,
      "models estimate": 73168,
      "learning new": 62847,
      "pos taggers": 88222,
      "leads significantly": 61959,
      "multisource transfer": 75791,
      "gaussian noise": 46547,
      "online learning": 81779,
      "learning finally": 62582,
      "using empirical": 129624,
      "age gender": 4283,
      "gender region": 46611,
      "linguistic variation": 64580,
      "variation online": 130896,
      "online social": 81803,
      "analysis effects": 5587,
      "speech features": 111687,
      "flemish dutch": 44646,
      "platform present": 87657,
      "shows combining": 108562,
      "essential understanding": 38573,
      "corpus presents": 23937,
      "presents methodology": 89867,
      "systematic study": 116687,
      "study variation": 114546,
      "nonstandard words": 80296,
      "addition approach": 3401,
      "method simultaneously": 69155,
      "simultaneously learn": 109682,
      "learn align": 61986,
      "using monolingual": 129897,
      "data smaller": 26462,
      "use new": 128172,
      "languages aligned": 60396,
      "aligned data": 4911,
      "data properties": 26287,
      "achieve state": 2553,
      "art results": 9086,
      "crosslingual text": 25022,
      "classification word": 17502,
      "word translation": 133607,
      "tasks political": 119383,
      "political speech": 88002,
      "speech generation": 111689,
      "generation report": 47595,
      "report present": 99026,
      "political speeches": 88003,
      "political party": 87998,
      "relies combination": 98648,
      "combination stateoftheart": 18584,
      "stateoftheart nlp": 112789,
      "nlp methods": 79636,
      "report include": 99007,
      "latent dirichlet": 61580,
      "dirichlet allocation": 31922,
      "sequences words": 107148,
      "words generated": 133970,
      "underlying models": 126689,
      "model takes": 72136,
      "takes care": 117528,
      "topic model": 122540,
      "model aims": 70650,
      "dataset contains": 26828,
      "present manual": 89543,
      "manual automated": 66980,
      "automated approach": 10429,
      "approach evaluate": 7540,
      "quality generated": 94672,
      "evaluation generated": 39219,
      "shown high": 108480,
      "assessment tool": 9531,
      "tool automatically": 122372,
      "automatically evaluating": 10765,
      "free online": 45759,
      "designed use": 29630,
      "web portal": 132243,
      "written response": 135144,
      "assessing contextual": 9507,
      "items automatically": 57108,
      "automatically analyzes": 10717,
      "provides detailed": 94018,
      "information scientific": 54957,
      "reasoning model": 96272,
      "model types": 72240,
      "models provided": 73836,
      "detailed analyses": 29752,
      "computational machinery": 20400,
      "complex using": 19892,
      "users need": 129148,
      "need know": 76824,
      "student responses": 114144,
      "use web": 128357,
      "explanations demonstrate": 41293,
      "trained human": 123156,
      "human raters": 50940,
      "teaching large": 119710,
      "large classes": 61046,
      "software architecture": 110514,
      "core concepts": 23318,
      "models important": 73362,
      "important word": 52293,
      "likelihood estimation": 64118,
      "parameters models": 84771,
      "models easily": 73113,
      "easily overfit": 34463,
      "data data": 25815,
      "data sparse": 26477,
      "popular solution": 88131,
      "model lacks": 71402,
      "propose framework": 92686,
      "models framework": 73261,
      "framework allows": 45420,
      "allows developers": 5140,
      "pair word": 83465,
      "strategy according": 113498,
      "according performance": 1994,
      "performance data": 86271,
      "carefully examine": 15525,
      "examine various": 39759,
      "various performance": 131164,
      "performance criteria": 86264,
      "criteria propose": 24796,
      "gives best": 48179,
      "best result": 13432,
      "extracting events": 42211,
      "events text": 39604,
      "events various": 39611,
      "various kinds": 131109,
      "discussed text": 32300,
      "textual documents": 121692,
      "various machine": 131124,
      "applications event": 6923,
      "systems particularly": 117053,
      "summarization medical": 115531,
      "medical domain": 68203,
      "twitter posts": 126117,
      "end paper": 36819,
      "challenges future": 16162,
      "parsing strategies": 85246,
      "based nlp": 11899,
      "systems general": 116915,
      "general architecture": 46634,
      "semantic resources": 105242,
      "mechanism performs": 68026,
      "information leading": 54720,
      "accurate interpretation": 2338,
      "interpretation input": 56262,
      "input text": 55451,
      "language variation": 60321,
      "dependence linguistic": 29056,
      "key task": 57603,
      "task analyzing": 117879,
      "dialectal variation": 30529,
      "approaches important": 8185,
      "models dependence": 73038,
      "types linguistic": 126313,
      "approaches apply": 8063,
      "problems approach": 91296,
      "approach builds": 7405,
      "hilbert space": 50444,
      "using synthetic": 130263,
      "synthetic data": 116615,
      "real datasets": 96062,
      "proposed test": 93568,
      "support robust": 115989,
      "range scenarios": 95599,
      "types data": 126269,
      "data survey": 26525,
      "web technology": 132263,
      "huge volume": 50729,
      "volume data": 131937,
      "data present": 26265,
      "lot data": 65321,
      "data generated": 25975,
      "exchanging ideas": 39949,
      "opinions social": 82133,
      "social networking": 110441,
      "networking sites": 77496,
      "like twitter": 64106,
      "twitter facebook": 126110,
      "rapidly gaining": 95730,
      "gaining popularity": 46378,
      "allow people": 5089,
      "people share": 85890,
      "express views": 41709,
      "discussion different": 32318,
      "lot work": 65333,
      "work field": 134528,
      "field sentiment": 43981,
      "focuses mainly": 44908,
      "analyze information": 5980,
      "information tweets": 55060,
      "cases paper": 15661,
      "survey comparative": 116164,
      "comparative analyses": 19200,
      "analyses existing": 5454,
      "existing techniques": 40311,
      "like machine": 64058,
      "approaches evaluation": 8150,
      "algorithms like": 4865,
      "like naive": 64068,
      "entropy support": 38172,
      "machine provide": 65846,
      "challenges applications": 16128,
      "applications sentiment": 7014,
      "discussed paper": 32298,
      "zipfs law": 135549,
      "task text": 118786,
      "levels text": 63566,
      "words letters": 134032,
      "units meaning": 127257,
      "methods segmentation": 69737,
      "prediction classification": 89040,
      "computational text": 20439,
      "analysis propose": 5778,
      "exhibit better": 39997,
      "generation model": 47477,
      "model words": 72328,
      "words texts": 134265,
      "topically homogeneous": 122600,
      "generative model": 47735,
      "model produces": 71798,
      "study suggests": 114526,
      "language recognition": 60026,
      "recognition evaluation": 96875,
      "describes systems": 29436,
      "technology university": 120047,
      "based ivectors": 11782,
      "extracted different": 42150,
      "features given": 43526,
      "language detection": 58951,
      "detection task": 30073,
      "including support": 53387,
      "vector machines": 131320,
      "machines svm": 66341,
      "linear discriminant": 64345,
      "discriminant analysis": 32190,
      "networks dnn": 77572,
      "whats missing": 132470,
      "vector embedding": 131304,
      "embedding learner": 35427,
      "method generating": 68861,
      "feature embeddings": 43268,
      "embeddings feature": 35720,
      "cooccurrence matrix": 23229,
      "information matrix": 54753,
      "matrix stochastic": 67482,
      "stochastic gradient": 113355,
      "gradient descent": 48600,
      "makes use": 66813,
      "use information": 128094,
      "requires computation": 100253,
      "process thousands": 91576,
      "rows columns": 103330,
      "predicted values": 88972,
      "results accurate": 101490,
      "embeddings achieved": 35541,
      "methods consider": 69393,
      "scale larger": 103734,
      "larger corpora": 61356,
      "sampling methods": 103603,
      "methods simple": 69757,
      "simple search": 109515,
      "search algorithms": 104296,
      "algorithms semantic": 4881,
      "learned language": 62224,
      "use recent": 128226,
      "modeling research": 72534,
      "research focused": 100518,
      "focused semantic": 44875,
      "task informative": 118299,
      "semantic memory": 105110,
      "representations semantic": 99871,
      "question representations": 95212,
      "relations learned": 98200,
      "rich representations": 102777,
      "data simple": 26454,
      "replicate human": 98960,
      "knowledge words": 58237,
      "role modeling": 103203,
      "human behavior": 50763,
      "range semantic": 95600,
      "unstructured documents": 127571,
      "wealth information": 132213,
      "information financial": 54617,
      "text extraction": 120946,
      "identify corresponding": 51489,
      "entities tasks": 37877,
      "ner entity": 77040,
      "entity resolution": 38099,
      "rulebased approach": 103358,
      "approach exploit": 7552,
      "typically represented": 126455,
      "create specialized": 24641,
      "evaluate effectiveness": 38827,
      "ner evaluation": 77042,
      "benefits limitations": 13011,
      "extraction knowledge": 42362,
      "knowledge proposed": 58127,
      "proposed solutions": 93553,
      "dictionaries attempt": 30857,
      "specialized knowledge": 111391,
      "corpus analysis": 23656,
      "analysis prior": 5769,
      "prior linguistic": 90716,
      "consider problem": 21222,
      "problem identifying": 91075,
      "identifying meaningful": 51604,
      "principles completely": 90685,
      "way using": 132141,
      "using kind": 129775,
      "methods identifying": 69538,
      "finding words": 44286,
      "automated way": 10480,
      "considered step": 21305,
      "step automatically": 113252,
      "corpus hope": 23826,
      "long run": 65086,
      "approach turn": 7988,
      "kind sequence": 57683,
      "sequence data": 106927,
      "primarily interested": 90641,
      "applications results": 7010,
      "obtained variety": 81430,
      "engine technology": 37035,
      "technology related": 120046,
      "related fields": 97864,
      "bidirectional neural": 13966,
      "propose train": 93115,
      "train bidirectional": 122905,
      "trained conventional": 123098,
      "conventional maximum": 22885,
      "did outperform": 30917,
      "sequential structure": 107254,
      "multiple levels": 75604,
      "sentences paragraphs": 106432,
      "infer meaning": 54101,
      "text paper": 121162,
      "lstm extension": 65620,
      "network lstm": 77307,
      "model incorporate": 71328,
      "incorporate contextual": 53459,
      "contextual features": 22465,
      "topics model": 122639,
      "tasks word": 119604,
      "prediction sentence": 89122,
      "selection sentence": 104826,
      "sentence topic": 106108,
      "prediction results": 89120,
      "corpora english": 23469,
      "english documents": 37119,
      "documents wikipedia": 33314,
      "google news": 48521,
      "indicate using": 53853,
      "words topics": 134272,
      "topics features": 122628,
      "features improves": 43555,
      "models baseline": 72819,
      "baseline lstm": 12244,
      "tasks example": 119111,
      "example sentence": 39792,
      "selection task": 104838,
      "accuracy improvements": 2182,
      "improvements 21": 52794,
      "wikipedia dataset": 132650,
      "dataset 18": 26710,
      "news dataset": 79327,
      "demonstrates significant": 28961,
      "benefit using": 12996,
      "using context": 129479,
      "language nl": 59718,
      "implications wide": 51998,
      "applications like": 6956,
      "answering sentence": 6695,
      "paraphrase generation": 84814,
      "utterance prediction": 130610,
      "dialog systems": 30594,
      "similarity learning": 109250,
      "similarity methods": 109265,
      "methods focus": 69514,
      "parts input": 85577,
      "semantic meanings": 105109,
      "sentences work": 106548,
      "model account": 70527,
      "similarities dissimilarities": 109179,
      "lexical semantics": 63817,
      "semantics sentences": 105466,
      "sentences model": 106398,
      "matching vector": 67440,
      "vector word": 131400,
      "component based": 19970,
      "cnn model": 18010,
      "model employed": 71062,
      "employed capture": 36296,
      "capture features": 15302,
      "composing similar": 20084,
      "similar dissimilar": 109084,
      "similarity score": 109299,
      "results model": 101932,
      "performance answer": 86138,
      "answer sentence": 6575,
      "task achieves": 117836,
      "achieves comparable": 2756,
      "comparable result": 19169,
      "identification task": 51449,
      "ultradense word": 126561,
      "orthogonal transformation": 82587,
      "generic representations": 47807,
      "representations useful": 99963,
      "useful nlp": 128912,
      "method learns": 68939,
      "embedding space": 35494,
      "focuses information": 44904,
      "information relevant": 54917,
      "relevant task": 98591,
      "embeddings generated": 35731,
      "reach state": 95900,
      "words annotated": 133816,
      "types lexical": 126312,
      "information sentiment": 54968,
      "analysis task": 5884,
      "information lost": 54741,
      "training order": 123766,
      "order magnitude": 82356,
      "magnitude efficient": 66380,
      "generally used": 46879,
      "search results": 104347,
      "results based": 101536,
      "closely match": 17858,
      "match source": 67366,
      "parallel passages": 84675,
      "passages large": 85615,
      "corpus propose": 23942,
      "orthographic variation": 82598,
      "key ideas": 57578,
      "word corpus": 132976,
      "words differ": 133911,
      "differ word": 30936,
      "pairs using": 83661,
      "parallel pairs": 84674,
      "18 million": 419,
      "empirical comparisons": 36156,
      "computer scientists": 20490,
      "apply ideas": 7177,
      "study literature": 114430,
      "use computational": 127949,
      "related phenomena": 97887,
      "study approach": 114324,
      "review use": 102586,
      "use sequence": 128268,
      "common technique": 18934,
      "large corpora": 61060,
      "texts especially": 121507,
      "especially suited": 38516,
      "extended languages": 41806,
      "languages question": 60829,
      "relation extraction": 97978,
      "textual evidence": 121705,
      "evidence existing": 39644,
      "existing knowledgebased": 40150,
      "knowledgebased question": 58254,
      "small annotated": 110134,
      "annotated training": 6247,
      "shallow methods": 107921,
      "methods like": 69597,
      "data scarcity": 26372,
      "representation methods": 99326,
      "like semantic": 64087,
      "answering questions": 6682,
      "questions involving": 95321,
      "involving multiple": 56918,
      "multiple constraints": 75526,
      "method additional": 68617,
      "additional evidence": 3512,
      "evidence wikipedia": 39677,
      "wikipedia present": 132669,
      "based relation": 11990,
      "relation extractor": 98035,
      "retrieve candidate": 102451,
      "candidate answers": 15098,
      "wikipedia validate": 132686,
      "answers experiments": 6728,
      "experiments webquestions": 41203,
      "answering dataset": 6618,
      "dataset method": 27018,
      "achieves f1": 2785,
      "improvement stateoftheart": 52765,
      "approach exploiting": 7553,
      "exploiting multiple": 41475,
      "embeddings sentence": 35928,
      "sentence classification": 105784,
      "classification introduce": 17236,
      "novel simple": 80729,
      "convolution neural": 23142,
      "cnn architecture": 17986,
      "multiple sets": 75692,
      "extracts features": 42578,
      "features input": 43569,
      "input embedding": 55328,
      "layer network": 61730,
      "form final": 45087,
      "final feature": 44104,
      "regularization strategy": 97755,
      "weights associated": 132372,
      "sets model": 107685,
      "model simpler": 72045,
      "alternative architectures": 5225,
      "time furthermore": 122031,
      "require input": 100150,
      "input word": 55472,
      "embeddings dimensionality": 35655,
      "consistently outperforms": 21429,
      "baseline models": 12261,
      "models multidomain": 73589,
      "multidomain neural": 75095,
      "generation spoken": 47630,
      "open domain": 81891,
      "domain difficult": 33508,
      "difficult number": 31648,
      "semantic input": 105083,
      "number domains": 80878,
      "domains important": 33790,
      "important leverage": 52179,
      "leverage existing": 63586,
      "domains facilitate": 33781,
      "adaptation paper": 3244,
      "procedure train": 91402,
      "rnn language": 102918,
      "outofdomain dataset": 82650,
      "dataset fine": 26935,
      "fine tuned": 44327,
      "set indomain": 107465,
      "discriminative objective": 32214,
      "proposed procedure": 93528,
      "achieve competitive": 2444,
      "significantly reducing": 109049,
      "reducing data": 97411,
      "data needed": 26160,
      "needed train": 76878,
      "new unseen": 79234,
      "unseen domains": 127522,
      "greatly improves": 49056,
      "performance small": 86724,
      "semantic nlp": 105126,
      "structure sentences": 113953,
      "structure expressed": 113860,
      "information loss": 54740,
      "directly address": 31859,
      "semantic applications": 104990,
      "representation designed": 99208,
      "associated tool": 9612,
      "tool extracting": 122386,
      "research extract": 100502,
      "entities relations": 37860,
      "specific domain": 111432,
      "domain small": 33641,
      "input user": 55465,
      "instances used": 55639,
      "extract patterns": 42102,
      "patterns output": 85766,
      "output set": 83122,
      "according relation": 1998,
      "relation type": 98079,
      "result obtained": 101388,
      "083 precision": 84,
      "multilingual media": 75278,
      "media analysis": 68078,
      "analysis large": 5671,
      "developments field": 30442,
      "negative opinions": 76938,
      "opinions expressed": 82122,
      "predictions likely": 89180,
      "future developments": 46268,
      "largescale media": 61462,
      "combining information": 18729,
      "information news": 54802,
      "articles written": 9230,
      "written different": 135121,
      "media posts": 68153,
      "media data": 68097,
      "examine data": 39739,
      "according multiple": 1990,
      "study analyzed": 114317,
      "analyzed corpus": 6022,
      "classification propose": 17353,
      "classification combines": 17148,
      "pretrained word": 90212,
      "embeddings ii": 35744,
      "convolution filters": 23138,
      "largescale twitter": 61516,
      "subjectivity classification": 114705,
      "namedentity disambiguation": 76135,
      "disambiguation named": 31973,
      "entity disambiguation": 37922,
      "disambiguation ned": 31977,
      "task linking": 118358,
      "wordsense disambiguation": 134329,
      "classifier built": 17527,
      "target entity": 117618,
      "entity mention": 37969,
      "set training": 107614,
      "training instances": 123665,
      "instances automatically": 55618,
      "derived wikipedia": 29367,
      "wikipedia provide": 132671,
      "provide empirical": 93812,
      "unified model": 127118,
      "model parsing": 71712,
      "sentence understanding": 106114,
      "treestructured neural": 125728,
      "networks exploit": 77596,
      "parse information": 84882,
      "suffer key": 115227,
      "problems make": 91341,
      "tasks usually": 119591,
      "sentences directly": 106280,
      "address issues": 3701,
      "issues introducing": 57054,
      "hybrid model": 51185,
      "model integrating": 71353,
      "parser model": 84971,
      "25 times": 819,
      "treestructured models": 125727,
      "models integrated": 73410,
      "data little": 26094,
      "little loss": 64820,
      "loss accuracy": 65250,
      "accuracy evaluate": 2153,
      "stanford nli": 112374,
      "entailment task": 37677,
      "task significantly": 118715,
      "study use": 114540,
      "selection methods": 104802,
      "different conditions": 31054,
      "features dynamic": 43471,
      "apply techniques": 7223,
      "experiments languages": 40975,
      "models higher": 73327,
      "higher accuracy": 50159,
      "works better": 134929,
      "systems benefit": 116757,
      "used select": 128741,
      "order maximize": 82369,
      "accuracy joint": 2190,
      "final results": 44124,
      "results represent": 102130,
      "art languages": 9053,
      "languages time": 60920,
      "time reducing": 122087,
      "reducing number": 97430,
      "number features": 80891,
      "running time": 103471,
      "prepositional attachment": 89309,
      "attachment disambiguation": 9696,
      "disambiguation using": 31993,
      "using bilingual": 129409,
      "alignments paper": 5016,
      "paper attempt": 83744,
      "attempt solve": 9753,
      "prepositional phrase": 89311,
      "phrase pp": 87362,
      "alignments parallel": 5017,
      "data language": 26069,
      "dual decomposition": 34232,
      "based algorithm": 11501,
      "experiments performed": 41057,
      "performance improved": 86441,
      "baseline baseline": 12194,
      "trained english": 123134,
      "english features": 37135,
      "features random": 43682,
      "random forest": 95490,
      "words derived": 133907,
      "al 2016": 4611,
      "forest algorithm": 45049,
      "corpusbased features": 24076,
      "features evaluate": 43489,
      "10fold cross": 206,
      "cross validation": 24867,
      "adjectives nouns": 3863,
      "classes present": 17071,
      "vector cosine": 131301,
      "classification binary": 17139,
      "following results": 44985,
      "results baseline": 101537,
      "al 2014": 4608,
      "2014 datasets": 575,
      "datasets proving": 27652,
      "learns semantic": 63235,
      "semantic relation": 105202,
      "simply learns": 109625,
      "levy et": 63716,
      "al 2015": 4610,
      "2015 second": 593,
      "negative examples": 76926,
      "drastically reduce": 34103,
      "reduce bias": 97314,
      "13 unsupervised": 291,
      "competitive stateoftheart": 19693,
      "technique used": 119819,
      "domainspecific language": 33906,
      "al 2010": 4601,
      "language generates": 59065,
      "directly translate": 31912,
      "generated approach": 47049,
      "article contains": 9118,
      "contains lot": 21827,
      "lines code": 64406,
      "code using": 18161,
      "using generic": 129705,
      "lstm based": 65597,
      "based conversation": 11605,
      "conversation models": 22962,
      "model incorporates": 71330,
      "incorporates context": 53503,
      "different architectures": 31003,
      "context information": 22139,
      "information long": 54737,
      "lstm language": 65630,
      "function language": 46031,
      "corpus model": 23884,
      "model capture": 70806,
      "capture multiple": 15347,
      "interaction participants": 55960,
      "traditional lstm": 122826,
      "model perplexity": 71745,
      "generated responses": 47115,
      "participant roles": 85308,
      "neural attention": 77842,
      "attention models": 9934,
      "models sequence": 74008,
      "sequence classification": 106923,
      "classification analysis": 17112,
      "key term": 57609,
      "extraction dialogue": 42304,
      "network architectures": 77148,
      "architectures combining": 8793,
      "attention mechanism": 9878,
      "mechanism neural": 68017,
      "attention model": 9930,
      "promising performance": 92285,
      "performance recently": 86659,
      "including speech": 53378,
      "recognition image": 96889,
      "image caption": 51760,
      "caption generation": 15252,
      "generation visual": 47697,
      "visual question": 131802,
      "model applied": 70670,
      "applied sequence": 7115,
      "classification tasks": 17438,
      "tasks dialogue": 119057,
      "detection key": 29977,
      "extraction sequence": 42485,
      "tasks model": 119298,
      "model input": 71347,
      "major difficulty": 66554,
      "irrelevant information": 56949,
      "information sequence": 54970,
      "treated equally": 125556,
      "classification performance": 17330,
      "performance attention": 86153,
      "entire sequence": 37714,
      "task experimental": 118172,
      "results attention": 101526,
      "improvements achieved": 52801,
      "labeling task": 58546,
      "paper domain": 83883,
      "understanding goal": 126855,
      "goal paper": 48377,
      "use multitask": 128157,
      "learning efficiently": 62532,
      "efficiently scale": 35156,
      "filling models": 44048,
      "handle multiple": 49393,
      "multiple target": 75721,
      "target tasks": 117724,
      "reducing training": 97436,
      "task proposed": 118578,
      "proposed multitask": 93503,
      "multitask model": 75875,
      "model delivers": 70963,
      "delivers better": 28622,
      "data leveraging": 26086,
      "learns tasks": 63239,
      "open vocabulary": 81946,
      "allows models": 5175,
      "models generalize": 73274,
      "generalize unseen": 46821,
      "unseen words": 127556,
      "words particularly": 134108,
      "little training": 64830,
      "data used": 26600,
      "used newly": 128660,
      "newly collected": 79265,
      "crowdsourced data": 25107,
      "covering different": 24534,
      "used demonstrate": 128480,
      "techniques characterlevel": 119850,
      "characterlevel neural": 16575,
      "traditional nlp": 122855,
      "recognition asr": 96809,
      "asr machine": 9433,
      "addresses novel": 3801,
      "problems arising": 91301,
      "asr transcripts": 9452,
      "various sources": 131201,
      "essential task": 38567,
      "address problems": 3751,
      "lowdimensional semantic": 65413,
      "representation capabilities": 99184,
      "sequence neural": 107022,
      "models enable": 73144,
      "enable joint": 36357,
      "joint multitask": 57300,
      "learning multilingual": 62812,
      "multilingual neural": 75306,
      "translation morphologically": 124972,
      "replace attention": 98928,
      "characterlevel wordlevel": 16590,
      "clustering problem": 17952,
      "problem tackled": 91258,
      "lowdimensional vectors": 65417,
      "vectors produced": 131465,
      "process results": 91565,
      "results paper": 102023,
      "paper novel": 84054,
      "automatic story": 10674,
      "new tag": 79191,
      "grammar tag": 48660,
      "like tamil": 64101,
      "time dependency": 122014,
      "resolution tag": 100788,
      "rich morphology": 102770,
      "tamil morphological": 117570,
      "knowledge extraction": 57930,
      "knowledge graph": 57954,
      "graph construction": 48773,
      "template based": 120078,
      "parsing analysis": 85070,
      "features verbs": 43782,
      "design hierarchical": 29542,
      "entity knowledge": 37946,
      "extraction finally": 42338,
      "finally experimental": 44182,
      "results prove": 102083,
      "prove method": 93700,
      "using sentencelevel": 130149,
      "growing body": 49165,
      "body research": 14397,
      "research statistical": 100629,
      "models event": 73179,
      "event sequences": 39542,
      "sequences allow": 107113,
      "systems operate": 117034,
      "produced nlp": 91962,
      "nlp pipeline": 79662,
      "systems recent": 117100,
      "recent recurrent": 96504,
      "models directly": 73079,
      "raw tokens": 95865,
      "predict sentences": 88931,
      "sentences finding": 106319,
      "predicting missing": 88995,
      "disfluency detection": 32357,
      "detection using": 30094,
      "using bidirectional": 129397,
      "bidirectional lstm": 13949,
      "bidirectional longshort": 13946,
      "memory neural": 68362,
      "network blstm": 77174,
      "word sequence": 133546,
      "features developed": 43454,
      "takes advantage": 117525,
      "advantage explicit": 4041,
      "states addition": 113037,
      "addition standard": 3476,
      "final output": 44115,
      "incorporate constraints": 53455,
      "structure experiments": 113859,
      "performance standard": 86733,
      "correction detection": 24146,
      "task analysis": 117876,
      "analysis shows": 5849,
      "model better": 70773,
      "tuning statistical": 125941,
      "using bayesian": 129383,
      "bayesian optimization": 12587,
      "bleu scores": 14333,
      "decoding speed": 28133,
      "constraint associated": 21579,
      "report overall": 99020,
      "reduction compared": 97442,
      "random search": 95508,
      "time speed": 122108,
      "small subset": 110213,
      "subset sentences": 114836,
      "multilingual partofspeech": 75331,
      "auxiliary loss": 10878,
      "memory bilstm": 68293,
      "bilstm networks": 14101,
      "networks recently": 77723,
      "successful various": 115173,
      "sequence modeling": 107012,
      "modeling tasks": 72559,
      "tasks little": 119271,
      "little known": 64814,
      "input representations": 55416,
      "representations target": 99927,
      "languages data": 60481,
      "set size": 107583,
      "size label": 109927,
      "label noise": 58403,
      "word character": 132941,
      "embeddings pos": 35866,
      "taggers languages": 117366,
      "data sizes": 26460,
      "bilstm model": 14098,
      "loss function": 65266,
      "rare words": 95749,
      "model obtains": 71603,
      "obtains stateoftheart": 81477,
      "languages works": 60980,
      "works especially": 134944,
      "especially morphologically": 38479,
      "morphologically complex": 74759,
      "complex languages": 19826,
      "languages analysis": 60401,
      "levels previously": 63557,
      "previously assumed": 90588,
      "verb arguments": 131518,
      "arguments using": 8998,
      "using dependencybased": 129562,
      "semeval 2015": 105497,
      "efforts build": 35190,
      "english verbs": 37331,
      "annotation process": 6356,
      "parsing subtasks": 85249,
      "report supervised": 99054,
      "features derived": 43444,
      "trees semantic": 125709,
      "shown approach": 108443,
      "sparsity issues": 111254,
      "characterize dataset": 16549,
      "dataset obtain": 27061,
      "learning longterm": 62696,
      "longterm goal": 65203,
      "learning research": 62963,
      "build intelligent": 14774,
      "intelligent dialog": 55847,
      "dialog agent": 30551,
      "focused learning": 44862,
      "fixed training": 44608,
      "training sets": 123855,
      "data supervision": 26520,
      "supervision word": 115930,
      "parsing tasks": 85260,
      "tasks sentence": 119487,
      "level question": 63496,
      "humans learn": 51086,
      "learned used": 62277,
      "used communication": 128436,
      "work study": 134825,
      "given naturally": 48067,
      "conversation study": 22974,
      "weston et": 132458,
      "largescale question": 61501,
      "evaluate set": 38925,
      "set baseline": 107373,
      "learning strategies": 63059,
      "tasks novel": 119345,
      "model incorporating": 71331,
      "approach learning": 7677,
      "surprising result": 116129,
      "learn answer": 61988,
      "questions correctly": 95292,
      "learner language": 62294,
      "language corpus": 58922,
      "corpus european": 23779,
      "studies present": 114263,
      "corpus swedish": 24030,
      "learner essays": 62293,
      "learners performance": 62299,
      "performance according": 86109,
      "framework reference": 45672,
      "native languages": 76227,
      "learner texts": 62295,
      "texts contain": 121482,
      "text genres": 121026,
      "genres topics": 47839,
      "a1 a2": 1455,
      "annotation automatic": 6280,
      "resource howto": 100847,
      "different attention": 31015,
      "phrase alignments": 87331,
      "work studies": 134821,
      "studies comparatively": 114194,
      "comparatively typical": 19216,
      "typical sentence": 126404,
      "tasks textual": 119559,
      "te answer": 119683,
      "selection observing": 104810,
      "contribute differently": 22749,
      "tasks address": 118923,
      "words single": 134224,
      "single word": 109814,
      "short phrase": 108228,
      "unrelated words": 127505,
      "work limitations": 134612,
      "limitations phrase": 64185,
      "phrase generation": 87349,
      "generation representation": 47596,
      "representation ii": 99253,
      "ii conducts": 51701,
      "conducts alignment": 20967,
      "alignment word": 5002,
      "phrase levels": 87355,
      "levels handcrafted": 63538,
      "features iii": 43544,
      "iii utilizes": 51721,
      "utilizes single": 130561,
      "single attention": 109707,
      "considering characteristics": 21316,
      "characteristics specific": 16533,
      "tasks limits": 119268,
      "effectiveness tasks": 34961,
      "propose architecture": 92553,
      "based gated": 11730,
      "gated recurrent": 46512,
      "recurrent unit": 97264,
      "unit supports": 127211,
      "supports representation": 116039,
      "learning phrases": 62890,
      "arbitrary granularity": 8582,
      "granularity ii": 48747,
      "ii taskspecific": 51716,
      "alignments sentences": 5018,
      "sentences attention": 106220,
      "attention pooling": 9973,
      "results te": 102259,
      "te match": 119686,
      "match observation": 67360,
      "embedding language": 35420,
      "language distance": 58973,
      "measure distance": 67791,
      "embedding called": 35387,
      "called word": 15067,
      "distribution words": 32690,
      "words languages": 134026,
      "measure perform": 67810,
      "perform language": 86020,
      "language dataset": 58935,
      "dataset collection": 26798,
      "variety language": 130984,
      "language families": 59041,
      "use parallel": 128189,
      "languages interestingly": 60647,
      "cases languages": 15652,
      "languages family": 60582,
      "languages perform": 60797,
      "12 different": 252,
      "similarity languages": 109248,
      "languages classification": 60445,
      "genre identification": 47831,
      "dialect identification": 30520,
      "evaluation translations": 39428,
      "translations word": 125508,
      "word ordering": 133386,
      "structure important": 113875,
      "order sentence": 82402,
      "sentence fact": 105872,
      "model simple": 72042,
      "simple heuristic": 109443,
      "strong results": 113707,
      "results task": 102253,
      "task furthermore": 118228,
      "model effective": 71041,
      "basic model": 12529,
      "model outperforming": 71634,
      "outperforming stateoftheart": 82832,
      "syntactic model": 116430,
      "additional data": 3507,
      "data larger": 26076,
      "yield gains": 135336,
      "search time": 104358,
      "time improving": 122036,
      "patent claim": 85662,
      "parsing dataset": 85094,
      "dataset experiments": 26914,
      "performs poorly": 87017,
      "patent claims": 85664,
      "corpora built": 23429,
      "articles web": 9228,
      "utilized train": 130548,
      "method adapting": 68615,
      "adapting existing": 3308,
      "speech tag": 111815,
      "correction proposed": 24165,
      "campaign organized": 15088,
      "public corpus": 94241,
      "corpus train": 24042,
      "train improved": 122941,
      "improved claim": 52598,
      "lessons learned": 63399,
      "use future": 128061,
      "future nlp": 46286,
      "performance improvement": 86442,
      "finally utility": 44247,
      "experiments showing": 41137,
      "showing improved": 108408,
      "utilized generate": 130543,
      "generate features": 46941,
      "features detecting": 43452,
      "context dependence": 22050,
      "exercise item": 39986,
      "item candidates": 57105,
      "selected corpora": 104724,
      "corpora explore": 23485,
      "factors influencing": 42894,
      "dependence single": 29057,
      "single sentences": 109797,
      "sentences larger": 106373,
      "textual context": 121680,
      "context order": 22204,
      "order automatically": 82281,
      "identify candidate": 51481,
      "candidate sentences": 15123,
      "sentences language": 106368,
      "learning exercises": 62561,
      "question previously": 95204,
      "selection candidate": 104772,
      "reducing time": 97435,
      "time required": 122089,
      "present set": 89693,
      "set relevant": 107563,
      "relevant aspects": 98528,
      "implemented rulebased": 51975,
      "algorithm using": 4813,
      "achieved average": 2594,
      "average precision": 11208,
      "related context": 97847,
      "dropped pronoun": 34181,
      "dropped pronouns": 34183,
      "prodrop languages": 91866,
      "like chinese": 64028,
      "chinese japanese": 16776,
      "work mainly": 134624,
      "mainly focused": 66487,
      "propose neural": 92815,
      "model recover": 71870,
      "feature engineering": 43271,
      "process experimental": 91479,
      "datasets experiment": 27465,
      "results chinese": 101570,
      "chinese zero": 16843,
      "zero pronoun": 135475,
      "resolution performance": 100775,
      "performance zp": 86868,
      "books written": 14426,
      "published year": 94377,
      "restricted language": 101340,
      "language interesting": 59217,
      "context existing": 22087,
      "analysis scientific": 5831,
      "production rules": 92060,
      "word distributions": 132997,
      "distributions vocabulary": 32775,
      "translation order": 125076,
      "rich language": 102755,
      "phenomena neural": 87236,
      "use large": 128114,
      "time large": 122042,
      "large memory": 61148,
      "memory usage": 68393,
      "usage paper": 127867,
      "alleviate issue": 5036,
      "issue introducing": 57002,
      "output vocabulary": 83150,
      "vocabulary sentence": 131900,
      "predict target": 88944,
      "target words": 117752,
      "words sentencelevel": 134196,
      "vocabulary reduce": 131896,
      "method simply": 69154,
      "account translation": 2041,
      "translation options": 125075,
      "target vocabulary": 117741,
      "wordtoword translation": 134342,
      "bilingual phrase": 14052,
      "traditional machine": 122827,
      "model experimental": 71127,
      "results largescale": 101885,
      "task method": 118386,
      "achieves better": 2744,
      "better translation": 13750,
      "translation performance": 125106,
      "performance bleu": 86186,
      "bleu point": 14299,
      "vocabulary neural": 131886,
      "web scale": 132251,
      "scale event": 103720,
      "summarization using": 115590,
      "present based": 89389,
      "based sequential": 12036,
      "massive document": 67330,
      "web given": 132236,
      "given event": 48030,
      "short text": 108255,
      "time unlike": 122129,
      "unlike previous": 127439,
      "able jointly": 1660,
      "jointly model": 57363,
      "model relevance": 71895,
      "f1 metrics": 42663,
      "taskspecific word": 119646,
      "learning use": 63146,
      "learning optimizing": 62864,
      "learned representations": 62250,
      "ranking function": 95668,
      "learned weight": 62285,
      "weight vector": 132331,
      "engineered feature": 37040,
      "different aspects": 31011,
      "corpus learning": 23860,
      "performance variety": 86838,
      "variety downstream": 130973,
      "corpus order": 23919,
      "inputs propose": 55491,
      "analysis learning": 5677,
      "certain patterns": 15951,
      "patterns training": 85787,
      "data test": 26552,
      "proven difficult": 93718,
      "difficult impossible": 31633,
      "limited datasets": 64228,
      "datasets contrast": 27383,
      "contrast human": 22694,
      "similar tasks": 109156,
      "tasks provide": 119421,
      "provide framework": 93835,
      "fail generalizing": 42962,
      "use framework": 128058,
      "inputs demonstrate": 55481,
      "demonstrate results": 28860,
      "feedforward neural": 43850,
      "producing large": 92021,
      "texts written": 121654,
      "computer processing": 20486,
      "memory natural": 68351,
      "human brains": 50772,
      "human communication": 50779,
      "particular paper": 85437,
      "paper consider": 83782,
      "model semantics": 71969,
      "based frequency": 11728,
      "formulated terms": 45287,
      "models semantic": 73989,
      "trees improve": 125694,
      "improve syntactic": 52558,
      "syntactic parses": 116443,
      "crucial nlp": 25156,
      "nlp pipelines": 79663,
      "worlds languages": 135059,
      "languages large": 60674,
      "corpora available": 23418,
      "available building": 10957,
      "parsers syntactic": 85048,
      "issue using": 57032,
      "corpora consisting": 23440,
      "resourcepoor resourcerich": 100916,
      "resourcerich language": 100923,
      "taking advantage": 117541,
      "languages project": 60817,
      "resourcepoor language": 100907,
      "methods suffer": 69779,
      "investigate possibility": 56795,
      "possibility using": 88372,
      "parallel annotated": 84585,
      "automatically detect": 10753,
      "patterns languages": 85752,
      "data necessary": 26158,
      "fullysupervised methods": 45989,
      "methods detection": 69440,
      "detection process": 30027,
      "demonstrate common": 28690,
      "common patterns": 18906,
      "automatically prior": 10811,
      "algorithms machine": 4866,
      "resources methods": 101013,
      "survey introduce": 116174,
      "mt evaluation": 74964,
      "contains manual": 21828,
      "methods traditional": 69808,
      "human evaluation": 50801,
      "evaluation criteria": 39164,
      "fluency adequacy": 44693,
      "human assessments": 50754,
      "methods categories": 69365,
      "categories including": 15740,
      "including lexical": 53310,
      "recall fmeasure": 96339,
      "features semantic": 43715,
      "features include": 43557,
      "include speech": 53234,
      "types sentence": 126356,
      "include named": 53230,
      "models deep": 73021,
      "models evaluation": 73177,
      "newly proposed": 79276,
      "introduce evaluation": 56421,
      "evaluation including": 39241,
      "including different": 53285,
      "correlation scores": 24244,
      "quality estimation": 94647,
      "estimation qe": 38662,
      "qe tasks": 94533,
      "existing works": 40340,
      "recent development": 96450,
      "different classifications": 31041,
      "hope work": 50646,
      "mt researchers": 74991,
      "researchers easily": 100688,
      "best suitable": 13459,
      "mt model": 74972,
      "model development": 70987,
      "evaluation research": 39362,
      "research developed": 100465,
      "hopefully work": 50650,
      "evaluation tasks": 39416,
      "tasks translation": 119575,
      "translation nlp": 125019,
      "nlp fields": 79616,
      "guided neural": 49256,
      "translation investigate": 124877,
      "use hierarchical": 128085,
      "hierarchical phrasebased": 49990,
      "phrasebased smt": 87389,
      "translation nmt": 125020,
      "complete translation": 19773,
      "translation hypotheses": 124853,
      "model score": 71950,
      "beamsearch decoder": 12613,
      "nmt decoding": 79853,
      "practical advantages": 88698,
      "extending nmt": 41818,
      "nmt large": 79883,
      "large input": 61118,
      "network text": 77453,
      "classification multitask": 17294,
      "learning neural": 62837,
      "methods obtained": 69648,
      "great progress": 49014,
      "variety natural": 130999,
      "tasks previous": 119396,
      "models learned": 73491,
      "insufficient training": 55727,
      "framework jointly": 45588,
      "jointly learn": 57347,
      "learn multiple": 62096,
      "multiple related": 75662,
      "related tasks": 97902,
      "tasks based": 118953,
      "based recurrent": 11977,
      "network propose": 77397,
      "propose different": 92624,
      "different mechanisms": 31251,
      "sharing information": 108131,
      "information model": 54768,
      "model text": 72162,
      "entire network": 37710,
      "network trained": 77458,
      "trained jointly": 123164,
      "jointly tasks": 57391,
      "tasks experiments": 119121,
      "experiments benchmark": 40791,
      "benchmark text": 12868,
      "tasks proposed": 119411,
      "proposed models": 93489,
      "models improve": 73365,
      "performance task": 86785,
      "task help": 118254,
      "help related": 49773,
      "siamese convolutional": 108650,
      "networks based": 77517,
      "based phonetic": 11926,
      "phonetic features": 87292,
      "features cognate": 43403,
      "cognate identification": 18258,
      "identification paper": 51415,
      "explore use": 41608,
      "use convolutional": 127967,
      "purpose cognate": 94424,
      "binary classifiers": 14124,
      "classifiers based": 17601,
      "based string": 12077,
      "string similarity": 113599,
      "families experiments": 43053,
      "networks achieve": 77501,
      "competitive results": 19679,
      "resources learning": 100996,
      "learning entity": 62551,
      "entity embeddings": 37930,
      "data recent": 26323,
      "vectorspace embeddings": 131498,
      "relational information": 98092,
      "information derived": 54473,
      "derived large": 29352,
      "large text": 61290,
      "corpora propose": 23561,
      "descriptions entities": 29480,
      "semantics order": 105442,
      "order derive": 82307,
      "models applying": 72761,
      "model results": 71923,
      "new stateoftheart": 79154,
      "wordnet dataset": 133769,
      "previous best": 90391,
      "faster convergence": 43168,
      "entity representations": 38097,
      "representations tradeoff": 99937,
      "regarding performance": 97658,
      "performance improvements": 86446,
      "modelling interaction": 72606,
      "interactions sentences": 56001,
      "sentences deep": 106272,
      "networks existing": 77592,
      "methods encode": 69471,
      "separate encoders": 106850,
      "sentence encoded": 105848,
      "little information": 64812,
      "information sentence": 54965,
      "sentence paper": 105973,
      "propose deep": 92615,
      "architecture model": 8694,
      "model strong": 72089,
      "specifically introduce": 111559,
      "ways model": 132166,
      "use dynamic": 128020,
      "select informative": 104702,
      "informative features": 55139,
      "experiments large": 40977,
      "large datasets": 61074,
      "datasets demonstrate": 27394,
      "demonstrate efficacy": 28727,
      "efficacy proposed": 35014,
      "proposed architecture": 93225,
      "stateoftheart automatic": 112590,
      "grammatical error": 48696,
      "parameter tuning": 84726,
      "m2 metric": 65732,
      "standard metric": 112262,
      "automatic grammar": 10549,
      "grammar error": 48632,
      "correction gec": 24151,
      "gec tasks": 46576,
      "framework investigate": 45585,
      "sparse features": 111226,
      "sparse feature": 111225,
      "feature weights": 43330,
      "smt setup": 110295,
      "outperforms previously": 82960,
      "previously published": 90614,
      "published results": 94370,
      "results conll2014": 101625,
      "conll2014 test": 21106,
      "set large": 107475,
      "large margin": 61138,
      "smt neural": 110287,
      "neural features": 77916,
      "features trained": 43764,
      "trained publicly": 123252,
      "available data": 10970,
      "data newly": 26170,
      "improve stateoftheart": 52549,
      "discourse corpora": 32049,
      "corpora dialogue": 23454,
      "dialogue translation": 30807,
      "proposed automatically": 93233,
      "construct parallel": 21646,
      "discourse corpus": 32050,
      "dialogue machine": 30704,
      "translation firstly": 124826,
      "subtitle data": 114990,
      "corresponding monolingual": 24295,
      "movie script": 74904,
      "data crawled": 25804,
      "speaker discourse": 111287,
      "data information": 26040,
      "retrieval approach": 102381,
      "approach order": 7759,
      "order map": 82367,
      "map monolingual": 67108,
      "bilingual texts": 14062,
      "texts evaluate": 121508,
      "evaluate mapping": 38851,
      "speaker information": 111291,
      "experiments proposed": 41067,
      "method achieve": 68583,
      "model adaptation": 70624,
      "05 bleu": 32,
      "points improvement": 87860,
      "improvement translation": 52779,
      "publicly release": 94329,
      "discourse data": 32051,
      "data manual": 26117,
      "integrating distributional": 55790,
      "contrast word": 22719,
      "novel vector": 80771,
      "salient features": 103550,
      "similarity improved": 109240,
      "significantly outperform": 108979,
      "outperform standard": 82746,
      "standard models": 112267,
      "models distinguish": 73097,
      "classes adjectives": 17058,
      "integrate lexical": 55760,
      "model novel": 71594,
      "novel embedding": 80555,
      "embedding outperforms": 35466,
      "predicting word": 89022,
      "variational neural": 130931,
      "translation discriminative": 124758,
      "distribution target": 32682,
      "target sentence": 117697,
      "given source": 48132,
      "propose variational": 93151,
      "variational model": 130929,
      "distribution neural": 32670,
      "variational encoderdecoder": 130922,
      "encoderdecoder model": 36599,
      "model generates": 71237,
      "generates target": 47182,
      "target translations": 117736,
      "hidden representations": 49906,
      "representations source": 99897,
      "source sentences": 110820,
      "model introduces": 71371,
      "introduces continuous": 56606,
      "continuous latent": 22620,
      "variable explicitly": 130838,
      "model underlying": 72242,
      "guide generation": 49238,
      "generation target": 47654,
      "translations order": 125482,
      "order perform": 82382,
      "perform efficient": 85986,
      "posterior inference": 88483,
      "largescale training": 61513,
      "build neural": 14791,
      "target sides": 117708,
      "variational lower": 130927,
      "lower bound": 65422,
      "german translation": 47920,
      "proposed variational": 93587,
      "translation achieves": 124622,
      "improvements vanilla": 52936,
      "vanilla neural": 130816,
      "translation baselines": 124668,
      "recursive autoencoders": 97283,
      "learning bilingual": 62411,
      "phrase embeddings": 87346,
      "embeddings paper": 35849,
      "based recursive": 11985,
      "recursive autoencoder": 97282,
      "interactions multiple": 55992,
      "phrase representations": 87367,
      "autoencoders generate": 10416,
      "generate tree": 47039,
      "tree structures": 125615,
      "embeddings different": 35648,
      "granularity words": 48753,
      "embeddings source": 35945,
      "attention network": 9944,
      "network learn": 77303,
      "learn interactions": 62071,
      "soft attention": 110485,
      "attention weight": 10036,
      "based learned": 11798,
      "representations use": 99959,
      "neural model": 78143,
      "incorporate semantic": 53485,
      "additional feature": 3514,
      "stateoftheart smt": 112956,
      "experiments nist": 41041,
      "nist chineseenglish": 79496,
      "chineseenglish test": 16858,
      "test sets": 120526,
      "achieves substantial": 2911,
      "163 bleu": 385,
      "points average": 87848,
      "average baseline": 11182,
      "scale using": 103762,
      "given test": 48152,
      "set equal": 107431,
      "generation nlp": 47519,
      "characteristics individual": 16526,
      "human intelligence": 50867,
      "nlp task": 79700,
      "paper demonstrate": 83800,
      "collecting large": 18449,
      "human responses": 50957,
      "model compares": 70854,
      "systems performance": 117062,
      "performance human": 86432,
      "able provide": 1676,
      "insight performance": 55524,
      "standard evaluation": 112236,
      "accuracy score": 2267,
      "score does": 104066,
      "reverse dictionary": 102525,
      "dictionary based": 30873,
      "word definitions": 132982,
      "architecture paper": 8722,
      "paper outline": 84061,
      "approach build": 7403,
      "input outputs": 55385,
      "words semantically": 134188,
      "problem use": 91280,
      "graph assess": 48758,
      "word input": 133322,
      "approach better": 7391,
      "approach sets": 7886,
      "sets new": 107686,
      "new performance": 79073,
      "performance baseline": 86166,
      "semeval2016 task": 105527,
      "task 12": 117815,
      "vs joint": 131965,
      "joint inference": 57280,
      "inference clinical": 54122,
      "temporal information": 120107,
      "identified text": 51463,
      "text spans": 121317,
      "time event": 122026,
      "clinical notes": 17781,
      "predicted relation": 88966,
      "time temporal": 122121,
      "extraction joint": 42360,
      "using structured": 130232,
      "structured prediction": 114024,
      "outperforms vanilla": 83043,
      "vanilla recurrent": 130818,
      "network incorporates": 77283,
      "incorporates word": 53514,
      "embeddings trained": 35983,
      "large clinical": 61048,
      "document sets": 33079,
      "sets document": 107664,
      "supervision rules": 115913,
      "rules predicting": 103431,
      "predicting relations": 89005,
      "events time": 39606,
      "time expressions": 122029,
      "expressions improves": 41754,
      "improves classification": 52960,
      "scale training": 103757,
      "data deep": 25819,
      "deep reinforcement": 28409,
      "reinforcement learning": 97791,
      "learning dialogue": 62500,
      "dialogue generation": 30673,
      "generation recent": 47582,
      "recent neural": 96475,
      "models dialogue": 73064,
      "great promise": 49018,
      "generating responses": 47257,
      "conversational agents": 22987,
      "utterances time": 130670,
      "influence future": 54310,
      "future direction": 46269,
      "nlp models": 79638,
      "paper integrate": 83985,
      "applying deep": 7235,
      "model future": 71216,
      "future reward": 46313,
      "dialogue model": 30709,
      "using policy": 130027,
      "policy gradient": 87950,
      "gradient methods": 48603,
      "answering related": 6689,
      "evaluate model": 38861,
      "model diversity": 71015,
      "showing proposed": 108427,
      "proposed algorithm": 93172,
      "conversation dialogue": 22948,
      "work marks": 134630,
      "models opendomain": 73664,
      "opendomain discourse": 81971,
      "coherence discourse": 18295,
      "strongly associated": 113742,
      "text quality": 121224,
      "quality making": 94714,
      "making important": 66845,
      "understanding existing": 126840,
      "existing models": 40203,
      "lexical overlap": 63789,
      "rhetorical structure": 102714,
      "structure entity": 113853,
      "narrow domains": 76190,
      "models discourse": 73081,
      "multiple aspects": 75496,
      "coherence existing": 18298,
      "generating new": 47236,
      "new sentences": 79127,
      "sentences study": 106506,
      "study discriminative": 114359,
      "discriminative models": 32211,
      "learn distinguish": 62027,
      "generative models": 47741,
      "models produce": 73807,
      "produce coherent": 91879,
      "text including": 121048,
      "including novel": 53342,
      "novel neural": 80664,
      "discourse dependencies": 32053,
      "dependencies sentences": 29113,
      "sentences text": 106520,
      "text work": 121413,
      "work achieves": 134351,
      "performance multiple": 86552,
      "initial step": 55226,
      "coherent texts": 18329,
      "pseudo training": 94186,
      "pronoun resolution": 92341,
      "resolution existing": 100761,
      "heavily relying": 49660,
      "relying annotated": 98755,
      "data released": 26334,
      "task organizers": 118482,
      "lack annotated": 58678,
      "data major": 26109,
      "major obstacle": 66575,
      "obstacle progress": 81248,
      "labeling data": 58492,
      "performance alleviate": 86133,
      "simple novel": 109487,
      "furthermore successfully": 46215,
      "clozestyle reading": 17902,
      "reading comprehension": 95973,
      "comprehension neural": 20206,
      "task propose": 118573,
      "propose twostep": 93126,
      "twostep training": 126175,
      "training mechanism": 123704,
      "data real": 26321,
      "approach significantly": 7897,
      "stateoftheart systems": 112985,
      "absolute improvements": 1740,
      "ontonotes 50": 81858,
      "50 data": 1073,
      "deep averaging": 28200,
      "crosslingual sentiment": 25011,
      "classification recent": 17369,
      "great success": 49028,
      "success achieved": 115062,
      "thanks availability": 121769,
      "annotated resources": 6224,
      "data tackle": 26535,
      "problem lowresource": 91113,
      "data propose": 26288,
      "propose adversarial": 92534,
      "averaging network": 11227,
      "transfer knowledge": 124064,
      "knowledge learned": 58046,
      "resourcerich source": 100937,
      "language lowresource": 59260,
      "unlabeled data": 127378,
      "data exists": 25916,
      "sentiment classifier": 106722,
      "adversarial language": 4129,
      "language discriminator": 58971,
      "shared feature": 107980,
      "feature extractor": 43282,
      "languages experiments": 60571,
      "arabic sentiment": 8551,
      "classification demonstrate": 17177,
      "systems neural": 117018,
      "translation external": 124821,
      "memory paper": 68364,
      "memory stores": 68386,
      "human experts": 50856,
      "tagging information": 117393,
      "information representation": 54920,
      "representation source": 99413,
      "strategy generate": 113518,
      "generate sequence": 47012,
      "sequence multiple": 107020,
      "multiple words": 75749,
      "approaches step": 8354,
      "incorporating external": 53537,
      "external knowledge": 42006,
      "knowledge neural": 58081,
      "generation mechanism": 47470,
      "translation shows": 125255,
      "bleu improvement": 14286,
      "generic neural": 47803,
      "discrete translation": 32180,
      "translation lexicons": 124910,
      "lexicons neural": 63931,
      "makes mistakes": 66794,
      "understanding meaning": 126885,
      "meaning sentence": 67687,
      "sentence propose": 106007,
      "method alleviate": 68628,
      "problem augmenting": 90944,
      "nmt systems": 79972,
      "efficiently encode": 35141,
      "lowfrequency words": 65470,
      "probability word": 90873,
      "using attention": 129351,
      "attention vector": 10035,
      "nmt model": 79894,
      "model select": 71957,
      "source word": 110864,
      "probabilities model": 90849,
      "model focus": 71197,
      "methods combine": 69376,
      "standard nmt": 112286,
      "linear interpolation": 64348,
      "experiments corpora": 40857,
      "convergence time": 22926,
      "time neural": 122056,
      "translation simultaneous": 125261,
      "simultaneous translation": 109665,
      "investigate potential": 56798,
      "novel decoding": 80533,
      "decoding algorithm": 28070,
      "algorithm called": 4702,
      "greedy decoding": 49069,
      "existing neural": 40224,
      "translating source": 124605,
      "approach unique": 7991,
      "translation jointly": 124881,
      "step building": 113254,
      "translation optimizing": 125074,
      "learning parsing": 62879,
      "number latent": 80906,
      "contrary common": 22677,
      "common belief": 18859,
      "methods parsing": 69666,
      "results significantly": 102182,
      "globally optimized": 48290,
      "interactions different": 55983,
      "empirical analysis": 36149,
      "consistently performs": 21437,
      "semi supervised": 105565,
      "definition extraction": 28522,
      "tool present": 122399,
      "use semi": 128261,
      "designed extract": 29590,
      "extract target": 42127,
      "corpus textual": 24041,
      "textual fragments": 121708,
      "conditional random": 20774,
      "random fields": 95483,
      "sequential labeling": 107231,
      "labeling algorithm": 58484,
      "bootstrapping approach": 14457,
      "enables model": 36392,
      "paper main": 84035,
      "automatic manual": 10584,
      "source necessary": 110799,
      "immediate use": 51845,
      "arabic neural": 8536,
      "research neural": 100568,
      "translation focused": 124827,
      "despite language": 29701,
      "language agnostic": 58822,
      "nature paper": 76666,
      "apply neural": 7196,
      "task arabic": 117890,
      "arabic translation": 8562,
      "compare standard": 19295,
      "phrasebased translation": 87399,
      "extensive comparison": 41866,
      "various configurations": 131065,
      "arabic script": 8547,
      "phrasebased neural": 87388,
      "perform comparably": 85961,
      "translation significantly": 125257,
      "outofdomain test": 82662,
      "making attractive": 66822,
      "wmt 16": 132768,
      "wmt 2016": 132771,
      "2016 shared": 617,
      "shared news": 107994,
      "news translation": 79385,
      "task building": 117948,
      "building neural": 14871,
      "systems language": 116966,
      "pairs trained": 83650,
      "englishczech englishgerman": 37359,
      "englishromanian englishrussian": 37424,
      "based attentional": 11534,
      "attentional encoderdecoder": 10045,
      "bpe subword": 14541,
      "subword segmentation": 115032,
      "openvocabulary translation": 82044,
      "fixed vocabulary": 44609,
      "experimented using": 40747,
      "using automatic": 129366,
      "backtranslations monolingual": 11345,
      "news corpus": 79324,
      "corpus additional": 23646,
      "models reported": 73926,
      "reported methods": 99065,
      "substantial improvements": 114859,
      "baseline systems": 12319,
      "systems human": 116937,
      "improve neural": 52437,
      "translation recently": 125203,
      "recently achieved": 96648,
      "using little": 129822,
      "way external": 132084,
      "external linguistic": 42030,
      "learning capability": 62418,
      "capability neural": 15184,
      "mt models": 74973,
      "models does": 73103,
      "does make": 33366,
      "features redundant": 43694,
      "easily incorporated": 34457,
      "provide improvements": 93849,
      "improvements performance": 52895,
      "performance generalize": 86414,
      "embedding layer": 35424,
      "layer encoder": 61711,
      "encoderdecoder architecture": 36591,
      "arbitrary features": 8581,
      "features addition": 43355,
      "word feature": 133283,
      "features partofspeech": 43649,
      "tags syntactic": 117484,
      "dependency labels": 29151,
      "labels input": 58613,
      "wmt16 training": 132836,
      "improve model": 52415,
      "quality according": 94584,
      "according metrics": 1989,
      "perplexity bleu": 87064,
      "opensource implementation": 82019,
      "implementation neural": 51948,
      "mt available": 74958,
      "external lexical": 42026,
      "information multilingual": 54775,
      "lexicons word": 63936,
      "useful improving": 128894,
      "statistical partofspeech": 113150,
      "partofspeech taggers": 85544,
      "compare performances": 19281,
      "systems datasets": 116831,
      "datasets covering": 27385,
      "covering 16": 24531,
      "languages systems": 60910,
      "approaches perform": 8279,
      "perform similarly": 86072,
      "reach stateoftheart": 95902,
      "better performances": 13666,
      "performances obtained": 86890,
      "featurebased models": 43337,
      "datasets morphologically": 27581,
      "results higher": 101823,
      "lexical variability": 63838,
      "designed features": 29592,
      "certain conditions": 15936,
      "competitive respect": 19676,
      "neural methods": 78141,
      "text representation": 121245,
      "learning propose": 62921,
      "new active": 78763,
      "active learning": 3109,
      "learning al": 62317,
      "al method": 4654,
      "classification convolutional": 17159,
      "networks cnns": 77535,
      "manually labeled": 67066,
      "model performance": 71722,
      "performance minimal": 86532,
      "minimal effort": 70158,
      "effort neural": 35178,
      "task hand": 118250,
      "al strategies": 4656,
      "models focus": 73258,
      "focus selecting": 44812,
      "selecting instances": 104752,
      "representations contrast": 99572,
      "contrast traditional": 22715,
      "uncertainty sampling": 126626,
      "containing words": 21812,
      "words embeddings": 133931,
      "learning discriminative": 62507,
      "taskspecific embeddings": 119624,
      "embeddings extend": 35715,
      "extend approach": 41780,
      "approach document": 7500,
      "classification jointly": 17240,
      "jointly considering": 57337,
      "constituent word": 21551,
      "representations models": 99757,
      "instances likely": 55629,
      "improve representations": 52523,
      "embeddings best": 35574,
      "knowledge work": 58238,
      "models text": 74182,
      "agreementbased learning": 4410,
      "nonparallel corpora": 80264,
      "corpora introduce": 23504,
      "corpora basic": 23423,
      "em algorithm": 35334,
      "algorithm jointly": 4760,
      "jointly training": 57395,
      "unidirectional models": 127096,
      "models efficiently": 73122,
      "chineseenglish dataset": 16847,
      "learning significantly": 63026,
      "representations present": 99813,
      "continuous bag": 22612,
      "bag words": 11360,
      "cbow model": 15848,
      "efficient estimation": 35072,
      "sentence embeddings": 105838,
      "proven surprisingly": 93733,
      "surprisingly successful": 116149,
      "efficient way": 35133,
      "obtaining sentence": 81447,
      "embeddings word": 36026,
      "optimized task": 82235,
      "task sentence": 118688,
      "handles problem": 49410,
      "problem training": 91267,
      "embeddings directly": 35657,
      "underlying neural": 126692,
      "network learns": 77305,
      "surrounding sentences": 116156,
      "model evaluating": 71113,
      "20 datasets": 499,
      "variety sources": 131016,
      "encoder decoder": 36504,
      "decoder architecture": 27994,
      "pivot based": 87579,
      "based sequence": 12023,
      "sequence generation": 106948,
      "generation interlingua": 47444,
      "based machine": 11821,
      "encode multiple": 36440,
      "multiple languages": 75591,
      "languages common": 60451,
      "common linguistic": 18890,
      "representation decode": 99203,
      "decode sentences": 27985,
      "sentences multiple": 106403,
      "representation work": 99473,
      "work explore": 134506,
      "explore idea": 41548,
      "context neural": 22196,
      "neural encoder": 77903,
      "decoder architectures": 27995,
      "end goal": 36810,
      "specifically consider": 111533,
      "consider case": 21191,
      "generating sequences": 47262,
      "available parallel": 11071,
      "available training": 11131,
      "available case": 10958,
      "solution elegant": 110553,
      "stage model": 112149,
      "model converts": 70920,
      "instead explore": 55662,
      "solution jointly": 110558,
      "jointly learns": 57358,
      "representation evaluate": 99228,
      "model tasks": 72147,
      "ii bridge": 51698,
      "report promising": 99029,
      "results applications": 101510,
      "right step": 102836,
      "response suggestion": 101246,
      "email paper": 35338,
      "investigate novel": 56791,
      "novel endtoend": 80556,
      "endtoend method": 36921,
      "generating short": 47263,
      "designed work": 29632,
      "largescale deep": 61425,
      "challenges faced": 16159,
      "response diversity": 101196,
      "usergenerated content": 129070,
      "university cambridge": 127348,
      "englishgerman wmt16": 37394,
      "submission wmt16": 114739,
      "complementary nature": 19750,
      "nmt different": 79854,
      "simple neural": 109481,
      "rescoring approach": 100389,
      "nmt search": 79958,
      "modified version": 74450,
      "especially using": 38526,
      "using multiple": 129919,
      "systems ensemble": 116871,
      "unsupervised rulebased": 127708,
      "approach multilingual": 7718,
      "main advantages": 66394,
      "advantages unsupervised": 4089,
      "unsupervised algorithms": 127600,
      "supervised models": 115807,
      "models behave": 72824,
      "black box": 14245,
      "robustness different": 103092,
      "introducing concept": 56631,
      "information form": 54623,
      "form universal": 45139,
      "universal dependencies": 127283,
      "data structured": 26510,
      "differently depending": 31594,
      "concerned experiments": 20686,
      "improvement existing": 52706,
      "existing unsupervised": 40327,
      "methods stateoftheart": 69770,
      "stateoftheart supervised": 112976,
      "models evaluating": 73176,
      "shared languages": 107987,
      "timeline generation": 122155,
      "generation complex": 47337,
      "complex events": 19811,
      "consider information": 21208,
      "information traditional": 55049,
      "traditional media": 122834,
      "ignoring rich": 51692,
      "social context": 110329,
      "context provided": 22238,
      "instead aim": 55652,
      "aim generate": 4458,
      "news article": 79285,
      "article summaries": 9162,
      "selected user": 104738,
      "user comments": 128971,
      "optimization framework": 82193,
      "framework designed": 45489,
      "coverage event": 24505,
      "automatic evaluations": 10541,
      "realworld datasets": 96158,
      "datasets cover": 27384,
      "produces informative": 91999,
      "summaries furthermore": 115440,
      "analysis approach": 5498,
      "approach online": 7755,
      "detection investigate": 29976,
      "novel task": 80743,
      "task online": 118477,
      "detection propose": 30028,
      "propose sentiment": 93038,
      "analysis solution": 5863,
      "solution problem": 110563,
      "problem aim": 90924,
      "aim identify": 4460,
      "use features": 128052,
      "features classifier": 43398,
      "classifier predicts": 17573,
      "discussion evaluate": 32319,
      "detection approaches": 29881,
      "newly created": 79267,
      "created corpus": 24659,
      "corpus wikipedia": 24066,
      "wikipedia talk": 132679,
      "talk page": 117558,
      "sentiment tagging": 106800,
      "features outperform": 43641,
      "outperform best": 82695,
      "best model": 13367,
      "achieves promising": 2836,
      "accuracy 080": 2061,
      "agreement disagreement": 4389,
      "online discussions": 81770,
      "sentiment lexicon": 106756,
      "lexicon study": 63906,
      "study problem": 114480,
      "detection online": 30010,
      "sequential model": 107237,
      "model proposed": 71813,
      "proposed make": 93329,
      "predictions sentence": 89191,
      "existing generalpurpose": 40138,
      "lexicons improve": 63927,
      "performance evaluate": 86345,
      "tagging model": 117407,
      "online discussion": 81767,
      "corpora wikipedia": 23621,
      "online debates": 81761,
      "shown outperform": 108500,
      "approaches datasets": 8115,
      "datasets example": 27459,
      "crf model": 24753,
      "f1 scores": 42692,
      "linear chain": 64330,
      "chain crf": 15976,
      "egyptian arabic": 35210,
      "arabic english": 8513,
      "dialectal arabic": 30527,
      "efforts focused": 35197,
      "arabic using": 8565,
      "using tools": 130296,
      "using stateoftheart": 130215,
      "model sparse": 72072,
      "features neural": 43624,
      "network joint": 77290,
      "joint model": 57289,
      "model unsupervised": 72249,
      "unsupervised transliteration": 127745,
      "hypothesis combination": 51262,
      "ranked second": 95649,
      "automatic postediting": 10623,
      "multimodal translation": 75463,
      "neural sequence": 78669,
      "sequence learning": 107000,
      "promising paradigm": 92283,
      "paradigm machine": 84538,
      "translation achieving": 124623,
      "achieving competitive": 2940,
      "phrasebased systems": 87398,
      "recently published": 96744,
      "published methods": 94360,
      "used neural": 128653,
      "sequential learning": 107233,
      "learning order": 62865,
      "order build": 82289,
      "shared tasks": 108110,
      "tasks automatic": 118947,
      "multimodal machine": 75439,
      "domain causal": 33477,
      "mechanisms needed": 68065,
      "meaningful search": 67726,
      "search inference": 104322,
      "domain distinct": 33510,
      "second use": 104462,
      "use corpus": 127972,
      "corpus investigate": 23847,
      "individual model": 53918,
      "model achieved": 70543,
      "achieved micro": 2654,
      "micro f1": 70035,
      "best performers": 13398,
      "outperform rulebased": 82741,
      "models performance": 73732,
      "sievebased architecture": 108663,
      "points evaluating": 87855,
      "existing corpora": 40094,
      "evaluation targeted": 39414,
      "targeted tasks": 117774,
      "domains twitter": 33878,
      "news comment": 79317,
      "informal words": 54347,
      "explicit text": 41356,
      "text normalization": 121149,
      "step possible": 113294,
      "variants propose": 130879,
      "propose metric": 92781,
      "spelling variant": 111913,
      "dataset collected": 26794,
      "collected using": 18438,
      "neural response": 78654,
      "response generation": 101198,
      "generation consider": 47341,
      "consider incorporating": 21205,
      "incorporating topic": 53568,
      "topic information": 122531,
      "information sequencetosequence": 54972,
      "sequencetosequence framework": 107159,
      "framework generate": 45552,
      "generate informative": 46956,
      "interesting responses": 56090,
      "chatbots end": 16669,
      "end propose": 36823,
      "propose topic": 93113,
      "model model": 71520,
      "knowledge human": 58003,
      "responses conversation": 101261,
      "information generation": 54638,
      "joint attention": 57264,
      "generation probability": 47560,
      "mechanism summarizes": 68040,
      "hidden vectors": 49918,
      "input message": 55364,
      "context vectors": 22321,
      "topic vectors": 122583,
      "vectors topic": 131479,
      "topic attention": 122498,
      "topic words": 122586,
      "words message": 134052,
      "obtained pretrained": 81394,
      "pretrained lda": 90118,
      "lda model": 61828,
      "vectors jointly": 131446,
      "words decoding": 133899,
      "responses model": 101284,
      "model modifies": 71529,
      "adding extra": 3377,
      "overall distribution": 83227,
      "distribution empirical": 32651,
      "study automatic": 114326,
      "metrics human": 69966,
      "generation models": 47482,
      "alignment score": 4987,
      "network word": 77475,
      "representations compute": 99556,
      "scores sentence": 104202,
      "unsupervised training": 127741,
      "training use": 123942,
      "scores given": 104175,
      "given target": 48144,
      "words decreasing": 133900,
      "words present": 134135,
      "align model": 4902,
      "model approach": 70674,
      "approach improves": 7623,
      "alignment accuracy": 4940,
      "estimation machine": 38651,
      "notoriously difficult": 80417,
      "realworld translation": 96187,
      "scenarios improved": 103852,
      "improved quality": 52631,
      "quality predictions": 94760,
      "uncertainty models": 126622,
      "evaluated terms": 39013,
      "probabilistic methods": 90818,
      "methods quality": 69704,
      "uncertainty estimates": 126617,
      "predictive distributions": 89201,
      "information useful": 55077,
      "aims capture": 4508,
      "translation workflows": 125435,
      "models biomedical": 72858,
      "extraction propose": 42450,
      "approach biomedical": 7392,
      "learning directly": 62504,
      "directly data": 31871,
      "data benefits": 25696,
      "rulebased approaches": 103359,
      "interpretability approach": 56220,
      "approach starts": 7930,
      "starts training": 112418,
      "statistical model": 113131,
      "features rules": 43709,
      "large body": 61042,
      "body work": 14400,
      "work machine": 134622,
      "produces interpretable": 92000,
      "interpretable model": 56246,
      "model directly": 70999,
      "experts evaluate": 41240,
      "results small": 102194,
      "small performance": 110184,
      "effort human": 35174,
      "model similar": 72040,
      "similar performance": 109118,
      "point representation": 87813,
      "representation texts": 99438,
      "texts complex": 121478,
      "mesoscopic approach": 68492,
      "approach statistical": 7934,
      "text analytics": 120642,
      "use simple": 128277,
      "simple word": 109547,
      "word count": 132978,
      "new paradigm": 79069,
      "set methods": 107493,
      "representations terms": 99930,
      "methods successfully": 69777,
      "syntactical features": 116501,
      "written texts": 135152,
      "represent important": 99112,
      "important aspects": 52105,
      "aspects textual": 9411,
      "topical structure": 122594,
      "level text": 63511,
      "text aspects": 120661,
      "semantical content": 105360,
      "analyze documents": 5966,
      "fashion proposed": 43121,
      "model limited": 71462,
      "represented nodes": 100032,
      "nodes connected": 80032,
      "model present": 71766,
      "present case": 89400,
      "case example": 15583,
      "structure document": 113840,
      "document modeled": 33038,
      "modeled network": 72359,
      "texts approach": 121457,
      "approach illustrated": 7615,
      "learning context": 62456,
      "texts classified": 121472,
      "detection social": 30056,
      "able react": 1678,
      "perform optimally": 86039,
      "text messages": 121112,
      "set rules": 107569,
      "rules detect": 103392,
      "direct indirect": 31787,
      "particularly context": 85475,
      "similar performances": 109121,
      "despite differences": 29685,
      "depends nature": 29288,
      "particularly problematic": 85491,
      "terms way": 120406,
      "applications work": 7040,
      "work focused": 134534,
      "approaches work": 8405,
      "applications need": 6974,
      "decision use": 27949,
      "task sharing": 118708,
      "art approaches": 9035,
      "recognition rely": 96987,
      "rely hand": 98701,
      "hand crafted": 49315,
      "crafted features": 24591,
      "corpora recently": 23571,
      "require annotated": 100111,
      "corpora annotated": 23409,
      "model allows": 70659,
      "sharing decoder": 108129,
      "decoder word": 28061,
      "level parameters": 63486,
      "parameters languages": 84761,
      "languages allowing": 60398,
      "language specifically": 60113,
      "specifically focus": 111551,
      "focus case": 44744,
      "case limited": 15592,
      "limited annotated": 64204,
      "available language": 11023,
      "architecture parameters": 8724,
      "l1 l2": 58370,
      "leads improved": 61936,
      "require hand": 100139,
      "features instead": 43572,
      "instead directly": 55657,
      "directly learns": 31884,
      "learns meaningful": 63222,
      "meaningful feature": 67716,
      "feature representations": 43303,
      "representations training": 99941,
      "data experiment": 25920,
      "experiment language": 40469,
      "corpora model": 23530,
      "model jointly": 71383,
      "jointly trained": 57394,
      "trained data": 123106,
      "language performs": 59809,
      "better model": 13629,
      "trained limited": 123189,
      "limited corpora": 64223,
      "corpora language": 23510,
      "language evaluating": 59017,
      "evaluating unsupervised": 39108,
      "dutch word": 34284,
      "embeddings linguistic": 35790,
      "embeddings recently": 35902,
      "recently seen": 96758,
      "strong performance": 113697,
      "tasks research": 119468,
      "tasks datasets": 119027,
      "datasets used": 27775,
      "used tasks": 128804,
      "tasks prove": 119420,
      "observations language": 81177,
      "demonstrate performance": 28814,
      "multiple types": 75740,
      "types embeddings": 126275,
      "embeddings created": 35623,
      "tasks relation": 119449,
      "relation evaluation": 97976,
      "evaluation dialect": 39180,
      "compare unsupervised": 19309,
      "research provide": 100598,
      "embeddings relation": 35905,
      "evaluation task": 39415,
      "task benchmark": 117927,
      "benchmark use": 12871,
      "research demonstrate": 100461,
      "unsupervised linguistic": 127656,
      "effectively used": 34859,
      "used downstream": 128501,
      "downstream task": 34029,
      "text comparison": 120805,
      "text sources": 121315,
      "using python": 130070,
      "view text": 131710,
      "sources text": 110924,
      "text summaries": 121342,
      "summaries source": 115460,
      "enables users": 36407,
      "explore text": 41596,
      "sources like": 110906,
      "map word": 67116,
      "capture linguistic": 15329,
      "semantic concepts": 105015,
      "concepts like": 20625,
      "using dimensionality": 129579,
      "words close": 133861,
      "technique uses": 119821,
      "uses word2vec": 129301,
      "word2vec model": 133681,
      "python library": 94480,
      "contextdependent word": 22356,
      "representation neural": 99353,
      "representations symbols": 99923,
      "translation continuous": 124711,
      "embedding vector": 35526,
      "multiple dimensions": 75541,
      "encoding meaning": 36709,
      "meaning word": 67706,
      "decoder recurrent": 28049,
      "translation need": 125003,
      "based context": 11601,
      "propose contextualize": 92600,
      "embedding vectors": 35527,
      "bagofwords representation": 11380,
      "additionally propose": 3621,
      "propose represent": 93020,
      "special tokens": 111372,
      "translated continuous": 124551,
      "continuous vectors": 22646,
      "vectors experiments": 131436,
      "enfr ende": 37011,
      "reveal proposed": 102504,
      "proposed approaches": 93224,
      "improves translation": 53056,
      "quality neural": 94728,
      "systems significantly": 117145,
      "application lowresource": 6858,
      "using crosslingual": 129519,
      "crosslingual projection": 24996,
      "cross lingual": 24864,
      "sources bias": 110880,
      "noise leading": 80057,
      "used directly": 128493,
      "directly paper": 31896,
      "approach sequence": 7880,
      "sequence tagging": 107075,
      "correct errors": 24099,
      "errors crosslingual": 38362,
      "using explicit": 129655,
      "explicit debiasing": 41313,
      "joint learning": 57284,
      "tagged gold": 117349,
      "tokens tagged": 122337,
      "exceeds stateoftheart": 39908,
      "simulated lowresource": 109638,
      "lowresource settings": 65562,
      "real lowresource": 96070,
      "formal models": 45172,
      "texts documents": 121502,
      "model texts": 72166,
      "texts using": 121645,
      "analysis particular": 5742,
      "constraints present": 21608,
      "present experimental": 89480,
      "bridge gap": 14591,
      "gap normative": 46469,
      "normative text": 80365,
      "representation approach": 99171,
      "uses dependency": 129222,
      "rules heuristics": 103409,
      "extracting relevant": 42233,
      "data converted": 25795,
      "translation improves": 124862,
      "improves neural": 52997,
      "order control": 82302,
      "complexity neural": 19928,
      "systems convert": 116818,
      "words outside": 134099,
      "luong et": 65725,
      "use multiple": 128156,
      "words unseen": 134286,
      "unseen training": 127552,
      "suffers noisy": 115262,
      "focus major": 44787,
      "entity ne": 37977,
      "propose translate": 93122,
      "level sequence": 63504,
      "alignment bilingual": 4948,
      "bilingual training": 14063,
      "translation alignment": 124634,
      "nmt able": 79820,
      "chinese english": 16763,
      "english task": 37303,
      "embedding words": 35531,
      "sentences character": 106236,
      "character ngrams": 16454,
      "ngrams present": 79468,
      "embeddings simple": 35940,
      "embed textual": 35350,
      "sequences word": 107146,
      "sentence represented": 106049,
      "using character": 129434,
      "count vector": 24415,
      "nonlinear transformation": 80226,
      "embedding use": 35522,
      "tasks evaluation": 119106,
      "evaluation word": 39442,
      "similarity sentence": 109307,
      "tagging demonstrate": 117382,
      "embeddings outperform": 35844,
      "outperform complex": 82699,
      "complex architectures": 19800,
      "architectures based": 8782,
      "based characterlevel": 11570,
      "characterlevel recurrent": 16580,
      "recurrent convolutional": 97164,
      "achieving new": 2960,
      "similarity tasks": 109318,
      "distributional statistics": 32738,
      "traditional semantic": 122862,
      "parsers map": 85031,
      "language compositional": 58903,
      "allows effectively": 5147,
      "effectively leverage": 34829,
      "leverage information": 63589,
      "information contained": 54443,
      "contained large": 21774,
      "bases kbs": 12503,
      "meaning language": 67644,
      "methods open": 69651,
      "learning execution": 62560,
      "arbitrary language": 8585,
      "using text": 130286,
      "corpus kind": 23849,
      "kind knowledge": 57679,
      "base prior": 11471,
      "prior approaches": 90691,
      "use kb": 128101,
      "kb models": 57501,
      "used approaches": 128393,
      "time semantic": 122098,
      "semantic parser": 105133,
      "representations language": 99713,
      "corpus limited": 23866,
      "demonstrate significantly": 28866,
      "stateoftheart baselines": 112593,
      "opendomain natural": 81972,
      "language question": 60016,
      "answering task": 6708,
      "task dependency": 118066,
      "dependency language": 29152,
      "models transitionbased": 74244,
      "approach improve": 7618,
      "accuracy strong": 2291,
      "strong transitionbased": 113728,
      "parser exploiting": 84954,
      "parsed corpus": 84896,
      "small number": 110176,
      "models parser": 73715,
      "parser demonstrate": 84933,
      "evaluate parser": 38880,
      "standard english": 112235,
      "english chinese": 37083,
      "data base": 25688,
      "parser achieve": 84911,
      "competitive accuracy": 19633,
      "accuracy scores": 2268,
      "stateoftheart accuracy": 112572,
      "accuracy chinese": 2114,
      "data competitive": 25757,
      "results english": 101753,
      "english data": 37104,
      "gained large": 46364,
      "absolute improvement": 1734,
      "point uas": 87817,
      "05 points": 33,
      "detection entity": 29939,
      "web documents": 132232,
      "numerous applications": 81013,
      "results time": 102274,
      "entities particular": 37836,
      "ability recognize": 1568,
      "problem widely": 91285,
      "propose paper": 92983,
      "paper method": 84044,
      "crossdocument coreference": 24881,
      "resolution named": 100772,
      "clusters used": 17970,
      "used additional": 128380,
      "additional signal": 3571,
      "turn used": 125980,
      "performances tasks": 86900,
      "tasks implemented": 119166,
      "implemented prototype": 51971,
      "report results": 99034,
      "collection news": 18481,
      "potential approach": 88537,
      "approach empirical": 7518,
      "document embedding": 32998,
      "generation recently": 47588,
      "extension word2vec": 41843,
      "word2vec mikolov": 133677,
      "mikolov et": 70079,
      "al 2013a": 4606,
      "documentlevel embeddings": 33139,
      "embeddings despite": 35645,
      "despite promising": 29716,
      "results original": 102016,
      "rigorous empirical": 102850,
      "tasks compare": 118989,
      "baselines stateoftheart": 12470,
      "stateoftheart document": 112645,
      "performs robustly": 87020,
      "using models": 129890,
      "trained large": 123173,
      "corpora improved": 23500,
      "improved using": 52653,
      "using pretrained": 130042,
      "provide recommendations": 93905,
      "hyperparameter settings": 51231,
      "settings general": 107804,
      "release source": 98472,
      "using trained": 130300,
      "models discriminating": 73085,
      "discriminating similar": 32200,
      "twitter using": 126131,
      "label propagation": 58407,
      "media messages": 68126,
      "step linguistic": 113278,
      "linguistic processing": 64529,
      "processing existing": 91669,
      "models twitter": 74253,
      "dissimilar language": 32397,
      "pairs propose": 83611,
      "propose label": 92739,
      "social graph": 110340,
      "tease apart": 119739,
      "languages results": 60851,
      "results stateoftheart": 102205,
      "task performance": 118528,
      "skipgram models": 109988,
      "representations way": 99975,
      "embeddings particular": 35855,
      "particular propose": 85443,
      "propose learning": 92745,
      "learning procedure": 62913,
      "function capture": 46006,
      "compose phrases": 20060,
      "improvement word": 52784,
      "phrase similarity": 87369,
      "syntactic tasks": 116491,
      "parsing using": 85277,
      "proposed joint": 93316,
      "joint models": 57297,
      "consists set": 21493,
      "relevant context": 98532,
      "context determining": 22055,
      "lexical units": 63837,
      "reflect effects": 97605,
      "effects context": 34977,
      "effect main": 34601,
      "analyze effects": 5969,
      "contextual variation": 22514,
      "specifically focused": 111552,
      "variation based": 130887,
      "conducted empirical": 20918,
      "analysis set": 5840,
      "variation different": 130889,
      "different phenomena": 31334,
      "concepts additionally": 20604,
      "second applied": 104384,
      "sequencetosequence learning": 107162,
      "learning mapping": 62708,
      "mapping utterances": 67155,
      "pragmatics paper": 88762,
      "grammars cfgs": 48672,
      "sequencetosequence seq2seq": 107202,
      "learning specifically": 63043,
      "generate dialogue": 46926,
      "using tool": 130295,
      "semantic pragmatic": 105181,
      "dialogue manager": 30707,
      "use grammatical": 128074,
      "surface utterance": 116080,
      "infer information": 54099,
      "using rnn": 130121,
      "information dialogue": 54482,
      "offline evaluation": 81611,
      "task demonstrate": 118057,
      "demonstrate trained": 28891,
      "accuracy knowledge": 2191,
      "knowledge usage": 58223,
      "seq2seq learning": 106895,
      "learning conversational": 62462,
      "neural sentence": 78665,
      "sentence ordering": 105962,
      "critical task": 24836,
      "works focused": 134949,
      "focused improving": 44860,
      "performance external": 86371,
      "summarization given": 115514,
      "importance propose": 52074,
      "propose study": 93093,
      "isolated task": 56975,
      "task collect": 117970,
      "collect large": 18382,
      "corpus academic": 23639,
      "academic texts": 1863,
      "texts derive": 121494,
      "data driven": 25870,
      "approach learn": 7675,
      "validate efficacy": 130720,
      "experiments source": 41147,
      "source codes": 110729,
      "codes dataset": 18199,
      "dataset paper": 27084,
      "paper publicly": 84389,
      "latent tree": 61621,
      "introduce latent": 56448,
      "encodes syntax": 36679,
      "semantics given": 105422,
      "tree word": 125620,
      "roles learning": 103232,
      "learning phase": 62887,
      "gibbs sampling": 47962,
      "infer tree": 54107,
      "does guarantee": 33355,
      "based dynamic": 11665,
      "provide comparison": 93781,
      "comparison algorithms": 19524,
      "kneserney language": 57722,
      "model linear": 71463,
      "experiments english": 40930,
      "english czech": 37102,
      "corpora significant": 23586,
      "significant perplexity": 108845,
      "model machine": 71484,
      "understanding semantic": 126952,
      "meanings different": 67743,
      "kept pace": 57529,
      "keywords lead": 57648,
      "end present": 36822,
      "present machine": 89538,
      "rich features": 102744,
      "features deep": 43439,
      "unsupervised feature": 127638,
      "base classifiers": 11447,
      "training ensemble": 123611,
      "ensemble method": 37599,
      "learning solution": 63035,
      "solution using": 110569,
      "using manual": 129856,
      "methods automatically": 69331,
      "detect semantic": 29816,
      "positions proposed": 88311,
      "finally ensemble": 44177,
      "final prediction": 44118,
      "precision current": 88792,
      "current position": 25322,
      "educational background": 34573,
      "synthetic language": 116629,
      "model validation": 72294,
      "generating synthetic": 47271,
      "synthetic languages": 116630,
      "future computational": 46261,
      "computational linguistic": 20390,
      "models methods": 73565,
      "methods thesis": 69804,
      "thesis extends": 121851,
      "linguistic sequence": 64553,
      "generation multiple": 47495,
      "word borrowing": 132935,
      "inference process": 54202,
      "process widely": 91587,
      "models connecting": 72960,
      "translation adaptation": 124624,
      "available statistical": 11115,
      "belong similar": 12717,
      "similar domains": 109087,
      "corpus enhance": 23776,
      "performance directly": 86305,
      "adaptation methods": 3232,
      "focus sentence": 44814,
      "fine grained": 44322,
      "data selection": 26383,
      "selection propose": 104820,
      "propose straightforward": 93085,
      "adaptation method": 3231,
      "phrase pair": 87360,
      "results phrase": 102036,
      "performance significantly": 86712,
      "comparison existing": 19543,
      "methods syntactic": 69788,
      "structural constraints": 113761,
      "primary goal": 90650,
      "goal thesis": 48392,
      "better syntactic": 13736,
      "bias language": 13811,
      "focus particular": 44803,
      "syntactic construction": 116384,
      "tree constructed": 125584,
      "avoid generating": 11236,
      "structures explore": 114077,
      "explore utilities": 41611,
      "theoretical empirical": 121793,
      "empirical examinations": 36165,
      "primary task": 90657,
      "task unsupervised": 118823,
      "input algorithm": 55294,
      "collection sentences": 18489,
      "extract salient": 42110,
      "salient patterns": 103553,
      "hard problem": 49478,
      "help improving": 49739,
      "performance effectively": 86329,
      "possible search": 88431,
      "model build": 70790,
      "build model": 14786,
      "model extending": 71147,
      "algorithm efficiently": 4723,
      "examine effectiveness": 39744,
      "methods particularly": 69668,
      "systems exploit": 116896,
      "exploit efficiently": 41412,
      "analogy corpus": 5417,
      "embedding methods": 35441,
      "methods proven": 69697,
      "useful tasks": 128939,
      "tasks nlp": 119343,
      "nlp natural": 79650,
      "investigated word": 56846,
      "embeddings english": 35691,
      "explore behavior": 41524,
      "stateoftheart word": 113028,
      "czech language": 25492,
      "language characterized": 58874,
      "morphology introduce": 74788,
      "corpus word": 24067,
      "analogy task": 5424,
      "morphosyntactic semantic": 74817,
      "experiment word2vec": 40520,
      "word2vec glove": 133675,
      "glove algorithms": 48298,
      "algorithms discuss": 4843,
      "discuss results": 32288,
      "results corpus": 101643,
      "corpus available": 23673,
      "research community": 100445,
      "largescale evaluation": 61434,
      "play critical": 87687,
      "attention recent": 9985,
      "semantics research": 105457,
      "research introduce": 100538,
      "evaluation resource": 39363,
      "human ratings": 50941,
      "semantic phenomena": 105178,
      "influence human": 54312,
      "human understanding": 50982,
      "significantly larger": 108972,
      "development test": 30431,
      "existing benchmarks": 40082,
      "robust evaluation": 103032,
      "learning architectures": 62370,
      "richer understanding": 102810,
      "diversity complexity": 32870,
      "systems effectively": 116861,
      "effectively represent": 34847,
      "poses major": 88267,
      "languages significant": 60872,
      "significant differences": 108752,
      "differences word": 30980,
      "order paper": 82379,
      "output experiments": 83075,
      "statistically significant": 113181,
      "using approach": 129340,
      "approach compared": 7432,
      "compared stateoftheart": 19453,
      "stateoftheart statistical": 112972,
      "incorporates prior": 53512,
      "focusing particular": 44928,
      "phenomena language": 87228,
      "analysis existing": 5608,
      "existing static": 40296,
      "using high": 129739,
      "dimensional spaces": 31752,
      "potentially solved": 88619,
      "outofvocabulary words": 82682,
      "words bilingual": 133840,
      "bilingual embeddings": 14036,
      "embeddings machine": 35796,
      "translation outofvocabulary": 125079,
      "large proportion": 61224,
      "errors machine": 38381,
      "systems especially": 116874,
      "used different": 128491,
      "different domain": 31104,
      "domain trained": 33678,
      "order alleviate": 82277,
      "model vocabulary": 72310,
      "expansion given": 40371,
      "word model": 133363,
      "list possible": 64696,
      "possible translations": 88440,
      "translations target": 125498,
      "uses word": 129298,
      "large unlabelled": 61316,
      "monolingual corpora": 74557,
      "fairly small": 42997,
      "input probabilistic": 55403,
      "translation obtain": 125069,
      "obtain consistent": 81270,
      "improvements translation": 52933,
      "englishspanish language": 37429,
      "improvement 39": 52676,
      "tested outofdomain": 120581,
      "algorithm designed": 4715,
      "designed task": 29627,
      "colloquial language": 18523,
      "bottomup strategy": 14503,
      "highlevel performance": 50247,
      "performance relatively": 86669,
      "basic data": 12519,
      "models feature": 73242,
      "space allows": 110969,
      "languages domains": 60515,
      "domains experiments": 33780,
      "experiments spanning": 41148,
      "19 different": 434,
      "stateoftheart model": 112719,
      "evaluation recent": 39354,
      "sharedtask data": 108122,
      "data places": 26238,
      "overall best": 83217,
      "segmentation algorithm": 104573,
      "classes multiple": 17066,
      "english domains": 37121,
      "domains including": 33793,
      "usergenerated text": 129077,
      "text performance": 121177,
      "release opensource": 98467,
      "opensource software": 82030,
      "sequence labelling": 106993,
      "understanding paper": 126912,
      "encoderdecoder attention": 36593,
      "attention sequence": 10006,
      "understanding introduce": 126868,
      "introduce bidirectional": 56387,
      "long short": 65092,
      "short term": 108247,
      "memory long": 68321,
      "memory networks": 68356,
      "model fully": 71209,
      "fully utilize": 45979,
      "labelling task": 58573,
      "output sequences": 83121,
      "sequences aligned": 107112,
      "aligned word": 4929,
      "word attention": 132930,
      "exact alignment": 39715,
      "alignment address": 4941,
      "address limitation": 3712,
      "limitation propose": 64165,
      "mechanism encoderdecoder": 67976,
      "framework experiments": 45535,
      "experiments standard": 41150,
      "atis dataset": 9679,
      "dataset showed": 27198,
      "stateoftheart outperforming": 112797,
      "outperforming standard": 82828,
      "based encoderdecoder": 11674,
      "robust speech": 103072,
      "recognition errors": 96871,
      "dataset evaluation": 26899,
      "extent semantic": 41983,
      "semantic category": 105000,
      "concept pairs": 20582,
      "cognitive psychology": 18278,
      "psychology research": 94219,
      "human semantic": 50960,
      "binary relation": 14134,
      "research existing": 100496,
      "existing largescale": 40158,
      "wordnet dbpedia": 133770,
      "native english": 76220,
      "english speakers": 37288,
      "diverse range": 32834,
      "crowdsourcing platform": 25120,
      "compare human": 19251,
      "human judgements": 50877,
      "automatic systems": 10685,
      "huge gap": 50721,
      "gap human": 46455,
      "distributional representation": 32715,
      "models substantial": 74112,
      "differences models": 30964,
      "models discuss": 73089,
      "improving semantic": 53161,
      "models overcome": 73693,
      "indicate future": 53836,
      "provides simple": 94076,
      "simple ways": 109544,
      "language emerging": 58998,
      "early development": 34388,
      "formal account": 45152,
      "conveyed language": 23123,
      "quantitative predictions": 94874,
      "predictions human": 89175,
      "understanding test": 126997,
      "test model": 120472,
      "model diverse": 71014,
      "diverse domains": 32809,
      "generic language": 47800,
      "causal language": 15807,
      "understanding work": 127019,
      "work opens": 134664,
      "opens door": 82007,
      "understanding precisely": 126921,
      "abstract knowledge": 1765,
      "language hierarchical": 59120,
      "identification social": 51443,
      "pose challenge": 88249,
      "challenge language": 16045,
      "introduce hierarchical": 56433,
      "wordlevel representations": 133753,
      "identification method": 51396,
      "method performs": 69046,
      "strong base": 113625,
      "automatically detecting": 10754,
      "detecting inappropriate": 29840,
      "content difficult": 21873,
      "task requiring": 118649,
      "understanding context": 126818,
      "identifying specific": 51628,
      "specific keywords": 111453,
      "large quantity": 61232,
      "online usergenerated": 81814,
      "content automatic": 21859,
      "increasingly necessary": 53704,
      "small segment": 110196,
      "crowdsourced annotations": 25105,
      "topic modelling": 122549,
      "suggest certain": 115305,
      "certain topics": 15961,
      "useful detecting": 128874,
      "finegrained analysis": 44333,
      "analysis sentence": 5835,
      "using auxiliary": 129374,
      "auxiliary prediction": 10882,
      "prediction tasks": 89138,
      "lot research": 65325,
      "variable length": 130840,
      "sentences fixed": 106320,
      "length vectors": 63382,
      "sentence meanings": 105946,
      "common methods": 18891,
      "methods include": 69544,
      "averaging word": 11228,
      "vectors representations": 131471,
      "based hidden": 11752,
      "hidden states": 49911,
      "states recurrent": 113054,
      "networks lstms": 77652,
      "sentence vectors": 106125,
      "vectors used": 131483,
      "subsequent machine": 114814,
      "learning tasks": 63089,
      "tasks pretraining": 119395,
      "learning known": 62666,
      "known properties": 58318,
      "properties encoded": 92449,
      "encoded sentence": 36482,
      "capture propose": 15364,
      "facilitates better": 42802,
      "better understanding": 13758,
      "understanding encoded": 126834,
      "encoded representations": 36481,
      "define prediction": 28489,
      "aspects sentence": 9407,
      "sentence length": 105921,
      "word content": 132965,
      "content word": 21975,
      "score representations": 104116,
      "ability train": 1573,
      "prediction task": 89133,
      "using representation": 130108,
      "representation input": 99264,
      "input demonstrate": 55318,
      "analyzing different": 6039,
      "analysis sheds": 5843,
      "sheds light": 108161,
      "relative strengths": 98377,
      "strengths different": 113573,
      "sentence embedding": 105832,
      "low level": 65372,
      "level prediction": 63490,
      "tasks effect": 119078,
      "resulting representations": 101468,
      "technology used": 120048,
      "used widely": 128848,
      "intelligence human": 55830,
      "explosive growth": 41672,
      "data challenges": 25724,
      "challenges natural": 16177,
      "process large": 91519,
      "data required": 26344,
      "processing architectures": 91623,
      "contains language": 21825,
      "processing components": 91642,
      "stanford corenlp": 112364,
      "study describes": 114351,
      "data finally": 25948,
      "advantages disadvantages": 4077,
      "processing performance": 91769,
      "performance fast": 86378,
      "efficient methods": 35092,
      "propose language": 92740,
      "onthefly present": 81825,
      "usage large": 127864,
      "method highly": 68872,
      "highly competitive": 50296,
      "lower memory": 65441,
      "memory requirements": 68381,
      "similar training": 109162,
      "neural approach": 77829,
      "nmt recently": 79953,
      "recently emerged": 96683,
      "englishgerman language": 37380,
      "syntactic differences": 116395,
      "analysis neural": 5717,
      "iwslt data": 57165,
      "data time": 26562,
      "analysis provides": 5784,
      "provides useful": 94093,
      "useful insights": 128897,
      "linguistic phenomena": 64521,
      "best modeled": 13376,
      "models reordering": 73923,
      "building using": 14898,
      "using sentiment": 130151,
      "sentiment dictionary": 106731,
      "media increasingly": 68120,
      "considered important": 21292,
      "issues sentiment": 57078,
      "difficult measure": 31645,
      "content usually": 21972,
      "short informal": 108211,
      "traditional sentiment": 122863,
      "sentiment words": 106819,
      "words remains": 134168,
      "available existing": 10994,
      "existing dictionaries": 40114,
      "propose build": 92579,
      "build sentiment": 14807,
      "analysis social": 5859,
      "media content": 68092,
      "laborious timeconsuming": 58669,
      "collect label": 18381,
      "comprehensive list": 20265,
      "approach leverage": 7681,
      "resources construct": 100957,
      "sentiment word": 106818,
      "word dictionary": 132990,
      "advantages using": 4090,
      "classification provide": 17359,
      "ease use": 34408,
      "existing sentiment": 40281,
      "sentiment learning": 106754,
      "learning start": 63052,
      "sequence architecture": 106917,
      "architecture sequence": 8744,
      "generation neural": 47504,
      "model potential": 71752,
      "relationship sentences": 98308,
      "consists parts": 21487,
      "parts encoder": 85572,
      "encoder reads": 36557,
      "reads source": 96045,
      "decoder generates": 28016,
      "word according": 132901,
      "output generated": 83078,
      "generated word": 47153,
      "cold start": 18349,
      "problem generating": 91064,
      "generating word": 47285,
      "existing work": 40336,
      "mainly use": 66495,
      "work learnable": 134606,
      "words start": 134237,
      "error accumulation": 38277,
      "word incorrectly": 133317,
      "generated paper": 47105,
      "paper proposed": 84346,
      "proposed novel": 93512,
      "architecture using": 8769,
      "generation short": 47619,
      "conversation proposed": 22967,
      "stateoftheart approach": 112579,
      "manual evaluations": 66996,
      "evaluations using": 39489,
      "using output": 129995,
      "output embedding": 83070,
      "embedding improve": 35414,
      "improve language": 52400,
      "models study": 74110,
      "weight matrix": 132327,
      "tying input": 126183,
      "analyze resulting": 6002,
      "similar way": 109170,
      "embedding input": 35417,
      "model offer": 71606,
      "offer new": 81568,
      "regularizing output": 97773,
      "variety neural": 131007,
      "models finally": 73248,
      "reduce size": 97358,
      "size neural": 109937,
      "application sentiment": 6881,
      "analysis apply": 5497,
      "apply natural": 7192,
      "analysis english": 5596,
      "book reviews": 14414,
      "distinct semantic": 32542,
      "cognitive evaluation": 18267,
      "analysis automatic": 5507,
      "subjectivity text": 114706,
      "text discuss": 120879,
      "analysis language": 5668,
      "applications improving": 6936,
      "representations distributional": 99604,
      "distributional inference": 32704,
      "inference semantic": 54214,
      "semantic composition": 105012,
      "composition distributional": 20092,
      "models derived": 73045,
      "corpus small": 24001,
      "small proportion": 110188,
      "plausible cooccurrences": 87681,
      "observed results": 81231,
      "sparse vector": 111240,
      "missing knowledge": 70329,
      "methods face": 69504,
      "face challenge": 42729,
      "resulting word": 101479,
      "hard model": 49467,
      "explore alternative": 41516,
      "alternative involves": 5231,
      "using distributional": 129600,
      "inference improves": 54151,
      "representations word": 99978,
      "similarity benchmarks": 109205,
      "benchmarks demonstrate": 12897,
      "demonstrate model": 28787,
      "model competitive": 70863,
      "language nlp": 59721,
      "world data": 135024,
      "radically benchmark": 95441,
      "benchmark corpora": 12746,
      "corpora use": 23610,
      "technologies real": 120027,
      "performance drops": 86325,
      "limited set": 64277,
      "varieties considered": 130954,
      "considered standard": 21304,
      "english newswire": 37223,
      "language genre": 59107,
      "texts differ": 121497,
      "factors clear": 42883,
      "clear best": 17720,
      "best current": 13324,
      "current practice": 25323,
      "practice training": 88735,
      "homogeneous data": 50614,
      "single domain": 109729,
      "language argue": 58844,
      "plain sight": 87624,
      "heterogeneous data": 49852,
      "robust models": 103048,
      "capable addressing": 15191,
      "help readers": 49769,
      "task automatically": 117914,
      "short descriptions": 108207,
      "collect dataset": 18377,
      "articles mention": 9198,
      "propose generate": 92701,
      "generate descriptions": 46923,
      "facts knowledge": 42915,
      "using sequencetosequence": 130159,
      "sequencetosequence recurrent": 107199,
      "f1 improvement": 42658,
      "point improvement": 87804,
      "generation improving": 47436,
      "correlation human": 24230,
      "integrating semantic": 55801,
      "vectors vector": 131485,
      "space methods": 111022,
      "cooccurrence frequencies": 23225,
      "statistical measures": 113124,
      "extend methods": 41786,
      "methods incorporating": 69548,
      "similarity based": 109201,
      "based human": 11758,
      "human curated": 50790,
      "representation results": 99398,
      "results measure": 101908,
      "contextual information": 22469,
      "space representation": 111050,
      "representation semantic": 99402,
      "knowledge biomedical": 57819,
      "biomedical ontology": 14201,
      "results incorporating": 101851,
      "incorporating semantic": 53556,
      "second order": 104436,
      "improves correlation": 52963,
      "method compares": 68712,
      "compares favorably": 19495,
      "various different": 131079,
      "methods recently": 69708,
      "standards used": 112354,
      "translation efforts": 124779,
      "translation spoken": 125285,
      "rich resource": 102779,
      "language translated": 60184,
      "explored previous": 41632,
      "research create": 100453,
      "larger bitext": 61354,
      "generate better": 46907,
      "better quality": 13685,
      "quality alignment": 94589,
      "languages key": 60662,
      "key problem": 57593,
      "typically using": 126468,
      "time information": 122039,
      "exploit information": 41419,
      "information develop": 54479,
      "original algorithm": 82502,
      "outperforms current": 82873,
      "current best": 25265,
      "training bitext": 123378,
      "yields improvement": 135419,
      "improvement bleu": 52687,
      "points absolute": 87843,
      "pmi matrix": 87772,
      "applications neural": 6975,
      "negative sampling": 76948,
      "function used": 46050,
      "used word2vec": 128853,
      "estimation nce": 38657,
      "effective learning": 34696,
      "representations unlike": 99954,
      "learning parameters": 62877,
      "model study": 72098,
      "information contexts": 54448,
      "predicted words": 88973,
      "words obtained": 134088,
      "nce language": 76702,
      "function provide": 46040,
      "main language": 66435,
      "function experimental": 46015,
      "results popular": 102041,
      "popular language": 88094,
      "comparable perplexity": 19166,
      "perplexity results": 87070,
      "reviews sentiment": 102629,
      "popular task": 88134,
      "processing work": 91859,
      "work goal": 134548,
      "goal predict": 48381,
      "carefully tuned": 15530,
      "train simple": 123020,
      "addition present": 3460,
      "present different": 89453,
      "methods deal": 69423,
      "highly skewed": 50347,
      "data common": 25750,
      "common problem": 18910,
      "models evaluated": 73173,
      "different original": 31313,
      "original ones": 82535,
      "classification methods": 17267,
      "accuracy translation": 2309,
      "classifier evaluated": 17542,
      "outside domain": 83198,
      "variety evaluation": 130976,
      "evaluation scenarios": 39375,
      "scenarios unsupervised": 103870,
      "unsupervised classification": 127611,
      "classification highly": 17222,
      "highly accurate": 50288,
      "suggest method": 115317,
      "method determining": 68772,
      "determining correct": 30146,
      "use labels": 128110,
      "suggest simple": 115332,
      "simple method": 109466,
      "challenging case": 16230,
      "fullyunsupervised method": 45993,
      "applied new": 7100,
      "present dependency": 89439,
      "single deep": 109724,
      "words directly": 133917,
      "directly generates": 31879,
      "dependencies labels": 29091,
      "unlike typical": 127454,
      "typical approaches": 126395,
      "approaches parsing": 8276,
      "model doesnt": 71024,
      "doesnt require": 33414,
      "additional supervision": 3576,
      "slavic languages": 110008,
      "languages universal": 60937,
      "dependencies treebank": 29119,
      "features characters": 43395,
      "parser accurate": 84910,
      "transition based": 124482,
      "trained perfect": 123238,
      "pos tags": 88241,
      "machine transliteration": 66313,
      "results machine": 101897,
      "transliteration process": 125526,
      "process automatically": 91427,
      "automatically transforming": 10830,
      "word source": 133577,
      "paper characterbased": 83760,
      "characterbased encoderdecoder": 16489,
      "proposed consists": 93242,
      "consists recurrent": 21490,
      "networks encoder": 77584,
      "encoder bidirectional": 36500,
      "bidirectional recurrent": 13968,
      "network encodes": 77242,
      "encodes sequence": 36677,
      "sequence symbols": 107073,
      "fixedlength vector": 44617,
      "representation decoder": 99204,
      "target sequence": 117705,
      "sequence using": 107094,
      "using attentionbased": 129355,
      "attentionbased recurrent": 10094,
      "network encoder": 77239,
      "decoder attention": 27996,
      "mechanism jointly": 68003,
      "trained maximize": 123196,
      "source sequence": 110825,
      "proposed encoderdecoder": 93265,
      "achieve significantly": 2543,
      "significantly higher": 108917,
      "traditional statistical": 122868,
      "ranks interpretations": 95703,
      "provides framework": 94034,
      "framework integrating": 45581,
      "structures language": 114080,
      "structures word": 114118,
      "morphology phonology": 74800,
      "interpretation models": 56268,
      "incremental processing": 53738,
      "linear time": 64380,
      "provide systematic": 93939,
      "design features": 29540,
      "features human": 43542,
      "properties single": 92482,
      "features framework": 43520,
      "framework provides": 45663,
      "evaluation realistic": 39352,
      "linguistic theories": 64566,
      "objective paper": 81105,
      "demonstrate unique": 28893,
      "grammar rank": 48653,
      "approach capturing": 7412,
      "complexity human": 19915,
      "italian paper": 57096,
      "set fast": 107441,
      "processing modules": 91715,
      "integrated existing": 55776,
      "event data": 39503,
      "social science": 110451,
      "approaches generating": 8173,
      "generating data": 47209,
      "presents overview": 89888,
      "data including": 26033,
      "experiments determine": 40903,
      "extraction political": 42436,
      "political events": 87994,
      "events news": 39587,
      "news text": 79378,
      "text minimally": 121117,
      "minimally supervised": 70187,
      "texttospeech tts": 121662,
      "tts automatic": 125914,
      "refers task": 97585,
      "task converting": 118022,
      "representation text": 99436,
      "speech applications": 111648,
      "applications text": 7025,
      "grammar used": 48663,
      "possible ways": 88446,
      "token given": 122255,
      "language statistical": 60123,
      "select appropriate": 104690,
      "pronunciation context": 92366,
      "context study": 22283,
      "examine tradeoffs": 39756,
      "using languagespecific": 129784,
      "produce set": 91931,
      "set possible": 107533,
      "corpus aligned": 23653,
      "ranking model": 95675,
      "model selects": 71961,
      "selects appropriate": 104860,
      "given context": 47998,
      "consider scenario": 21225,
      "provide set": 93916,
      "aligned corpus": 4910,
      "corpus consider": 23715,
      "spoken corresponding": 111971,
      "english russian": 37266,
      "russian semantic": 103500,
      "deep residual": 28417,
      "networks propose": 77710,
      "novel semantic": 80714,
      "multilingual semantic": 75357,
      "tagger using": 117362,
      "using deep": 129538,
      "networks resnets": 77738,
      "character representations": 16461,
      "includes novel": 53259,
      "architecture evaluate": 8660,
      "task semantic": 118672,
      "semantic tags": 105324,
      "outperforms prior": 82963,
      "prior results": 90728,
      "english universal": 37325,
      "deep multitask": 28340,
      "learning shared": 63022,
      "models achieved": 72677,
      "deep architectures": 28199,
      "architectures trained": 8853,
      "jointly multiple": 57370,
      "tasks specifically": 119512,
      "specifically augment": 111529,
      "augment neural": 10261,
      "model external": 71158,
      "external memory": 42034,
      "proposed architectures": 93227,
      "architectures improve": 8809,
      "amrtotext generation": 5403,
      "problem task": 91260,
      "task amrtotext": 117873,
      "generation generate": 47415,
      "meaning given": 67636,
      "given amr": 47981,
      "amr graph": 5388,
      "generating translation": 47283,
      "trained estimate": 123139,
      "used optimized": 128673,
      "largescale machine": 61460,
      "arabic hebrew": 8516,
      "available corpora": 10965,
      "initial results": 55222,
      "hebrew far": 49671,
      "far limited": 43092,
      "limited lack": 64248,
      "lack parallel": 58731,
      "political cultural": 87991,
      "importance language": 52066,
      "work relied": 134772,
      "pivoting english": 87590,
      "scalable accurate": 103686,
      "work compare": 134419,
      "neural systems": 78695,
      "translation experiment": 124808,
      "external tools": 42047,
      "subword modeling": 115027,
      "improved translation": 52650,
      "advantage neural": 4052,
      "network encoders": 77241,
      "encoders neural": 36659,
      "nmt heavily": 79871,
      "heavily relies": 49655,
      "representations input": 99695,
      "natural word": 76625,
      "delimiters chinese": 28613,
      "chinese input": 16775,
      "conventional nmt": 22896,
      "encoder nmt": 36550,
      "handle issues": 49385,
      "issues propose": 57067,
      "wordlattice based": 133724,
      "rnn encoders": 102916,
      "generalize standard": 46818,
      "standard rnn": 112295,
      "learn generate": 62056,
      "states preceding": 113050,
      "time steps": 122114,
      "based encoders": 11678,
      "alleviate negative": 5041,
      "negative impact": 76929,
      "impact tokenization": 51895,
      "tokenization errors": 122276,
      "sentences experiment": 106299,
      "results chineseenglish": 101571,
      "translation demonstrate": 124741,
      "factorized model": 42877,
      "model transitive": 72232,
      "semantics present": 105447,
      "semantics model": 105433,
      "model representation": 71907,
      "constructions model": 21704,
      "subject verb": 114686,
      "vectors combined": 131414,
      "representation simple": 99411,
      "vector operations": 131334,
      "construction model": 21693,
      "outperforms recent": 82972,
      "recent previous": 96493,
      "work unsupervised": 134869,
      "alignment lowresource": 4966,
      "languages lowresource": 60707,
      "languages spoken": 60890,
      "resources likely": 101001,
      "likely annotated": 64134,
      "annotated translations": 6251,
      "translations transcriptions": 125501,
      "translated speech": 124563,
      "data potentially": 26253,
      "potentially valuable": 88622,
      "endangered languages": 36841,
      "languages training": 60926,
      "training speech": 123875,
      "systems step": 117162,
      "step making": 113280,
      "words translations": 134281,
      "translations present": 125486,
      "present model": 89559,
      "dyer et": 34290,
      "et als": 38744,
      "ibm model": 51302,
      "kmeans clustering": 57715,
      "clustering using": 17959,
      "using dynamic": 129610,
      "dynamic time": 34329,
      "time warping": 122137,
      "distance metric": 32428,
      "components trained": 20049,
      "jointly using": 57401,
      "using expectationmaximization": 129652,
      "extremely lowresource": 42603,
      "lowresource scenario": 65557,
      "scenario model": 103827,
      "better neural": 13635,
      "strong baseline": 113626,
      "baseline aligning": 12186,
      "text streams": 121330,
      "multiple sources": 75697,
      "sources multiple": 110910,
      "opened new": 81983,
      "new research": 79106,
      "crosslingual knowledge": 24963,
      "knowledge discovery": 57862,
      "discovery paper": 32146,
      "paper aim": 83720,
      "advance stateoftheart": 3933,
      "finegrained information": 44359,
      "information units": 55068,
      "entities events": 37777,
      "utilize network": 130518,
      "network structures": 77447,
      "structures capture": 114062,
      "keywords corpus": 57644,
      "propose effective": 92639,
      "effective approach": 34629,
      "approach construct": 7453,
      "incorporating novel": 53550,
      "criteria based": 24791,
      "based multidimensional": 11860,
      "structure experimental": 113857,
      "approach accurately": 7294,
      "high confidence": 50048,
      "run parallel": 103463,
      "huge amounts": 50714,
      "streaming data": 113554,
      "evaluation rnn": 39373,
      "rnn architectures": 102902,
      "task recurrent": 118615,
      "networks achieved": 77502,
      "results problems": 102056,
      "problems nlp": 91352,
      "nlp popular": 79665,
      "pooling model": 88044,
      "paper hybrid": 83973,
      "hybrid architecture": 51176,
      "architecture proposed": 8733,
      "present empirical": 89465,
      "study using": 114543,
      "performance rnn": 86686,
      "rnn structures": 102937,
      "max pooling": 67497,
      "achieves best": 2736,
      "performance datasets": 86273,
      "does outperform": 33376,
      "outperform models": 82716,
      "learning sentence": 63000,
      "human attention": 50756,
      "attention recently": 9988,
      "recently progress": 96736,
      "progress learning": 92157,
      "learning generalpurpose": 62611,
      "generalpurpose sentence": 46892,
      "domains existing": 33776,
      "models typically": 74255,
      "typically treat": 126465,
      "treat word": 125554,
      "sentence equally": 105857,
      "extensive studies": 41957,
      "human read": 50944,
      "improve sentence": 52537,
      "different weights": 31564,
      "component words": 20014,
      "novel attention": 80500,
      "models attention": 72782,
      "attention weights": 10037,
      "predictors human": 89212,
      "human reading": 50948,
      "reading time": 96035,
      "demonstrate proposed": 28831,
      "methods significantly": 69750,
      "stateoftheart sentence": 112947,
      "models inducing": 73398,
      "networks work": 77818,
      "work focuses": 134536,
      "annotation tools": 6397,
      "languages experiment": 60568,
      "crosslingual annotation": 24927,
      "annotation projection": 6358,
      "using recurrent": 130087,
      "networks rnn": 77742,
      "rnn models": 102927,
      "distinctive feature": 32558,
      "multilingual word": 75402,
      "corpus source": 24004,
      "following characteristics": 44964,
      "assume knowledge": 9647,
      "foreign languages": 45043,
      "languages makes": 60717,
      "applicable wide": 6830,
      "languages provides": 60825,
      "truly multilingual": 125862,
      "uni bidirectional": 127085,
      "bidirectional rnn": 13975,
      "method include": 68895,
      "external information": 42004,
      "information instance": 54693,
      "level information": 63460,
      "information pos": 54847,
      "demonstrate validity": 28905,
      "corpora obtained": 23547,
      "crosslingual pos": 24992,
      "vocabulary selection": 131899,
      "selection strategies": 104831,
      "strategies neural": 113477,
      "space possible": 111042,
      "possible outputs": 88416,
      "selecting subset": 104761,
      "based input": 11772,
      "sentence recent": 106016,
      "work improving": 134564,
      "improving efficiency": 53093,
      "efficiency neural": 35035,
      "models adopted": 72726,
      "candidates given": 15136,
      "source paper": 110801,
      "paper experiment": 83908,
      "methods extend": 69499,
      "extend previous": 41792,
      "work examining": 134500,
      "speed accuracy": 111862,
      "decoding time": 28141,
      "wmt15 englishgerman": 132827,
      "wmt16 englishromanian": 132830,
      "tasks negligible": 119336,
      "decode state": 27986,
      "art neural": 9069,
      "single cpu": 109721,
      "sentence segmentation": 106058,
      "tests using": 120614,
      "networks automated": 77515,
      "tools based": 122440,
      "based natural": 11878,
      "sentence boundary": 105780,
      "boundary segmentation": 14524,
      "direct application": 31782,
      "application nlp": 6872,
      "methods rely": 69722,
      "taggers parsers": 117368,
      "steps taken": 113335,
      "based narrative": 11877,
      "presenting new": 89808,
      "new automatic": 78807,
      "automatic sentence": 10651,
      "segmentation method": 104599,
      "speech model": 111708,
      "uses recurrent": 129269,
      "speech pos": 111724,
      "embeddings evaluated": 35700,
      "evaluated intrinsically": 38981,
      "spontaneous speech": 112035,
      "speech normal": 111716,
      "speech presents": 111728,
      "respectively used": 101175,
      "study results": 114505,
      "suggest model": 115318,
      "code parameters": 18129,
      "binary ternary": 14137,
      "errorcorrecting codes": 38349,
      "data syntactic": 26526,
      "world languages": 135036,
      "languages study": 60902,
      "functions substantial": 46080,
      "language change": 58871,
      "entailment relations": 37671,
      "syntactic parameters": 116433,
      "improve code": 52352,
      "large sets": 61271,
      "syntactic data": 116390,
      "data gain": 25967,
      "languages explore": 60574,
      "sequence basic": 106919,
      "translation related": 125210,
      "languages use": 60944,
      "syllable level": 116303,
      "level translation": 63514,
      "outperforms models": 82918,
      "basic units": 12538,
      "units word": 127270,
      "word morpheme": 133365,
      "training small": 123868,
      "representations lexical": 99738,
      "sets contain": 107658,
      "contain words": 21771,
      "properties lexical": 92461,
      "semantics investigate": 105427,
      "distance lexical": 32420,
      "sets different": 107662,
      "properties particular": 92473,
      "particular case": 85387,
      "study lexical": 114427,
      "vector models": 131330,
      "spatial distance": 111264,
      "discovered automatically": 32123,
      "computational approach": 20359,
      "automatic prediction": 10628,
      "introduce automatic": 56384,
      "identifying text": 51630,
      "tweets labeled": 126033,
      "labeled using": 58477,
      "using hashtags": 129728,
      "supervision classifiers": 115873,
      "classifiers use": 17642,
      "use set": 128271,
      "stylistic features": 114628,
      "features detect": 43451,
      "present quantitative": 89658,
      "evidence text": 39674,
      "text contains": 120817,
      "provide largest": 93866,
      "quality phrasebased": 94752,
      "include hierarchical": 53227,
      "phrasebased mt": 87387,
      "performed recently": 86930,
      "united nations": 127226,
      "nations parallel": 76218,
      "second paper": 104437,
      "investigate aspects": 56719,
      "aspects translation": 9414,
      "translation speed": 125284,
      "efficient neural": 35097,
      "decoder demonstrate": 28002,
      "demonstrate current": 28695,
      "current neural": 25307,
      "meteor using": 68575,
      "presents approach": 89818,
      "applied evaluation": 7072,
      "evaluation machine": 39258,
      "metric meteor": 69888,
      "reference translation": 97542,
      "experiments framework": 40949,
      "metrics task": 70011,
      "task wmt": 118846,
      "wmt 2014": 132769,
      "representations good": 99670,
      "bring interesting": 14631,
      "available github": 11001,
      "github page": 47976,
      "embeddings demonstrated": 35637,
      "benefit nlp": 12984,
      "representations current": 99581,
      "embeddings typically": 35996,
      "noise propose": 80063,
      "novel models": 80649,
      "improve word": 52584,
      "embeddings unsupervised": 36001,
      "order yield": 82431,
      "salient information": 103551,
      "noise original": 80060,
      "original word": 82563,
      "embeddings based": 35565,
      "deep feedforward": 28233,
      "results benchmark": 101539,
      "benchmark tasks": 12863,
      "outperform original": 82722,
      "developing automated": 30334,
      "dominant paradigm": 33934,
      "smt translations": 110297,
      "models parameters": 73714,
      "parameters learned": 84764,
      "like make": 64062,
      "data possible": 26249,
      "train better": 122904,
      "mobile devices": 70467,
      "computing power": 20523,
      "widely available": 132539,
      "despite successful": 29737,
      "mobile applications": 70465,
      "applications relying": 7006,
      "clientserver architecture": 17756,
      "limited use": 64297,
      "mobile device": 70466,
      "main challenge": 66400,
      "devices memory": 30458,
      "thesis investigate": 121853,
      "standard translation": 112332,
      "proposed alternatives": 93177,
      "translation work": 125432,
      "limited memory": 64252,
      "translation lowresource": 124917,
      "present research": 89665,
      "particular present": 85442,
      "hindienglish machine": 50489,
      "lowresource language": 65500,
      "especially speech": 38512,
      "suitable training": 115410,
      "applied variety": 7139,
      "variety methods": 130992,
      "extraction web": 42536,
      "manual annotation": 66975,
      "data developed": 25835,
      "automatically augmenting": 10724,
      "augmenting training": 10346,
      "synthetically generated": 116655,
      "generated variants": 47152,
      "ccg supertagging": 15860,
      "task assign": 117901,
      "categories word": 15757,
      "sentence previous": 106001,
      "methods use": 69829,
      "use fixed": 128056,
      "context window": 22324,
      "window sizes": 132713,
      "usually rely": 130449,
      "rely different": 98685,
      "mechanism local": 68009,
      "local contexts": 64912,
      "contexts applying": 22377,
      "applying dropout": 7239,
      "embeddings use": 36002,
      "use approach": 127896,
      "approach demonstrate": 7476,
      "demonstrate stateoftheart": 28868,
      "standard test": 112320,
      "modelling sentence": 72621,
      "pairs treestructured": 83655,
      "recursive neural": 97285,
      "networks sequential": 77762,
      "sequential recurrent": 107248,
      "structure propose": 113937,
      "propose way": 93158,
      "way incorporate": 132093,
      "incorporate attention": 53450,
      "tree topology": 125618,
      "given pair": 48072,
      "encoding sentence": 36724,
      "proposed attentive": 93230,
      "question selection": 95217,
      "outperforms baselines": 82848,
      "baselines achieves": 12349,
      "tasks leveraging": 119253,
      "social norms": 110445,
      "rules govern": 103406,
      "work investigate": 134589,
      "investigate task": 56817,
      "automatically identifying": 10786,
      "discourse goal": 32058,
      "leverage power": 63612,
      "power recurrent": 88650,
      "multimodal information": 75433,
      "propose predictive": 92994,
      "predictive model": 89204,
      "model recognize": 71866,
      "implications work": 51999,
      "empirical exploration": 36168,
      "skip connections": 109979,
      "sequential tagging": 107256,
      "paper empirically": 83889,
      "empirically explore": 36232,
      "explore effects": 41544,
      "connections stacked": 21149,
      "stacked bidirectional": 112122,
      "lstms sequential": 65716,
      "tagging investigate": 117395,
      "lstm cells": 65608,
      "internal states": 56173,
      "comprehensive experiments": 20262,
      "furthermore observe": 46195,
      "observe using": 81215,
      "using gated": 129690,
      "based novel": 11902,
      "successfully train": 115194,
      "train deep": 122919,
      "models obtain": 73654,
      "comparable results": 19170,
      "extensively used": 41971,
      "gain insights": 46345,
      "logs contain": 65056,
      "signal used": 108694,
      "used learn": 128607,
      "information syntactic": 55023,
      "test hypothesis": 120460,
      "labels derived": 58588,
      "auxiliary task": 10886,
      "task multitask": 118425,
      "results shallow": 102166,
      "data come": 25746,
      "distinct sources": 32543,
      "models significantly": 74038,
      "trained text": 123311,
      "complex ideas": 19818,
      "humans variety": 51119,
      "contexts research": 22424,
      "focus designing": 44750,
      "collaborate humans": 18352,
      "abstract propose": 1779,
      "situated language": 109884,
      "robust language": 103042,
      "language agents": 58820,
      "agents able": 4319,
      "prediction propose": 89109,
      "neuralnetwork based": 78738,
      "prediction network": 89086,
      "network designed": 77224,
      "similarity conjuncts": 109211,
      "coordination phrase": 23265,
      "coherent sentences": 18324,
      "sentences modeling": 106399,
      "lstm networks": 65650,
      "trained solely": 123286,
      "annotations treebank": 6474,
      "external resources": 42038,
      "improvements predicting": 52897,
      "improvement previous": 52745,
      "fast scalable": 43151,
      "using opensource": 129991,
      "software developed": 110517,
      "nlp community": 79587,
      "commercial use": 18842,
      "decoding efficient": 28088,
      "efficient utilization": 35129,
      "paper reexamine": 84396,
      "decoder implementation": 28021,
      "speed scalability": 111877,
      "dropin replacement": 34173,
      "times faster": 122172,
      "estimation using": 38670,
      "submission shared": 114735,
      "wordphrase level": 133791,
      "level quality": 63493,
      "translation wmt16": 125427,
      "task predict": 118541,
      "predict given": 88889,
      "translation given": 124840,
      "approach word": 8019,
      "model rnnlm": 71936,
      "rnnlm architecture": 102953,
      "different natural": 31277,
      "space language": 111011,
      "modeling different": 72417,
      "different nlp": 31300,
      "nlp problems": 79671,
      "predicts label": 89217,
      "translated text": 124568,
      "phrase level": 87354,
      "use output": 128185,
      "output word": 83151,
      "nmt new": 79931,
      "progress recent": 92179,
      "years recent": 135292,
      "nmt generally": 79870,
      "produces fluent": 91994,
      "fluent inadequate": 44709,
      "inadequate translations": 53199,
      "tu et": 125921,
      "al 2016b": 4618,
      "al 2016a": 4617,
      "al 2017": 4619,
      "contrast conventional": 22690,
      "conventional statistical": 22901,
      "leverage advantages": 63574,
      "advantages models": 4084,
      "models better": 72845,
      "translations work": 125509,
      "propose incorporate": 92721,
      "smt model": 110285,
      "model nmt": 71587,
      "nmt framework": 79869,
      "framework specifically": 45693,
      "decoding step": 28135,
      "generated words": 47154,
      "based decoding": 11625,
      "information nmt": 54805,
      "partial translation": 85300,
      "translation attention": 124651,
      "attention history": 9855,
      "classifier score": 17577,
      "score smt": 104124,
      "gating function": 46534,
      "function combine": 46007,
      "nmt architecture": 79834,
      "architecture endtoend": 8655,
      "endtoend manner": 36916,
      "manner experimental": 66946,
      "translation proposed": 125154,
      "significant consistent": 108742,
      "stateoftheart nmt": 112792,
      "systems multiple": 117008,
      "development neural": 30405,
      "nmt significantly": 79961,
      "quality automatic": 94599,
      "translation sentences": 125239,
      "sentences accurate": 106196,
      "fluent translations": 44723,
      "systems cases": 116774,
      "completely different": 19780,
      "especially case": 38436,
      "translation shown": 125252,
      "shown significant": 108522,
      "significant gains": 108757,
      "used example": 128529,
      "input target": 55449,
      "language neural": 59712,
      "hypothesis using": 51281,
      "output phrasebased": 83105,
      "translation pbmt": 125104,
      "output source": 83123,
      "sentence evaluate": 105859,
      "evaluate technique": 38935,
      "able outperform": 1667,
      "baseline neural": 12270,
      "mt bleu": 74961,
      "analyzed influence": 6026,
      "influence quality": 54316,
      "initial final": 55215,
      "final result": 44123,
      "pure neural": 94410,
      "systems online": 117030,
      "demonstration neural": 28987,
      "development recently": 30415,
      "production systems": 92062,
      "nmt engines": 79860,
      "replace existing": 98929,
      "systems large": 116970,
      "number training": 80985,
      "training process": 123793,
      "systems usually": 117225,
      "release online": 98465,
      "covering large": 24544,
      "large variety": 61322,
      "12 languages": 255,
      "32 language": 929,
      "explore different": 41539,
      "framework data": 45479,
      "data preparation": 26260,
      "architecture additional": 8610,
      "discuss evaluation": 32250,
      "evaluation methodology": 39273,
      "methodology present": 69266,
      "present findings": 89495,
      "finally outline": 44211,
      "outline work": 82636,
      "ultimate goal": 126554,
      "build competitive": 14755,
      "generic translation": 47812,
      "translation aim": 124632,
      "collaborative framework": 18361,
      "foster research": 45338,
      "research efforts": 100483,
      "industry usecase": 54065,
      "integrated real": 55780,
      "technology allow": 120031,
      "allow build": 5076,
      "build translation": 14814,
      "translation engines": 124790,
      "outperforming current": 82806,
      "systems chinese": 116779,
      "chinese restaurant": 16814,
      "cognate clustering": 18257,
      "approach motivated": 7716,
      "process purpose": 91556,
      "approach yields": 8022,
      "yields similar": 135435,
      "similar results": 109138,
      "process fast": 91485,
      "applied language": 7081,
      "language family": 59042,
      "syntactic transfer": 116493,
      "transfer limited": 124145,
      "simple effective": 109402,
      "effective method": 34703,
      "method crosslingual": 68752,
      "scenario large": 103826,
      "translation data": 124732,
      "available method": 11046,
      "steps method": 113323,
      "method deriving": 68766,
      "crosslingual word": 25043,
      "used multilingual": 128642,
      "information target": 55028,
      "language source": 60104,
      "method integrating": 68918,
      "integrating steps": 55803,
      "projection method": 92236,
      "experiments improvements": 40959,
      "stateoftheart languages": 112687,
      "work setting": 134796,
      "source translation": 110859,
      "smaller corpus": 110229,
      "europarl corpus": 38764,
      "work results": 134786,
      "data additional": 25575,
      "improvements results": 52906,
      "results 38": 101487,
      "dependencies corpora": 29071,
      "byte pair": 14997,
      "pair encoding": 83427,
      "using byte": 129421,
      "bpe units": 14542,
      "languages compare": 60453,
      "currently best": 25400,
      "sequences basic": 107114,
      "syllables used": 116310,
      "used languages": 128605,
      "languages writing": 60983,
      "extensive experimentation": 41890,
      "spanning multiple": 111189,
      "multiple language": 75588,
      "output propose": 83110,
      "model hierarchical": 71281,
      "encoder takes": 36578,
      "individual characters": 53901,
      "characters input": 16609,
      "hierarchical representation": 50000,
      "representation character": 99187,
      "character encoder": 16429,
      "reduces computational": 97380,
      "decoder model": 28031,
      "common words": 18949,
      "single embedding": 109730,
      "words names": 134071,
      "character character": 16421,
      "systems opendomain": 117033,
      "humancomputer conversation": 51005,
      "attracted attention": 10142,
      "attention field": 9837,
      "field nlp": 43973,
      "opendomain conversation": 81958,
      "usually requires": 130452,
      "datadriven approaches": 26657,
      "systems search": 117127,
      "utterance called": 130591,
      "generative approaches": 47717,
      "approaches typically": 8385,
      "typically based": 126413,
      "networks rnns": 77743,
      "utterances paper": 130659,
      "novel ensemble": 80561,
      "systems open": 117031,
      "addition original": 3457,
      "model aware": 70719,
      "new candidate": 78818,
      "results ensemble": 101759,
      "ensemble outperforms": 37617,
      "outperforms single": 82983,
      "single large": 109750,
      "comparing traditional": 19521,
      "sequencetosequence models": 107177,
      "encoderdecoder neural": 36609,
      "models compare": 72934,
      "compare wellknown": 19316,
      "established methods": 38603,
      "methods represent": 69726,
      "represent different": 99104,
      "traditional approaches": 122793,
      "approaches applied": 8062,
      "monotone sequencetosequence": 74640,
      "sequencetosequence tasks": 107210,
      "ocr postcorrection": 81528,
      "spelling correction": 111901,
      "practical relevance": 88709,
      "relevance various": 98520,
      "research fields": 100513,
      "fields including": 43998,
      "digital humanities": 31723,
      "text correction": 120838,
      "recognition investigate": 96897,
      "approaches adapt": 8046,
      "tasks perform": 119377,
      "perform comparison": 85967,
      "models entity": 73162,
      "representations entity": 99632,
      "entities short": 37868,
      "search query": 104343,
      "entity description": 37919,
      "goal develop": 48348,
      "effective model": 34711,
      "model make": 71491,
      "related entities": 97858,
      "entities similar": 37869,
      "similar query": 109128,
      "propose kinds": 92735,
      "relationship entity": 98294,
      "analyze strengths": 6008,
      "learning strategy": 63060,
      "validate methods": 130725,
      "methods public": 69701,
      "public datasets": 94248,
      "datasets contain": 27380,
      "movies tv": 74916,
      "indicate proposed": 53849,
      "methods adapt": 69294,
      "adapt different": 3173,
      "types entity": 126278,
      "outperform current": 82702,
      "based keyword": 11784,
      "word2vec models": 133682,
      "methods trained": 69810,
      "easily extended": 34449,
      "key component": 57551,
      "paper demonstrates": 83804,
      "demonstrates neural": 28951,
      "neural sequencetosequence": 78675,
      "art close": 9039,
      "close state": 17833,
      "make machine": 66697,
      "new arabic": 78800,
      "dataset trained": 27241,
      "trained models": 123202,
      "novel architecture": 80496,
      "architecture machine": 8689,
      "machine reading": 65849,
      "noisy sources": 80123,
      "sources order": 110914,
      "text machine": 121096,
      "information expressed": 54553,
      "expressed human": 41714,
      "hypotheses single": 51259,
      "single coherent": 109715,
      "global constraints": 48228,
      "domain work": 33698,
      "work focus": 134531,
      "focus task": 44829,
      "clusters related": 17969,
      "related news": 97885,
      "articles labels": 9194,
      "previous machine": 90421,
      "values occur": 130798,
      "occur frequently": 81498,
      "missing incorrect": 70327,
      "model noisy": 71590,
      "noisy nature": 80113,
      "nature data": 76652,
      "issues models": 57060,
      "achieve improvement": 2491,
      "f1 previous": 42671,
      "work despite": 134470,
      "despite using": 29740,
      "using far": 129666,
      "factor graph": 42858,
      "graph constraints": 48772,
      "belief propagation": 12688,
      "resulting relative": 101466,
      "previous stateoftheart": 90469,
      "german twitter": 47923,
      "despite substantial": 29733,
      "substantial progress": 114874,
      "developing new": 30354,
      "new sentiment": 79129,
      "methods english": 69476,
      "task transferring": 118806,
      "approaches languages": 8206,
      "open paper": 81920,
      "problem systematically": 91257,
      "common english": 18871,
      "algorithms applied": 4829,
      "applied directly": 7063,
      "german data": 47884,
      "data evaluate": 25906,
      "annotated tweets": 6254,
      "domain social": 33642,
      "media experiments": 68112,
      "outperform automatic": 82687,
      "macroaveraged f1score": 66363,
      "techniques produce": 119956,
      "produce better": 91876,
      "compared corpusbased": 19353,
      "approaches best": 8086,
      "document vector": 33108,
      "tasks document": 119070,
      "document commonly": 32976,
      "term frequencyinverse": 120219,
      "frequencyinverse document": 45855,
      "tfidf feature": 121760,
      "ignores word": 51681,
      "word orders": 133387,
      "words document": 133923,
      "important nlp": 52196,
      "proposes novel": 93609,
      "novel distributed": 80547,
      "distributed vector": 32630,
      "representation document": 99214,
      "document simple": 33081,
      "rnnlm long": 102954,
      "documents task": 33300,
      "represent document": 99105,
      "new document": 78886,
      "believe new": 12700,
      "vectors capture": 131411,
      "capture highlevel": 15310,
      "sequential information": 107229,
      "information documents": 54507,
      "documents current": 33216,
      "current document": 25278,
      "document representations": 33067,
      "representations fail": 99654,
      "fail capture": 42952,
      "vectors evaluated": 131433,
      "classification documents": 17192,
      "brown corpus": 14721,
      "artificially created": 9272,
      "classification performances": 17332,
      "performances compared": 86874,
      "compared performance": 19408,
      "performance tfidf": 86801,
      "model paragraph": 71701,
      "performance faster": 86379,
      "faster decoding": 43171,
      "common effective": 18867,
      "effective way": 34777,
      "way train": 132136,
      "train translation": 123047,
      "languages consider": 60464,
      "sentences resulting": 106481,
      "time increase": 122037,
      "data format": 25956,
      "format representing": 45208,
      "investigate different": 56742,
      "different choices": 31036,
      "decoder parameters": 28042,
      "parameters data": 84742,
      "impact decoding": 51865,
      "translation accuracy": 124617,
      "improve decoding": 52369,
      "time little": 122046,
      "little impact": 64810,
      "impact translation": 51897,
      "context dependent": 22052,
      "automatic response": 10643,
      "generation building": 47327,
      "building chatbot": 14827,
      "drawn lot": 34134,
      "lot attention": 65316,
      "limited understanding": 64296,
      "consider linguistic": 21213,
      "linguistic context": 64453,
      "task challenging": 117956,
      "study social": 114521,
      "conversation data": 22945,
      "crawled web": 24598,
      "identifying context": 51592,
      "weak supervision": 132183,
      "supervision propose": 115908,
      "using long": 129831,
      "lstm network": 65649,
      "classifier method": 17557,
      "learning unified": 63133,
      "framework experimental": 45533,
      "baseline methods": 12248,
      "methods accuracy": 69276,
      "commonsense inference": 18987,
      "established discourse": 38601,
      "rarely stated": 95763,
      "stated explicitly": 112543,
      "explicitly propose": 41389,
      "based extension": 11706,
      "framework extracting": 45541,
      "extracting commonsense": 42204,
      "commonsense knowledge": 18990,
      "corpora used": 23611,
      "used construct": 128449,
      "construct dataset": 21624,
      "task train": 118798,
      "train neural": 122977,
      "sequencetosequence model": 107165,
      "model dataset": 70949,
      "dataset use": 27254,
      "use score": 128257,
      "generate possible": 46989,
      "previously established": 90596,
      "established datasets": 38600,
      "annotation protocol": 6362,
      "order analyze": 82278,
      "polysemous word": 88025,
      "vectorspace word": 131501,
      "representations paraphrases": 99801,
      "bring noise": 14632,
      "prior works": 90766,
      "works use": 134987,
      "overall learning": 83235,
      "equally paper": 38231,
      "method improves": 68888,
      "improves word": 53062,
      "word makes": 133355,
      "makes approach": 66755,
      "easier use": 34425,
      "conventional methods": 22889,
      "experiments empirical": 40928,
      "study vietnamese": 114550,
      "vietnamese dependency": 131674,
      "presents empirical": 89843,
      "comparison different": 19536,
      "results neural": 101983,
      "parsers perform": 85038,
      "better traditional": 13745,
      "report highest": 99004,
      "labeled attachment": 58424,
      "score las": 104089,
      "unlabeled attachment": 127369,
      "cogalexv shared": 18254,
      "distributional measures": 32708,
      "semantically related": 105372,
      "pairs according": 83480,
      "according semantic": 2001,
      "solid performance": 110542,
      "poor performance": 88060,
      "performance second": 86693,
      "second subtask": 104458,
      "scores reported": 104199,
      "multiple semantic": 75681,
      "paragraph vectors": 84574,
      "models called": 72872,
      "vector used": 131398,
      "learn stateoftheart": 62155,
      "representations documents": 99607,
      "documents inspired": 33241,
      "inspired work": 55585,
      "models simple": 74050,
      "binary codes": 14126,
      "vectors outperform": 131457,
      "using fewer": 129676,
      "fewer bits": 43886,
      "learning settings": 63018,
      "codes inferred": 18205,
      "experiments indicate": 40964,
      "domainspecific documents": 33898,
      "documents finally": 33231,
      "model simultaneously": 72049,
      "simultaneously learns": 109684,
      "representations model": 99755,
      "short list": 108215,
      "highly relevant": 50339,
      "documents large": 33248,
      "large document": 61084,
      "document collection": 32974,
      "general class": 46640,
      "class language": 17043,
      "models treat": 74248,
      "architecture allows": 8615,
      "models create": 72997,
      "mentions entities": 68451,
      "entities attributes": 37747,
      "recipe generation": 96788,
      "internal state": 56172,
      "models aware": 72800,
      "words experiments": 133949,
      "experiments tasks": 41170,
      "tasks shows": 119500,
      "model variants": 72295,
      "deterministic attention": 30161,
      "latent attention": 61574,
      "program synthesis": 92115,
      "language descriptions": 58946,
      "problem work": 91288,
      "work consider": 134432,
      "textual descriptions": 121690,
      "devise novel": 30464,
      "architecture task": 8757,
      "train endtoend": 122932,
      "endtoend specifically": 36965,
      "weights words": 132387,
      "twostage process": 126164,
      "better leveraging": 13620,
      "relevant parts": 98567,
      "reduces error": 97383,
      "prior art": 90692,
      "propose oneshot": 92981,
      "oneshot learning": 81715,
      "existing dataset": 40103,
      "dataset demonstrate": 26849,
      "scenario outperforms": 103828,
      "outperforms original": 82929,
      "closing gap": 17885,
      "gap model": 46466,
      "data neural": 26163,
      "nmt achieved": 79821,
      "achieved remarkable": 2681,
      "remarkable progress": 98839,
      "progress past": 92171,
      "years suffers": 135306,
      "major drawback": 66555,
      "generated nmt": 47100,
      "systems lack": 116965,
      "widely observed": 132549,
      "observed nmt": 81226,
      "nmt tends": 79988,
      "translate source": 124542,
      "framework nmt": 45630,
      "incorporated nmt": 53499,
      "manages reconstruct": 66909,
      "reconstruct input": 97108,
      "input source": 55439,
      "sentence hidden": 105897,
      "hidden layer": 49892,
      "output target": 83133,
      "information source": 54991,
      "proposed framework": 93282,
      "improves adequacy": 52946,
      "adequacy nmt": 3826,
      "nmt output": 79937,
      "achieves superior": 2918,
      "translation result": 125221,
      "result stateoftheart": 101409,
      "nmt statistical": 79965,
      "generation problem": 47561,
      "representing key": 100052,
      "events large": 39583,
      "large pool": 61209,
      "documents paper": 33263,
      "lack standard": 58745,
      "standard dataset": 112220,
      "problem present": 91167,
      "make publicly": 66713,
      "available new": 11053,
      "entities entity": 37775,
      "entity provide": 37989,
      "provide gold": 93838,
      "timeline set": 122156,
      "articles propose": 9207,
      "rouge evaluation": 103292,
      "dataset showing": 27199,
      "results outperform": 102017,
      "tv series": 126006,
      "online video": 81816,
      "platforms like": 87667,
      "reviews important": 102612,
      "viewers producers": 131718,
      "rely heavily": 98707,
      "quality result": 94777,
      "automatically classifying": 10735,
      "classifying reviews": 17675,
      "according different": 1974,
      "different requirements": 31385,
      "popular research": 88123,
      "research topic": 100641,
      "daily life": 25509,
      "life paper": 63979,
      "successfully trained": 115195,
      "trained generic": 123153,
      "results showed": 102171,
      "showed promising": 108383,
      "generalization different": 46774,
      "elements texts": 35286,
      "texts standard": 121621,
      "data ner": 26161,
      "ner applied": 77020,
      "persons locations": 87156,
      "domain dependent": 33502,
      "used entity": 128519,
      "entity categories": 37912,
      "general set": 46713,
      "set named": 107502,
      "entities usually": 37893,
      "persons organizations": 87157,
      "ner data": 77034,
      "results discussion": 101732,
      "development web": 30436,
      "web collection": 132225,
      "finnish newspapers": 44559,
      "collection contains": 18465,
      "use material": 128138,
      "ocr errors": 81526,
      "ner tagger": 77085,
      "rulebased tagger": 103375,
      "results limited": 101889,
      "research group": 100527,
      "aalto university": 1461,
      "research reports": 100615,
      "published large": 94358,
      "scale results": 103753,
      "results ner": 101982,
      "collection results": 18488,
      "ner results": 77075,
      "results languages": 101879,
      "noisy data": 80098,
      "detection natural": 29999,
      "spatiotemporal patterns": 111274,
      "paper generalize": 83966,
      "work considering": 134433,
      "particle filtering": 85372,
      "algorithm exploits": 4733,
      "algorithm extensive": 4736,
      "improvement target": 52773,
      "dataless classification": 26692,
      "approach classify": 7420,
      "documents language": 33246,
      "language english": 59005,
      "label space": 58416,
      "space text": 111070,
      "data approach": 25614,
      "approach crosslingual": 7467,
      "mapping english": 67128,
      "category description": 15784,
      "wikipediabased semantic": 132690,
      "use target": 128312,
      "language wikipedia": 60342,
      "performance suffer": 86771,
      "wikipedia target": 132681,
      "language small": 60099,
      "small paper": 110179,
      "focus languages": 44779,
      "use wordlevel": 128366,
      "convert documents": 23094,
      "thousands languages": 121925,
      "languages develop": 60495,
      "ranking algorithm": 95662,
      "algorithm makes": 4768,
      "language similarity": 60095,
      "similarity metrics": 109270,
      "automatically select": 10826,
      "select good": 104699,
      "possible new": 88414,
      "new recurrent": 79098,
      "neural crf": 77881,
      "crf learning": 24752,
      "features conditional": 43418,
      "random field": 95475,
      "field crf": 43942,
      "achieved success": 2709,
      "models gain": 73267,
      "dense features": 29014,
      "features globally": 43528,
      "globally normalized": 48288,
      "models mainly": 73543,
      "undirected graphs": 127047,
      "graphs edge": 48947,
      "features prove": 43676,
      "prove important": 93699,
      "important crf": 52140,
      "prediction work": 89154,
      "work introduce": 134574,
      "features makes": 43604,
      "compare model": 19256,
      "model different": 70993,
      "different neural": 31290,
      "experiments model": 41011,
      "chinese word": 16834,
      "network named": 77365,
      "recognition chinese": 96830,
      "chinese social": 16820,
      "media focus": 68117,
      "focus named": 44792,
      "ner chinese": 77027,
      "unlabeled text": 127411,
      "text quite": 121226,
      "quite limited": 95402,
      "limited labelled": 64246,
      "propose semisupervised": 93032,
      "network advantage": 77133,
      "advantage traditional": 4063,
      "traditional methods": 122836,
      "methods ner": 69635,
      "transition probability": 124494,
      "model bridge": 70784,
      "label accuracy": 58382,
      "accuracy fscore": 2166,
      "construct model": 21639,
      "directly trained": 31909,
      "accuracy propose": 2248,
      "propose integrated": 92726,
      "method train": 69191,
      "integrated model": 55778,
      "result knowledge": 101379,
      "knowledge enhanced": 57899,
      "hybrid neural": 51187,
      "text matching": 121102,
      "brings big": 14643,
      "big challenge": 13989,
      "challenge semantic": 16097,
      "complicated semantic": 19958,
      "tackle challenge": 117285,
      "consider using": 21237,
      "using prior": 130056,
      "knowledge help": 57998,
      "help identify": 49736,
      "identify useful": 51572,
      "filter noise": 44072,
      "propose knowledge": 92736,
      "model fuses": 71214,
      "knowledge word": 58233,
      "representations knowledge": 99710,
      "sentences given": 106335,
      "recurrent units": 97270,
      "units gru": 127243,
      "processed convolutional": 91594,
      "network generate": 77266,
      "generate high": 46949,
      "matching features": 67404,
      "matching score": 67431,
      "multilayer perceptron": 75180,
      "model extends": 71148,
      "extends existing": 41825,
      "matching words": 67441,
      "words local": 134039,
      "global context": 48229,
      "results extensive": 101786,
      "experiments public": 41089,
      "public data": 94243,
      "sets question": 107702,
      "matching models": 67416,
      "models particularly": 73718,
      "pairs long": 83577,
      "using pointer": 130020,
      "pointer network": 87824,
      "network sentence": 77425,
      "nlp previous": 79669,
      "using pairwise": 129998,
      "models incorporate": 73384,
      "contextual sentence": 22497,
      "sentence information": 105907,
      "information addition": 54364,
      "addition error": 3427,
      "using pipeline": 130017,
      "propose endtoend": 92649,
      "approach address": 7333,
      "problem uses": 91281,
      "uses pointer": 129263,
      "network ptrnet": 77399,
      "alleviate error": 5034,
      "error propagation": 38315,
      "propagation problem": 92397,
      "utilize contextual": 130499,
      "information experimental": 54549,
      "results effectiveness": 101743,
      "model source": 72070,
      "paper available": 83752,
      "scarce resources": 103794,
      "national library": 76213,
      "digitized historical": 31733,
      "historical newspapers": 50531,
      "published finland": 94356,
      "contains approximately": 21817,
      "collection consists": 18463,
      "billion words": 14082,
      "words national": 134072,
      "web service": 132256,
      "service known": 107323,
      "freely downloadable": 45788,
      "university helsinki": 127353,
      "text resources": 121260,
      "style information": 114580,
      "test collection": 120428,
      "collection produced": 18487,
      "important topic": 52282,
      "general usability": 46727,
      "method assess": 68646,
      "quality large": 94701,
      "large collections": 61052,
      "used approximate": 128394,
      "quality paper": 94744,
      "discusses different": 32309,
      "different corpus": 31067,
      "style methods": 114584,
      "methods approximate": 69319,
      "overall lexical": 83236,
      "collection methods": 18480,
      "error rates": 38338,
      "frequency analysis": 45837,
      "analysis words": 5939,
      "words comparisons": 133869,
      "data aim": 25587,
      "quality analysis": 94590,
      "twofold firstly": 126141,
      "present state": 89715,
      "methods build": 69356,
      "build compact": 14754,
      "quality assessment": 94596,
      "assessment new": 9523,
      "post correction": 88452,
      "different analyses": 30990,
      "analyses word": 5474,
      "word document": 132999,
      "network approaches": 77138,
      "approaches data": 8113,
      "representation fundamental": 99245,
      "fundamental task": 46118,
      "learning representation": 62959,
      "representation data": 99201,
      "data affects": 25585,
      "long history": 65075,
      "history representation": 50566,
      "researchers aim": 100676,
      "designing better": 29637,
      "better features": 13588,
      "features specific": 43738,
      "tasks recently": 119443,
      "development deep": 30380,
      "brought new": 14714,
      "various domains": 131084,
      "processing widely": 91856,
      "used feature": 128549,
      "feature representation": 43302,
      "bagofwords model": 11377,
      "model data": 70947,
      "problem word": 91286,
      "order information": 82348,
      "information features": 54614,
      "tagging complex": 117378,
      "syntax features": 116535,
      "fit specific": 44582,
      "thesis focuses": 121852,
      "focuses word": 44919,
      "systems present": 117070,
      "model generating": 71243,
      "embeddings make": 35799,
      "comprehensive comparisons": 20254,
      "comparisons existing": 19591,
      "existing word": 40333,
      "models terms": 74175,
      "models skipgram": 74059,
      "experiments analyze": 40771,
      "analyze key": 5981,
      "key points": 57592,
      "including model": 53324,
      "construction training": 21698,
      "design evaluate": 29537,
      "embeddings types": 35995,
      "types tasks": 126366,
      "tasks argue": 118941,
      "embedding second": 35484,
      "chinese character": 16742,
      "character word": 16476,
      "introduce joint": 56440,
      "existing document": 40118,
      "including recursive": 53359,
      "models present": 73769,
      "model recurrent": 71871,
      "terms recognition": 120370,
      "recognition methods": 96908,
      "automatically recognized": 10822,
      "used various": 128841,
      "domainspecific texts": 33919,
      "methods best": 69344,
      "best suited": 13460,
      "particular settings": 85451,
      "believe main": 12698,
      "main reasons": 66457,
      "reasons lack": 96329,
      "order address": 82271,
      "issues present": 57066,
      "methods automatic": 69327,
      "automatic terminology": 10691,
      "pipeline text": 87557,
      "document preprocessing": 33052,
      "ranking highly": 95670,
      "highly scalable": 50343,
      "support automatic": 115958,
      "open datasets": 81890,
      "datasets average": 27326,
      "experimental comparison": 40526,
      "method demonstrates": 68763,
      "best average": 13307,
      "datasets available": 27325,
      "best methods": 13366,
      "methods exploiting": 69496,
      "dependency annotations": 29132,
      "minimum spanning": 70218,
      "typically require": 126456,
      "require large": 100154,
      "indirect supervision": 53888,
      "rules help": 103408,
      "direct supervision": 31804,
      "annotations provides": 6458,
      "provides strong": 94082,
      "adapt unsupervised": 3199,
      "learn partial": 62108,
      "24 hours": 800,
      "annotation obtain": 6344,
      "17 absolute": 399,
      "unlabeled dependency": 127394,
      "scores english": 104169,
      "spanish respectively": 111171,
      "respectively compared": 101132,
      "compared parser": 19406,
      "using universal": 130338,
      "aims provide": 4564,
      "modeling representation": 72533,
      "data digital": 25851,
      "databases constructed": 26650,
      "constructed manually": 21666,
      "manually semiautomatically": 67077,
      "propose systematic": 93103,
      "set modeling": 107498,
      "modeling principles": 72514,
      "existing future": 40137,
      "history words": 50574,
      "words meanings": 134050,
      "language digital": 58965,
      "data focus": 25952,
      "focus identifying": 44771,
      "offered tei": 81575,
      "tei guidelines": 120059,
      "representation lexical": 99316,
      "semisupervised approach": 105591,
      "approach identify": 7613,
      "context consisting": 22036,
      "consisting english": 21451,
      "translated language": 124553,
      "language expression": 59036,
      "meaning automatically": 67618,
      "expressions corpus": 41750,
      "corpus extraction": 23799,
      "extraction pipeline": 42433,
      "statistical approaches": 113081,
      "employ different": 36258,
      "detect different": 29800,
      "pos tagged": 88217,
      "language initially": 59207,
      "regular expression": 97725,
      "narrow search": 76192,
      "space certain": 110979,
      "patterns like": 85757,
      "verbs word": 131565,
      "word sequences": 133547,
      "series linguistic": 107276,
      "false positives": 43044,
      "using wordnet": 130389,
      "order detect": 82308,
      "features constituent": 43422,
      "words statistical": 134238,
      "methods applied": 69312,
      "applied detect": 7058,
      "automatically extracted": 10770,
      "definition modeling": 28525,
      "learning define": 62491,
      "embeddings natural": 35822,
      "language distributed": 58974,
      "words shown": 134220,
      "capture lexical": 15327,
      "demonstrated effectiveness": 28913,
      "tasks tasks": 119550,
      "tasks evaluate": 119102,
      "evaluate lexical": 38848,
      "paper study": 84441,
      "study possible": 114472,
      "possible utilize": 88444,
      "utilize distributed": 130503,
      "representations generate": 99667,
      "definitions words": 28540,
      "words direct": 133916,
      "modeling task": 72558,
      "task generating": 118240,
      "generating definition": 47210,
      "embedding present": 35469,
      "present definition": 89436,
      "definition model": 28524,
      "model architectures": 70684,
      "models multiple": 73593,
      "multiple data": 75531,
      "dependencies word": 29125,
      "better characterlevel": 13535,
      "convolution layer": 23139,
      "layer designed": 61708,
      "wordlevel embeddings": 133733,
      "embeddings finally": 35722,
      "model provide": 71822,
      "polysemous words": 88026,
      "alternative method": 5235,
      "instead vector": 55694,
      "representations estimated": 99635,
      "use previous": 128212,
      "work uses": 134874,
      "probabilistic method": 90817,
      "robust small": 103071,
      "new neural": 79050,
      "using lexicon": 129810,
      "continuous bagofwords": 22615,
      "way alleviates": 132057,
      "proposed neural": 93506,
      "trained using": 123324,
      "using negative": 129945,
      "probabilities target": 90851,
      "words distinguishing": 133919,
      "compare proposed": 19284,
      "improving previous": 53146,
      "semantic tasks": 105326,
      "size corpus": 109917,
      "recognition engine": 96865,
      "used speech": 128774,
      "prediction accuracy": 89025,
      "accuracy machine": 2205,
      "larger models": 61371,
      "large model": 61150,
      "memory intensive": 68318,
      "results high": 101821,
      "power consumption": 88630,
      "leads high": 61934,
      "total cost": 122697,
      "data center": 25721,
      "energy efficient": 36997,
      "model size": 72054,
      "negligible loss": 76977,
      "loss prediction": 65293,
      "processing propose": 91777,
      "compressed model": 20297,
      "finally design": 44166,
      "architecture named": 8701,
      "model implemented": 71300,
      "working directly": 134896,
      "3x faster": 1003,
      "titan gpu": 122218,
      "efficiency compared": 35022,
      "cpu gpu": 24576,
      "event dataset": 39504,
      "dataset generation": 26955,
      "terms data": 120302,
      "data acquisition": 25567,
      "acquisition process": 3051,
      "coding text": 18246,
      "improvements opensource": 52889,
      "opensource natural": 82022,
      "content paper": 21916,
      "dataset named": 27047,
      "dataset includes": 26977,
      "collection process": 18485,
      "processing pipeline": 91770,
      "pipeline necessary": 87548,
      "integrated crisis": 55773,
      "crisis early": 24783,
      "early warning": 34397,
      "complementary entity": 19741,
      "recognition knowledge": 96900,
      "reviews product": 102623,
      "reviews contain": 102602,
      "contain lot": 21744,
      "information product": 54876,
      "product features": 92037,
      "opinions important": 82123,
      "work reviewed": 134787,
      "reviewed product": 102590,
      "complementary entities": 19740,
      "ones paper": 81700,
      "problem complementary": 90971,
      "recognition cer": 96826,
      "existing method": 40173,
      "method solve": 69157,
      "unsupervised method": 127665,
      "utilize syntactic": 130533,
      "dependency paths": 29215,
      "knowledge complementary": 57832,
      "entities using": 37892,
      "using general": 129697,
      "unlabeled reviews": 127406,
      "knowledge helps": 57999,
      "method adapt": 68614,
      "improves precision": 53025,
      "cer task": 15928,
      "advantage proposed": 4056,
      "method does": 68783,
      "require labeled": 100152,
      "data training": 26570,
      "experiments popular": 41059,
      "reviews total": 102634,
      "effective discovering": 34664,
      "task inferring": 118295,
      "customer service": 25444,
      "service propose": 107324,
      "probabilistic classifier": 90799,
      "information approach": 54379,
      "chat logs": 16658,
      "chinese ecommerce": 16762,
      "outperforms heuristic": 82909,
      "heuristic baselines": 49872,
      "general knowledge": 46666,
      "knowledge machine": 58058,
      "specific knowledge": 111454,
      "sources external": 110889,
      "linked open": 64646,
      "open data": 81888,
      "data resource": 26353,
      "used modern": 128639,
      "modern nlp": 74407,
      "nlp techniques": 79779,
      "techniques word": 120013,
      "conceptnet knowledge": 20597,
      "graph connects": 48770,
      "labeled edges": 58456,
      "knowledge collected": 57826,
      "represent general": 99110,
      "language improving": 59158,
      "improving natural": 53122,
      "better understand": 13754,
      "use conceptnet": 127953,
      "combined word": 18674,
      "resources wordnet": 101068,
      "results intrinsic": 101866,
      "intrinsic evaluations": 56358,
      "evaluations word": 39491,
      "word relatedness": 133431,
      "applications word": 7038,
      "vectors including": 131442,
      "paragraph sentence": 84571,
      "sentence alignment": 105767,
      "alignment comparable": 4949,
      "corpora parallel": 23555,
      "driven great": 34152,
      "progress field": 92151,
      "simplification sentence": 109591,
      "alignment algorithms": 4944,
      "offer limited": 81567,
      "limited range": 64268,
      "documents address": 33176,
      "problem introducing": 91091,
      "introducing new": 56645,
      "new set": 79132,
      "entities question": 37854,
      "classification using": 17482,
      "using distant": 129590,
      "community question": 19090,
      "information products": 54877,
      "product descriptions": 92034,
      "frequently addressed": 45868,
      "yesno question": 135326,
      "question does": 95152,
      "windows 10": 132718,
      "problem extracting": 91050,
      "yesno questions": 135327,
      "twostage framework": 126160,
      "framework perform": 45651,
      "questions second": 95365,
      "classifier combining": 17531,
      "combining distant": 18723,
      "positive unlabeled": 88345,
      "binary classifier": 14123,
      "classifier second": 17578,
      "answers using": 6748,
      "using human": 129742,
      "human annotated": 50739,
      "data conduct": 25771,
      "conduct experiments": 20846,
      "method effective": 68789,
      "translation pairs": 125084,
      "difficult timeconsuming": 31666,
      "translation attractive": 124655,
      "produce scores": 91925,
      "systems model": 117003,
      "model specific": 72074,
      "agreement long": 4395,
      "long distances": 65069,
      "novel words": 80781,
      "core idea": 23324,
      "specific type": 111511,
      "type error": 126199,
      "largescale data": 61416,
      "wmt englishgerman": 132787,
      "englishgerman translation": 37387,
      "simple rules": 109510,
      "rules report": 103434,
      "results number": 101999,
      "systems recently": 117103,
      "characterlevel nmt": 16579,
      "bytepair encoding": 15004,
      "encoding bpe": 36683,
      "bpe segmentation": 14539,
      "automatic prosodic": 10630,
      "utterance level": 130603,
      "global local": 48249,
      "level usually": 63520,
      "serves represent": 107317,
      "features related": 43695,
      "classes obtained": 17069,
      "bottomup way": 14505,
      "f0 energy": 42642,
      "mean variance": 67614,
      "automatic annotation": 10491,
      "annotation comprises": 6286,
      "extraction unsupervised": 42526,
      "phrase boundaries": 87339,
      "boundaries local": 14516,
      "prosodic event": 93641,
      "rates building": 95816,
      "building robust": 14883,
      "lexicon resource": 63905,
      "polarity lexicons": 87915,
      "labor intensive": 58663,
      "automatically translating": 10836,
      "resourceful languages": 100897,
      "requires indomain": 100279,
      "rely large": 98714,
      "large quantities": 61228,
      "propose replace": 93019,
      "transferring words": 124247,
      "words lexicon": 134035,
      "embeddings aligned": 35550,
      "languages simple": 60876,
      "simple linear": 109456,
      "linear transform": 64382,
      "approach leads": 7668,
      "degradation compared": 28549,
      "compared machine": 19380,
      "classification tweets": 17473,
      "languages neural": 60762,
      "translation address": 124626,
      "address data": 3671,
      "selection problem": 104818,
      "problem statistical": 91248,
      "task new": 118456,
      "new data": 78852,
      "selection method": 104800,
      "classifier present": 17574,
      "method provides": 69083,
      "quality compared": 94613,
      "stateoftheart method": 112699,
      "cross entropy": 24859,
      "exploring different": 41662,
      "different dimensions": 31096,
      "dimensions attention": 31764,
      "uncertainty detection": 126616,
      "detection neural": 30002,
      "networks attention": 77512,
      "attention mechanisms": 9919,
      "detection particular": 30020,
      "introducing external": 56633,
      "attention novel": 9959,
      "novel architectures": 80499,
      "standard approaches": 112186,
      "sequence information": 106961,
      "information compare": 54427,
      "new state": 79146,
      "perform similar": 86071,
      "similar stateoftheart": 109150,
      "model biomedical": 70781,
      "uses large": 129239,
      "set linguistic": 107480,
      "features grammar": 43529,
      "known morphological": 58312,
      "morphological complexity": 74693,
      "linguistics research": 64623,
      "answer question": 6559,
      "grammar context": 48626,
      "aspect mood": 9338,
      "using soft": 130196,
      "soft clustering": 110489,
      "present feature": 89493,
      "overcomes shortcomings": 83305,
      "representations widely": 99976,
      "used text": 128808,
      "semantic contexts": 105021,
      "contexts words": 22436,
      "express complex": 41696,
      "multitopic documents": 75906,
      "multiclass multilabel": 75056,
      "tasks outperform": 119355,
      "liu et": 64840,
      "topic coherence": 122504,
      "achieve significant": 2534,
      "reduction training": 97463,
      "training prediction": 123783,
      "times compared": 122168,
      "best worlds": 13473,
      "performance lower": 86505,
      "lower time": 65449,
      "time space": 122104,
      "space complexity": 110982,
      "bias removal": 13841,
      "score prediction": 104107,
      "prediction text": 89142,
      "text reviews": 121269,
      "reviews recently": 102626,
      "recently gained": 96690,
      "gained lot": 46365,
      "recommendation systems": 97089,
      "systems major": 116992,
      "major problem": 66577,
      "problem models": 91132,
      "presence noise": 89351,
      "scores propose": 104194,
      "simple statistical": 109520,
      "methods remove": 69724,
      "remove noise": 98876,
      "noise improve": 80053,
      "compared methods": 19383,
      "multiple classifiers": 75518,
      "user model": 129013,
      "uses single": 129279,
      "predict review": 88929,
      "empirically evaluate": 36231,
      "reviews dataset": 102605,
      "obtain improved": 81293,
      "text feature": 120949,
      "representations continuous": 99571,
      "language vectors": 60329,
      "models multilingual": 73590,
      "treat language": 125551,
      "predictions language": 89177,
      "contrast propose": 22709,
      "using continuous": 129488,
      "improve inference": 52393,
      "inference language": 54156,
      "language varieties": 60324,
      "seen training": 104545,
      "explore capacity": 41527,
      "capacity multilingual": 15235,
      "relationships languages": 98322,
      "using multinomial": 129918,
      "multinomial logistic": 75471,
      "regression model": 97703,
      "model work": 72329,
      "participation shared": 85365,
      "task detecting": 118076,
      "forum information": 45305,
      "retrieval evaluation": 102391,
      "detection method": 29989,
      "variety features": 130981,
      "sentences pair": 106422,
      "performance evaluated": 86346,
      "evaluated test": 39014,
      "set released": 107562,
      "11 teams": 229,
      "teams participated": 119728,
      "participated shared": 85345,
      "languages hindi": 60620,
      "hindi punjabi": 50471,
      "languages participated": 60791,
      "task1 task2": 118863,
      "languages overall": 60773,
      "overall average": 83214,
      "average performance": 11203,
      "performance including": 86456,
      "second highest": 104412,
      "highest score": 50235,
      "score systems": 104130,
      "networks representation": 77734,
      "representation erasure": 99227,
      "come cost": 18760,
      "interpretability paper": 56227,
      "general methodology": 46677,
      "methodology analyze": 69243,
      "model observing": 71598,
      "effects model": 34994,
      "various parts": 131162,
      "intermediate hidden": 56132,
      "units input": 127250,
      "approaches analyzing": 8060,
      "computing relative": 20526,
      "relative difference": 98348,
      "using reinforcement": 130101,
      "set input": 107467,
      "words order": 134095,
      "models decision": 73017,
      "comprehensive analysis": 20245,
      "analysis multiple": 5709,
      "multiple nlp": 75629,
      "including linguistic": 53311,
      "classification sentencelevel": 17390,
      "sentencelevel sentiment": 106173,
      "analysis document": 5581,
      "document level": 33032,
      "level sentiment": 63503,
      "sentiment aspect": 106684,
      "aspect prediction": 9342,
      "prediction proposed": 89111,
      "proposed methodology": 93408,
      "methodology offers": 69264,
      "offers clear": 81586,
      "explanations neural": 41297,
      "provides way": 94095,
      "way conduct": 132067,
      "unsupervised task": 127732,
      "task supervised": 118767,
      "tasks introduce": 119221,
      "lower layers": 65436,
      "final layer": 44105,
      "task auxiliary": 117918,
      "shows improvements": 108588,
      "percentage points": 85924,
      "media mining": 68127,
      "recent approaches": 96429,
      "approaches sentiment": 8334,
      "capture latent": 15325,
      "properties embeddings": 92448,
      "performance given": 86419,
      "contextual words": 22527,
      "problem exploiting": 91045,
      "taskspecific representations": 119640,
      "representations induced": 99691,
      "embedding subspace": 35508,
      "subspace projection": 114844,
      "property model": 92501,
      "learns suitable": 63238,
      "lexicons model": 63930,
      "baselines low": 12424,
      "low training": 65400,
      "data regimes": 26328,
      "lexiconbased sentiment": 63919,
      "sentiment classifiers": 106723,
      "similar resources": 109137,
      "performances comparable": 86872,
      "comparable supervised": 19186,
      "models shortcut": 74023,
      "stacked rnns": 112131,
      "hard train": 49483,
      "different layers": 31222,
      "common way": 18946,
      "ease training": 34407,
      "architecture propose": 8732,
      "gating mechanism": 46535,
      "lstm cell": 65607,
      "present extensive": 89487,
      "empirical experiments": 36166,
      "makes training": 66812,
      "improves generalization": 52985,
      "propose various": 93155,
      "explore effectiveness": 41543,
      "based architecture": 11521,
      "architecture obtain": 8715,
      "implicit sentiment": 52019,
      "rapid growth": 95719,
      "growth social": 49195,
      "opinions products": 82127,
      "data creates": 25807,
      "need automated": 76783,
      "real time": 96081,
      "opinion extraction": 82088,
      "current research": 25330,
      "research devoted": 100468,
      "devoted study": 30480,
      "subjective sentences": 114702,
      "limited work": 64301,
      "work reported": 134777,
      "sentiment paper": 106764,
      "paper fuzzy": 83964,
      "based knowledge": 11787,
      "knowledge engineering": 57896,
      "engineering model": 37050,
      "model developed": 70986,
      "developed sentiment": 30300,
      "special group": 111363,
      "sentences including": 106352,
      "including change": 53268,
      "rich source": 102787,
      "reviews different": 102606,
      "main conclusion": 66406,
      "conclusion study": 20725,
      "study order": 114462,
      "increase accuracy": 53587,
      "accuracy level": 2197,
      "mining systems": 70261,
      "objective sentences": 81115,
      "topic modeling": 122542,
      "modeling classification": 72391,
      "discrete categories": 32162,
      "loss information": 65279,
      "type labels": 126207,
      "ability understand": 1574,
      "apply machine": 7181,
      "methods short": 69745,
      "narrative text": 76179,
      "text descriptions": 120869,
      "meaningful latent": 67722,
      "classes term": 17078,
      "latent classes": 61576,
      "modeling methods": 72474,
      "methods produce": 69688,
      "produce use": 91944,
      "use topic": 128332,
      "topic distributions": 122511,
      "topics replicate": 122653,
      "nuances linked": 80825,
      "target characteristics": 117588,
      "tools methods": 122455,
      "topics provide": 122652,
      "provide unique": 93946,
      "crosslingual dependency": 24937,
      "truly lowresource": 125859,
      "languages crosslingual": 60475,
      "dependency annotation": 29131,
      "present endtoend": 89470,
      "graphbased neural": 48904,
      "scores directly": 104168,
      "alignments approach": 5008,
      "achieves absolute": 2731,
      "10 languages": 138,
      "languages compared": 60454,
      "previous state": 90465,
      "attention researchers": 9995,
      "researchers field": 100691,
      "field paper": 43975,
      "study nmt": 114455,
      "model persian": 71746,
      "language best": 58859,
      "hyper parameters": 51204,
      "persian dataset": 87078,
      "dataset yields": 27285,
      "modified loss": 74446,
      "enhance word": 37500,
      "new loss": 78997,
      "point improvements": 87805,
      "improvements terms": 52930,
      "score translation": 104136,
      "personalized response": 87148,
      "generation domain": 47376,
      "generation conversational": 47346,
      "learning especially": 62554,
      "framework propose": 45659,
      "propose twophase": 93123,
      "twophase approach": 126153,
      "initialization adaptation": 55234,
      "adaptation model": 3233,
      "evaluation propose": 39337,
      "novel human": 80593,
      "models online": 73662,
      "offline human": 81612,
      "human judgement": 50876,
      "responses generated": 101269,
      "personalized models": 87147,
      "indicates proposed": 53864,
      "generating personalized": 47246,
      "multifaceted evaluation": 75104,
      "evaluation neural": 39307,
      "language directions": 58967,
      "conduct multifaceted": 20884,
      "evaluation compare": 39149,
      "outputs produced": 83179,
      "number dimensions": 80873,
      "effect sentence": 34611,
      "different error": 31131,
      "error categories": 38291,
      "translations produced": 125488,
      "produced neural": 91961,
      "systems considerably": 116807,
      "considerably different": 21268,
      "fluent accurate": 44703,
      "terms word": 120407,
      "systems accurate": 116709,
      "inflected forms": 54280,
      "long sentences": 65089,
      "network distinguishing": 77227,
      "task achieve": 117831,
      "high performance": 50095,
      "performance nlp": 86573,
      "systems notoriously": 117027,
      "patternbased methods": 85724,
      "relations paper": 98231,
      "lexicosyntactic patterns": 63942,
      "patterns syntactic": 85783,
      "addition lexical": 3447,
      "information successfully": 55014,
      "classification experiments": 17204,
      "performance prior": 86622,
      "parsing universal": 85272,
      "dependencies training": 29118,
      "training propose": 123799,
      "dependencies ud": 29121,
      "rules features": 103402,
      "leaf nodes": 61976,
      "parser requires": 84992,
      "requires training": 100326,
      "linguistically sound": 64598,
      "crosslingual parsing": 24990,
      "used baseline": 128406,
      "robust domain": 103031,
      "change languages": 16364,
      "languages generating": 60607,
      "generating highquality": 47225,
      "models sequencetosequence": 74010,
      "models applied": 72757,
      "conversation response": 22968,
      "problem source": 91238,
      "conversation history": 22956,
      "history target": 50571,
      "generation long": 47463,
      "long informative": 65076,
      "informative coherent": 55129,
      "coherent diverse": 18316,
      "diverse responses": 32838,
      "hard task": 49481,
      "task work": 118854,
      "focus single": 44818,
      "single turn": 109807,
      "selfattention decoder": 104876,
      "responses propose": 101291,
      "propose practical": 92993,
      "scaling large": 103773,
      "datasets introduce": 27530,
      "introduce stochastic": 56543,
      "trained combined": 123095,
      "combined data": 18650,
      "evaluation studies": 39409,
      "compared baseline": 19328,
      "baseline sequencetosequence": 12307,
      "models explicit": 73207,
      "produces better": 91987,
      "better responses": 13700,
      "user generated": 128985,
      "generated comments": 47062,
      "largescale social": 61509,
      "media news": 68130,
      "popular news": 88111,
      "varying quality": 131267,
      "higher probability": 50197,
      "original article": 82503,
      "approach presented": 7811,
      "corpus adding": 23644,
      "adding new": 3389,
      "accuracy time": 2303,
      "time deep": 122011,
      "approaches usually": 8396,
      "usually use": 130464,
      "use preprocessing": 128202,
      "information additional": 54365,
      "implementation work": 51953,
      "use deep": 127981,
      "learn simple": 62150,
      "choice motivated": 16891,
      "different networks": 31289,
      "networks compared": 77541,
      "standard multilayer": 112269,
      "convolutional network": 23162,
      "network experiments": 77248,
      "experiments czech": 40862,
      "networks significantly": 77769,
      "baseline method": 12247,
      "uses rich": 129273,
      "rich set": 102785,
      "set features": 107443,
      "best results": 13435,
      "2016 evaluation": 607,
      "newly emerged": 79271,
      "arabicenglish language": 8571,
      "trained strong": 123294,
      "adaptation techniques": 3275,
      "using fine": 129677,
      "mt trained": 75001,
      "data different": 25841,
      "different genres": 31164,
      "applying ensemble": 7243,
      "ensemble models": 37609,
      "beat strong": 12617,
      "significant bleu": 108728,
      "similar gains": 109093,
      "combination phrasebased": 18577,
      "nmt outputs": 79938,
      "outputs neural": 83176,
      "understanding nlu": 126906,
      "nlu tasks": 79815,
      "parsing text": 85263,
      "meaningful chunks": 67711,
      "sentence current": 105817,
      "current deep": 25272,
      "network dnn": 77228,
      "dnn based": 32931,
      "consider tasks": 21232,
      "tasks sequence": 119494,
      "labeling problem": 58525,
      "labels paper": 58628,
      "propose alternative": 92543,
      "alternative approach": 5222,
      "approach investigating": 7650,
      "models chunk": 72903,
      "filling tasks": 44062,
      "tasks deep": 119029,
      "deep memory": 28335,
      "consider task": 21231,
      "entities text": 37879,
      "separate subtasks": 106861,
      "subtasks target": 114985,
      "target detection": 117600,
      "entity mentioned": 37971,
      "mentioned text": 68439,
      "text explicitly": 120938,
      "explicitly implicitly": 41372,
      "sentiment identified": 106744,
      "memory network": 68353,
      "detection provide": 30029,
      "provide clues": 93775,
      "representations share": 99887,
      "proposed deep": 93255,
      "consider interactions": 21209,
      "including conventional": 53278,
      "conventional machine": 22881,
      "stateoftheart deep": 112630,
      "models leveraging": 73500,
      "cognitive features": 18268,
      "features sentiment": 43720,
      "analysis sentiments": 5839,
      "usergenerated short": 129075,
      "text sentences": 121282,
      "pragmatic levels": 88753,
      "address propose": 3754,
      "propose augment": 92563,
      "augment traditional": 10267,
      "used sentiment": 128749,
      "analysis sarcasm": 5829,
      "sarcasm detection": 103640,
      "detection cognitive": 29899,
      "eyemovement patterns": 42636,
      "statistical classification": 113082,
      "using enhanced": 129630,
      "enhanced feature": 37507,
      "feature set": 43310,
      "set improves": 107463,
      "performance fscore": 86394,
      "polarity detection": 87908,
      "datasets systems": 27744,
      "use traditional": 128334,
      "features perform": 43652,
      "perform feature": 86007,
      "significance analysis": 108713,
      "analysis experiment": 5609,
      "handle complex": 49377,
      "word clustering": 132956,
      "research method": 100555,
      "method generates": 68860,
      "clusters basis": 17964,
      "contextual similarity": 22498,
      "importance word": 52092,
      "spell checker": 111893,
      "implemented methods": 51968,
      "methods word": 69849,
      "lack resources": 58742,
      "resources word": 101067,
      "research word": 100664,
      "based preceding": 11939,
      "preceding words": 88779,
      "words key": 134019,
      "word efficient": 133002,
      "efficient result": 35105,
      "trying implement": 125903,
      "5gram model": 1140,
      "observe best": 81189,
      "using machine": 129845,
      "learning technique": 63091,
      "research generate": 100524,
      "analyze clusters": 5954,
      "testing different": 120592,
      "temporal evolution": 120102,
      "wikipedia article": 132640,
      "pairs wikipedia": 83665,
      "wikipedia articles": 132641,
      "articles representing": 9213,
      "entity topic": 38110,
      "lead different": 61850,
      "different points": 31336,
      "information propagated": 54878,
      "wikipedia language": 132662,
      "provide important": 93847,
      "support quality": 115985,
      "quality control": 94617,
      "novel webbased": 80775,
      "interface provides": 56102,
      "provides overview": 94059,
      "similarities differences": 109177,
      "learning survey": 63079,
      "survey natural": 116179,
      "symbolic representation": 116320,
      "human knowledge": 50888,
      "knowledge recent": 58135,
      "recent advances": 96393,
      "advances machine": 3994,
      "learning ml": 62749,
      "ml natural": 70436,
      "discrete symbols": 32178,
      "clearer understanding": 17735,
      "new deep": 78876,
      "learning networks": 62834,
      "networks paper": 77687,
      "paper make": 84037,
      "symbolic representations": 116321,
      "predicting target": 89015,
      "nmt models": 79905,
      "information sequential": 54973,
      "phrase attachment": 87333,
      "aims answer": 4501,
      "questions does": 95304,
      "does explicitly": 33347,
      "explicitly modeling": 41382,
      "help nmt": 49759,
      "multitask training": 75895,
      "training introduce": 123666,
      "introduce syntactic": 56547,
      "sequence results": 107044,
      "results wmt": 102336,
      "wmt data": 132782,
      "data explicitly": 25929,
      "quality germanenglish": 94677,
      "lowresource pair": 65550,
      "tight coupling": 121967,
      "quality multitask": 94727,
      "adding sourceside": 3395,
      "sourceside dependency": 110935,
      "layer obtain": 61737,
      "09 bleu": 98,
      "12 bleu": 248,
      "bleu romanianenglish": 14320,
      "distillation neural": 32516,
      "knowledge distillation": 57863,
      "student network": 114143,
      "network perform": 77387,
      "better learning": 13617,
      "translating sentence": 124604,
      "sentence neural": 105956,
      "engine time": 37036,
      "having smaller": 49569,
      "smaller model": 110237,
      "process demonstrate": 91455,
      "demonstrate transfer": 28892,
      "quality ensemble": 94644,
      "ensemble oracle": 37616,
      "network single": 77435,
      "single nmt": 109775,
      "nmt present": 79947,
      "present translation": 89750,
      "translation improvements": 124861,
      "network training": 77461,
      "student model": 114139,
      "model expensive": 71124,
      "introduce data": 56405,
      "data filtering": 25947,
      "filtering method": 44084,
      "knowledge teacher": 58204,
      "teacher model": 119696,
      "training leads": 123683,
      "decoding process": 28127,
      "cnn rnn": 18022,
      "processing deep": 91656,
      "revolutionized field": 102665,
      "field natural": 43967,
      "nlp convolutional": 79591,
      "cnn recurrent": 18018,
      "dnn architectures": 32930,
      "widely explored": 132544,
      "handle various": 49405,
      "modeling units": 72578,
      "art nlp": 9072,
      "cnns rnns": 18045,
      "rnns work": 102986,
      "systematic comparison": 116664,
      "representative nlp": 100014,
      "basic guidance": 12523,
      "selection knowledgegrounded": 104791,
      "neural conversation": 77873,
      "conversation model": 22960,
      "models capable": 72874,
      "capable generating": 15200,
      "models demonstrate": 73033,
      "incorporate content": 53456,
      "content form": 21882,
      "factual information": 42934,
      "taskoriented conversational": 118881,
      "conversational applications": 22993,
      "novel fully": 80577,
      "fully datadriven": 45953,
      "model aimed": 70649,
      "external facts": 42001,
      "allowing model": 5114,
      "opendomain setting": 81980,
      "setting approach": 107732,
      "yields significant": 135428,
      "improvements competitive": 52833,
      "seq2seq baseline": 106891,
      "baseline human": 12231,
      "significantly informative": 108971,
      "morphology present": 74801,
      "semitic languages": 105637,
      "unsupervised approaches": 127605,
      "approaches harness": 8177,
      "syntacticosemantic information": 116524,
      "information distributed": 54500,
      "long standing": 65132,
      "standing problem": 112356,
      "prove validity": 93708,
      "automatically annotated": 10719,
      "corpus named": 23893,
      "recognition text": 97029,
      "using largescale": 129793,
      "namedentity recognition": 76136,
      "sentences obtained": 106417,
      "constructed largescale": 21665,
      "algorithm extract": 4737,
      "extract relevant": 42107,
      "domain information": 33548,
      "information semantic": 54962,
      "automated processes": 10461,
      "content specific": 21952,
      "noise reduction": 80064,
      "types person": 126337,
      "loc org": 64901,
      "org misc": 82457,
      "construct different": 21629,
      "different dataset": 31076,
      "annotations comparing": 6416,
      "make datasets": 66650,
      "datasets publicly": 27654,
      "ner text": 77092,
      "feature machine": 43292,
      "confusion network": 21055,
      "combination approach": 18548,
      "approach additional": 7332,
      "trained neural": 123217,
      "network work": 77478,
      "work motivated": 134639,
      "motivated fact": 74848,
      "models assign": 72778,
      "impact input": 51873,
      "individual systems": 53933,
      "systems low": 116983,
      "words seen": 134185,
      "different outputs": 31315,
      "systems different": 116852,
      "word positions": 133412,
      "positions sentence": 88312,
      "use distributed": 128008,
      "textual similarity": 121737,
      "contributions paper": 22791,
      "based distributed": 11650,
      "words combine": 133864,
      "combine different": 18623,
      "finally obtain": 44209,
      "obtain overall": 81307,
      "overall f1": 83230,
      "level challenging": 63426,
      "challenging corpus": 16235,
      "universal semantic": 127330,
      "syntactic representation": 116457,
      "aim advancing": 4443,
      "applications recent": 7002,
      "work shows": 134805,
      "logical forms": 65005,
      "forms work": 45267,
      "work limited": 134616,
      "dependency graphs": 29146,
      "handling complex": 49413,
      "maps natural": 67166,
      "language logical": 59257,
      "perform experiments": 85995,
      "experiments question": 41100,
      "german spanish": 47913,
      "facilitate multilingual": 42782,
      "multilingual evaluation": 75254,
      "outperforms strong": 83024,
      "strong baselines": 113639,
      "baselines languages": 12419,
      "languages datasets": 60484,
      "datasets english": 27448,
      "english achieves": 37064,
      "code data": 18076,
      "scope language": 104040,
      "able handle": 1650,
      "complex semantic": 19873,
      "english promising": 37249,
      "presence complex": 89349,
      "phenomena universal": 87247,
      "learning concept": 62452,
      "concept embeddings": 20567,
      "embeddings efficient": 35676,
      "explicit concept": 41306,
      "models proven": 73832,
      "space concepts": 110983,
      "main ideas": 66428,
      "representation suffers": 99428,
      "suffers data": 115250,
      "low similarity": 65398,
      "similarity scores": 109301,
      "scores similar": 104205,
      "similar texts": 109160,
      "learn continuous": 62009,
      "concept vectors": 20591,
      "vectors learned": 131450,
      "propose efficient": 92643,
      "generate fully": 46946,
      "representations evaluate": 99636,
      "tasks measuring": 119290,
      "entity semantic": 38100,
      "improvement correlation": 52699,
      "concept categorization": 20562,
      "performance reduce": 86661,
      "categorization error": 15764,
      "rate compared": 95788,
      "word entity": 133263,
      "entity embedding": 37928,
      "classification models": 17278,
      "representations addition": 99493,
      "methods operate": 69652,
      "representation vector": 99461,
      "learning different": 62502,
      "tasks image": 119164,
      "image classification": 51775,
      "embedding recently": 35479,
      "appealing performance": 6792,
      "performance reported": 86671,
      "specific natural": 111468,
      "tasks comparison": 118996,
      "comparison approaches": 19527,
      "popularity word": 88157,
      "embedding task": 35511,
      "task mapping": 118376,
      "mapping words": 67158,
      "vector paper": 131335,
      "learning embed": 62533,
      "english version": 37333,
      "wikipedia contains": 132646,
      "contains million": 21829,
      "considered concept": 21288,
      "concept concepts": 20563,
      "concepts correspond": 20609,
      "embedding ambiguous": 35372,
      "different vectors": 31554,
      "surface form": 116060,
      "form different": 45081,
      "different mentions": 31254,
      "approaches evaluated": 8149,
      "evaluated performance": 39000,
      "performance based": 86165,
      "based concept": 11588,
      "approaches performance": 8280,
      "comparable cases": 19136,
      "higher stateoftheart": 50208,
      "translation relatively": 125212,
      "relatively little": 98397,
      "incorporating linguistic": 53546,
      "linguistic prior": 64524,
      "prior neural": 90721,
      "translation previous": 125138,
      "considering linguistic": 21326,
      "propose hybrid": 92713,
      "model called": 70800,
      "combining recurrent": 18746,
      "translation approach": 124644,
      "encourages neural": 36774,
      "prior training": 90739,
      "experiments language": 40969,
      "pairs effectiveness": 83523,
      "parallel meaning": 84670,
      "corpus translations": 24050,
      "meaning representations": 67679,
      "representations parallel": 99800,
      "bank corpus": 11419,
      "11 million": 224,
      "german italian": 47899,
      "italian dutch": 57087,
      "based crosslingual": 11618,
      "automatically produced": 10816,
      "manually corrected": 67050,
      "semantic annotations": 104989,
      "annotations english": 6429,
      "semantic annotation": 104986,
      "annotation consists": 6287,
      "consists main": 21481,
      "main steps": 66465,
      "sentences lexical": 106380,
      "compositional semantic": 20130,
      "analysis based": 5510,
      "based discourse": 11647,
      "representation theory": 99440,
      "semisupervised manner": 105613,
      "using semisupervised": 130144,
      "learning proposed": 62925,
      "algorithmic approach": 4821,
      "sense word": 105705,
      "electronic data": 35246,
      "different communication": 31046,
      "use words": 128368,
      "approach detects": 7487,
      "abusive words": 1847,
      "procedure real": 91397,
      "real life": 96069,
      "complete word": 19776,
      "abbreviated forms": 1463,
      "approach detect": 7484,
      "forms using": 45263,
      "procedure using": 91406,
      "using synset": 130254,
      "concept analysis": 20560,
      "word evaluated": 133275,
      "evaluated automated": 38951,
      "massive text": 67337,
      "fundamental tasks": 46122,
      "analysis phrase": 5750,
      "mining aims": 70221,
      "aims extracting": 4528,
      "phrases text": 87460,
      "important various": 52290,
      "taxonomy construction": 119665,
      "modeling existing": 72427,
      "rely complex": 98679,
      "trained linguistic": 123190,
      "unsatisfactory performance": 127513,
      "corpora new": 23540,
      "domains genres": 33788,
      "recently datadriven": 96673,
      "methods developed": 69443,
      "text stateoftheart": 121328,
      "models fully": 73264,
      "require human": 100147,
      "designing rules": 29647,
      "produced human": 91958,
      "experts paper": 41246,
      "leverages large": 63647,
      "large highquality": 61103,
      "limited human": 64239,
      "human labeled": 50889,
      "addition develop": 3422,
      "segmentation model": 104601,
      "information partofspeech": 54830,
      "enhance performance": 37483,
      "performance pos": 86609,
      "support language": 115977,
      "language long": 59258,
      "base wikipedia": 11491,
      "language available": 58853,
      "tagger compared": 117357,
      "effectiveness realworld": 34940,
      "datasets different": 27424,
      "domains languages": 33800,
      "languages dependencybased": 60493,
      "dependencybased neural": 29257,
      "reordering model": 98900,
      "involves translating": 56907,
      "correct word": 24128,
      "translated words": 124579,
      "help determine": 49709,
      "determine correct": 30125,
      "predict translations": 88949,
      "translated sentence": 124561,
      "sentence experiments": 105864,
      "sets compared": 107656,
      "uses sparse": 129280,
      "scientific discourse": 103999,
      "prediction using": 89150,
      "model identifying": 71296,
      "labeling approach": 58485,
      "parts experiment": 85573,
      "experiment dataset": 40457,
      "dataset consists": 26819,
      "open access": 81879,
      "information result": 54926,
      "pilot annotation": 87506,
      "representations computed": 99557,
      "mechanism involves": 68002,
      "rnn compare": 102911,
      "model lstms": 71483,
      "input layer": 55355,
      "layer simple": 61748,
      "simple attention": 109367,
      "model furthermore": 71211,
      "work useful": 134873,
      "extraction scientific": 42474,
      "harmonic grammar": 49508,
      "theory syntax": 121828,
      "czech word": 25499,
      "order work": 82430,
      "presents systematic": 89907,
      "algorithms proposed": 4876,
      "proposed learning": 93321,
      "algorithms able": 4824,
      "able compare": 1621,
      "results additional": 101501,
      "surface word": 116081,
      "sentences compare": 106248,
      "entropy model": 38169,
      "model addition": 70630,
      "addition showing": 3474,
      "theoretically appealing": 121803,
      "shows performance": 108605,
      "upper bound": 127819,
      "capable accurately": 15189,
      "observed variation": 81240,
      "biography generation": 14158,
      "generation text": 47677,
      "increasingly common": 53694,
      "requires manual": 100296,
      "manual postediting": 67011,
      "manual editing": 66988,
      "particular challenge": 85389,
      "settings present": 107831,
      "analysing human": 5476,
      "human postediting": 50932,
      "combined manual": 18657,
      "helps identify": 49816,
      "efficiency accuracy": 35021,
      "dialogue models": 30712,
      "models diversity": 73100,
      "diversity present": 32890,
      "present dialogue": 89452,
      "possible responses": 88428,
      "responses given": 101271,
      "given input": 48043,
      "generates diverse": 47160,
      "diverse outputs": 32832,
      "evolutionary model": 39695,
      "model emergence": 71055,
      "diversity language": 32881,
      "people communicate": 85870,
      "learns language": 63215,
      "languages developed": 60496,
      "high number": 50089,
      "number languages": 80905,
      "relative size": 98376,
      "power law": 88634,
      "generate onesentence": 46979,
      "wikipedia biographies": 132644,
      "slotvalue pairs": 110096,
      "pairs train": 83649,
      "train recurrent": 123004,
      "network sequencetosequence": 77429,
      "model attention": 70701,
      "facts generate": 42914,
      "generate textual": 47034,
      "textual summaries": 121750,
      "summaries model": 115450,
      "generates sentences": 47172,
      "achieves bleu": 2753,
      "vanilla sequencetosequence": 130823,
      "model scoring": 71952,
      "simple template": 109530,
      "human preference": 50933,
      "evaluation suggests": 39411,
      "suggests model": 115376,
      "nearly good": 76734,
      "manual analysis": 66974,
      "suggesting model": 115348,
      "ability infer": 1535,
      "infer knowledge": 54100,
      "generation people": 47540,
      "people speak": 85891,
      "speak different": 111277,
      "different situations": 31429,
      "specific general": 111444,
      "approach gives": 7591,
      "ability approach": 1486,
      "approach involves": 7651,
      "distillation model": 32512,
      "produced model": 91959,
      "retrain model": 102369,
      "model remaining": 71900,
      "remaining dataset": 98782,
      "dataset dialogue": 26872,
      "trained different": 123115,
      "different degrees": 31087,
      "degrees data": 28579,
      "models choose": 72902,
      "choose best": 16924,
      "level specificity": 63506,
      "compared original": 19404,
      "higherquality responses": 50219,
      "context research": 22250,
      "constitutes specific": 21559,
      "specific case": 111413,
      "involving training": 56921,
      "training multiple": 123723,
      "single dataset": 109722,
      "differences specific": 30974,
      "model set": 71991,
      "use reinforcement": 128236,
      "learning build": 62413,
      "different input": 31183,
      "input contexts": 55312,
      "test time": 120544,
      "time utilizing": 122136,
      "similarity related": 109293,
      "related lowresource": 97880,
      "pivotbased translation": 87588,
      "using related": 130103,
      "substantially better": 114882,
      "models highly": 73334,
      "competitive best": 19647,
      "direct translation": 31808,
      "combining multiple": 18737,
      "language pivot": 59813,
      "models rival": 73963,
      "translation units": 125401,
      "pivot languages": 87581,
      "compensate lack": 19601,
      "lack direct": 58700,
      "corpus inherent": 23841,
      "inherent biases": 55187,
      "pattern learning": 85716,
      "model relatively": 71892,
      "network recurrent": 77402,
      "recurrent layer": 97179,
      "biases learning": 13876,
      "learning mimic": 62747,
      "biases human": 13873,
      "human learning": 50905,
      "patterns learned": 85755,
      "learning experiments": 62565,
      "models capturing": 72883,
      "requires use": 100331,
      "features representation": 43701,
      "features recurrent": 43691,
      "soft label": 110494,
      "language inference": 59168,
      "multiple labels": 75587,
      "noise work": 80075,
      "investigate using": 56825,
      "soft labels": 110495,
      "data improve": 26020,
      "improve generalization": 52389,
      "generalization machine": 46778,
      "training deep": 123574,
      "networks dnns": 77573,
      "labels large": 58615,
      "sets propose": 107697,
      "propose soft": 93072,
      "finetuning approach": 44445,
      "labels provided": 58634,
      "provided human": 93969,
      "true label": 125851,
      "noise experiments": 80050,
      "demonstrate improved": 28760,
      "generalization performance": 46790,
      "inference nli": 54175,
      "nli task": 79552,
      "task experiments": 118179,
      "performance baselines": 86168,
      "speech twitter": 111840,
      "twitter present": 126118,
      "relations different": 98138,
      "data fully": 25962,
      "approaches unsupervised": 8389,
      "unsupervised statistical": 127724,
      "classification based": 17131,
      "discuss benefits": 32236,
      "challenges approaches": 16132,
      "approaches report": 8321,
      "report empirical": 98990,
      "evidence obtained": 39655,
      "graphbased dependency": 48890,
      "parsing dependency": 85097,
      "parsing important": 85126,
      "task popular": 118534,
      "approach dependency": 7480,
      "structured perceptron": 114023,
      "parsing time": 85264,
      "time complexity": 121997,
      "slow training": 110102,
      "training deal": 123572,
      "deal problem": 27865,
      "propose parallel": 92985,
      "parallel algorithm": 84584,
      "lot training": 65331,
      "time based": 121990,
      "based experiments": 11702,
      "experiments observe": 41045,
      "faster training": 43194,
      "training speed": 123876,
      "traditional structured": 122874,
      "using 10": 129304,
      "accuracy novel": 2220,
      "comprehensive approach": 20246,
      "works paper": 134967,
      "paper focuses": 83953,
      "estimate semantic": 38635,
      "new hybrid": 78950,
      "utilize new": 130521,
      "method propose": 69078,
      "based wordnet": 12169,
      "designed test": 29628,
      "obviously improve": 81488,
      "approach comparing": 7435,
      "artificial data": 9245,
      "previous measures": 90423,
      "random vectors": 95516,
      "usual vector": 130400,
      "model construct": 70900,
      "similarities terms": 109188,
      "semantically significant": 105376,
      "word selection": 133512,
      "selection automatic": 104768,
      "language easily": 58989,
      "time new": 122059,
      "segmentation word": 104653,
      "processing languages": 91690,
      "languages having": 60617,
      "significant challenges": 108734,
      "researched problem": 100671,
      "reasonable results": 96214,
      "results meeting": 101912,
      "present solution": 89712,
      "solution based": 110547,
      "clusters using": 17971,
      "results achieve": 101492,
      "difficult learn": 31641,
      "relative difficulty": 98349,
      "particular kinds": 85426,
      "kind data": 57677,
      "implemented learning": 51965,
      "theory grammatical": 121818,
      "grammatical theory": 48725,
      "simplified versions": 109610,
      "english case": 37081,
      "training recurrent": 123806,
      "model long": 71471,
      "long term": 65136,
      "difficult propose": 31657,
      "knowledge explicit": 57915,
      "used augment": 128398,
      "resulting graph": 101445,
      "directed acyclic": 31812,
      "introduce model": 56459,
      "model encodes": 71072,
      "use model": 128147,
      "model coreference": 70929,
      "coreference relations": 23359,
      "apply model": 7188,
      "text comprehension": 120808,
      "comprehension tasks": 20237,
      "tasks achieve": 118915,
      "achieve new": 2500,
      "results considered": 101630,
      "benchmarks including": 12907,
      "including cnn": 53270,
      "qa tasks": 94523,
      "model solves": 72067,
      "15 20": 339,
      "20 tasks": 529,
      "1000 training": 177,
      "examples task": 39887,
      "analysis learned": 5675,
      "representations demonstrates": 99587,
      "demonstrates ability": 28939,
      "ability model": 1543,
      "model encode": 71069,
      "entity information": 37945,
      "learning applied": 62336,
      "typically associated": 126411,
      "vision systems": 131768,
      "systems today": 117198,
      "recently cnns": 96671,
      "applied problems": 7108,
      "interesting results": 56091,
      "paper try": 84481,
      "different variations": 31550,
      "explores problem": 41651,
      "authorship attribution": 10390,
      "approaches methods": 8233,
      "methods explored": 69498,
      "scheme uses": 103941,
      "stylistic language": 114629,
      "discriminative features": 32208,
      "features second": 43712,
      "positive samples": 88342,
      "samples unlabeled": 103597,
      "set using": 107631,
      "using nearest": 129944,
      "ground truth": 49096,
      "data effectiveness": 25878,
      "ranked lists": 95647,
      "compositionality detection": 20141,
      "language refers": 60029,
      "refers meaning": 97583,
      "meaning phrase": 67658,
      "meaning constituents": 67626,
      "constituents combined": 21554,
      "based premise": 11942,
      "version phrase": 131617,
      "words replaced": 134170,
      "ways representing": 132172,
      "vectors language": 131447,
      "similarity propose": 109287,
      "method represents": 69105,
      "method approximates": 68643,
      "using range": 130080,
      "correlation metrics": 24239,
      "method completely": 68718,
      "completely unsupervised": 19784,
      "unsupervised experiments": 127635,
      "experiments publicly": 41096,
      "available dataset": 10975,
      "shows compared": 108564,
      "compared strong": 19467,
      "strong supervised": 113719,
      "supervised baselines": 115732,
      "baselines approach": 12354,
      "using distance": 129589,
      "pattern discovery": 85713,
      "discovery massive": 32142,
      "textual patterns": 121724,
      "news tweets": 79392,
      "mining nlp": 70246,
      "research previous": 100584,
      "adopt dependency": 3891,
      "discovery approach": 32134,
      "context entities": 22081,
      "process costly": 91447,
      "scale study": 103754,
      "structure called": 113813,
      "certain context": 15937,
      "efficient framework": 35079,
      "massive corpora": 67325,
      "develops contextaware": 30449,
      "determine boundaries": 30122,
      "avoids costly": 11249,
      "distributions entities": 32760,
      "inflectional morphology": 54295,
      "discrete continuous": 32164,
      "set feature": 107442,
      "feature values": 43325,
      "feature value": 43324,
      "vector sum": 131394,
      "inner product": 55275,
      "labelling tasks": 58574,
      "words produced": 134142,
      "produced automatic": 91950,
      "asr systems": 9446,
      "systems typically": 117210,
      "texts input": 121530,
      "input available": 55300,
      "available asr": 10950,
      "asr output": 9442,
      "output paper": 83104,
      "technique jointly": 119796,
      "jointly modeling": 57366,
      "tasks punctuation": 119423,
      "punctuation capitalization": 94394,
      "tasks method": 119293,
      "method extended": 68831,
      "joint modeling": 57294,
      "tasks improving": 119174,
      "improving neural": 53126,
      "translation conditional": 124704,
      "sequence generative": 106955,
      "generative adversarial": 47707,
      "adversarial nets": 4149,
      "proposes approach": 93595,
      "adversarial net": 4148,
      "generator discriminator": 47768,
      "discriminator generator": 32223,
      "generator aims": 47766,
      "aims generate": 4531,
      "generate sentences": 47011,
      "target sentences": 117702,
      "sentences discriminator": 106282,
      "models play": 73746,
      "game achieve": 46418,
      "sentencelevel bleu": 106145,
      "generation high": 47428,
      "evaluate generated": 38839,
      "generated sentences": 47120,
      "guide learning": 49241,
      "model consistently": 70890,
      "emerged stateoftheart": 36054,
      "stateoftheart transformer": 113011,
      "englishgerman chineseenglish": 37372,
      "tasks endtoend": 119096,
      "endtoend optimization": 36950,
      "visually grounded": 131845,
      "systems endtoend": 116867,
      "design dialogue": 29531,
      "recently popular": 96729,
      "powerful tools": 88680,
      "encoderdecoder architectures": 36592,
      "architectures sequencetosequence": 8846,
      "learning current": 62478,
      "dialogue management": 30706,
      "aiming predicting": 4492,
      "history dialogue": 50552,
      "making context": 66832,
      "context dialogue": 22058,
      "answering tasks": 6709,
      "far using": 43102,
      "endtoend architectures": 36873,
      "architectures paper": 8836,
      "introduce deep": 56409,
      "taskoriented dialogues": 118898,
      "dialogues based": 30823,
      "based policy": 11935,
      "gradient algorithm": 48596,
      "approach tested": 7969,
      "tested dataset": 120566,
      "dialogues collected": 30824,
      "generating natural": 47233,
      "natural dialogues": 76250,
      "task discovering": 118103,
      "semeval2017 task": 105529,
      "task attentionbased": 117904,
      "similarity model": 109271,
      "model performed": 71733,
      "semeval 2017": 105500,
      "crosslingual semantic": 25003,
      "semantic textual": 105328,
      "similarity sts": 109315,
      "employs attentionbased": 36328,
      "model optimizes": 71618,
      "similarity paper": 109278,
      "measures similarity": 67897,
      "spanish arabic": 111147,
      "increase available": 53588,
      "available information": 11016,
      "information context": 54447,
      "reading understanding": 96037,
      "understanding flow": 126846,
      "people understand": 85899,
      "applications based": 6900,
      "based combinatorial": 11584,
      "combinatorial optimization": 18606,
      "compression msc": 20315,
      "method reduces": 69096,
      "optimization methods": 82196,
      "study improve": 114403,
      "method msc": 68973,
      "msc using": 74945,
      "graph theory": 48875,
      "informative sentences": 55153,
      "sentences maintaining": 106391,
      "maintaining grammaticality": 66516,
      "better stateoftheart": 13726,
      "framenet 17": 45394,
      "hierarchical rnn": 50002,
      "sentencelevel attention": 106144,
      "speaker change": 111283,
      "change detection": 16359,
      "detection speaker": 30059,
      "task dialog": 118087,
      "dialog modeling": 30573,
      "useful various": 128945,
      "various scenarios": 131187,
      "scenarios example": 103845,
      "dialog transcripts": 30615,
      "speaker identities": 111289,
      "matching problem": 67424,
      "propose hierarchical": 92709,
      "hierarchical recurrent": 49991,
      "consistently achieve": 21408,
      "approaches attentionbased": 8069,
      "attentionbased model": 10081,
      "data commercial": 25749,
      "customer feedback": 25438,
      "products services": 92073,
      "need structured": 76861,
      "services like": 107330,
      "terms sentiment": 120381,
      "provides finegrained": 94030,
      "like social": 64093,
      "media twitter": 68176,
      "forums quora": 45318,
      "general sentiment": 46711,
      "prediction model": 89079,
      "model objective": 71597,
      "objective generating": 81087,
      "opinion sentiment": 82099,
      "like described": 64034,
      "classification problems": 17351,
      "classification approaches": 17120,
      "approaches study": 8357,
      "study aims": 114312,
      "overhead manual": 83326,
      "manual labelling": 67007,
      "preprocessing feature": 89323,
      "extraction research": 42470,
      "research project": 100591,
      "aims design": 4514,
      "used predict": 128688,
      "tracking dataset": 122752,
      "dataset english": 26891,
      "present dataset": 89430,
      "words written": 134325,
      "articles october": 9203,
      "token annotated": 122246,
      "hard create": 49456,
      "potential user": 88595,
      "nontrivial task": 80315,
      "task adapting": 117842,
      "adapting stateoftheart": 3324,
      "produced dataset": 91953,
      "dataset allows": 26731,
      "research going": 100526,
      "ensuring quality": 37647,
      "usage data": 127856,
      "data enables": 25889,
      "novel insights": 80602,
      "wikipedia using": 132685,
      "translation started": 125289,
      "generating text": 47275,
      "certain linguistic": 15947,
      "simplification using": 109597,
      "quality simplified": 94792,
      "simplified text": 109605,
      "text terms": 121365,
      "meteor bleu": 68569,
      "end report": 36830,
      "examples text": 39889,
      "metric machine": 69884,
      "topics natural": 122640,
      "nlp literature": 79630,
      "evaluate mt": 38873,
      "makes improvement": 66788,
      "improvement traditional": 52778,
      "expensive timeconsuming": 40435,
      "automatic mt": 10611,
      "tend perform": 120158,
      "pairs english": 83526,
      "english target": 37301,
      "used source": 128771,
      "rely additional": 98674,
      "additional linguistic": 3542,
      "features achieve": 43348,
      "performance makes": 86520,
      "apply language": 7179,
      "pairs easily": 83522,
      "popular metrics": 88102,
      "low performance": 65378,
      "thesis address": 121847,
      "address existing": 3679,
      "design novel": 29554,
      "novel mt": 80653,
      "methods investigate": 69562,
      "performances different": 86878,
      "firstly design": 44563,
      "factors yield": 42903,
      "accurate evaluation": 2335,
      "evaluation model": 39298,
      "model weighting": 72313,
      "optimized according": 82230,
      "according characteristics": 1971,
      "methods design": 69436,
      "using partofspeech": 130009,
      "methods yield": 69857,
      "yield higher": 135340,
      "higher performance": 50192,
      "finally introduce": 44199,
      "performance metrics": 86531,
      "workshop shared": 135016,
      "methods robust": 69735,
      "robust different": 103030,
      "mt using": 75006,
      "using reference": 130098,
      "reference translations": 97543,
      "models naive": 73598,
      "machine svm": 65870,
      "classification algorithms": 17111,
      "affective meaning": 4239,
      "expansion using": 40377,
      "induction methods": 54026,
      "representations building": 99527,
      "similarity graph": 109238,
      "implemented evaluated": 51961,
      "metrics comprehensive": 69939,
      "comprehensive evaluation": 20258,
      "evaluation approaches": 39121,
      "set demonstrating": 107414,
      "demonstrating methods": 28976,
      "methods generate": 69522,
      "significant number": 108828,
      "number new": 80923,
      "error mean": 38308,
      "obtained combining": 81352,
      "distributional features": 32700,
      "features outperformed": 43642,
      "approached supervised": 8034,
      "model grounded": 71268,
      "grounded language": 49107,
      "understanding present": 126922,
      "model pragmatic": 71757,
      "network classifiers": 77186,
      "pragmatic reasoning": 88755,
      "reasoning framework": 96249,
      "color descriptions": 18530,
      "improvement results": 52759,
      "reasoning helps": 96252,
      "cases model": 15656,
      "model distinguish": 71011,
      "corpus human": 23827,
      "human utterances": 50986,
      "color reference": 18531,
      "reference games": 97519,
      "speaker model": 111293,
      "model reproduces": 71911,
      "neural crosslingual": 77883,
      "transfer paradigm": 124172,
      "paradigm completion": 84533,
      "novel crosslingual": 80518,
      "transfer method": 124150,
      "completion task": 19796,
      "mapping lemma": 67137,
      "lemma inflected": 63328,
      "model state": 72085,
      "art monolingual": 9062,
      "monolingual task": 74622,
      "task use": 118824,
      "use labeled": 128108,
      "data highresource": 26004,
      "highresource language": 50422,
      "increase performance": 53609,
      "performance lowresource": 86506,
      "language experiments": 59029,
      "pairs different": 83518,
      "obtain 58": 81253,
      "transfer zeroshot": 124212,
      "learning possible": 62896,
      "degree language": 28572,
      "language relatedness": 60032,
      "transfer morphological": 124156,
      "morphological knowledge": 74713,
      "knowledge sentiment": 58170,
      "using word2vec": 130387,
      "task scientific": 118664,
      "scientific paper": 104013,
      "paper analysis": 83731,
      "existing machine": 40165,
      "analysis focusing": 5627,
      "corpus automatic": 23671,
      "automatic feature": 10544,
      "applied sentiment": 7114,
      "analysis short": 5845,
      "short texts": 108261,
      "work conducted": 134431,
      "research question": 100603,
      "work sentiment": 134793,
      "classifying positive": 17671,
      "formed feature": 45225,
      "space examined": 110999,
      "sentence mapped": 105939,
      "mapped features": 67123,
      "input classifiers": 55306,
      "using 10crossvalidation": 129307,
      "10crossvalidation scheme": 202,
      "scheme evaluation": 103921,
      "evaluation conducted": 39156,
      "conducted set": 20948,
      "set annotated": 107358,
      "embeddings effective": 35674,
      "features performed": 43654,
      "performed better": 86913,
      "better overall": 13642,
      "overall classification": 83219,
      "embeddings important": 35746,
      "important problems": 52215,
      "problems machine": 91339,
      "evaluate similarity": 38926,
      "similarity translation": 109325,
      "hypotheses different": 51253,
      "different surface": 31467,
      "surface forms": 116061,
      "segment level": 104562,
      "level propose": 63492,
      "perform word": 86098,
      "performed experiments": 86920,
      "evaluated proposed": 39002,
      "methods various": 69842,
      "various translation": 131229,
      "translation datasets": 124735,
      "datasets experimental": 27467,
      "methods combining": 69378,
      "combining lexical": 18731,
      "texts news": 121562,
      "information event": 54544,
      "manner information": 66951,
      "information seeking": 54960,
      "applications information": 6940,
      "extraction question": 42457,
      "assume text": 9649,
      "empirically test": 36244,
      "international relations": 56187,
      "sports science": 112040,
      "clearly indicate": 17739,
      "news texts": 79379,
      "texts study": 121623,
      "study fact": 114380,
      "motivate development": 74833,
      "development supervised": 30429,
      "label large": 58402,
      "large training": 61296,
      "corpus task": 24034,
      "features manually": 43605,
      "data compare": 25754,
      "performance domainspecific": 86315,
      "domainspecific classifiers": 33892,
      "classifiers trained": 17639,
      "data given": 25981,
      "given news": 48069,
      "news domain": 79335,
      "domain general": 33539,
      "general classifier": 46641,
      "data domains": 25866,
      "prediction experiments": 89061,
      "varies depending": 130948,
      "depending domain": 29271,
      "domain naive": 33587,
      "naive annotators": 76010,
      "domain label": 33563,
      "texts typically": 121641,
      "classification prediction": 17336,
      "acyclic graph": 3158,
      "graph parser": 48853,
      "crosslinguistically applicable": 25058,
      "representation builds": 99182,
      "rapid annotation": 95710,
      "annotation ucca": 6399,
      "challenge existing": 16032,
      "existing parsing": 40252,
      "corresponding complex": 24278,
      "uses novel": 129258,
      "novel transition": 80757,
      "parsing ability": 85056,
      "ability handle": 1526,
      "graph structures": 48872,
      "parsers semantic": 85043,
      "structures languages": 114082,
      "frequently use": 45880,
      "structures neural": 114091,
      "word segmenter": 133510,
      "speech tagger": 111816,
      "models potentially": 73760,
      "tai et": 117495,
      "forget gates": 45057,
      "bias term": 13847,
      "mechanism experiment": 67980,
      "translation lattices": 124896,
      "report consistent": 98984,
      "improvements baselines": 52812,
      "tweet representations": 126016,
      "research analysis": 100403,
      "analysis microblogging": 5697,
      "microblogging platforms": 70050,
      "number works": 80997,
      "models applications": 72756,
      "like sentiment": 64090,
      "analysis semantic": 5834,
      "similarity computation": 109210,
      "prediction performance": 89102,
      "traditional baselines": 122800,
      "baselines tasks": 12476,
      "representations work": 99995,
      "work better": 134399,
      "certain tasks": 15957,
      "tasks work": 119609,
      "vector embeddings": 131305,
      "traditional feature": 122818,
      "effective application": 34628,
      "evaluate representations": 38914,
      "accuracy model": 2209,
      "model properties": 71808,
      "length presence": 63374,
      "presence particular": 89355,
      "extensive study": 41958,
      "study supervised": 114527,
      "representations popular": 99806,
      "popular textual": 88138,
      "textual social": 121745,
      "encode textual": 36462,
      "model low": 71479,
      "resource settings": 100879,
      "degradation reduction": 28553,
      "embedding size": 35492,
      "draw interesting": 34116,
      "interesting insights": 56078,
      "methods semantic": 69738,
      "sentences estimate": 106294,
      "combine methods": 18631,
      "methods unsupervised": 69825,
      "unsupervised supervised": 127728,
      "supervised way": 115854,
      "best run": 13444,
      "ranked 1st": 95634,
      "annotations linear": 6443,
      "using iterative": 129768,
      "orthogonal procrustes": 82586,
      "complex language": 19825,
      "corpora training": 23605,
      "embeddings relatively": 35906,
      "models input": 73404,
      "analogy tests": 5427,
      "tests results": 120612,
      "poorly experiments": 88067,
      "combine word": 18645,
      "ensemble model": 37602,
      "model leads": 71425,
      "relative improvements": 98359,
      "initial models": 55221,
      "report annotator": 98976,
      "radlex terms": 95454,
      "goal explore": 48351,
      "radiology reports": 95450,
      "brief paper": 14614,
      "paper explain": 83910,
      "embeddings explore": 35713,
      "explore ability": 41515,
      "capture semantic": 15378,
      "morphological tag": 74740,
      "compose representation": 20061,
      "train models": 122960,
      "models uses": 74296,
      "evaluating models": 39074,
      "conversation modeling": 22961,
      "lstm paper": 65658,
      "media using": 68180,
      "conversation structure": 22973,
      "experiments task": 41169,
      "predicting popularity": 88999,
      "comments reddit": 18825,
      "analyses benefit": 5441,
      "benefit model": 12979,
      "course discussion": 24465,
      "improving detection": 53087,
      "early late": 34391,
      "bidirectional tree": 13984,
      "state updates": 112538,
      "user engagement": 128979,
      "rhetorical devices": 102711,
      "public discourse": 94256,
      "analysis work": 5940,
      "study analyze": 114316,
      "regression analysis": 97691,
      "build models": 14787,
      "models recognize": 73890,
      "sentences conclude": 106255,
      "potential implications": 88561,
      "media websites": 68182,
      "websites electronic": 132300,
      "electronic newspapers": 35256,
      "newspapers internet": 79409,
      "internet forums": 56190,
      "forums allow": 45311,
      "allow visitors": 5099,
      "visitors leave": 131778,
      "leave comments": 63259,
      "comments read": 18821,
      "read interact": 95932,
      "interact exchange": 55931,
      "exchange free": 39941,
      "free participants": 45761,
      "participants malicious": 85321,
      "malicious intentions": 66885,
      "users adopt": 129094,
      "strategies handle": 113467,
      "users paper": 129151,
      "research propose": 100593,
      "jointly predicts": 57384,
      "crucial aspects": 25132,
      "intention disclosure": 55914,
      "response strategy": 101245,
      "new annotated": 78772,
      "annotated dataset": 6173,
      "dataset containing": 26825,
      "containing excerpts": 21789,
      "conversations involving": 23058,
      "trolls interactions": 125838,
      "interactions users": 56005,
      "users hope": 129130,
      "hope useful": 50645,
      "task extending": 118190,
      "embeddings multilingual": 35814,
      "relational knowledge": 98093,
      "2017 task": 641,
      "task multilingual": 118415,
      "multilingual crosslingual": 75228,
      "based conceptnet": 11589,
      "multilingual knowledge": 75264,
      "submission semeval": 114734,
      "work builds": 134405,
      "highquality multilingual": 50397,
      "embeddings combination": 35598,
      "took place": 122367,
      "subtasks ranked": 114977,
      "crosslingual language": 24964,
      "require considerable": 100123,
      "data produced": 26283,
      "allows use": 5195,
      "use larger": 128118,
      "larger training": 61384,
      "shows method": 108595,
      "generally outperforms": 46871,
      "method previous": 69071,
      "model chinese": 70821,
      "chinese discourse": 16760,
      "coherence modeling": 18304,
      "modeling chinese": 72389,
      "modeling remains": 72532,
      "need feature": 76809,
      "sophisticated features": 110668,
      "features capture": 43388,
      "coherence evaluation": 18297,
      "based current": 11619,
      "english discourse": 37115,
      "model specifically": 72075,
      "results sentence": 102153,
      "translation coherence": 124695,
      "rating task": 95821,
      "task effectiveness": 118130,
      "existing strong": 40299,
      "characterbased word": 16503,
      "tagging mt": 117412,
      "robustness nlp": 103116,
      "neural approaches": 77830,
      "approaches achieve": 8040,
      "accuracy certain": 2112,
      "sensitive small": 105746,
      "changes input": 16383,
      "stability robustness": 112105,
      "desired properties": 29669,
      "content humans": 21889,
      "humans easily": 51077,
      "conditions paper": 20819,
      "study impact": 114401,
      "impact noisy": 51880,
      "noisy input": 80106,
      "consider different": 21198,
      "type noise": 126216,
      "distributions training": 32772,
      "training testing": 123916,
      "evaluate robustness": 38916,
      "different models": 31265,
      "models convolutional": 72988,
      "different basic": 31025,
      "units characters": 127236,
      "characters byte": 16597,
      "units different": 127238,
      "tagging machine": 117402,
      "model large": 71413,
      "achieved promising": 2670,
      "approaches statistical": 8351,
      "despite recent": 29718,
      "larger vocabulary": 61388,
      "vocabulary training": 131910,
      "training complexity": 123388,
      "complexity decoding": 19906,
      "decoding complexity": 28080,
      "number target": 80976,
      "words problem": 134140,
      "problem translating": 91269,
      "patent documents": 85665,
      "documents contain": 33209,
      "contain technical": 21765,
      "phrases contain": 87415,
      "allows proposed": 5180,
      "nmt applied": 79831,
      "applied translation": 7134,
      "task language": 118333,
      "languagespecific knowledge": 60994,
      "technical term": 119754,
      "term identification": 120223,
      "tokens training": 122344,
      "translation table": 125323,
      "smt evaluation": 110282,
      "sentence translation": 106112,
      "translation proved": 125157,
      "proved effectiveness": 93713,
      "phrases selected": 87451,
      "baseline nmt": 12275,
      "translation errors": 124795,
      "model translation": 72234,
      "sentences large": 106371,
      "terms using": 120402,
      "represented single": 100034,
      "method enables": 68801,
      "nmt translate": 79998,
      "train nmt": 122990,
      "nmt bilingual": 79844,
      "data technical": 26547,
      "terms use": 120400,
      "translated sentences": 124562,
      "tokens experiments": 122307,
      "sentences proposed": 106458,
      "improvement 31": 52675,
      "31 bleu": 919,
      "traditional smt": 122867,
      "nmt proposed": 79950,
      "technique crosslingual": 119779,
      "abstract meaning": 1768,
      "representation parsing": 99369,
      "parsing abstract": 85057,
      "representation amr": 99165,
      "amr annotation": 5384,
      "annotation efforts": 6302,
      "focused english": 44853,
      "english order": 37230,
      "order train": 82420,
      "parsers languages": 85029,
      "based annotation": 11507,
      "language parallel": 59797,
      "italian spanish": 57102,
      "spanish german": 111160,
      "german chinese": 47881,
      "languages evaluating": 60558,
      "datasets propose": 27631,
      "method exploits": 68829,
      "gold annotations": 48430,
      "require access": 100108,
      "access gold": 1909,
      "annotations target": 6469,
      "process new": 91538,
      "new english": 78899,
      "evaluated existing": 38974,
      "standard graph": 112246,
      "graph convolutional": 48779,
      "convolutional encoders": 23156,
      "syntaxaware neural": 116578,
      "approach incorporating": 7634,
      "incorporating syntactic": 53562,
      "structure neural": 113915,
      "neural attentionbased": 77849,
      "attentionbased encoderdecoder": 10068,
      "networks gcns": 77609,
      "networks developed": 77566,
      "graphstructured data": 48979,
      "use predicted": 128201,
      "trees source": 125714,
      "sentences produce": 106454,
      "produce representations": 91922,
      "words hidden": 133980,
      "states encoder": 113039,
      "sensitive syntactic": 105749,
      "produce word": 91946,
      "representations output": 99793,
      "networks evaluate": 77588,
      "observe substantial": 81213,
      "data crosslingual": 25812,
      "years seen": 135297,
      "development largescale": 30397,
      "evaluation datasets": 39171,
      "datasets snli": 27720,
      "inference problem": 54198,
      "relation pairs": 98060,
      "inference effectively": 54135,
      "english paper": 37232,
      "advance research": 3930,
      "evaluation end": 39201,
      "languages arabic": 60408,
      "arabic french": 8515,
      "french spanish": 45824,
      "spanish russian": 111173,
      "experiment set": 40508,
      "set baselines": 107375,
      "translation best": 124672,
      "best scores": 13446,
      "average accuracy": 11180,
      "research multilingual": 100558,
      "sentences important": 106347,
      "important semantic": 52249,
      "units natural": 127259,
      "language generic": 59106,
      "representation sentences": 99405,
      "sentences capture": 106234,
      "multiple downstream": 75546,
      "sentences word": 106546,
      "representations given": 99669,
      "semeval datasets": 105517,
      "datasets standard": 27731,
      "standard deviation": 112230,
      "represent sentence": 99138,
      "vectors unsupervised": 131481,
      "unsupervised representation": 127703,
      "tasks 19": 118910,
      "sophisticated neural": 110673,
      "mining using": 70267,
      "using tree": 130320,
      "language tree": 60191,
      "grammars tags": 48683,
      "provide ample": 93755,
      "represents special": 100075,
      "special challenge": 111354,
      "challenge computational": 16010,
      "agglutinative morphology": 4340,
      "difficult argument": 31609,
      "structure modelling": 113908,
      "syntax morphology": 116541,
      "morphology using": 74809,
      "using tag": 130265,
      "interesting problem": 56084,
      "extent semantics": 41984,
      "sentence order": 105961,
      "semantic property": 105188,
      "property use": 92506,
      "inhouse developed": 55205,
      "algorithm given": 4747,
      "relation tags": 98073,
      "shall present": 107910,
      "present brief": 89398,
      "adversarial multitask": 4145,
      "learning text": 63102,
      "classification neural": 17306,
      "models shown": 74025,
      "learning focus": 62587,
      "focus learning": 44780,
      "layers extract": 61779,
      "shared features": 107982,
      "taskspecific features": 119626,
      "latent feature": 61587,
      "feature spaces": 43317,
      "conduct extensive": 20864,
      "16 different": 377,
      "different text": 31487,
      "tasks demonstrates": 119044,
      "benefits approach": 13005,
      "approach shared": 7889,
      "shared knowledge": 107985,
      "learned proposed": 62246,
      "model regarded": 71882,
      "transferred new": 124233,
      "datasets 16": 27288,
      "learning external": 62574,
      "problem predicting": 91166,
      "comments online": 18818,
      "discussion forum": 32320,
      "learning particularly": 62881,
      "addressing challenges": 3808,
      "language state": 60120,
      "state action": 112420,
      "action spaces": 3081,
      "state representation": 112511,
      "particular point": 85440,
      "world events": 135030,
      "events available": 39565,
      "making feasible": 66841,
      "feasible search": 43242,
      "action space": 3080,
      "reddit communities": 97297,
      "showing methods": 108414,
      "methods improve": 69540,
      "improve previous": 52502,
      "previous reported": 90445,
      "reported results": 99071,
      "task neural": 118451,
      "generates fluent": 47161,
      "usually better": 130409,
      "better nmt": 13637,
      "nmt translation": 79999,
      "translation adequacy": 124629,
      "promising direction": 92269,
      "combine advantages": 18615,
      "smt paper": 110294,
      "leveraging multisource": 63692,
      "multisource nmt": 75789,
      "final translation": 44134,
      "translation extensive": 124819,
      "task model": 118394,
      "53 bleu": 1101,
      "points best": 87850,
      "best single": 13452,
      "single output": 109777,
      "stateoftheart traditional": 113009,
      "methods deep": 69425,
      "present deep": 89432,
      "parses sentences": 85054,
      "dependency graph": 29145,
      "improve state": 52547,
      "art semantic": 9089,
      "using handengineered": 129725,
      "handengineered features": 49363,
      "syntax explore": 116534,
      "explore multitask": 41565,
      "shares parameters": 108126,
      "graphs jointly": 48960,
      "approaches improve": 8186,
      "art code": 9040,
      "code opensource": 18125,
      "opensource available": 82016,
      "factored model": 42861,
      "categories syntactic": 15754,
      "structures explicitly": 114076,
      "english japanese": 37178,
      "parsing learning": 85144,
      "learning create": 62468,
      "openvocabulary neural": 82043,
      "models fail": 73231,
      "fail account": 42946,
      "characterlevel language": 16569,
      "models offer": 73659,
      "hierarchical lstm": 49972,
      "mechanism learns": 68008,
      "previously generated": 90599,
      "construct new": 21641,
      "modeling corpus": 72410,
      "corpus multilingual": 23888,
      "multilingual wikipedia": 75401,
      "typologically diverse": 126497,
      "diverse languages": 32820,
      "languages demonstrate": 60489,
      "effectiveness model": 34913,
      "model range": 71841,
      "range languages": 95575,
      "dialogue agents": 30638,
      "dynamic knowledge": 34312,
      "graph embeddings": 48798,
      "embeddings study": 35956,
      "common goal": 18882,
      "goal openended": 48371,
      "openended dialogue": 81987,
      "dialogue state": 30743,
      "poses new": 88269,
      "new challenges": 78823,
      "challenges existing": 16156,
      "existing dialogue": 40110,
      "collected dataset": 18415,
      "humanhuman dialogues": 51033,
      "model structured": 72093,
      "structured knowledge": 114006,
      "knowledge unstructured": 58222,
      "model dynamic": 71033,
      "dialogue progresses": 30728,
      "automatic human": 10550,
      "evaluations model": 39469,
      "effective achieving": 34622,
      "achieving goal": 2947,
      "new graphbased": 78941,
      "graphbased approach": 48888,
      "dictionaries word": 30871,
      "embeddings build": 35582,
      "second apply": 104385,
      "sense induction": 105681,
      "ambiguous words": 5301,
      "words finally": 133958,
      "ambiguous input": 5292,
      "input graph": 55347,
      "clustering algorithm": 17933,
      "algorithm perform": 4782,
      "clustering graph": 17945,
      "approach shows": 7892,
      "shows excellent": 108577,
      "excellent results": 39920,
      "results outperforming": 102018,
      "outperforming competitive": 82804,
      "methods terms": 69797,
      "terms fscore": 120329,
      "standard datasets": 112221,
      "derived largescale": 29353,
      "largescale manually": 61461,
      "useful knowledge": 128898,
      "applications language": 6950,
      "processing knowledge": 91688,
      "representation important": 99257,
      "wikipedia pages": 132668,
      "categories according": 15727,
      "labels used": 58652,
      "used information": 128590,
      "improvements entity": 52847,
      "entity ranking": 37990,
      "makes difficult": 66770,
      "recognize important": 97053,
      "important ones": 52201,
      "descriptions paper": 29489,
      "categories observe": 15743,
      "texts wikipedia": 121650,
      "wikipedia links": 132665,
      "links corresponding": 64678,
      "overall agreement": 83213,
      "compared human": 19376,
      "challenge set": 16098,
      "set approach": 107362,
      "approach evaluating": 7543,
      "evaluating machine": 39069,
      "forward translation": 45329,
      "address questions": 3758,
      "set consists": 107400,
      "approach present": 7809,
      "set use": 107625,
      "use analyze": 127894,
      "systems resulting": 117122,
      "insight linguistic": 55523,
      "remain reach": 98774,
      "develop streaming": 30236,
      "based canonical": 11561,
      "skipgram negative": 109989,
      "algorithm implemented": 4751,
      "algorithm word2vec": 4816,
      "cosine similarity": 24338,
      "applying algorithm": 7229,
      "provide partial": 93891,
      "validation approach": 130738,
      "discuss potential": 32277,
      "failure modes": 42989,
      "suggest directions": 115308,
      "directions future": 31840,
      "sampling sgns": 103607,
      "sgns word": 107904,
      "embedding model": 35447,
      "model known": 71397,
      "viewed problem": 131715,
      "standard way": 112336,
      "framework optimize": 45636,
      "propose algorithm": 92538,
      "algorithm optimizes": 4775,
      "objective using": 81126,
      "using riemannian": 130120,
      "embeddings detecting": 35646,
      "speech transcripts": 111829,
      "features mainly": 43601,
      "used detect": 128485,
      "suitable largescale": 115401,
      "precision automatic": 88791,
      "automatic correction": 10505,
      "transcripts paper": 124012,
      "enriched word": 37562,
      "better represent": 13695,
      "texts produced": 121583,
      "network measurements": 77320,
      "measurements applied": 67846,
      "classifiers automatically": 17600,
      "binary classification": 14117,
      "task comparison": 117988,
      "comparison performance": 19561,
      "performance traditional": 86805,
      "words bow": 133841,
      "features datasets": 43437,
      "using complex": 129460,
      "networks support": 77787,
      "machine superior": 65869,
      "using linguistic": 129816,
      "dataset manually": 27012,
      "taken results": 117514,
      "indicate complex": 53834,
      "duluth semeval2017": 34259,
      "describes duluth": 29399,
      "duluth systems": 34263,
      "participated semeval2017": 85342,
      "task detection": 118078,
      "participated subtasks": 85349,
      "methods included": 69545,
      "disambiguation measures": 31969,
      "survey neural": 116181,
      "extraction techniques": 42514,
      "techniques using": 120011,
      "architectures research": 8845,
      "research questions": 100604,
      "focus stateoftheart": 44824,
      "techniques proven": 119962,
      "useful tools": 128941,
      "tools language": 122453,
      "tasks semisupervised": 119486,
      "semisupervised sequence": 105621,
      "bidirectional language": 13930,
      "models pretrained": 73779,
      "learned unlabeled": 62273,
      "standard component": 112213,
      "component neural": 19994,
      "recurrent network": 97184,
      "representations produce": 99825,
      "produce context": 91881,
      "context sensitive": 22258,
      "representations trained": 99939,
      "little labeled": 64817,
      "general semisupervised": 46710,
      "context embeddings": 22074,
      "embeddings bidirectional": 35578,
      "apply sequence": 7216,
      "datasets named": 27587,
      "results surpassing": 102244,
      "surpassing previous": 116117,
      "previous systems": 90507,
      "learning additional": 62311,
      "additional labeled": 3533,
      "data task": 26544,
      "task specific": 118728,
      "conditional variational": 20792,
      "variational framework": 130923,
      "dialog generation": 30564,
      "generation deep": 47357,
      "variable models": 130844,
      "shown facilitate": 108471,
      "generation opendomain": 47521,
      "responses paper": 101288,
      "framework allowing": 45419,
      "generation based": 47323,
      "based specific": 12062,
      "manually assigned": 67038,
      "dialog states": 30591,
      "order reflect": 82398,
      "framework different": 45497,
      "different scenarios": 31399,
      "states respectively": 113055,
      "experiment result": 40488,
      "dependency parses": 29170,
      "inject linguistic": 55258,
      "tasks practitioners": 119389,
      "efficiently parse": 35153,
      "advances gpu": 3990,
      "gpu hardware": 48578,
      "fail leverage": 42967,
      "sequential processing": 107244,
      "processing sentence": 91790,
      "response propose": 101230,
      "parsing graph": 85120,
      "efficient endtoend": 35071,
      "english penn": 37237,
      "treebank benchmark": 125629,
      "perform par": 86042,
      "par best": 84517,
      "best neural": 13383,
      "model transfer": 72224,
      "tagging lowresource": 117400,
      "crosslingual model": 24976,
      "popular method": 88100,
      "method predicting": 69061,
      "corpora provide": 23564,
      "data readily": 26318,
      "approaches address": 8050,
      "address drawbacks": 3678,
      "framework takes": 45710,
      "advantage crosslingual": 4038,
      "solely high": 110531,
      "high coverage": 50055,
      "dictionary propose": 30889,
      "model joint": 71378,
      "training sources": 123871,
      "sources data": 110882,
      "embeddings substantial": 35957,
      "empirical improvements": 36170,
      "baseline techniques": 12322,
      "techniques propose": 119957,
      "propose active": 92526,
      "competitive benchmark": 19644,
      "benchmark methods": 12840,
      "methods data": 69417,
      "data augmentation": 25628,
      "lowresource neural": 65543,
      "corpora lowresource": 23521,
      "resulting poor": 101464,
      "novel data": 80524,
      "augmentation approach": 10272,
      "words generating": 133971,
      "new sentence": 79126,
      "pairs containing": 83498,
      "settings method": 107820,
      "quality 29": 94583,
      "points baseline": 87849,
      "32 bleu": 928,
      "bleu backtranslation": 14278,
      "efficient natural": 35095,
      "language response": 60060,
      "computationally efficient": 20449,
      "method natural": 68981,
      "using ngram": 129967,
      "embedding features": 35407,
      "features encode": 43478,
      "pairs high": 83558,
      "search finds": 104317,
      "largescale commercial": 61410,
      "compared sequencetosequence": 19439,
      "sequencetosequence approach": 107151,
      "achieves quality": 2840,
      "small fraction": 110151,
      "fraction computational": 45364,
      "computational requirements": 20422,
      "effectiveness feature": 34892,
      "set augmentation": 107367,
      "augmentation using": 10315,
      "empirically shown": 36242,
      "important performance": 52204,
      "improvements various": 52937,
      "tasks despite": 119053,
      "despite importance": 29694,
      "hyperparameters like": 51236,
      "like number": 64079,
      "understand role": 126771,
      "evaluate effect": 38826,
      "segmentation classification": 104579,
      "task convolutional": 118023,
      "uses convolutional": 129211,
      "sentiment label": 106751,
      "tweet network": 126015,
      "network consists": 77210,
      "convolutional layer": 23158,
      "layer followed": 61721,
      "fullyconnected layer": 45984,
      "layer softmax": 61749,
      "network initialized": 77284,
      "inputs different": 55482,
      "different initializations": 31182,
      "combine results": 18640,
      "label given": 58398,
      "ranked fourth": 95644,
      "distillation text": 32523,
      "task classifying": 117966,
      "model distillation": 71010,
      "adapts extends": 3343,
      "originally proposed": 82569,
      "model compression": 70872,
      "compression using": 20329,
      "probabilistic predictions": 90827,
      "corpus documents": 23764,
      "documents train": 33304,
      "train classifiers": 122910,
      "new languages": 78981,
      "available adversarial": 10938,
      "feature adaptation": 43248,
      "adaptation technique": 3274,
      "applied model": 7093,
      "german french": 47892,
      "unlabeled target": 127409,
      "languages proposed": 60821,
      "comparable performance": 19155,
      "stateofart methods": 112553,
      "methods supervised": 69785,
      "learning universal": 63135,
      "universal sentence": 127331,
      "inference data": 54131,
      "rely word": 98752,
      "previously trained": 90626,
      "trained unsupervised": 123321,
      "unsupervised manner": 127661,
      "manner large": 66952,
      "base features": 11457,
      "obtain embeddings": 81279,
      "embeddings larger": 35775,
      "chunks text": 16969,
      "learning unsupervised": 63143,
      "unsupervised representations": 127706,
      "representations sentences": 99880,
      "satisfactory performance": 103652,
      "performance widely": 86853,
      "widely adopted": 132536,
      "supervised data": 115746,
      "data stanford": 26501,
      "stanford natural": 112370,
      "inference datasets": 54133,
      "consistently outperform": 21424,
      "outperform unsupervised": 82769,
      "range transfer": 95612,
      "transfer tasks": 124192,
      "obtain features": 81282,
      "work tends": 134847,
      "inference transfer": 54242,
      "learning nlp": 62848,
      "tasks encoder": 119093,
      "encoding models": 36714,
      "results study": 102219,
      "study model": 114439,
      "representations convolutional": 99575,
      "prevalent approach": 90358,
      "maps input": 67162,
      "sequence variable": 107096,
      "output sequence": 83120,
      "networks introduce": 77629,
      "introduce architecture": 56381,
      "based entirely": 11685,
      "entirely convolutional": 37730,
      "compared recurrent": 19432,
      "recurrent models": 97182,
      "parallelized training": 84706,
      "input length": 55359,
      "use gated": 128062,
      "gated linear": 46508,
      "linear units": 64387,
      "decoder layer": 28027,
      "layer separate": 61747,
      "separate attention": 106845,
      "attention module": 9937,
      "accuracy deep": 2130,
      "deep lstm": 28330,
      "lstm setup": 65666,
      "wmt14 englishgerman": 132813,
      "englishgerman wmt14": 37391,
      "wmt14 englishfrench": 132811,
      "faster speed": 43189,
      "gpu cpu": 48576,
      "datatotext generation": 27813,
      "generation benchmarks": 47324,
      "recently datasets": 96674,
      "data text": 26556,
      "text created": 120841,
      "unclear extent": 126633,
      "surface realisation": 116071,
      "challenging datasets": 16242,
      "datasets provide": 27649,
      "provide variety": 93954,
      "encourage development": 36751,
      "datasets important": 27510,
      "claim using": 17013,
      "metrics manual": 69984,
      "benchmark help": 12830,
      "help better": 49698,
      "better support": 13735,
      "support development": 115962,
      "development evaluation": 30386,
      "approach vietnamese": 8013,
      "short message": 108218,
      "message service": 68498,
      "service sms": 107326,
      "systems detect": 116843,
      "use machine": 128132,
      "content messages": 21906,
      "classify vietnamese": 17667,
      "appropriate preprocessing": 8426,
      "preprocessing method": 89325,
      "method existing": 68821,
      "existing tools": 40318,
      "good accuracy": 48459,
      "accuracy dataset": 2123,
      "dataset experiment": 26907,
      "model problem": 71791,
      "problem achieves": 90915,
      "accuracy 94": 2085,
      "misclassification rate": 70292,
      "result compared": 101364,
      "compared english": 19363,
      "baseline future": 12227,
      "future development": 46267,
      "compositional neural": 20121,
      "tree structure": 125613,
      "networks proven": 77715,
      "spite success": 111938,
      "success existing": 115080,
      "models suffer": 74126,
      "use shared": 128273,
      "shared compositional": 107967,
      "compositional function": 20113,
      "compositional process": 20125,
      "capture richness": 15376,
      "compositionality paper": 20145,
      "dynamically generated": 34347,
      "compositional rules": 20129,
      "results typical": 102290,
      "tasks effectiveness": 119081,
      "parsing transitionbased": 85268,
      "produce certain": 91878,
      "propagation paper": 92394,
      "enables direct": 36383,
      "leverage lexical": 63598,
      "information directly": 54494,
      "beam size": 12609,
      "reduce error": 97325,
      "transition systems": 124496,
      "treebank dependency": 125633,
      "task english": 118147,
      "modeling content": 72402,
      "content discourse": 21874,
      "discourse relations": 32076,
      "present joint": 89522,
      "modeling approach": 72375,
      "identify salient": 51548,
      "spoken meetings": 112012,
      "treated latent": 125557,
      "variables experimental": 130855,
      "corpora joint": 23507,
      "model outperform": 71625,
      "approaches phrasebased": 8282,
      "discourse relation": 32074,
      "relation prediction": 98064,
      "better predictive": 13675,
      "predictive performance": 89206,
      "political discourse": 87993,
      "vectors paper": 131460,
      "learning used": 63147,
      "discourse vectors": 32101,
      "produced training": 91974,
      "lexicon large": 63891,
      "presents simple": 89898,
      "simple implementation": 109445,
      "presidential elections": 89949,
      "number reasons": 80950,
      "suggestions improve": 115367,
      "internal representations": 56170,
      "learned endtoend": 62214,
      "optimization deep": 82189,
      "questionanswering qa": 95256,
      "qa task": 94522,
      "using basic": 129382,
      "basic concepts": 12517,
      "linguistic theory": 64567,
      "interpretability proposed": 56230,
      "performs comparably": 86991,
      "comparably stateoftheart": 19198,
      "stateoftheart squad": 112969,
      "internal representation": 56169,
      "product representation": 92042,
      "encode word": 36465,
      "attention overall": 9961,
      "trained model": 123201,
      "model support": 72118,
      "grammatical roles": 48718,
      "al 2003": 4592,
      "significant aspects": 108725,
      "unannotated text": 126588,
      "questions answers": 95273,
      "given model": 48063,
      "inductive bias": 54047,
      "approximately discrete": 8465,
      "deep investigation": 28239,
      "methods paper": 69663,
      "paper deep": 83796,
      "parallel comparable": 84590,
      "collections documents": 18503,
      "documents multiple": 33256,
      "characteristics different": 16522,
      "genres languages": 47836,
      "texts investigate": 121533,
      "methods language": 69572,
      "granularities text": 48744,
      "units order": 127262,
      "document styles": 33084,
      "styles languages": 114619,
      "matching based": 67398,
      "variety text": 131023,
      "proposes simple": 93617,
      "model augmented": 70707,
      "output lstm": 83091,
      "fed softmax": 43808,
      "softmax layer": 110508,
      "make prediction": 66708,
      "base model": 11463,
      "model order": 71619,
      "order enhance": 82316,
      "performance proposed": 86628,
      "proposed techniques": 93567,
      "integration multiple": 55812,
      "ensemble based": 37578,
      "model averaging": 70718,
      "snli dataset": 110312,
      "techniques effective": 119875,
      "effective boosting": 34638,
      "biomedical event": 14182,
      "event trigger": 39550,
      "trigger identification": 125784,
      "events complex": 39569,
      "complex interactions": 19823,
      "various biomedical": 131054,
      "biomedical entities": 14179,
      "methods current": 69415,
      "approaches rely": 8318,
      "complex handcrafted": 19813,
      "features consider": 43420,
      "window paper": 132710,
      "features present": 43664,
      "present sentence": 89688,
      "hidden state": 49908,
      "features avoid": 43375,
      "avoid relying": 11242,
      "features generated": 43524,
      "generated using": 47148,
      "nlp toolkits": 79784,
      "experiments shown": 41138,
      "shown achieve": 108442,
      "longterm goals": 65204,
      "goals artificial": 48418,
      "agent communicate": 4298,
      "language existing": 59024,
      "work natural": 134645,
      "learning relies": 62956,
      "relies heavily": 98651,
      "heavily training": 49661,
      "dataset annotated": 26735,
      "annotated labels": 6204,
      "external training": 42048,
      "data essentially": 25905,
      "representation knowledge": 99272,
      "agent trained": 4313,
      "trained way": 123333,
      "generalization behavior": 46767,
      "behavior different": 12653,
      "process humans": 91501,
      "learning agent": 62315,
      "agent learns": 4306,
      "teacher learning": 119695,
      "feedback learning": 43832,
      "improving language": 53107,
      "language skills": 60098,
      "achieve goal": 2463,
      "goal propose": 48384,
      "imitation reinforcement": 51840,
      "validate effectiveness": 130715,
      "subject specific": 114684,
      "data stream": 26507,
      "microblogging service": 70051,
      "source data": 110739,
      "algorithm accurately": 4687,
      "accurately classify": 2378,
      "mutually exclusive": 75990,
      "algorithm run": 4793,
      "high efficiency": 50068,
      "composition model": 20098,
      "model convolutional": 70921,
      "networks dependency": 77564,
      "present transitionbased": 89749,
      "compose word": 20062,
      "representations characters": 99541,
      "characters character": 16600,
      "shows great": 108583,
      "great improvement": 49006,
      "model especially": 71094,
      "better using": 13763,
      "extra data": 42058,
      "greedy parser": 49072,
      "parser ballesteros": 84922,
      "geometry word": 47870,
      "embeddings help": 35736,
      "classification case": 17141,
      "based representations": 11993,
      "representations investigate": 99706,
      "algebraic topology": 4680,
      "methods enable": 69470,
      "set vectors": 107633,
      "space work": 111088,
      "work evaluate": 134494,
      "evaluate utility": 38941,
      "document clustering": 32972,
      "embeddings benefit": 35571,
      "worse simple": 135068,
      "simple techniques": 109529,
      "techniques like": 119922,
      "does provide": 33381,
      "topic sentiment": 122574,
      "datasets analysis": 27307,
      "analysis effect": 5585,
      "information predicateargument": 54854,
      "structure analysis": 113807,
      "zero anaphora": 135471,
      "resolution paper": 100774,
      "analysis pasa": 5745,
      "resolution zar": 100794,
      "straightforward approach": 113430,
      "directly predicts": 31900,
      "predicts relationships": 89219,
      "relationships predicates": 98328,
      "arguments semantic": 8996,
      "features designed": 43449,
      "compensate absence": 19600,
      "information using": 55080,
      "using features": 129671,
      "used dependency": 128482,
      "dependencies crucial": 29074,
      "world evaluating": 135029,
      "evaluating word": 39110,
      "methods exploit": 69494,
      "exploit word": 41447,
      "widespread use": 132619,
      "use modern": 128149,
      "modern natural": 74399,
      "encode necessary": 36441,
      "representations predict": 99809,
      "human participants": 50924,
      "fail encode": 42957,
      "prediction errors": 89056,
      "approaches help": 8179,
      "advanced neural": 3953,
      "python toolkit": 94488,
      "toolkit based": 122409,
      "sequencetosequence architectures": 107154,
      "training inference": 123660,
      "new architecture": 78801,
      "submissions wmt": 114757,
      "wmt multimodal": 132801,
      "translation news": 125018,
      "2016 2017": 602,
      "assisted analysis": 9583,
      "vowel length": 131957,
      "algorithms automatic": 4830,
      "automatic analysis": 10490,
      "create new": 24631,
      "research opportunities": 100575,
      "empirical approaches": 36153,
      "larger set": 61381,
      "conditioning factors": 20811,
      "paper illustrates": 83976,
      "challenge automatic": 15998,
      "automatic methods": 10593,
      "methods analysis": 69307,
      "language subsaharan": 60133,
      "propose multiple": 92803,
      "multiple features": 75566,
      "different factors": 31143,
      "tokens proposed": 122328,
      "proposed features": 93281,
      "learning adaptive": 62309,
      "understanding concept": 126816,
      "important language": 52177,
      "understanding lu": 126880,
      "issue paper": 57015,
      "relationship concepts": 98287,
      "hierarchical semantic": 50003,
      "representation concepts": 99194,
      "concepts proposed": 20636,
      "proposed semantic": 93544,
      "concepts based": 20605,
      "based new": 11897,
      "new hierarchical": 78942,
      "representation transfer": 99446,
      "approaches developed": 8125,
      "developed adaptive": 30250,
      "applied tasks": 7125,
      "value set": 130787,
      "thorough empirical": 121881,
      "studies validate": 114299,
      "efficiency effectiveness": 35023,
      "performance f1score": 86376,
      "unwritten languages": 127778,
      "communications social": 19047,
      "media colloquial": 68089,
      "method inducing": 68905,
      "expert knowledge": 41225,
      "amounts monolingual": 5351,
      "data investigate": 26053,
      "darijaenglish translation": 25541,
      "lexicon induced": 63886,
      "translation statistical": 125292,
      "tagger based": 117355,
      "networks cnn": 77534,
      "cnn used": 18028,
      "vectors encoding": 131432,
      "information cnn": 54420,
      "tagging tasks": 117454,
      "tuning hyperparameters": 125935,
      "tagging morphological": 117410,
      "outofvocabulary problem": 82679,
      "years recurrent": 135294,
      "rnns proved": 102979,
      "effective nlp": 34725,
      "despite great": 29689,
      "labeling limited": 58506,
      "models proved": 73830,
      "effective domain": 34665,
      "propose solution": 93073,
      "far simpler": 43100,
      "simpler effective": 109555,
      "input network": 55372,
      "embeddings way": 36019,
      "way words": 132144,
      "words compare": 133867,
      "elman jordan": 35308,
      "rnn lstm": 102924,
      "lstm gru": 65626,
      "tasks spoken": 119518,
      "understanding slu": 126958,
      "label embeddings": 58394,
      "layer proposed": 61742,
      "uses parameters": 129260,
      "far fewer": 43079,
      "similarity semantic": 109306,
      "measures important": 67870,
      "general use": 46728,
      "use perform": 128196,
      "perform specific": 86074,
      "domains study": 33863,
      "study introduce": 114411,
      "word2vec word": 133686,
      "used semantic": 128743,
      "similarity calculation": 109206,
      "lexicon based": 63874,
      "methods prove": 69696,
      "prove proposed": 93702,
      "generic corpus": 47794,
      "trained domain": 123122,
      "use lexical": 128123,
      "methods augment": 69325,
      "prove text": 93705,
      "text lemmatization": 121085,
      "table retrieval": 117260,
      "unstructured text": 127582,
      "retrieval given": 102395,
      "given query": 48095,
      "task relevant": 118631,
      "progress improving": 92153,
      "improving area": 53071,
      "powerful models": 88672,
      "richer training": 102809,
      "training evaluation": 123616,
      "remedy present": 98860,
      "present ranking": 89659,
      "ranking based": 95665,
      "carefully designed": 15520,
      "architectures measure": 8823,
      "relevance query": 98514,
      "content table": 21956,
      "release opendomain": 98466,
      "opendomain dataset": 81964,
      "conduct comprehensive": 20830,
      "experiments real": 41104,
      "synthetic datasets": 116626,
      "datasets results": 27685,
      "results verify": 102317,
      "verify effectiveness": 131580,
      "present challenges": 89404,
      "challenges task": 16215,
      "embed words": 35351,
      "context syntactic": 22288,
      "tasks present": 119391,
      "models embedding": 73130,
      "context surrounding": 22287,
      "words models": 134059,
      "models refer": 73899,
      "token embeddings": 122251,
      "word specific": 133582,
      "specific given": 111445,
      "semantic role": 105244,
      "explore simple": 41587,
      "simple efficient": 109419,
      "efficient token": 35121,
      "standard neural": 112277,
      "architectures learn": 8812,
      "embeddings large": 35774,
      "large unannotated": 61309,
      "text evaluate": 120921,
      "taggers dependency": 117364,
      "parsers trained": 85049,
      "smaller amounts": 110226,
      "embeddings consistently": 35609,
      "model proven": 71819,
      "representations capturing": 99534,
      "capturing sentence": 15490,
      "propose suite": 93094,
      "validate hypothesis": 130722,
      "hypothesis given": 51265,
      "given current": 48010,
      "sentence provide": 106010,
      "provide similar": 93919,
      "model second": 71955,
      "decoder help": 28017,
      "help model": 49751,
      "model generalize": 71228,
      "generalize better": 46803,
      "better semantic": 13717,
      "embedding initialization": 35416,
      "learning better": 62406,
      "better sentence": 13719,
      "train model": 122955,
      "sentences evaluate": 106295,
      "evaluate trained": 38937,
      "model supervised": 72116,
      "detection text": 30081,
      "classification benchmarks": 17136,
      "empirically proposed": 36240,
      "powerful alternative": 88663,
      "model classification": 70828,
      "questions learning": 95326,
      "similarity measurements": 109260,
      "extracting learning": 42218,
      "learning material": 62711,
      "used classify": 128424,
      "domains cognitive": 33740,
      "classify questions": 17655,
      "experiment designed": 40461,
      "investigate semantic": 56812,
      "semantic relationship": 105220,
      "verbs used": 131563,
      "obtain accurate": 81255,
      "accurate classification": 2328,
      "verbs collected": 131554,
      "allows measure": 5170,
      "proposed solution": 93552,
      "able classify": 1620,
      "using confusion": 129473,
      "confusion matrix": 21054,
      "90 f1": 1367,
      "research suggests": 100632,
      "questions based": 95278,
      "childdirected speech": 16720,
      "increases function": 53644,
      "age child": 4281,
      "controlled paper": 22844,
      "significant negative": 108827,
      "negative correlation": 76923,
      "negative log": 76932,
      "log probability": 64983,
      "nonprojective parsing": 80280,
      "parsing restricted": 85221,
      "shown beneficial": 108444,
      "parser previous": 84987,
      "lack information": 58718,
      "variants recently": 130881,
      "dynamic oracle": 34319,
      "loss experiments": 65264,
      "experiments datasets": 40867,
      "majority languages": 66605,
      "languages finegrained": 60590,
      "translation compare": 124699,
      "compare approaches": 19224,
      "translation pure": 125160,
      "finegrained manual": 44365,
      "error annotation": 38288,
      "error types": 38345,
      "types annotation": 126247,
      "multidimensional quality": 75069,
      "quality metrics": 94718,
      "annotation performed": 6350,
      "annotators interannotator": 6493,
      "results best": 101550,
      "performing neural": 86963,
      "errors produced": 38400,
      "alzheimers disease": 5259,
      "transcribed speech": 123985,
      "utterance text": 130616,
      "disease ad": 32333,
      "different versions": 31556,
      "text semantic": 121280,
      "naturally applicable": 76637,
      "method computing": 68721,
      "conduct comparison": 20829,
      "diagnostic classification": 30507,
      "task different": 118096,
      "small significant": 110204,
      "derived word": 29368,
      "embedding clustering": 35388,
      "results considerably": 101629,
      "certain properties": 15952,
      "representation model": 99328,
      "model formal": 71205,
      "language case": 58868,
      "text natural": 121139,
      "using dependency": 129554,
      "structures obtained": 114093,
      "stateoftheart stanford": 112971,
      "components result": 20047,
      "data structure": 26509,
      "sentence split": 106090,
      "process fully": 91492,
      "documents different": 33217,
      "domains report": 33848,
      "report initial": 99008,
      "initial evaluation": 55213,
      "creation strategies": 24723,
      "translation training": 125389,
      "usually uses": 130466,
      "shorter sentences": 108296,
      "sentence efficient": 105831,
      "efficient computation": 35066,
      "computation previous": 20353,
      "work noted": 134657,
      "length making": 63365,
      "processing speed": 91800,
      "essential step": 38565,
      "step nmt": 113288,
      "nmt training": 79996,
      "training widely": 123950,
      "used nmt": 128663,
      "nmt toolkits": 79989,
      "disparate strategies": 32371,
      "work investigates": 134596,
      "large effect": 61086,
      "compared simple": 19441,
      "improving text": 53176,
      "reduced precision": 97373,
      "precision paper": 88795,
      "presents analysis": 89817,
      "quality text": 94810,
      "representing data": 100045,
      "memory footprint": 68315,
      "hardware platforms": 49498,
      "corpora using": 23613,
      "account results": 2036,
      "accuracy cases": 2111,
      "reduction general": 97448,
      "bahasa indonesia": 11385,
      "overcome data": 83278,
      "language different": 58963,
      "languages result": 60850,
      "result different": 101370,
      "different effects": 31118,
      "russian language": 103493,
      "language shown": 60090,
      "used train": 128818,
      "words extracted": 133956,
      "wikipedia far": 132656,
      "far know": 43087,
      "train set": 123019,
      "set used": 107626,
      "perplexity values": 87075,
      "interesting note": 56081,
      "model modified": 71528,
      "technique outperforms": 119806,
      "consistently improving": 21423,
      "ngram order": 79450,
      "nmt encoderdecoder": 79858,
      "networks used": 77809,
      "used nlp": 128661,
      "graphbased models": 48903,
      "transitionbased models": 124508,
      "components achieve": 20016,
      "parsing constituent": 85086,
      "constituent parsing": 21545,
      "parsing respectively": 85220,
      "work empirically": 134487,
      "apply simple": 7217,
      "simple encoderdecoder": 109424,
      "achieving comparable": 2939,
      "results parser": 102029,
      "parser dyer": 84944,
      "vinyals et": 131729,
      "synonym discovery": 116345,
      "discovery knowledge": 32138,
      "text crucial": 120846,
      "crucial task": 25175,
      "corpora news": 23541,
      "scientific papers": 104017,
      "challenging current": 16238,
      "current systems": 25377,
      "systems entity": 116872,
      "input names": 55370,
      "ignoring fact": 51689,
      "refer multiple": 97497,
      "multiple entities": 75553,
      "methods require": 69728,
      "require training": 100202,
      "manually created": 67053,
      "domain experts": 33528,
      "supervisedlearning systems": 115861,
      "problem automatic": 90946,
      "identifying synonyms": 51629,
      "base entities": 11454,
      "entities given": 37788,
      "domainspecific corpus": 33894,
      "base form": 11458,
      "form set": 45123,
      "set strings": 107590,
      "disambiguate meaning": 31941,
      "determine important": 30128,
      "important features": 52164,
      "features task": 43753,
      "integrate kinds": 55754,
      "corpuslevel statistics": 24086,
      "statistics textual": 113206,
      "contexts particular": 22416,
      "jointly optimizes": 57374,
      "enhance training": 37497,
      "training stage": 123878,
      "inference stage": 54223,
      "discover synonyms": 32119,
      "given entities": 48028,
      "entities experimental": 37779,
      "prove effectiveness": 93697,
      "field machine": 43964,
      "nmt suffers": 79971,
      "suffers problem": 115264,
      "problem repeating": 91205,
      "missing words": 70338,
      "words translation": 134280,
      "2017 proposed": 637,
      "nmt using": 80006,
      "using backtranslation": 129381,
      "backtranslation method": 11338,
      "bahdanau et": 11387,
      "bidirectional translation": 13982,
      "model finetuning": 71192,
      "offers significant": 81592,
      "alleviates problem": 5061,
      "task addition": 117844,
      "effectiveness pretraining": 34924,
      "translation backtranslation": 124663,
      "e2e dataset": 34358,
      "dataset new": 27054,
      "generation paper": 47527,
      "data new": 26168,
      "dataset training": 27242,
      "training endtoend": 123608,
      "endtoend datadriven": 36882,
      "datadriven natural": 26671,
      "restaurant domain": 101319,
      "frequently used": 45881,
      "used datasets": 128472,
      "dataset poses": 27097,
      "human reference": 50952,
      "reference texts": 97539,
      "including discourse": 53287,
      "discourse phenomena": 32072,
      "selection learning": 104795,
      "learning dataset": 62483,
      "establish baseline": 38577,
      "baseline dataset": 12207,
      "difficulties associated": 31679,
      "data generating": 25976,
      "newly formed": 79273,
      "play crucial": 87689,
      "success failure": 115081,
      "computational method": 20404,
      "entities use": 37889,
      "set diverse": 107420,
      "rnns syntactic": 102980,
      "semantic arguments": 104992,
      "built using": 14945,
      "achieves competitive": 2767,
      "performance calls": 86194,
      "parser introduce": 84965,
      "treebank training": 125647,
      "training multitask": 123724,
      "multitask objective": 75886,
      "required training": 100226,
      "time syntactic": 122117,
      "cheaper alternative": 16689,
      "continuous lexical": 22625,
      "semantic protoroles": 105189,
      "better multiple": 13633,
      "baselines improving": 12413,
      "tweets present": 126049,
      "present future": 89499,
      "helps solving": 49831,
      "detection microblog": 29991,
      "model handle": 71272,
      "poses challenges": 88261,
      "challenges short": 16213,
      "short length": 108214,
      "tweets work": 126074,
      "based objective": 11904,
      "literature provide": 64767,
      "provide interesting": 93861,
      "advancing field": 4029,
      "field building": 43933,
      "building highquality": 14844,
      "evaluating natural": 39080,
      "automated metrics": 10456,
      "metrics bleu": 69932,
      "bleu widely": 14343,
      "evaluating dialogue": 39047,
      "dialogue response": 30733,
      "generation previous": 47558,
      "work dialogue": 134474,
      "generation shown": 47621,
      "metrics correlate": 69943,
      "correlate strongly": 24202,
      "strongly human": 113745,
      "human judgment": 50880,
      "dialogue responses": 30735,
      "exhibit lower": 40003,
      "task consists": 118011,
      "dialogue acts": 30634,
      "conduct empirical": 20837,
      "findings indicate": 44304,
      "stronger correlation": 113735,
      "setting compared": 107738,
      "correlate better": 24191,
      "better datasets": 13560,
      "provide multiple": 93877,
      "reference sentences": 97533,
      "sentences addition": 106200,
      "solved simple": 110630,
      "simple models": 109472,
      "character embedding": 16426,
      "word formation": 133288,
      "words combined": 133865,
      "combined form": 18654,
      "form new": 45110,
      "sequencetosequence s2s": 107200,
      "explicitly use": 41397,
      "use additional": 127881,
      "additional phonetic": 3557,
      "phonetic information": 87293,
      "word lists": 133353,
      "candidate generation": 15104,
      "generation strategy": 47638,
      "strategy specifically": 113541,
      "approach superior": 7947,
      "superior stateoftheart": 115696,
      "accuracy human": 2177,
      "evaluation context": 39159,
      "context aware": 22017,
      "achieved excellent": 2619,
      "mechanism generates": 67989,
      "weights word": 132386,
      "contribution context": 22770,
      "networks context": 77547,
      "achieve similar": 2547,
      "vectors efficient": 131428,
      "analysis context": 5545,
      "weights shows": 132382,
      "result deep": 101368,
      "good translation": 48505,
      "translation current": 124726,
      "approaches crosslingual": 8112,
      "try leverage": 125898,
      "labeled english": 58457,
      "bilingual lexicons": 14046,
      "bilingual vector": 14067,
      "possible use": 88441,
      "use single": 128279,
      "single linear": 109755,
      "linear transformation": 64383,
      "capture finegrained": 15303,
      "finegrained sentiment": 44381,
      "crosslingual setting": 25013,
      "apply crosslingual": 7169,
      "sentiment models": 106761,
      "models diverse": 73099,
      "set tasks": 107601,
      "effectively leveraging": 34830,
      "leveraging english": 63675,
      "english sentiment": 37274,
      "sentiment knowledge": 106750,
      "knowledge need": 58079,
      "need accurate": 76775,
      "accurate translation": 2369,
      "translation analyze": 124638,
      "extract features": 42083,
      "languages scarce": 60856,
      "scarce data": 103791,
      "data low": 26097,
      "low cost": 65353,
      "cost making": 24362,
      "evaluation event": 39206,
      "impact various": 51901,
      "systems automatic": 116745,
      "using popular": 130032,
      "quantify impact": 94851,
      "impact training": 51896,
      "learning classifiers": 62437,
      "quality automatically": 94602,
      "learning using": 63150,
      "using computational": 129462,
      "paradigms language": 84559,
      "learning interactive": 62657,
      "interactive learning": 56020,
      "crosssituational learning": 25083,
      "role social": 103219,
      "analyze relationship": 5998,
      "new mixed": 79023,
      "test algorithms": 120420,
      "algorithms experiments": 4851,
      "feedback experiments": 43826,
      "new algorithms": 78769,
      "algorithms perform": 4873,
      "respect traditional": 101108,
      "aspects neural": 9399,
      "generation work": 47701,
      "work neural": 134651,
      "neural natural": 78223,
      "content generated": 21884,
      "generated text": 47137,
      "text experiment": 120932,
      "stylistic aspects": 114626,
      "aspects generated": 9384,
      "text addition": 120629,
      "addition content": 3415,
      "content method": 21907,
      "desired content": 29664,
      "stylistic parameters": 114630,
      "parameters serve": 84783,
      "sentences corresponding": 106266,
      "corresponding required": 24300,
      "style content": 114571,
      "architecture text": 8760,
      "learning multitask": 62818,
      "learning leverages": 62689,
      "common features": 18879,
      "features yield": 43795,
      "works consider": 134932,
      "failing model": 42978,
      "propose multitask": 92807,
      "types recurrent": 126345,
      "fuse information": 46223,
      "information multiple": 54776,
      "tasks architecture": 118939,
      "various interactions": 131105,
      "interactions tasks": 56003,
      "datasets text": 27757,
      "improve performances": 52495,
      "tasks additional": 118921,
      "greedy dependency": 49071,
      "parser bidirectional": 84925,
      "nonprojective dependency": 80278,
      "parsing bidirectional": 85080,
      "lstm approach": 65591,
      "goldberg 2016": 48449,
      "mitigate error": 70367,
      "participated conll": 85335,
      "conll 2017": 21077,
      "2017 ud": 642,
      "ud shared": 126532,
      "task spite": 118733,
      "using ensemble": 129633,
      "parser obtained": 84974,
      "languages ranking": 60831,
      "ranking 7th": 95660,
      "33 teams": 943,
      "code available": 18074,
      "qualitative data": 94558,
      "qualitative quantitative": 94570,
      "individual documents": 53907,
      "framework enables": 45517,
      "data analysts": 25598,
      "use stateoftheart": 128294,
      "stateoftheart natural": 112767,
      "media studies": 68167,
      "usage framework": 127859,
      "political science": 87999,
      "various topics": 131225,
      "topics preserving": 122651,
      "important contexts": 52136,
      "contexts case": 22381,
      "style writing": 114612,
      "analyzed using": 6029,
      "author piece": 10366,
      "piece text": 87495,
      "automatic tools": 10698,
      "tools help": 122449,
      "writing text": 135109,
      "discriminative approach": 32206,
      "calculate values": 15018,
      "metrics indicate": 69974,
      "preserving semantics": 89940,
      "text finally": 120952,
      "random noise": 95504,
      "hypernymy extraction": 51221,
      "extraction based": 42270,
      "learning present": 62902,
      "embeddings contrast": 35617,
      "pairs natural": 83588,
      "negative training": 76957,
      "supervised relation": 115826,
      "explicit negative": 41338,
      "used regularization": 128722,
      "intelligent assistants": 55845,
      "interact users": 55935,
      "various components": 131063,
      "components automatic": 20017,
      "recognition natural": 96917,
      "useful signal": 128930,
      "signal user": 108695,
      "method predict": 69060,
      "using user": 130348,
      "user logs": 129010,
      "logs commercial": 65055,
      "results demonstrated": 101708,
      "designed detect": 29587,
      "component improve": 19983,
      "cause detection": 15821,
      "problem detecting": 91007,
      "model taking": 72139,
      "initial experiments": 55214,
      "testing procedure": 120600,
      "abstract linguistic": 1767,
      "features linguistic": 43596,
      "linguistic complexity": 64443,
      "negative results": 76945,
      "results analyzing": 101508,
      "interesting patterns": 56083,
      "patterns scientific": 85778,
      "submissions wmt17": 114758,
      "task englishgerman": 118149,
      "attentive neural": 10120,
      "competitive scores": 19688,
      "obtained ensembling": 81368,
      "systems exploiting": 116897,
      "target monolingual": 117676,
      "corpora backtranslation": 23421,
      "quantity quality": 94900,
      "best entry": 13328,
      "16 bleu": 376,
      "document retrieval": 33069,
      "paper challenges": 83759,
      "formal writing": 45181,
      "writing target": 135107,
      "target documents": 117604,
      "writing task": 135108,
      "task query": 118593,
      "query sentence": 94988,
      "sentence extracted": 105867,
      "target document": 117603,
      "establish strong": 38594,
      "reranking approach": 100366,
      "improve initial": 52396,
      "semantic structures": 105311,
      "structures generated": 114078,
      "shows improvement": 108587,
      "hierarchical representations": 50001,
      "representations attentionbased": 99510,
      "hierarchical attentional": 49937,
      "model focuses": 71198,
      "local global": 64921,
      "global semantic": 48268,
      "treebased encoder": 125669,
      "likelihood target": 64128,
      "mechanism used": 68045,
      "attentive information": 10115,
      "information lexical": 54728,
      "phrase vectors": 87377,
      "rare word": 95747,
      "word encoding": 133259,
      "encoding proposed": 36719,
      "model extended": 71146,
      "outofvocabulary oov": 82673,
      "oov problem": 81865,
      "results reveal": 102137,
      "outperforms sequencetosequence": 82980,
      "treebased neural": 125670,
      "text datasets": 120856,
      "datasets visualizing": 27794,
      "visualizing relevant": 131842,
      "dataset important": 26970,
      "important explore": 52159,
      "applying advanced": 7228,
      "advanced machine": 3945,
      "algorithms best": 4833,
      "documents respect": 33286,
      "belonging different": 12721,
      "propose extract": 92673,
      "collection texts": 18493,
      "texts summarize": 121625,
      "summarize contents": 115599,
      "contents documents": 21987,
      "word clouds": 132955,
      "compare methods": 19255,
      "words demonstrate": 133904,
      "providing overview": 94128,
      "dataset scientific": 27178,
      "scientific publications": 104019,
      "discovering trending": 32129,
      "trending topics": 125744,
      "new york": 79257,
      "york times": 135453,
      "times article": 122166,
      "article snippets": 9159,
      "simple language": 109451,
      "models training": 74236,
      "training estimate": 123613,
      "conditional probabilities": 20771,
      "time specifically": 122107,
      "aspect approach": 9315,
      "approach models": 7714,
      "trained simple": 123277,
      "simple negative": 109479,
      "pairwise ranking": 83680,
      "ranking methods": 95674,
      "methods basis": 69339,
      "used discriminative": 128496,
      "discriminative training": 32218,
      "training approaches": 123363,
      "structure prediction": 113934,
      "prediction problems": 89108,
      "problem ranking": 91194,
      "enables simple": 36403,
      "translation framework": 124828,
      "framework directly": 45498,
      "fit given": 44580,
      "furthermore propose": 46205,
      "loss functions": 65275,
      "experiments largescale": 40979,
      "largescale chineseenglish": 61404,
      "describes model": 29416,
      "model automatic": 70709,
      "meanings time": 67750,
      "word employ": 133257,
      "discover groups": 32115,
      "vector based": 131298,
      "based svm": 12097,
      "classifier learns": 17556,
      "learns recognize": 63228,
      "rulebased methods": 103369,
      "structural semantic": 113793,
      "effective possibly": 34731,
      "correlate results": 24200,
      "competition task": 19627,
      "effects using": 35007,
      "word statistics": 133586,
      "task employing": 118140,
      "pun recognition": 94388,
      "automatic interpretation": 10569,
      "thesaurus implementation": 121840,
      "main semantic": 66461,
      "fields semantic": 44006,
      "rulebased model": 103370,
      "words definitions": 133902,
      "semeval task": 105518,
      "searching target": 104372,
      "distributional feature": 32699,
      "feature overlap": 43299,
      "distributional method": 32709,
      "method modeling": 68969,
      "role fillers": 103173,
      "features candidate": 43387,
      "method consistently": 68728,
      "stateoftheart achieves": 112574,
      "better comparable": 13539,
      "reported literature": 99064,
      "literature unsupervised": 64779,
      "unsupervised systems": 127731,
      "systems provides": 117088,
      "provides explicit": 94029,
      "explicit representation": 41343,
      "representation features": 99239,
      "learning predicting": 62900,
      "online content": 81753,
      "content using": 21971,
      "attention span": 10015,
      "content news": 21913,
      "major factor": 66558,
      "designed predict": 29618,
      "distinct datasets": 32533,
      "datasets news": 27597,
      "articles news": 9202,
      "news videos": 79396,
      "media contain": 68091,
      "datasets approach": 27313,
      "shallow approaches": 107914,
      "margin 15": 67184,
      "additionally using": 3637,
      "vectors embedding": 131429,
      "layer improves": 61725,
      "models especially": 73165,
      "especially training": 38519,
      "set small": 107585,
      "small knowledge": 110163,
      "knowledge attempt": 57757,
      "popularity prediction": 88151,
      "using textual": 130289,
      "information title": 55046,
      "embeddings improve": 35747,
      "performance neural": 86564,
      "model cost": 70934,
      "increasing model": 53676,
      "sublexical level": 114722,
      "level approach": 63424,
      "approach quite": 7833,
      "taskspecific training": 119642,
      "training optimize": 123765,
      "report interesting": 99009,
      "results variety": 102308,
      "using subword": 130235,
      "generalization lexical": 46777,
      "space using": 111075,
      "embeddings downstream": 35670,
      "downstream nlp": 34013,
      "tasks limited": 119265,
      "oov words": 81869,
      "exist paper": 40028,
      "approach generating": 7588,
      "oov word": 81868,
      "unlike prior": 127448,
      "embedding corpus": 35391,
      "corpus instead": 23842,
      "instead learning": 55667,
      "type level": 126209,
      "evaluations demonstrate": 39458,
      "demonstrate power": 28821,
      "23 languages": 790,
      "performance wordbased": 86861,
      "wordbased baseline": 133696,
      "morphosyntactic attributes": 74815,
      "characterbased model": 16493,
      "model lowresource": 71481,
      "gender using": 46618,
      "using characterlevel": 129436,
      "compared method": 19382,
      "bayes logistic": 12569,
      "ngrams features": 79464,
      "features evaluated": 43490,
      "evaluated models": 38992,
      "models dataset": 73014,
      "common use": 18944,
      "names names": 76148,
      "names using": 76153,
      "accuracy results": 2265,
      "better ones": 13638,
      "classical machine": 17089,
      "twitter social": 126124,
      "media users": 68179,
      "users make": 129142,
      "make explicit": 66667,
      "author expresses": 10362,
      "popular beliefs": 88078,
      "question build": 95134,
      "build corpus": 14759,
      "corpus tweets": 24052,
      "tweets annotated": 126018,
      "wisdom crowds": 132737,
      "users explicit": 129116,
      "data past": 26226,
      "sentiment tweet": 106811,
      "used measure": 128626,
      "measure reliability": 67820,
      "hierarchical embeddings": 49958,
      "hypernymy detection": 51220,
      "learn hierarchical": 62060,
      "embeddings shown": 35935,
      "shown limitations": 108491,
      "unsupervised measure": 127663,
      "specific order": 111472,
      "hierarchy model": 50029,
      "able generalize": 1643,
      "hypernymy pairs": 51224,
      "sets training": 107719,
      "unsupervised measures": 127664,
      "measures embedding": 67860,
      "wikipedia editors": 132654,
      "content wikipedia": 21974,
      "problem determining": 91008,
      "span wikipedia": 111139,
      "articles approach": 9169,
      "problem classifying": 90962,
      "classifying textual": 17682,
      "propose sequence": 93041,
      "classification approach": 17118,
      "finegrained level": 44362,
      "provide thorough": 93941,
      "thorough experimental": 121886,
      "scientific domain": 104004,
      "improvement evaluation": 52705,
      "spell checking": 111895,
      "tools automated": 122436,
      "automated text": 10471,
      "analysis widely": 5936,
      "means extract": 67762,
      "complex word": 19894,
      "word features": 133284,
      "features like": 43594,
      "verb tense": 131527,
      "language features": 59045,
      "word form": 133286,
      "verbs nouns": 131558,
      "nouns different": 80441,
      "dictionary known": 30883,
      "known words": 58333,
      "words morphological": 134062,
      "morphological rules": 74731,
      "spell checkers": 111894,
      "designed generate": 29594,
      "generate valid": 47042,
      "checking tools": 16698,
      "gender number": 46606,
      "derive words": 29343,
      "cap 2017": 15163,
      "2017 challenge": 629,
      "challenge challenge": 16004,
      "problem named": 91135,
      "tweets written": 126075,
      "written french": 135129,
      "dataset released": 27155,
      "tweets challenging": 126022,
      "problem especially": 91035,
      "especially number": 38489,
      "number entities": 80880,
      "examples statistics": 39881,
      "statistics data": 113196,
      "challenge teams": 16109,
      "methods employed": 69469,
      "scores achieved": 104150,
      "terms f1": 120324,
      "constructed dataset": 21658,
      "dataset comprising": 26815,
      "entities best": 37751,
      "knowledge dataset": 57852,
      "dataset french": 26944,
      "works learn": 134957,
      "learn lexicon": 62081,
      "corpus improve": 23831,
      "embeddings model": 35811,
      "model lexicon": 71459,
      "update neural": 127797,
      "networks corpus": 77557,
      "corpus lexicon": 23864,
      "difference corpus": 30941,
      "lexicon best": 63875,
      "embeddings learning": 35785,
      "words use": 134290,
      "context evaluate": 22084,
      "method compare": 68708,
      "trained proposed": 123247,
      "model control": 70911,
      "works word": 134993,
      "model provides": 71825,
      "better embeddings": 13568,
      "words improves": 133992,
      "classification learning": 17251,
      "time search": 122096,
      "systems focused": 116910,
      "providing relevant": 94131,
      "hundreds years": 51158,
      "intent user": 55911,
      "retrieving relevant": 102472,
      "relevant historical": 98553,
      "content significant": 21945,
      "significant challenge": 108733,
      "challenge common": 16008,
      "introduce temporal": 56553,
      "temporal relationship": 120125,
      "relationship model": 98300,
      "identifying given": 51600,
      "words relate": 134163,
      "algorithmic framework": 4822,
      "framework task": 45712,
      "image text": 51798,
      "classification novel": 17310,
      "model widely": 72316,
      "used computer": 128443,
      "collection documents": 18472,
      "documents represented": 33283,
      "represented word": 100039,
      "vector using": 131399,
      "kmeans algorithm": 57714,
      "algorithm applied": 4695,
      "applied word": 7146,
      "vectors order": 131456,
      "obtain fixedsize": 81284,
      "related word": 97911,
      "documents assigned": 33186,
      "cluster centroid": 17926,
      "end document": 36805,
      "document represented": 33068,
      "represented bag": 100027,
      "building single": 14887,
      "documents propose": 33276,
      "kind representation": 57681,
      "results text": 102269,
      "mining tasks": 70263,
      "categorization topic": 15771,
      "topic polarity": 122560,
      "networks output": 77686,
      "output constraints": 83062,
      "networks increasingly": 77626,
      "complex problems": 19861,
      "role labeling": 103182,
      "output structures": 83128,
      "problems require": 91368,
      "constraints output": 21606,
      "capture properties": 15363,
      "network able": 77119,
      "inference method": 54165,
      "method neural": 68985,
      "performing rulebased": 86967,
      "gradientbased training": 48611,
      "model weights": 72314,
      "inference procedure": 54200,
      "generates output": 47168,
      "satisfies constraints": 103656,
      "parsing sequence": 85232,
      "sequence transduction": 107090,
      "improves accuracy": 52945,
      "network stateoftheart": 77445,
      "determining semantic": 30153,
      "similarity core": 109213,
      "research subject": 100631,
      "models sentence": 73999,
      "use shallow": 128272,
      "information capturing": 54407,
      "logical semantic": 65019,
      "capture deeper": 15286,
      "relations sentence": 98252,
      "experiments able": 40756,
      "systems features": 116904,
      "learning textual": 63105,
      "ensemble deep": 37592,
      "task build": 117947,
      "build break": 14749,
      "task workshop": 118856,
      "workshop building": 135002,
      "building linguistically": 14858,
      "linguistically generalizable": 64585,
      "generalizable nlp": 46757,
      "automated sentiment": 10467,
      "sentiment analyzer": 106681,
      "mining sentiment": 70258,
      "media research": 68162,
      "scientific community": 103995,
      "problems particular": 91356,
      "languages dialects": 60498,
      "approach construction": 7458,
      "algerian dialect": 4683,
      "table containing": 117253,
      "different prefixes": 31345,
      "prefixes suffixes": 89257,
      "different grammatical": 31167,
      "tagging labeling": 117396,
      "deception detection": 27923,
      "lack large": 58724,
      "detection studies": 30065,
      "collection method": 18478,
      "based social": 12055,
      "quickly identify": 95393,
      "identify highquality": 51506,
      "deceptive truthful": 27925,
      "online reviews": 81799,
      "reviews amazon": 102595,
      "product domains": 92036,
      "dataset explore": 26917,
      "effective general": 34681,
      "general features": 46658,
      "perform domains": 85983,
      "domains demonstrate": 33759,
      "complexity scores": 19937,
      "detection performance": 30021,
      "improved adding": 52593,
      "adding additional": 3370,
      "domains training": 33876,
      "training finally": 123635,
      "level evaluation": 63441,
      "evaluation gives": 39226,
      "insight different": 55521,
      "writing styles": 135101,
      "weakly supervised": 132194,
      "supervised approach": 115723,
      "approach train": 7977,
      "temporal relation": 120119,
      "relation classifiers": 97962,
      "classifiers acquire": 17598,
      "event pairs": 39532,
      "pairs simultaneously": 83639,
      "capabilities detecting": 15167,
      "detecting temporal": 29855,
      "temporal relations": 120122,
      "events benefit": 39567,
      "benefit applications": 12959,
      "applications existing": 6926,
      "existing temporal": 40312,
      "trained supervised": 123297,
      "supervised manner": 115796,
      "relation despite": 97966,
      "various contexts": 131066,
      "rich contexts": 102733,
      "contexts used": 22432,
      "train contextual": 122914,
      "relation classifier": 97961,
      "recognize new": 97057,
      "contexts identify": 22398,
      "identify new": 51524,
      "pairs learns": 83574,
      "learns contextual": 63201,
      "quality contain": 94615,
      "contain rich": 21756,
      "rich commonsense": 102728,
      "knowledge addition": 57745,
      "supervised trained": 115845,
      "classifier achieves": 17520,
      "curriculum learning": 25422,
      "translation examine": 124801,
      "effects particular": 35001,
      "pairs online": 83594,
      "nmt focus": 79868,
      "focus types": 44836,
      "sentences similar": 106495,
      "sentence types": 106113,
      "types training": 126377,
      "training progresses": 123798,
      "experiments internal": 40965,
      "effect training": 34615,
      "approach determining": 7488,
      "job description": 57240,
      "learned word": 62286,
      "seen tremendous": 104551,
      "tremendous success": 125736,
      "numerous natural": 81023,
      "tasks recent": 119439,
      "main contribution": 66409,
      "develop technique": 30241,
      "technique called": 119775,
      "applies machine": 7154,
      "architecture inspired": 8675,
      "al 2013": 4605,
      "new vector": 79238,
      "experiment evaluation": 40466,
      "evaluation semantic": 39382,
      "measures meaning": 67880,
      "meaning similarity": 67692,
      "sentences applications": 106211,
      "applications include": 6937,
      "summarization generation": 115513,
      "generation question": 47579,
      "answering qa": 6665,
      "short answer": 108200,
      "answer grading": 6535,
      "semantic search": 105267,
      "dialog conversational": 30555,
      "task focuses": 118223,
      "crosslingual pairs": 24989,
      "task obtained": 118472,
      "analysis highlights": 5639,
      "common errors": 18874,
      "errors providing": 38405,
      "providing insight": 94118,
      "limitations existing": 64173,
      "models support": 74134,
      "work semantic": 134792,
      "introduced new": 56591,
      "new shared": 79134,
      "set carefully": 107387,
      "carefully selected": 15528,
      "task data": 118044,
      "verbs arguments": 131553,
      "limited coverage": 64225,
      "verb classes": 131519,
      "automatically induced": 10788,
      "adapted specific": 3293,
      "specific corpora": 111423,
      "verbs multiple": 131557,
      "argument types": 8964,
      "produce high": 91898,
      "argument type": 8963,
      "predicting sarcasm": 89008,
      "sarcasm sentiment": 103643,
      "tasks common": 118986,
      "use recurrent": 128230,
      "networks lstm": 77651,
      "embedded representations": 35359,
      "cooccurrence patterns": 23231,
      "graphical model": 48929,
      "linearchain crf": 64393,
      "presents method": 89866,
      "representations latent": 99719,
      "latent output": 61594,
      "data model": 26134,
      "embedded representation": 35358,
      "longterm label": 65206,
      "output labels": 83084,
      "demonstrate accuracy": 28654,
      "improvements interpretable": 52861,
      "latent structure": 61613,
      "complex task": 19887,
      "based conll": 11594,
      "university edinburghs": 127350,
      "describes university": 29444,
      "edinburghs submissions": 34528,
      "wmt17 shared": 132842,
      "biomedical translation": 14211,
      "czech german": 25490,
      "latvian russian": 61676,
      "russian turkish": 103507,
      "chinese biomedical": 16741,
      "task submitted": 118751,
      "systems english": 116868,
      "german polish": 47907,
      "systems trained": 117202,
      "bpebased models": 14544,
      "models parallel": 73713,
      "parallel backtranslated": 84587,
      "backtranslated monolingual": 11324,
      "monolingual training": 74625,
      "layer normalization": 61736,
      "extensive ablative": 41851,
      "ensembling techniques": 37633,
      "dependency grammar": 29141,
      "induction neural": 54028,
      "data study": 26513,
      "big models": 13996,
      "data terms": 26551,
      "terms training": 120396,
      "dependency model": 29155,
      "extension neural": 41837,
      "neural dependency": 77886,
      "training corpora": 123400,
      "good model": 48486,
      "achieves result": 2847,
      "competitive current": 19651,
      "stateoftheart domain": 112646,
      "neural dialog": 77890,
      "domains domain": 33767,
      "domain case": 33476,
      "technique using": 119822,
      "models sutskever": 74142,
      "sutskever et": 116226,
      "domain classifier": 33480,
      "classifier model": 17558,
      "features current": 43434,
      "current utterance": 25387,
      "previous utterances": 90516,
      "relevant responses": 98575,
      "metrics compare": 69937,
      "seq2seq model": 106899,
      "exploiting linguistic": 41469,
      "resources neural": 101018,
      "using multitask": 129924,
      "linguistic annotations": 64424,
      "learning successful": 63071,
      "easy approach": 34474,
      "introduce additional": 56377,
      "additional knowledge": 3532,
      "knowledge endtoend": 57895,
      "training natural": 123725,
      "able leverage": 1662,
      "leverage common": 63578,
      "information improve": 54669,
      "performance individual": 86462,
      "task analyze": 117878,
      "analyze impact": 5977,
      "training training": 123934,
      "training schedule": 123828,
      "parameter sharing": 84722,
      "sharing tasks": 108142,
      "tasks defined": 119033,
      "architecture experiments": 8662,
      "task additional": 117845,
      "pos information": 88209,
      "experiments translation": 41185,
      "improved 15": 52591,
      "15 bleu": 341,
      "points lowresource": 87863,
      "learning scheme": 62983,
      "fundamental component": 46089,
      "component nlp": 19995,
      "recognition machine": 96905,
      "translation popular": 125123,
      "popular models": 88104,
      "directly applicable": 31861,
      "highly agglutinative": 50291,
      "propose syllablebased": 93098,
      "using convolutional": 129496,
      "meaningful representation": 67724,
      "words compared": 133868,
      "skipgram embeddings": 109983,
      "embeddings results": 35918,
      "results quite": 102106,
      "comparison neural": 19559,
      "compare language": 19252,
      "based attentionbased": 11535,
      "attentionbased sequencetosequence": 10100,
      "models evaluate": 73172,
      "large german": 61100,
      "german wmt": 47926,
      "set significantly": 107581,
      "models novel": 73650,
      "work terms": 134848,
      "terms speed": 120384,
      "speed quality": 111876,
      "translation phrases": 125120,
      "role natural": 103205,
      "al 2002": 4591,
      "al 2005": 4597,
      "word work": 133664,
      "model dynamically": 71035,
      "generates relevant": 47170,
      "relevant target": 98590,
      "model reads": 71857,
      "nmt decoder": 79851,
      "translation updates": 125410,
      "decoding state": 28134,
      "generates word": 47192,
      "word vocabulary": 133656,
      "baseline various": 12335,
      "various test": 131218,
      "memoryaugmented neural": 68399,
      "achieved notable": 2659,
      "notable success": 80377,
      "success recent": 115125,
      "recent times": 96551,
      "limitations handling": 64177,
      "handling infrequent": 49415,
      "infrequent words": 55176,
      "pairs paper": 83599,
      "memoryaugmented nmt": 68400,
      "nmt mnmt": 79893,
      "stores knowledge": 113395,
      "words usually": 134298,
      "use memory": 128140,
      "memory mechanism": 68348,
      "rules learned": 103414,
      "nmt propose": 79949,
      "tasks demonstrated": 119043,
      "architecture outperformed": 8718,
      "nmt baseline": 79838,
      "tasks respectively": 119470,
      "respectively additionally": 101122,
      "compared competitive": 19345,
      "competitive methods": 19655,
      "youtube videos": 135463,
      "reviews paper": 102621,
      "paper target": 84471,
      "phenomenon introduce": 87251,
      "introduce dataset": 56406,
      "product review": 92044,
      "model aspect": 70689,
      "aspect extraction": 9327,
      "joint aspect": 57263,
      "extraction sentiment": 42482,
      "provides stateoftheart": 94079,
      "performance aspect": 86151,
      "semeval absa": 105516,
      "joint task": 57325,
      "task dataset": 118046,
      "baseline tasks": 12321,
      "tasks observe": 119346,
      "models comparison": 72939,
      "experiments domain": 40922,
      "discussed extensively": 32296,
      "domain product": 33618,
      "finegrained opinion": 44370,
      "recent trends": 96554,
      "methods employ": 69468,
      "layers learn": 61786,
      "results domains": 101739,
      "domains recently": 33847,
      "recently variety": 96774,
      "variety model": 130994,
      "model designs": 70974,
      "designs methods": 29655,
      "methods blossomed": 69353,
      "blossomed context": 14378,
      "context natural": 22191,
      "nlp paper": 79655,
      "review significant": 102578,
      "learning related": 62953,
      "numerous nlp": 81026,
      "summarize compare": 115597,
      "compare contrast": 19231,
      "contrast various": 22718,
      "past present": 85648,
      "neural speaker": 78684,
      "speaker modeling": 111294,
      "multiparty conversation": 75480,
      "conversation task": 22978,
      "dataset models": 27036,
      "systems attracting": 116743,
      "attracting increasing": 10165,
      "increasing attention": 53658,
      "attention academia": 9788,
      "academia industry": 1852,
      "industry recently": 54063,
      "recently researchers": 96753,
      "modeling neural": 72493,
      "speaker classification": 111284,
      "task general": 118233,
      "massive data": 67326,
      "facilitate research": 42789,
      "models speakers": 74072,
      "speakers propose": 111324,
      "experiments speaker": 41149,
      "hybrid models": 51186,
      "outperform single": 82744,
      "single component": 109717,
      "neural statistical": 78688,
      "methods leveraging": 69594,
      "discuss different": 32246,
      "meta information": 68531,
      "quality focus": 94670,
      "information proposed": 54882,
      "methods extended": 69500,
      "available input": 11018,
      "text automatically": 120677,
      "automatically predicted": 10809,
      "predicted using": 88970,
      "text content": 120818,
      "content main": 21902,
      "main novelty": 66445,
      "network methods": 77324,
      "methods tackle": 69789,
      "observe translation": 81214,
      "quality improvements": 94684,
      "score text": 104135,
      "text categories": 120696,
      "character vocabulary": 16475,
      "vocabulary large": 131880,
      "languages chinese": 60440,
      "models huge": 73340,
      "languages explored": 60575,
      "model sentiment": 71978,
      "radicals chinese": 95444,
      "chinese characters": 16743,
      "model composed": 70871,
      "feature encoder": 43270,
      "document feature": 33008,
      "results achieved": 101493,
      "embeddingbased models": 35536,
      "models close": 72911,
      "close stateoftheart": 17835,
      "word embeddingbased": 133045,
      "smaller vocabulary": 110246,
      "fewer parameters": 43889,
      "respectively results": 101162,
      "embeddingbased approach": 35533,
      "learning chinese": 62430,
      "making sense": 66863,
      "embeddings present": 35870,
      "sense embeddings": 105676,
      "contrast existing": 22691,
      "techniques directly": 119872,
      "learn sense": 62142,
      "sense representations": 105696,
      "representations corpora": 99576,
      "sense inventories": 105687,
      "resources approach": 100945,
      "induce sense": 54004,
      "sense inventory": 105688,
      "embeddings clustering": 35597,
      "mechanism enables": 67974,
      "sense vectors": 105704,
      "gives rise": 48207,
      "experiments performance": 41056,
      "method comparable": 68707,
      "comparable stateoftheart": 19178,
      "unsupervised wsd": 127759,
      "wsd systems": 135172,
      "corpora automatic": 23415,
      "identification discourse": 51381,
      "relations challenging": 98119,
      "processing discourse": 91663,
      "discourse connectives": 32046,
      "informative cues": 55134,
      "explicit relations": 41342,
      "relations discourse": 98140,
      "discourse parsers": 32070,
      "relations signaled": 98256,
      "effectively paper": 34835,
      "method leverage": 68941,
      "leverage parallel": 63610,
      "corpora text": 23600,
      "simplification lexical": 109585,
      "resources automatically": 100949,
      "signal discourse": 108685,
      "simple wikipedia": 109546,
      "automatic discovery": 10520,
      "boundary detection": 14519,
      "analysis involves": 5664,
      "processing unit": 91847,
      "paper verify": 84493,
      "induction method": 54025,
      "works best": 134926,
      "best sentence": 13448,
      "task specifically": 118731,
      "syntactic rules": 116461,
      "classification rely": 17377,
      "rely parsers": 98726,
      "dependent language": 29263,
      "language treebank": 60193,
      "treebank guidelines": 125638,
      "require adaptation": 100109,
      "adaptation especially": 3222,
      "especially multilingual": 38482,
      "multilingual scenarios": 75356,
      "challenge context": 16011,
      "portuguese spanish": 88202,
      "model available": 70717,
      "available annotated": 10940,
      "create release": 24638,
      "release publicly": 98470,
      "available commercial": 10962,
      "service corpus": 107320,
      "intelligent virtual": 55859,
      "text deemed": 120861,
      "deemed unnecessary": 28191,
      "second round": 104451,
      "annotation determine": 6299,
      "resulting corpus": 101436,
      "corpus valuable": 24061,
      "valuable resource": 130769,
      "resource improving": 100848,
      "usage language": 127863,
      "removal language": 98870,
      "positive effect": 88318,
      "demonstrating need": 28980,
      "automated methods": 10454,
      "largescale domain": 61429,
      "teacherstudent learning": 119704,
      "learning high": 62632,
      "large transcribed": 61304,
      "transcribed data": 123984,
      "data supervised": 26518,
      "supervised training": 115846,
      "absence data": 1715,
      "data domain": 25864,
      "acoustic model": 3014,
      "performed high": 86923,
      "requires significant": 100320,
      "significant labeled": 108820,
      "data target": 26539,
      "target domain": 117605,
      "transcriptions instead": 124006,
      "unlabeled parallel": 127403,
      "consisting pairs": 21460,
      "samples source": 103591,
      "source domain": 110747,
      "welltrained model": 132442,
      "domain perform": 33610,
      "perform adaptation": 85942,
      "employ teacherstudent": 36288,
      "teacherstudent ts": 119706,
      "ts learning": 125907,
      "probabilities generated": 90842,
      "labels train": 58649,
      "scenarios adapting": 103839,
      "noisy speech": 80124,
      "children speech": 16728,
      "speech significant": 111804,
      "improvements accuracy": 52800,
      "accuracy obtained": 2222,
      "reductions word": 97473,
      "original source": 82547,
      "source model": 110791,
      "domain increasing": 33546,
      "model robustness": 71943,
      "particularly beneficial": 85471,
      "beneficial using": 12956,
      "agree disagree": 4383,
      "detection dual": 29927,
      "presents models": 89870,
      "models detecting": 73057,
      "discussions work": 32331,
      "using siamese": 130168,
      "architecture encode": 8652,
      "longer need": 65158,
      "need rely": 76849,
      "structure evaluate": 113854,
      "model existing": 71120,
      "existing online": 40246,
      "lexical word": 63842,
      "achieves state": 2873,
      "score model": 104093,
      "dataset performs": 27095,
      "performs competitively": 86993,
      "annotated datasets": 6178,
      "arabic multidialect": 8532,
      "arabic word": 8568,
      "variety nlp": 131009,
      "paper compare": 83764,
      "major arabic": 66534,
      "arabic dialects": 8508,
      "using thousand": 130290,
      "approaches involve": 8199,
      "ranking problem": 95682,
      "best segmentation": 13447,
      "segmentation sequence": 104634,
      "bilstm rnn": 14103,
      "segment words": 104568,
      "words able": 133800,
      "segmentation results": 104629,
      "limited training": 64291,
      "modern standard": 74415,
      "standard arabic": 112187,
      "arabic data": 8501,
      "improve overall": 52449,
      "overall results": 83254,
      "wmt 2017": 132773,
      "human automatic": 50759,
      "nmt strong": 79969,
      "strong smt": 113714,
      "baseline discuss": 12211,
      "endtoend trainable": 36980,
      "trainable neural": 123055,
      "belief tracking": 12692,
      "tracking taskoriented": 122763,
      "taskoriented dialog": 118883,
      "model taskoriented": 72146,
      "able track": 1695,
      "track dialog": 122725,
      "dialog state": 30585,
      "issue api": 56989,
      "api calls": 6777,
      "base kb": 11461,
      "query results": 94985,
      "responses successfully": 101297,
      "successfully complete": 115182,
      "complete taskoriented": 19772,
      "taskoriented dialogs": 118888,
      "responses jointly": 101279,
      "jointly learning": 57353,
      "result processing": 101393,
      "dialog history": 30565,
      "restaurant search": 101325,
      "search domain": 104309,
      "domain using": 33692,
      "second dialog": 104401,
      "state tracking": 112520,
      "tracking challenge": 122747,
      "challenge dstc2": 16026,
      "corpus experiment": 23787,
      "given dialog": 48014,
      "history model": 50561,
      "model demonstrates": 70967,
      "appropriate responses": 8431,
      "outperforming prior": 82823,
      "learning paraphrase": 62878,
      "answering question": 6679,
      "qa systems": 94521,
      "systems sensitive": 117134,
      "sensitive different": 105736,
      "ways natural": 132167,
      "information need": 54792,
      "need paper": 76843,
      "knowledge present": 58113,
      "framework learns": 45598,
      "method trained": 69192,
      "endtoend using": 36990,
      "using questionanswer": 130075,
      "questionanswer pairs": 95246,
      "pairs supervision": 83644,
      "supervision signal": 115916,
      "question paraphrases": 95195,
      "serve input": 107296,
      "neural scoring": 78659,
      "scoring model": 104232,
      "linguistic expressions": 64476,
      "correct answers": 24093,
      "approach qa": 7830,
      "results datasets": 101658,
      "datasets framework": 27489,
      "framework consistently": 45464,
      "consistently improves": 21418,
      "performance achieving": 86117,
      "despite use": 29738,
      "simple qa": 109495,
      "qa models": 94508,
      "cold fusion": 18348,
      "training seq2seq": 123840,
      "seq2seq models": 106902,
      "tasks involve": 119227,
      "sentences machine": 106388,
      "translation image": 124854,
      "improved leveraging": 52613,
      "leveraging unlabeled": 63711,
      "data form": 25955,
      "form language": 45099,
      "fusion method": 46241,
      "method leverages": 68942,
      "leverages pretrained": 63651,
      "pretrained language": 90045,
      "effectiveness speech": 34951,
      "recognition task": 97022,
      "task seq2seq": 118695,
      "able better": 1614,
      "better utilize": 13765,
      "better generalization": 13594,
      "transfer new": 124165,
      "new domain": 78888,
      "translation homographs": 124851,
      "homographs words": 50619,
      "systems difficult": 116853,
      "difficult select": 31660,
      "select correct": 104694,
      "account global": 2025,
      "sentential context": 106555,
      "problem alleviated": 90927,
      "existing nmt": 40238,
      "significant problems": 108849,
      "translating ambiguous": 124585,
      "literature model": 64761,
      "model context": 70908,
      "encoder experiments": 36526,
      "pairs demonstrate": 83512,
      "demonstrate models": 28801,
      "performance nmt": 86577,
      "systems terms": 117192,
      "accuracy translating": 2308,
      "analysis requires": 5807,
      "global model": 48251,
      "naturally generalizes": 76643,
      "generalizes unseen": 46842,
      "automatic training": 10700,
      "annotate data": 6126,
      "capture long": 15334,
      "long range": 65080,
      "context compare": 22031,
      "compare baseline": 19226,
      "baseline logistic": 12241,
      "model corresponding": 70933,
      "corresponding current": 24280,
      "train evaluate": 122933,
      "important fields": 52165,
      "network baseline": 77167,
      "harder task": 49489,
      "f1 compared": 42652,
      "story telling": 113419,
      "serves different": 107314,
      "different social": 31434,
      "learn social": 62151,
      "stories told": 113401,
      "work suggests": 134832,
      "suggests information": 115373,
      "present algorithms": 89362,
      "deep representation": 28414,
      "representation story": 99420,
      "experiments test": 41175,
      "direct speech": 31801,
      "tedious timeconsuming": 120056,
      "infeasible paper": 54091,
      "paper utilizes": 84490,
      "utilizes rich": 130558,
      "narrative representation": 76175,
      "representation modeling": 99332,
      "modeling dialogue": 72416,
      "generation engine": 47387,
      "expands translation": 40369,
      "translation tool": 125380,
      "representation original": 99363,
      "perform character": 85958,
      "character substitution": 16469,
      "implement evaluate": 51931,
      "potential application": 88535,
      "application dialogue": 6844,
      "generating dialogue": 47212,
      "pilot study": 87510,
      "study human": 114400,
      "characters using": 16629,
      "using direct": 129581,
      "personality types": 87135,
      "types time": 126373,
      "recognition disambiguation": 96863,
      "widely researched": 132552,
      "significant growth": 108761,
      "systems crucial": 116826,
      "crucial natural": 25151,
      "tasks summarization": 119537,
      "summarization understanding": 115588,
      "translation standard": 125288,
      "making good": 66843,
      "systems end": 116866,
      "architecture enables": 8650,
      "provides standardized": 94078,
      "possible combine": 88391,
      "synonymy graph": 116354,
      "structure input": 113882,
      "substantially reduce": 114911,
      "reduce quality": 97353,
      "quality extracted": 94664,
      "performs preprocessing": 87018,
      "missing edges": 70323,
      "clusters evaluate": 17966,
      "evaluate approaches": 38797,
      "datasets russian": 27691,
      "discuss impact": 32256,
      "methods finally": 69511,
      "finally perform": 44213,
      "extensive error": 41874,
      "approach discuss": 7499,
      "discuss prominent": 32279,
      "alternative methods": 5236,
      "theorem prover": 121785,
      "automated theorem": 10475,
      "specially designed": 111399,
      "preserve linguistic": 89924,
      "large extent": 61088,
      "forms easily": 45242,
      "easily obtainable": 34461,
      "syntactic trees": 116496,
      "entailment datasets": 37659,
      "achieves high": 2797,
      "high results": 50134,
      "dataset finegrained": 26936,
      "finegrained domain": 44345,
      "outofdomain data": 82649,
      "data work": 26629,
      "study task": 114532,
      "particularly challenging": 85474,
      "represents hybrid": 100072,
      "cover different": 24483,
      "different market": 31245,
      "market sectors": 67242,
      "different properties": 31364,
      "standard techniques": 112319,
      "techniques semisupervised": 119979,
      "learning domain": 62517,
      "adaptation limited": 3227,
      "limited effectiveness": 64232,
      "suggests need": 115378,
      "need improve": 76819,
      "techniques release": 119969,
      "release dataset": 98448,
      "relatedness semantic": 97926,
      "similarity meaning": 109255,
      "judgments semantic": 57451,
      "extensively studied": 41969,
      "english using": 37329,
      "datasets generated": 27494,
      "kind available": 57675,
      "available russian": 11098,
      "present language": 89525,
      "resources different": 100965,
      "computing semantic": 20528,
      "terms semantic": 120378,
      "organize shared": 82473,
      "task russian": 118661,
      "19 teams": 438,
      "best approaches": 13306,
      "distributional thesaurus": 32740,
      "including largescale": 53309,
      "largescale crowdsourcing": 61414,
      "crowdsourcing study": 25124,
      "study involving": 114417,
      "native speakers": 76233,
      "indicate high": 53837,
      "accuracy challenging": 2113,
      "role improving": 103177,
      "arabic nlp": 8538,
      "nlp application": 79565,
      "accuracy offtheshelf": 2224,
      "offtheshelf tools": 81631,
      "morphological segmentation": 74732,
      "subword units": 115041,
      "tagging methods": 117406,
      "achieve close": 2438,
      "surpass stateoftheart": 116096,
      "performance analysis": 86136,
      "target tokens": 117733,
      "optimal performance": 82168,
      "features captured": 43389,
      "algorithms natural": 4869,
      "tasks traditionally": 119567,
      "rely manually": 98720,
      "manually defined": 67055,
      "tasks models": 119305,
      "learn features": 62048,
      "features automatically": 43373,
      "explicit feature": 41321,
      "domains neural": 33823,
      "natural choice": 76241,
      "specifically limited": 111566,
      "data known": 26062,
      "comes cost": 18776,
      "define different": 28484,
      "investigate ability": 56711,
      "ability bidirectional": 1490,
      "capturing contextual": 15463,
      "features particular": 43648,
      "particular analyze": 85379,
      "perform comprehensive": 85971,
      "general biomedical": 46636,
      "domain datasets": 33500,
      "datasets experiments": 27474,
      "focus important": 44772,
      "important contextual": 52137,
      "analyze various": 6016,
      "various feature": 131097,
      "feature types": 43322,
      "types investigate": 126301,
      "used error": 128520,
      "analysis review": 5821,
      "review rating": 102565,
      "various text": 131220,
      "unstructured information": 127572,
      "explore using": 41610,
      "statistical dependence": 113090,
      "textual classification": 121676,
      "text word": 121411,
      "scale free": 103721,
      "free text": 45765,
      "comments written": 18830,
      "used dataset": 128471,
      "dataset compared": 26806,
      "compared textual": 19476,
      "textual representations": 121732,
      "frequency word": 45850,
      "representation term": 99435,
      "relationship word": 98313,
      "vectors provide": 131467,
      "provide greater": 93840,
      "greater accuracy": 49036,
      "explored using": 41637,
      "comparable accuracy": 19126,
      "testing text": 120605,
      "analysis feature": 5620,
      "feature reduction": 43301,
      "reduction techniques": 97462,
      "used textual": 128811,
      "hypothesis testing": 51279,
      "embeddings introduce": 35765,
      "effective means": 34701,
      "distributional similarity": 32736,
      "similarity different": 109223,
      "different vector": 31552,
      "models additionally": 72720,
      "provide quantitative": 93902,
      "measure linguistic": 67806,
      "linguistic similarity": 64555,
      "limited word": 64300,
      "embeddings extended": 35716,
      "representations learning": 99734,
      "used far": 128548,
      "estimate salience": 38634,
      "specifically learn": 111564,
      "input accurately": 55292,
      "accurately predict": 2387,
      "predict words": 88953,
      "words appear": 133818,
      "appear sentence": 6802,
      "comparably better": 19195,
      "approaches representing": 8324,
      "sentences benchmark": 106227,
      "datasets sentence": 27698,
      "fraction training": 45368,
      "prior methods": 90718,
      "concreteness imageability": 20745,
      "perceived humans": 85912,
      "measure performance": 67811,
      "parsing statistical": 85245,
      "learning problems": 62912,
      "problems area": 91297,
      "including sequence": 53372,
      "sequence segmentation": 107045,
      "prediction methods": 89078,
      "optimization process": 82207,
      "resulting performance": 101463,
      "performance accuracy": 86110,
      "accuracy demonstrate": 2134,
      "constituency parsing": 21528,
      "optimization respect": 82210,
      "trees neural": 125700,
      "words entities": 133940,
      "study addresses": 114306,
      "identifying meaning": 51603,
      "respect word": 101111,
      "embedding approaches": 35374,
      "output layers": 83088,
      "layers neural": 61793,
      "entity representation": 38096,
      "representation used": 99459,
      "kobayashi et": 58338,
      "copy mechanism": 23288,
      "mechanism proposed": 68030,
      "gu et": 49209,
      "addition construct": 3413,
      "dataset called": 26775,
      "modeling evaluating": 72425,
      "evaluating ability": 39033,
      "ability capture": 1491,
      "capture word": 15422,
      "novel dataset": 80530,
      "dataset proposed": 27116,
      "model outperformed": 71628,
      "outperformed baseline": 82776,
      "furthermore experiments": 46173,
      "output layer": 83086,
      "model predict": 71758,
      "entities input": 37797,
      "words following": 133962,
      "terminological expressions": 120253,
      "bases neural": 12504,
      "focuses translation": 44916,
      "semantically structured": 105380,
      "ontologies knowledge": 81830,
      "knowledge graphs": 57980,
      "ontology labels": 81845,
      "highly specific": 50350,
      "specific vocabulary": 111516,
      "information guide": 54649,
      "translation translate": 125394,
      "domain challenges": 33478,
      "quality domainspecific": 94634,
      "financial domain": 44255,
      "statistical neural": 113142,
      "adaptation translation": 3277,
      "furthermore perform": 46199,
      "experiments observed": 41046,
      "observed significant": 81232,
      "significant advantage": 108721,
      "resource medical": 100864,
      "subword models": 115028,
      "models wordbased": 74355,
      "selection language": 104792,
      "intelligent selection": 55856,
      "data effective": 25876,
      "structural problems": 113787,
      "problems method": 91342,
      "indomain outofdomain": 53974,
      "does allow": 33328,
      "corpora similar": 23587,
      "extreme case": 42593,
      "come distribution": 18762,
      "distribution sentences": 32678,
      "selected sentences": 104732,
      "able model": 1664,
      "model indomain": 71336,
      "indomain data": 53960,
      "data cover": 25801,
      "optimal number": 82164,
      "number sentences": 80963,
      "building systems": 14893,
      "generally efficient": 46856,
      "goal using": 48395,
      "using vocabulary": 130363,
      "following properties": 44979,
      "corpora differ": 23455,
      "does involve": 33360,
      "information gained": 54629,
      "text combining": 120799,
      "latent topic": 61618,
      "mortality prediction": 74821,
      "intensive care": 55887,
      "care units": 15506,
      "joint endtoend": 57271,
      "architectures combine": 8792,
      "simultaneously train": 109696,
      "topic interpretability": 122532,
      "experiments mimiciii": 41009,
      "mimiciii dataset": 70136,
      "outperform prior": 82732,
      "prior models": 90720,
      "based lda": 11797,
      "lda topics": 61832,
      "achieve limited": 2494,
      "limited success": 64284,
      "interpreting topics": 56292,
      "topics trained": 122661,
      "style topic": 114599,
      "topic neural": 122557,
      "generation propose": 47570,
      "simple flexible": 109435,
      "training decoding": 123573,
      "decoding methods": 28107,
      "encoderdecoder based": 36594,
      "variety applications": 130963,
      "systems successful": 117169,
      "need produce": 76847,
      "produce language": 91904,
      "specific style": 111492,
      "generate responses": 47002,
      "neural generation": 77918,
      "model decoding": 70954,
      "decoding method": 28105,
      "output quality": 83112,
      "quality conversational": 94619,
      "conversational tasks": 23039,
      "datadriven dialogue": 26659,
      "systems social": 117149,
      "systems tackle": 117180,
      "social conversations": 110332,
      "conversations argue": 23048,
      "argue need": 8924,
      "need data": 76791,
      "approach includes": 7630,
      "chit chat": 16880,
      "incorporates different": 53504,
      "index large": 53804,
      "corpora social": 23589,
      "including twitter": 53402,
      "twitter conversations": 126099,
      "conversations online": 23061,
      "dialogues friends": 30834,
      "blog posts": 14368,
      "data retrieval": 26361,
      "retrieval modules": 102412,
      "perform tasks": 86087,
      "sentiment style": 106798,
      "style analysis": 114567,
      "analysis topic": 5903,
      "personal assistants": 87114,
      "capturing longrange": 15478,
      "longrange contextual": 65178,
      "contextual dependencies": 22452,
      "successful applications": 115150,
      "applications broad": 6902,
      "range nlp": 95590,
      "fields crfs": 43994,
      "model local": 71469,
      "features important": 43552,
      "benefits terms": 13022,
      "limits ability": 64312,
      "capture longrange": 15338,
      "longrange dependencies": 65179,
      "inference work": 54248,
      "integrating external": 55793,
      "taking inspiration": 117549,
      "incorporate information": 53475,
      "steps experiments": 113318,
      "tasks substantial": 119532,
      "improvements strong": 52920,
      "lstm baselines": 65601,
      "purely neural": 94416,
      "systems submitted": 117165,
      "submitted wmt": 114780,
      "2017 shared": 639,
      "systems built": 116764,
      "using opennmt": 129990,
      "opennmt opensource": 82005,
      "models lstm": 73535,
      "resulting models": 101456,
      "models according": 72651,
      "according evaluation": 1976,
      "evaluation test": 39420,
      "generalpurpose neural": 46889,
      "model solve": 72065,
      "solve wide": 110627,
      "variety problems": 131012,
      "classification ranking": 17363,
      "ranking tasks": 95691,
      "collaborative filteringbased": 18360,
      "graphs learning": 48964,
      "level embeddings": 63438,
      "embeddings case": 35588,
      "embedding entities": 35401,
      "discrete features": 32165,
      "features comparing": 43414,
      "task empirical": 118135,
      "competitive existing": 19654,
      "generally applicable": 46847,
      "applicable new": 6827,
      "new cases": 78820,
      "methods human": 69535,
      "help detect": 49708,
      "expressions paper": 41760,
      "words phrase": 134115,
      "low entropy": 65361,
      "performed russian": 86931,
      "response selection": 101237,
      "multiparty conversations": 75481,
      "conversations challenging": 23050,
      "challenging complex": 16234,
      "multiple speakers": 75701,
      "different roles": 31393,
      "challenge propose": 16088,
      "speaker embeddings": 111288,
      "novel dialog": 80544,
      "task sequence": 118696,
      "sequence prediction": 107030,
      "prediction problem": 89107,
      "problem experimental": 91042,
      "perspective paper": 87169,
      "use text": 128326,
      "variation english": 130890,
      "translation corpus": 124715,
      "corpus purpose": 23953,
      "motivated features": 74849,
      "features representing": 43703,
      "representing texts": 100066,
      "combination partofspeech": 18575,
      "bigrams trigrams": 14009,
      "classification method": 17265,
      "used paper": 128676,
      "bayesian classifier": 12579,
      "carry extensive": 15553,
      "extensive feature": 41937,
      "feature analysis": 43250,
      "analysis main": 5689,
      "main difference": 66415,
      "methods translation": 69815,
      "translation review": 125224,
      "evaluation techniques": 39419,
      "social dialogue": 110335,
      "systems contrast": 116814,
      "goaloriented dialogue": 48411,
      "task success": 118756,
      "hard paper": 49474,
      "review current": 102540,
      "current evaluation": 25282,
      "metrics conclude": 69941,
      "ignore context": 51664,
      "lack grounding": 58714,
      "decoder neural": 28034,
      "sequencetosequence networks": 107186,
      "remarkable performance": 98834,
      "information timestep": 55045,
      "prediction attention": 89031,
      "targetside context": 117796,
      "solely based": 110530,
      "model practice": 71756,
      "recency bias": 96385,
      "words address": 133809,
      "network decoding": 77219,
      "attention previous": 9977,
      "prediction word": 89153,
      "residual learning": 100739,
      "information distant": 54498,
      "wider context": 132607,
      "context proposed": 22236,
      "outperforms neural": 82926,
      "selfattention network": 104893,
      "analysis attention": 5505,
      "attention learned": 9868,
      "templates generating": 120086,
      "restaurant reviews": 101324,
      "make language": 66693,
      "lost translation": 65314,
      "engines paper": 37060,
      "explore method": 41559,
      "negative reviews": 76946,
      "reviews restaurant": 102627,
      "domain goal": 33542,
      "stylistic variation": 114633,
      "available natural": 11051,
      "language generator": 59105,
      "generator learn": 47774,
      "review sentences": 102574,
      "evaluate learned": 38847,
      "templates traditional": 120087,
      "using subjective": 130233,
      "results learned": 101887,
      "score highly": 104081,
      "finally analyze": 44145,
      "analyze linguistic": 5984,
      "linguistic categories": 64434,
      "plan use": 87631,
      "use learned": 128122,
      "twitter customer": 126101,
      "conversations using": 23076,
      "using finegrained": 129678,
      "finegrained dialogue": 44343,
      "acts given": 3143,
      "increasing popularity": 53681,
      "service dialogue": 107322,
      "twitter analysis": 126093,
      "understand trends": 126781,
      "customer agent": 25437,
      "interactions work": 56009,
      "frequently observed": 45872,
      "model conversation": 70917,
      "conversation flow": 22950,
      "predicting dialogue": 88982,
      "characterize differences": 16550,
      "conversations investigate": 23057,
      "investigate effect": 56745,
      "different customer": 31072,
      "finally use": 44245,
      "datadriven approach": 26656,
      "approach predict": 7805,
      "predict important": 88894,
      "important conversation": 52139,
      "customer satisfaction": 25443,
      "type location": 126211,
      "acts conversation": 3140,
      "conversation significant": 22972,
      "significant effect": 108753,
      "based findings": 11719,
      "attention word": 10041,
      "word discovery": 132994,
      "discovery task": 32149,
      "examine extent": 39745,
      "extent neural": 41979,
      "networks applied": 77507,
      "applied task": 7124,
      "language scenario": 60067,
      "scenario small": 103831,
      "small corpora": 110144,
      "corpora limited": 23519,
      "limited annotations": 64206,
      "annotations available": 6410,
      "available investigate": 11020,
      "limited supervision": 64286,
      "frequent words": 45864,
      "possible retrieve": 88430,
      "bayesian nonparametric": 12585,
      "nonparametric model": 80273,
      "approach advantage": 7341,
      "alignments used": 5023,
      "used create": 128462,
      "create bilingual": 24604,
      "bilingual lexicon": 14043,
      "work directly": 134479,
      "directly speech": 31907,
      "learning event": 62557,
      "prediction paper": 89098,
      "predicting event": 88985,
      "event description": 39506,
      "preceding sentence": 88777,
      "sentence text": 106104,
      "text approach": 120654,
      "approach explores": 7556,
      "multilayer recurrent": 75183,
      "approach substantially": 7939,
      "substantially outperforms": 114906,
      "datasets derived": 27420,
      "respectively bleu": 101127,
      "second evaluation": 104407,
      "exploits rich": 41493,
      "methods verbal": 69843,
      "received relatively": 96372,
      "model accurately": 70531,
      "paper makes": 84039,
      "british national": 14660,
      "order create": 82305,
      "evaluated study": 39008,
      "words approach": 133822,
      "models classification": 72906,
      "modelling verbal": 72630,
      "vietnamese word": 131701,
      "based single": 12049,
      "single classification": 109712,
      "ripple rules": 102862,
      "rules stored": 103442,
      "stored exception": 113384,
      "exception structure": 39924,
      "structure new": 113918,
      "new rules": 79114,
      "rules added": 103378,
      "added correct": 3357,
      "segmentation errors": 104584,
      "errors given": 38371,
      "given existing": 48032,
      "existing rules": 40275,
      "rules experimental": 103398,
      "machine translationbased": 66308,
      "diacritic restoration": 30495,
      "study machine": 114431,
      "problem including": 91081,
      "work applies": 134373,
      "method problem": 69072,
      "thorough comparison": 121880,
      "problem large": 91100,
      "phrasebased approach": 87380,
      "neuralbased approach": 78734,
      "slightly lower": 110025,
      "lower accuracy": 65420,
      "method terms": 69183,
      "inference speed": 54221,
      "room future": 103259,
      "future improvement": 46278,
      "incorporating pretrained": 53551,
      "collecting training": 18452,
      "suggestion mining": 115357,
      "datasets work": 27802,
      "propose evaluate": 92661,
      "supervision obtained": 115903,
      "obtained large": 81382,
      "large silver": 61272,
      "dataset constructed": 26821,
      "architecture learn": 8683,
      "use silver": 128275,
      "dataset approach": 26740,
      "approach directly": 7493,
      "trains classifier": 123962,
      "approach learns": 7680,
      "best classification": 13317,
      "learning domainspecific": 62523,
      "domainspecific word": 33922,
      "texts word": 121651,
      "embedding natural": 35461,
      "technique automatically": 119770,
      "automatically maps": 10801,
      "real numbers": 96074,
      "space widely": 111082,
      "recognition syntactic": 97016,
      "parsing sentiment": 85230,
      "classic word": 17083,
      "methods word2vec": 69854,
      "work given": 134545,
      "input texts": 55454,
      "specialized domains": 111389,
      "methods fail": 69506,
      "fail produce": 42970,
      "produce highquality": 91902,
      "addition domain": 3424,
      "texts method": 121553,
      "diverse types": 32859,
      "domain vocabulary": 33694,
      "relations specifically": 98261,
      "specifically propose": 111579,
      "annotations develop": 6422,
      "novel word": 80778,
      "word annotation": 132923,
      "types text": 126367,
      "annotations word": 6482,
      "description corpus": 29456,
      "corpus common": 23698,
      "noisecontrastive estimation": 80079,
      "models scale": 73974,
      "method allows": 68632,
      "allows fast": 5155,
      "learning large": 62680,
      "models missing": 73570,
      "paper showed": 84428,
      "successful approach": 115151,
      "approach neural": 7736,
      "learning rate": 62936,
      "rate schedule": 95805,
      "models outperforms": 73690,
      "stateoftheart singlemodel": 112955,
      "singlemodel methods": 109835,
      "methods popular": 69674,
      "popular benchmark": 88079,
      "benchmark challenging": 12738,
      "neural dialogue": 77892,
      "natural data": 76248,
      "networks fail": 77601,
      "phenomena natural": 87234,
      "dialogue proceeds": 30726,
      "incrementally wordbyword": 53750,
      "approaches dialogue": 8126,
      "order avoid": 82283,
      "trained clean": 123090,
      "clean data": 17700,
      "dialogue data": 30654,
      "corpus called": 23689,
      "incremental dialogue": 53727,
      "dialogues dataset": 30825,
      "explore performance": 41572,
      "retrieval model": 102409,
      "results semantic": 102145,
      "semantic accuracy": 104971,
      "learn process": 62118,
      "impractical training": 52312,
      "linguistically informed": 64588,
      "systems minimal": 117002,
      "minimal data": 70156,
      "investigate endtoend": 56755,
      "automatically inducing": 10789,
      "learning rl": 62973,
      "decision problem": 27939,
      "shown previously": 108510,
      "knowledge grammar": 57953,
      "combinatorially large": 18608,
      "number dialogue": 80866,
      "model process": 71793,
      "dialogues addition": 30820,
      "addition process": 3463,
      "corpus created": 23736,
      "model stateoftheart": 72086,
      "shows poor": 108607,
      "mitigating impact": 70385,
      "impact speech": 51891,
      "chatbot using": 16666,
      "model apply": 70672,
      "model mitigate": 71517,
      "mitigate impact": 70369,
      "errors open": 38397,
      "endtoend dialog": 36887,
      "cast task": 15679,
      "adaptation problem": 3253,
      "asr transcriptions": 9451,
      "text different": 120873,
      "model includes": 71326,
      "domain data": 33498,
      "data make": 26112,
      "states similar": 113056,
      "decoder predict": 28044,
      "predict dialog": 88880,
      "text method": 121115,
      "method shows": 69131,
      "text pair": 121159,
      "pair having": 83433,
      "having meaning": 49562,
      "errors experimental": 38368,
      "cornell movie": 23389,
      "movie dialog": 74894,
      "dialog dataset": 30559,
      "demonstrate domain": 28701,
      "domain adaption": 33461,
      "text answers": 120650,
      "association measures": 9623,
      "measures presented": 67889,
      "yelp reviews": 135323,
      "introduced machine": 56580,
      "based method": 11837,
      "method characterize": 68691,
      "main approach": 66396,
      "svm model": 116243,
      "sentiment tendency": 106804,
      "tendency review": 120169,
      "word frequency": 133291,
      "svm models": 116244,
      "word special": 133581,
      "tend express": 120148,
      "distinction different": 32554,
      "corpus extension": 23795,
      "resources improve": 100988,
      "performance language": 86480,
      "train machine": 122951,
      "use synthetic": 128311,
      "methods extending": 69501,
      "lowresource corpus": 65489,
      "corpus apply": 23665,
      "model showed": 72005,
      "showed improvement": 108379,
      "performance corpus": 86260,
      "extension using": 41842,
      "method specifically": 69160,
      "make better": 66627,
      "including use": 53403,
      "multisource model": 75785,
      "model efficient": 71049,
      "methods lowresource": 69604,
      "furthermore methods": 46190,
      "used better": 128412,
      "better machine": 13624,
      "learning explain": 62567,
      "15 years": 356,
      "prior studies": 90735,
      "studies focus": 114230,
      "propose dual": 92634,
      "dual encoder": 34234,
      "wordlevel encoder": 133734,
      "encoder learns": 36541,
      "learns representation": 63231,
      "representation context": 99196,
      "encoder learn": 36540,
      "hidden representation": 49905,
      "representation target": 99432,
      "model produce": 71796,
      "produce reasonable": 91918,
      "prosodic features": 93643,
      "features large": 43586,
      "age acquisition": 4279,
      "children acquire": 16723,
      "acquire language": 3033,
      "language widely": 60340,
      "studied phenomenon": 114168,
      "word learning": 133334,
      "learning remains": 62957,
      "active research": 3120,
      "raw input": 95855,
      "largescale multimodal": 61478,
      "multimodal corpus": 75421,
      "investigate utility": 56826,
      "demonstrate utility": 28903,
      "context multimodal": 22186,
      "multiple datasets": 75534,
      "data large": 26075,
      "nlp algorithms": 79562,
      "datasets order": 27605,
      "order ensure": 82318,
      "consistent performance": 21395,
      "performance heterogeneous": 86426,
      "setups multiple": 107865,
      "multiple comparisons": 75519,
      "pose significant": 88256,
      "challenges traditional": 16217,
      "lead erroneous": 61851,
      "conclusions paper": 20729,
      "analysis framework": 5629,
      "statistically sound": 113191,
      "tasks discuss": 119068,
      "discuss theoretical": 32290,
      "framework current": 45478,
      "demonstrate empirical": 28735,
      "parsing multilingual": 85164,
      "crossdomain sentiment": 24900,
      "architecture joint": 8679,
      "joint segmentation": 57319,
      "segmentation parsing": 104609,
      "real estate": 96063,
      "text using": 121396,
      "segmentation plain": 104617,
      "entities ii": 37793,
      "establish relations": 38592,
      "effective neural": 34721,
      "neural joint": 77932,
      "stateoftheart works": 113032,
      "works focus": 134947,
      "structured description": 113999,
      "tasks identifying": 119163,
      "identifying important": 51601,
      "important entities": 52155,
      "format work": 45211,
      "able tackle": 1694,
      "tackle tasks": 117311,
      "tasks simultaneously": 119504,
      "fashion ii": 43118,
      "extensive comparative": 41864,
      "study pipeline": 114469,
      "pipeline methods": 87545,
      "new proposed": 79088,
      "tree propose": 125607,
      "attention methods": 9929,
      "encourage model": 36758,
      "experimentally demonstrate": 40737,
      "neural architectures": 77838,
      "architectures proposed": 8839,
      "points edge": 87854,
      "labeling classification": 58487,
      "sparse attention": 111221,
      "attention neural": 9953,
      "networks sentencelevel": 77757,
      "sentencelevel classification": 106146,
      "classification sequential": 17400,
      "labeling fundamental": 58502,
      "tasks language": 119239,
      "understanding tasks": 126983,
      "usually modeled": 130438,
      "intent classification": 55893,
      "classification slot": 17408,
      "topic classification": 122500,
      "classification namedentity": 17297,
      "potential benefits": 88539,
      "propose jointly": 92733,
      "simultaneously long": 109685,
      "model predicts": 71762,
      "predicts sentencelevel": 89221,
      "output hidden": 83080,
      "representations lstm": 99744,
      "novel mechanism": 80625,
      "mechanism sparse": 68038,
      "words differently": 133914,
      "semantic relevance": 105225,
      "classification proposed": 17357,
      "datasets evaluation": 27457,
      "evaluation chinese": 39144,
      "dialogue technology": 30803,
      "evaluation scheme": 39376,
      "collect annotate": 18374,
      "test evaluation": 120453,
      "evaluation includes": 39240,
      "user intent": 128999,
      "task divided": 118110,
      "problems using": 91380,
      "applications developed": 6915,
      "developed industry": 30279,
      "data provided": 26301,
      "paper publish": 84391,
      "results present": 102049,
      "current performance": 25321,
      "tasks chinese": 118975,
      "technology analyze": 120032,
      "dialogue evaluation": 30670,
      "task universal": 118820,
      "semantically informative": 105369,
      "argue task": 8930,
      "task independent": 118289,
      "independent nature": 53776,
      "contributes better": 22762,
      "text present": 121192,
      "present initial": 89517,
      "provide semantically": 93915,
      "information suitable": 55018,
      "application semantic": 6880,
      "formal lexical": 45167,
      "annotate small": 6133,
      "corpus semantic": 23985,
      "new baseline": 78809,
      "baseline result": 12296,
      "speaker role": 111297,
      "modeling language": 72455,
      "understanding dialogue": 126827,
      "dialogue policy": 30722,
      "policy learning": 87958,
      "lu dialogue": 65721,
      "essential components": 38549,
      "contextual model": 22481,
      "model consider": 70885,
      "different speaker": 31444,
      "speaker roles": 111298,
      "based various": 12152,
      "multiturn dialogues": 75918,
      "behavioral patterns": 12673,
      "contextual encoding": 22463,
      "improves language": 52991,
      "graphs semantic": 48971,
      "vectorspace models": 131499,
      "embeddings neural": 35828,
      "unclear paper": 126636,
      "specific tree": 111510,
      "graph structure": 48869,
      "efficient training": 35123,
      "inference algorithms": 54113,
      "algorithms based": 4832,
      "space demonstrate": 110990,
      "representation training": 99445,
      "graphs evaluating": 48951,
      "unsupervised semantic": 127710,
      "induction semantic": 54033,
      "similarity natural": 109273,
      "messages text": 68517,
      "text messaging": 121113,
      "used form": 128558,
      "previous findings": 90411,
      "shown linguistic": 108492,
      "linguistic factors": 64477,
      "messages existing": 68508,
      "existing research": 40269,
      "research examined": 100491,
      "gender affect": 46579,
      "sample sizes": 103566,
      "aims address": 4497,
      "issues using": 57082,
      "dataset text": 27237,
      "collected large": 18424,
      "varied set": 130943,
      "participants using": 85329,
      "messaging application": 68523,
      "male female": 66880,
      "datasets statistical": 27733,
      "translation explore": 124817,
      "robust respect": 103067,
      "respect specific": 101103,
      "allow better": 5075,
      "better parameter": 13647,
      "tuning paper": 125938,
      "objective functions": 81086,
      "translations learning": 125475,
      "selected based": 104722,
      "length particular": 63373,
      "sentences achieve": 106198,
      "improvements bleu": 52816,
      "visual reasoning": 131806,
      "provides widely": 94096,
      "widely accessible": 132535,
      "robotic agents": 103010,
      "language inputs": 59209,
      "reasoning language": 96261,
      "language vision": 60333,
      "open problem": 81921,
      "receiving increasing": 96382,
      "attention existing": 9833,
      "existing data": 40099,
      "sets focus": 107674,
      "reasoning comparisons": 96236,
      "simple task": 109526,
      "language visual": 60334,
      "images paired": 51809,
      "paired descriptive": 83473,
      "true given": 125849,
      "corpus current": 23738,
      "current work": 25391,
      "collecting real": 18451,
      "vision data": 131760,
      "attentive convolution": 10106,
      "rnns attention": 102960,
      "hypothesize attention": 51288,
      "cnns mainly": 18042,
      "attentive pooling": 10124,
      "pooling applied": 88039,
      "word taking": 133600,
      "fixedsize context": 44620,
      "propose attentive": 92562,
      "convolution network": 23140,
      "nonlocal context": 80234,
      "context attention": 22013,
      "used rnns": 128735,
      "external contexts": 41997,
      "experiments sentence": 41124,
      "sentence modeling": 105953,
      "claim verification": 17015,
      "context particular": 22216,
      "strong competitor": 113659,
      "attentive rnns": 10126,
      "building chatbots": 14828,
      "forum data": 45302,
      "model selection": 71960,
      "using question": 130074,
      "qa data": 94501,
      "data web": 26626,
      "web forums": 132235,
      "dialog training": 30614,
      "data extract": 25937,
      "extract pairs": 42101,
      "pairs question": 83622,
      "question answer": 95015,
      "answer sentences": 6579,
      "longer texts": 65165,
      "texts questions": 121589,
      "train seq2seq": 123015,
      "models efficient": 73121,
      "way improve": 132090,
      "parameter optimization": 84717,
      "optimization using": 82218,
      "selection strategy": 104832,
      "strategy based": 113505,
      "based qa": 11962,
      "use extrinsic": 128047,
      "extrinsic evaluation": 42615,
      "extrinsic task": 42626,
      "task answer": 117881,
      "answer correctly": 6522,
      "similar questions": 109130,
      "questions asked": 95277,
      "questions conversational": 95290,
      "processing historical": 91677,
      "portuguese corpora": 88195,
      "report describes": 98987,
      "framework used": 45731,
      "processing large": 91691,
      "large portuguese": 61212,
      "corpora corpora": 23444,
      "corpora contain": 23442,
      "historical portuguese": 50532,
      "collection containing": 18464,
      "containing texts": 21809,
      "16th early": 395,
      "early 20th": 34384,
      "report presents": 99027,
      "preprocessing methods": 89326,
      "annotation corpora": 6289,
      "published research": 94368,
      "papers using": 84514,
      "using corpora": 129505,
      "framework computerassisted": 45463,
      "computerassisted translation": 20505,
      "translation underresourced": 125398,
      "project develop": 92214,
      "translation computerassisted": 124702,
      "translation basic": 124669,
      "translation constraint": 124707,
      "sentence sequence": 106068,
      "words associated": 133828,
      "structure necessary": 113913,
      "semantics despite": 105406,
      "despite number": 29710,
      "number nlp": 80925,
      "nlp studies": 79693,
      "studies dedicated": 114204,
      "related task": 97901,
      "verb argument": 131516,
      "structured distributional": 114000,
      "models implicitly": 73358,
      "implicitly assuming": 52024,
      "structured representation": 114036,
      "representation events": 99230,
      "experimental evidence": 40541,
      "evidence suggests": 39671,
      "suggests human": 115372,
      "human processing": 50936,
      "type event": 126200,
      "event representation": 39536,
      "representation predict": 99380,
      "input paper": 55387,
      "structured model": 114017,
      "model adapt": 70623,
      "different hypotheses": 31174,
      "event knowledge": 39524,
      "relative performance": 98361,
      "task argument": 117893,
      "successfully shown": 115193,
      "shown using": 108538,
      "using additional": 129321,
      "help improve": 49737,
      "improve results": 52525,
      "results natural": 101974,
      "sentence multiple": 105954,
      "languages corpora": 60470,
      "primarily used": 90644,
      "useful developing": 128876,
      "developing multilingual": 30350,
      "multilingual sequence": 75364,
      "multilingual experiments": 75255,
      "approaches present": 8285,
      "present work": 89767,
      "provide training": 93943,
      "training development": 123581,
      "test splits": 120538,
      "hope help": 50628,
      "speed pace": 111872,
      "multilingual nlp": 75314,
      "nlp researchers": 79684,
      "researchers obtain": 100703,
      "obtain results": 81315,
      "public research": 94271,
      "present manually": 89544,
      "lexicon encoding": 63880,
      "designed capture": 29585,
      "verb context": 131520,
      "event described": 39505,
      "resource developed": 100836,
      "provides evidence": 94025,
      "aspect iii": 9334,
      "verbs different": 131555,
      "patterns particular": 85769,
      "infinitival clauses": 54273,
      "data internet": 26050,
      "changed way": 16380,
      "way information": 132095,
      "microblogs twitter": 70053,
      "creating new": 24704,
      "new opportunities": 79067,
      "opportunities study": 82144,
      "study public": 114496,
      "data quickly": 26313,
      "business research": 14980,
      "research various": 100662,
      "various fields": 131099,
      "marketing political": 67244,
      "questions like": 95327,
      "like people": 64080,
      "like new": 64077,
      "questions requires": 95362,
      "people express": 85874,
      "express social": 41708,
      "given rise": 48107,
      "growth field": 49190,
      "especially popular": 38491,
      "variety topics": 131026,
      "topics discussed": 122624,
      "public access": 94233,
      "present overview": 89634,
      "dependencyparsed corpus": 29261,
      "english including": 37169,
      "million documents": 70096,
      "billion tokens": 14080,
      "entity tagger": 38107,
      "various applications": 131037,
      "applications ranging": 7000,
      "training syntaxbased": 123896,
      "embeddings open": 35840,
      "answering built": 6607,
      "sentences linguistic": 106384,
      "corpus demonstrate": 23747,
      "utility corpus": 130476,
      "similarity task": 109316,
      "distributional model": 32711,
      "trained corpus": 123101,
      "results models": 101953,
      "smaller corpora": 110228,
      "corpora like": 23518,
      "like wikipedia": 64109,
      "outperforms state": 82991,
      "art models": 9061,
      "based contextual": 11602,
      "contextual word": 22516,
      "generating suggestions": 47267,
      "related distinct": 97854,
      "task making": 118374,
      "making accurate": 66819,
      "accurate predictions": 2354,
      "predictions task": 89192,
      "offline training": 81615,
      "effective use": 34771,
      "pretraining natural": 90295,
      "inference neural": 54174,
      "open questions": 81926,
      "questions performance": 95340,
      "performance pretrained": 86614,
      "empirically using": 36247,
      "specifically compare": 111532,
      "pretrained glove": 90039,
      "glove word2vec": 48306,
      "information pretrained": 54864,
      "pretrained embeddings": 90027,
      "outperform random": 82734,
      "corpus experiments": 23792,
      "experiments controlled": 40856,
      "embeddings useful": 36006,
      "rest model": 101314,
      "yields gains": 135410,
      "present framework": 89496,
      "framework machine": 45605,
      "setting goal": 107754,
      "given reference": 48102,
      "information reference": 54906,
      "representations fed": 99655,
      "experiment benchmark": 40452,
      "datasets wmt": 27799,
      "wmt metrics": 132798,
      "metrics shared": 70007,
      "obtain best": 81260,
      "perform series": 86065,
      "analyze understand": 6013,
      "different components": 31049,
      "components network": 20037,
      "network evaluate": 77246,
      "semantic embeddings": 105040,
      "convolutional recurrent": 23212,
      "framework flexible": 45547,
      "efficient learning": 35086,
      "metric correlates": 69875,
      "correlates human": 24217,
      "par state": 84523,
      "pretrained subword": 90192,
      "subword embeddings": 115018,
      "subword unit": 115040,
      "entity typing": 38125,
      "bet ter": 13494,
      "document corpus": 32984,
      "20 paper": 518,
      "documents automatic": 33187,
      "automatic document": 10521,
      "language composed": 58902,
      "documents provided": 33279,
      "czech news": 25496,
      "news agency": 79284,
      "comparison document": 19539,
      "czech data": 25488,
      "data particularly": 26223,
      "approaches document": 8132,
      "document usually": 33106,
      "label information": 58401,
      "document classes": 32960,
      "annotated morphological": 6209,
      "morphological layer": 74714,
      "layer paper": 61740,
      "shows results": 108622,
      "offer possibility": 81569,
      "approaches learning": 8212,
      "fundamental issue": 46099,
      "semantics natural": 105438,
      "efficiently learn": 35150,
      "entailment using": 37683,
      "framework modelling": 45614,
      "latent vector": 61639,
      "vectors investigate": 131445,
      "best previous": 13415,
      "words unsupervised": 134288,
      "unsupervised semisupervised": 127711,
      "challenges sentiment": 16211,
      "media decade": 68104,
      "specific topic": 111505,
      "called sentiment": 15064,
      "emotion analysis": 36095,
      "develop systems": 30240,
      "systems automatically": 116748,
      "automatically understand": 10837,
      "relevant content": 98531,
      "previous efforts": 90407,
      "january 2016": 57180,
      "presidential election": 89948,
      "election social": 35237,
      "source news": 110800,
      "news information": 79348,
      "million tweets": 70110,
      "tweets natural": 126039,
      "define novel": 28488,
      "novel textual": 80752,
      "task requires": 118639,
      "requires inference": 100280,
      "sentences present": 106450,
      "dataset task": 27233,
      "lexical inferences": 63769,
      "everyday events": 39627,
      "presents challenging": 89832,
      "challenging setting": 16311,
      "strong neural": 113693,
      "neural baselines": 77854,
      "task differs": 118097,
      "differs standard": 31603,
      "standard textual": 112325,
      "neural nets": 78231,
      "nets combining": 77116,
      "combining textual": 18754,
      "textual visual": 121753,
      "visual features": 131794,
      "features recent": 43687,
      "optical character": 82156,
      "character recognition": 16458,
      "recognition ocr": 96950,
      "preservation document": 89916,
      "document contexts": 32981,
      "automatically separate": 10828,
      "scanned images": 103784,
      "developed novel": 30292,
      "text features": 120951,
      "architecture achieves": 8605,
      "accuracy 93": 2084,
      "stateoftheart task": 112999,
      "task deep": 118049,
      "monolingual word": 74628,
      "surge social": 116087,
      "media use": 68178,
      "brings huge": 14647,
      "huge demand": 50720,
      "texts languages": 121538,
      "differences languages": 30962,
      "embeddings separately": 35932,
      "spaces shared": 111104,
      "shared embedding": 107976,
      "space finally": 111005,
      "finally train": 44240,
      "paradigm effective": 84535,
      "stateoftheart baseline": 112591,
      "terms classification": 120291,
      "attention attention": 9796,
      "attention distributions": 9822,
      "generated translations": 47144,
      "network translation": 77463,
      "soft alignments": 110484,
      "alignments input": 5014,
      "output tokens": 83138,
      "tokens work": 122354,
      "use attention": 127900,
      "output translations": 83143,
      "present strategies": 89720,
      "strategies using": 113495,
      "bad translations": 11356,
      "translations large": 125474,
      "selecting best": 104745,
      "best translation": 13465,
      "hybrid setup": 51195,
      "different translation": 31515,
      "systems manual": 116996,
      "evaluation indicated": 39243,
      "confidence score": 20996,
      "score human": 104082,
      "translation finegrained": 124824,
      "finegrained prediction": 44374,
      "prediction syntactic": 89128,
      "discovering latent": 32126,
      "learning predict": 62899,
      "novel language": 80611,
      "language given": 59108,
      "corpus partofspeech": 23927,
      "direct objects": 31795,
      "typological properties": 126489,
      "problem usually": 91283,
      "surface features": 116059,
      "features correlate": 43430,
      "structure latent": 113894,
      "latent trees": 61624,
      "trees experiment": 125691,
      "given small": 48127,
      "real languages": 96068,
      "data robust": 26364,
      "include noise": 53232,
      "task outperforms": 118492,
      "baseline large": 12238,
      "multiturn dialogue": 75915,
      "dialogue dataset": 30655,
      "dataset develop": 26869,
      "develop highquality": 30204,
      "aspects language": 9390,
      "dataset reflect": 27148,
      "cover various": 24491,
      "manually label": 67065,
      "developed dataset": 30264,
      "dataset communication": 26802,
      "emotion information": 36105,
      "information evaluate": 54543,
      "evaluate existing": 38834,
      "dataset hope": 26966,
      "benefit research": 12988,
      "systems word": 117240,
      "translation parallel": 125097,
      "data stateoftheart": 26502,
      "learning crosslingual": 62471,
      "dictionaries parallel": 30863,
      "corpora recent": 23568,
      "need parallel": 76844,
      "characterlevel information": 16566,
      "information methods": 54762,
      "showed encouraging": 108375,
      "supervised counterparts": 115745,
      "sharing common": 108128,
      "work build": 134403,
      "aligning monolingual": 4937,
      "spaces unsupervised": 111107,
      "unsupervised way": 127747,
      "character information": 16435,
      "supervised methods": 115803,
      "methods crosslingual": 69414,
      "crosslingual tasks": 25020,
      "pairs experiments": 83542,
      "demonstrate method": 28779,
      "distant language": 32451,
      "pairs like": 83575,
      "finally experiments": 44184,
      "limited parallel": 64260,
      "data potential": 26252,
      "method fully": 68848,
      "fully unsupervised": 45978,
      "unsupervised machine": 127658,
      "translation code": 124694,
      "code embeddings": 18099,
      "dictionaries publicly": 30867,
      "local models": 64929,
      "structure pas": 113931,
      "prediction models": 89082,
      "models local": 73523,
      "features global": 43527,
      "global models": 48252,
      "models designed": 73049,
      "globally optimal": 48289,
      "tend employ": 120147,
      "simple local": 109460,
      "designing local": 29641,
      "model demonstrated": 70966,
      "study showing": 114517,
      "showing performance": 108421,
      "model considerably": 70887,
      "considerably improved": 21271,
      "improved recent": 52632,
      "feature embedding": 43267,
      "methods feature": 69509,
      "feature combination": 43256,
      "models f1": 73226,
      "common benchmark": 18860,
      "building vietnamese": 14899,
      "resource language": 100853,
      "annotation guidelines": 6322,
      "experiments pos": 41060,
      "vietnamese language": 131678,
      "script events": 104253,
      "events narrative": 39586,
      "narrative texts": 76180,
      "script knowledge": 104255,
      "knowledge plays": 58103,
      "plays central": 87720,
      "central role": 15906,
      "recent datasets": 96444,
      "provide rich": 93911,
      "rich general": 102745,
      "general representation": 46705,
      "sets introduce": 107678,
      "introduce task": 56549,
      "mapping event": 67129,
      "event mentions": 39527,
      "event types": 39555,
      "types present": 126338,
      "model task": 72143,
      "task exploits": 118185,
      "representations information": 99693,
      "information temporal": 55034,
      "ordering results": 82439,
      "paying attention": 85813,
      "expressions neural": 41759,
      "nmt struggles": 79970,
      "overcome paper": 83295,
      "experiments investigating": 40967,
      "nmt attention": 79836,
      "improvement strategies": 52769,
      "strategies explored": 113463,
      "bilingual pairs": 14050,
      "pairs automatically": 83492,
      "added parallel": 3365,
      "nmt sentences": 79959,
      "automated evaluation": 10439,
      "point increase": 87806,
      "approach second": 7869,
      "minimal improvements": 70164,
      "provide opensource": 93888,
      "vectors lexical": 131451,
      "postprocessing method": 88507,
      "known isa": 58307,
      "injecting external": 55264,
      "linguistic constraints": 64450,
      "space proposed": 111047,
      "distance measure": 32423,
      "effectiveness robustness": 34944,
      "robustness proposed": 103127,
      "analysis spanish": 5864,
      "spanish tweets": 111177,
      "using svm": 130253,
      "svm cnn": 116236,
      "cnn article": 17988,
      "svm convolutional": 116237,
      "tweets sentiment": 126056,
      "analysis classifier": 5528,
      "classifier best": 17525,
      "performance general": 86410,
      "cnn use": 18027,
      "embeddings particularly": 35856,
      "classifiers performance": 17624,
      "task exploring": 118188,
      "exploring neural": 41665,
      "architectures multilingual": 8828,
      "feedback analysis": 43822,
      "task customer": 118041,
      "cnn bidirectional": 17991,
      "lstm architectures": 65593,
      "facebooks fasttext": 42746,
      "model best": 70768,
      "performing model": 86958,
      "respectively task": 101172,
      "task outperformed": 118490,
      "outperformed models": 82785,
      "using micro": 129882,
      "french task": 45827,
      "label embedding": 58393,
      "embedding text": 35515,
      "works treat": 134985,
      "onehot vectors": 81676,
      "cause loss": 15825,
      "information makes": 54747,
      "models jointly": 73433,
      "classification semantic": 17388,
      "semantic vectors": 105353,
      "original tasks": 82551,
      "vector matching": 131328,
      "matching tasks": 67436,
      "supervised semisupervised": 115831,
      "semisupervised models": 105617,
      "models multitask": 73594,
      "embedding utilizing": 35525,
      "tasks making": 119288,
      "effectively improve": 34819,
      "adaptation neural": 3238,
      "models building": 72868,
      "conversation agent": 22936,
      "challenging owing": 16287,
      "owing lack": 83389,
      "training paper": 123770,
      "problem proposing": 91187,
      "proposing multitask": 93624,
      "approach training": 7979,
      "models leverages": 73499,
      "data speakers": 26488,
      "data pertaining": 26235,
      "leads significant": 61953,
      "speaking styles": 111340,
      "model offers": 71607,
      "offers benefits": 81582,
      "simple easy": 109401,
      "easy implement": 34479,
      "relying large": 98761,
      "quantities data": 94896,
      "data representing": 26342,
      "specific individual": 111449,
      "semantic understanding": 105342,
      "language traditional": 60176,
      "verb representations": 131526,
      "framenet propbank": 45396,
      "patterns represent": 85776,
      "corresponds single": 24320,
      "single semantic": 109790,
      "description length": 29464,
      "high effectiveness": 50067,
      "text coherence": 120794,
      "coherence analysis": 18291,
      "coherence model": 18303,
      "architecture capture": 8635,
      "coherence text": 18310,
      "problem investigated": 91093,
      "perspective learning": 87164,
      "particular model": 85432,
      "captures interactions": 15449,
      "representations easily": 99611,
      "easily trained": 34465,
      "endtoend fashion": 36900,
      "model evaluated": 71110,
      "standard sentence": 112301,
      "showing significant": 108431,
      "stateoftheart wide": 113026,
      "wide margin": 132494,
      "way determine": 132071,
      "unrelated sentences": 127503,
      "generating sentence": 47260,
      "embeddings dense": 35639,
      "dense vectors": 29021,
      "vectors context": 131420,
      "context meaning": 22173,
      "embeddings produced": 35884,
      "multiple methods": 75615,
      "try evaluate": 125895,
      "algorithms trained": 4884,
      "plain english": 87623,
      "giving sense": 48217,
      "linking tweets": 64676,
      "monolingual crosslingual": 74563,
      "news using": 79394,
      "transformed word": 124284,
      "embeddings social": 35943,
      "media platforms": 68142,
      "spread information": 112048,
      "diverse sources": 32849,
      "information discuss": 54497,
      "new insights": 78958,
      "gap word": 46483,
      "content tweets": 21964,
      "propose transformation": 93120,
      "transformation framework": 124263,
      "articles languages": 9196,
      "languages leveraging": 60686,
      "leveraging word": 63713,
      "using framework": 129686,
      "embeddings extracted": 35717,
      "tweets news": 126041,
      "comparison new": 19560,
      "new findings": 78919,
      "findings reported": 44309,
      "parsing nonlocal": 85177,
      "avoids need": 11250,
      "long sequences": 65091,
      "resulting parser": 101462,
      "parser outperforms": 84977,
      "original version": 82562,
      "best accuracy": 13295,
      "accuracy stanford": 2285,
      "stanford dependencies": 112366,
      "conversion penn": 23087,
      "treebank greedy": 125637,
      "greedy transitionbased": 49075,
      "transitionbased algorithms": 124500,
      "simple model": 109471,
      "model multilingual": 71539,
      "classification does": 17193,
      "require parallel": 100183,
      "based traditional": 12123,
      "machine classifier": 65744,
      "embeddings character": 35590,
      "ngrams model": 79466,
      "overall ranking": 83253,
      "ranking 1st": 95657,
      "12 teams": 258,
      "analysis languages": 5669,
      "french japanese": 45817,
      "japanese spanish": 57200,
      "impact coreference": 51862,
      "filling paper": 44049,
      "demonstrate importance": 28756,
      "importance coreference": 52051,
      "task illustrate": 118273,
      "resolution systems": 100787,
      "systems provide": 117087,
      "provide experimental": 93819,
      "results improve": 101841,
      "performance slot": 86721,
      "endtoend setting": 36963,
      "source corpus": 110738,
      "researchers working": 100718,
      "learning grammar": 62621,
      "learning machines": 62701,
      "identify patterns": 51533,
      "investigate effectiveness": 56748,
      "effectiveness inducing": 34897,
      "distinguish strings": 32574,
      "regular language": 97727,
      "language utilize": 60316,
      "algorithm extracting": 4738,
      "results regarding": 102119,
      "encoder sequence": 36575,
      "model opendomain": 71610,
      "opendomain dialogue": 81965,
      "dialogue modeling": 30710,
      "successful application": 115149,
      "application sequence": 6882,
      "problem domains": 91019,
      "domains recent": 33846,
      "work investigated": 134594,
      "use neural": 128167,
      "architectures modeling": 8826,
      "opendomain conversational": 81960,
      "conversational dialogue": 23004,
      "model dialogue": 70989,
      "onetomany mapping": 81727,
      "conversation paper": 22965,
      "propose tackle": 93104,
      "previous conversational": 90401,
      "conversational context": 22996,
      "context terms": 22296,
      "latent representations": 61600,
      "representations dialogue": 99594,
      "inject latent": 55257,
      "context representations": 22249,
      "representations sequence": 99882,
      "dialog acts": 30550,
      "acts using": 3148,
      "second encoder": 104406,
      "enhance quality": 37485,
      "quality coherence": 94611,
      "main task": 66467,
      "task research": 118650,
      "variables capture": 130853,
      "capture discourse": 15293,
      "does result": 33399,
      "responses compared": 101259,
      "compared conventional": 19350,
      "conventional sequence": 22900,
      "sequence models": 107018,
      "personalized word": 87149,
      "representations carrying": 99536,
      "semantics learned": 105430,
      "representations shown": 99890,
      "application tasks": 6888,
      "huge corpora": 50718,
      "carry semantic": 15557,
      "information words": 55101,
      "words known": 134021,
      "user language": 129009,
      "language patterns": 59805,
      "patterns different": 85738,
      "wording habits": 133722,
      "imply kind": 52042,
      "differently different": 31595,
      "different users": 31546,
      "users example": 129113,
      "example word": 39800,
      "universal word": 127337,
      "vectors trained": 131480,
      "people paper": 85882,
      "framework train": 45717,
      "train different": 122927,
      "vectors different": 131426,
      "users based": 129096,
      "based successful": 12083,
      "continuous skipgram": 22634,
      "network data": 77215,
      "data posted": 26250,
      "individual users": 53939,
      "vectors use": 131482,
      "quality personalized": 94750,
      "vectors obtained": 131455,
      "user prediction": 129018,
      "vectors shown": 131475,
      "shown carry": 108451,
      "japaneseenglish parallel": 57211,
      "corpus covering": 23732,
      "domain conversational": 33489,
      "consists 32": 21466,
      "32 million": 931,
      "million examples": 70099,
      "making largest": 66851,
      "dataset kind": 26991,
      "kind corpus": 57676,
      "incorporates number": 53509,
      "number novel": 80927,
      "ensure high": 37638,
      "quality using": 94823,
      "baseline machine": 12245,
      "systems machine": 116988,
      "spoken dialects": 111974,
      "swiss german": 116274,
      "german goal": 47897,
      "goal work": 48400,
      "work design": 134469,
      "mt lowresource": 74970,
      "widely spoken": 132554,
      "number parallel": 80932,
      "words identified": 133986,
      "promising data": 92268,
      "compared strategies": 19466,
      "input order": 55378,
      "best solution": 13454,
      "36 bleu": 966,
      "testing data": 120590,
      "normalization techniques": 80349,
      "german dialects": 47886,
      "approach design": 7482,
      "based reading": 11970,
      "reading text": 96033,
      "attacks recent": 9716,
      "recent significant": 96514,
      "significant advances": 108720,
      "advances deep": 3980,
      "learning dl": 62514,
      "address paper": 3725,
      "proposes new": 93606,
      "captioning architecture": 15256,
      "framework developed": 45493,
      "structures rules": 114103,
      "generation network": 47502,
      "vectors words": 131493,
      "including long": 53313,
      "novelty approach": 80789,
      "generate sentence": 47010,
      "sentence extract": 105866,
      "obtained unsupervised": 81426,
      "generating textual": 47279,
      "base triples": 11488,
      "web data": 132229,
      "data directly": 25853,
      "understand underlying": 126785,
      "help make": 49749,
      "explore problem": 41578,
      "summaries semantic": 115456,
      "data nontrivial": 26181,
      "context address": 22002,
      "problem explore": 91046,
      "encodes information": 36669,
      "information set": 54976,
      "set triples": 107617,
      "fixed dimensionality": 44595,
      "textual summary": 121751,
      "summary conditioning": 115634,
      "conditioning output": 20813,
      "models corpora": 72990,
      "dbpedia wikidata": 27849,
      "results improving": 101847,
      "forced decoding": 45015,
      "decoding compared": 28079,
      "compared traditional": 19477,
      "method combine": 68701,
      "advantages traditional": 4088,
      "smt nmt": 110291,
      "existing phrasebased": 40253,
      "model compute": 70877,
      "decoding cost": 28083,
      "cost nmt": 24368,
      "using cost": 129512,
      "rerank nbest": 100361,
      "space standard": 111066,
      "algorithm search": 4794,
      "quality different": 94631,
      "pairs text": 83646,
      "text annotation": 120646,
      "complex natural": 19840,
      "phenomena paper": 87242,
      "introduces new": 56620,
      "software tool": 110524,
      "annotating text": 6273,
      "text text": 121367,
      "provides functionality": 94035,
      "tools including": 122450,
      "annotations annotation": 6407,
      "sequential context": 107215,
      "context text": 22297,
      "text users": 121394,
      "similar structures": 109151,
      "annotated documents": 6188,
      "extraction tasks": 42508,
      "tasks large": 119245,
      "database biomedical": 26638,
      "support wide": 116018,
      "tasks domain": 119072,
      "software available": 110515,
      "available httpsgithubcom": 11010,
      "corpora neural": 23537,
      "requires largescale": 100290,
      "largescale parallel": 61491,
      "data scarce": 26370,
      "domains language": 33798,
      "address new": 3720,
      "translation scenario": 125228,
      "method translation": 69201,
      "pairs derived": 83513,
      "pairs monolingual": 83586,
      "training method": 123705,
      "different generation": 31162,
      "generation strategies": 47637,
      "strategies designed": 113456,
      "aligned unaligned": 4927,
      "function designed": 46011,
      "designed model": 29610,
      "model partially": 71713,
      "parts experiments": 85574,
      "relatively good": 98390,
      "result translation": 101417,
      "embeddings deep": 35631,
      "compositional code": 20108,
      "models require": 73932,
      "require massive": 100180,
      "number parameters": 80933,
      "embeddings resulting": 35917,
      "resulting large": 101448,
      "large storage": 61280,
      "neural nlp": 78615,
      "models mobile": 73573,
      "embeddings significant": 35936,
      "purpose propose": 94438,
      "propose construct": 92596,
      "construct embeddings": 21631,
      "embeddings basis": 35569,
      "basis vectors": 12549,
      "word composition": 132962,
      "maximize compression": 67509,
      "compression rate": 20319,
      "approach instead": 7641,
      "composed multiple": 20069,
      "limited fixed": 64235,
      "propose directly": 92626,
      "learn discrete": 62023,
      "tasks performance": 119379,
      "performance loss": 86502,
      "loss tasks": 65299,
      "method improve": 68883,
      "compared approaches": 19324,
      "approaches characterlevel": 8094,
      "characterlevel segmentation": 16584,
      "languageindependent does": 60361,
      "multitask multilingual": 75881,
      "multilingual modelling": 75284,
      "time learning": 122045,
      "language advantage": 58817,
      "languages speak": 60887,
      "learning multiple": 62816,
      "multiple tasks": 75723,
      "defined study": 28502,
      "computational processing": 20421,
      "language types": 60196,
      "share similarities": 107962,
      "traditional approach": 122792,
      "approach nlp": 7742,
      "consider single": 21230,
      "single task": 109801,
      "task single": 118720,
      "single language": 109747,
      "language time": 60169,
      "data multiple": 26144,
      "tasks languages": 119243,
      "languages simultaneously": 60877,
      "approach explore": 7555,
      "improving reliability": 53156,
      "nlp especially": 79611,
      "especially lowresource": 38470,
      "advantage relevant": 4058,
      "relevant data": 98536,
      "languages benefit": 60426,
      "advances nlp": 4008,
      "nlp currently": 79596,
      "highresource languages": 50425,
      "minority languages": 70276,
      "answering specific": 6700,
      "specific research": 111485,
      "questions posed": 95341,
      "use nlp": 128175,
      "inflected language": 54283,
      "research focusing": 100521,
      "focusing english": 44924,
      "original english": 82516,
      "english word2vec": 37344,
      "language created": 58927,
      "basic evaluation": 12520,
      "popular word": 88145,
      "data corpus": 25797,
      "corpus tested": 24038,
      "able create": 1626,
      "create meaningful": 24626,
      "representation research": 99395,
      "quality resulting": 94778,
      "model dialog": 70988,
      "topics using": 122663,
      "generative methods": 47734,
      "methods topic": 69807,
      "topic detection": 122506,
      "simple question": 109496,
      "handled separate": 49408,
      "separate modules": 106858,
      "scheme developed": 103917,
      "effective generation": 34684,
      "model general": 71225,
      "sentence encoding": 105854,
      "draw inspiration": 34115,
      "seen formal": 104531,
      "formal linguistics": 45169,
      "linguistics shown": 64624,
      "shown empirical": 108467,
      "composition functions": 20096,
      "models yield": 74367,
      "yield significant": 135351,
      "improvements existing": 52852,
      "compositional approaches": 20107,
      "composition function": 20095,
      "uses global": 129230,
      "operation based": 82052,
      "larger number": 61372,
      "model stanford": 72081,
      "multigenre nli": 75115,
      "stanford sentiment": 112381,
      "sentiment treebank": 106810,
      "best known": 13357,
      "emotions associated": 36127,
      "understand peoples": 126766,
      "paper summarize": 84453,
      "summarize data": 115602,
      "india explore": 53813,
      "automatically extracts": 10772,
      "allocation lda": 5070,
      "lda based": 61827,
      "based topic": 12121,
      "model identifies": 71294,
      "topics different": 122621,
      "categories additionally": 15729,
      "tweets related": 126054,
      "related event": 97860,
      "furthermore use": 46219,
      "use evaluation": 128034,
      "evaluation measure": 39263,
      "select best": 104692,
      "lda models": 61829,
      "models obtained": 73657,
      "discussion topics": 32327,
      "tagging vietnamese": 117463,
      "vietnamese paper": 131687,
      "vietnamese partofspeech": 131689,
      "text pipeline": 121179,
      "consider output": 21220,
      "ii joint": 51709,
      "strategy predict": 113532,
      "make comparison": 66636,
      "comparison stateoftheart": 19577,
      "stateoftheart sota": 112960,
      "featurebased neural": 43338,
      "models benchmark": 72826,
      "nguyen et": 79472,
      "results pipeline": 102037,
      "better scores": 13715,
      "text joint": 121067,
      "highest accuracy": 50222,
      "using featurebased": 129670,
      "featurebased model": 43336,
      "model classical": 70827,
      "prediction losses": 89074,
      "learning recent": 62943,
      "work training": 134857,
      "models sequencelevel": 74009,
      "linear models": 64357,
      "perform surprisingly": 86083,
      "search optimization": 104333,
      "report new": 99014,
      "iwslt14 germanenglish": 57171,
      "germanenglish translation": 47941,
      "abstractive summarization": 1808,
      "larger wmt14": 61392,
      "neural process": 78637,
      "networks understanding": 77807,
      "language requires": 60050,
      "explicitly stated": 41393,
      "networks understand": 77806,
      "procedural text": 91386,
      "text neural": 121144,
      "dynamics model": 34353,
      "entity tracking": 38111,
      "model reason": 71859,
      "provide accurate": 93750,
      "information understanding": 55065,
      "representations existing": 99642,
      "nlpcc 2017": 79794,
      "baselines paper": 12439,
      "characterenhanced word": 16510,
      "unsupervised morphological": 127674,
      "small datasets": 110148,
      "datasets improving": 27514,
      "improving word": 53182,
      "independent unsupervised": 53781,
      "method building": 68681,
      "building word": 14901,
      "text model": 121126,
      "model handles": 71273,
      "problem data": 90991,
      "yields improved": 135415,
      "artificially generated": 9273,
      "small sized": 110208,
      "sets test": 107718,
      "seven languages": 107871,
      "english evaluated": 37131,
      "set standard": 107588,
      "method improved": 68887,
      "human agent": 50738,
      "approach detecting": 7485,
      "agent responses": 4312,
      "f1score 20": 42710,
      "textual features": 121706,
      "addition features": 3431,
      "features common": 43409,
      "sentence set": 106069,
      "develop crowdsourcing": 30187,
      "100000 questions": 182,
      "detailed qualitative": 29783,
      "analysis demonstrates": 5563,
      "pairs cover": 83505,
      "cover vast": 24492,
      "vast majority": 131276,
      "datasets including": 27517,
      "propbank nombank": 92405,
      "arguments relations": 8995,
      "data annotation": 25603,
      "code publicly": 18140,
      "work best": 134398,
      "phenomena automatically": 87219,
      "automatically extracting": 10771,
      "materials science": 67446,
      "approaches achieved": 8042,
      "achieved recent": 2676,
      "available supervised": 11118,
      "inorganic materials": 55288,
      "language narratives": 59703,
      "scientific journal": 104009,
      "journal articles": 57407,
      "synthesis information": 116595,
      "extracted text": 42186,
      "text order": 121155,
      "order enable": 82314,
      "extracting structured": 42241,
      "structured representations": 114037,
      "extracted scientific": 42178,
      "entities evaluate": 37776,
      "approaches extracting": 8160,
      "entities results": 37863,
      "data directions": 25852,
      "area research": 8881,
      "research fast": 100504,
      "fast unsupervised": 43155,
      "initialization parameters": 55238,
      "parameters using": 84796,
      "using variational": 130349,
      "twostep method": 126173,
      "runtime comparable": 103478,
      "comparable translation": 19189,
      "distantly related": 32488,
      "robust flexible": 103035,
      "representations important": 99682,
      "areas language": 8892,
      "proposed early": 93261,
      "way representing": 132127,
      "sequences events": 107121,
      "understanding recently": 126937,
      "recently attracted": 96660,
      "renewed attention": 98893,
      "effective representations": 34738,
      "representations modeling": 99756,
      "challenging requires": 16304,
      "semantics propose": 105452,
      "method creating": 68750,
      "representations method": 99753,
      "method captures": 68689,
      "subtle semantic": 115000,
      "semantic interactions": 105084,
      "representations effective": 99613,
      "effective multiple": 34714,
      "tasks continuous": 119013,
      "continuous representations": 22628,
      "generation method": 47471,
      "method analysis": 68633,
      "subtle differences": 114998,
      "differences surface": 30977,
      "surface realizations": 116077,
      "wordbyword translation": 133703,
      "focus evaluating": 44761,
      "evaluating quality": 39088,
      "systems introduce": 116958,
      "translations based": 125454,
      "blacklisted words": 14259,
      "likely translation": 64145,
      "translation error": 124794,
      "perform evaluation": 85993,
      "evaluation stateoftheart": 39405,
      "chineseenglish neural": 16857,
      "evaluation confirms": 39157,
      "common error": 18873,
      "error type": 38344,
      "effective identifying": 34689,
      "relative importance": 98356,
      "relationships different": 98317,
      "remains uncertain": 98822,
      "quantitative studies": 94884,
      "studies date": 114203,
      "means identify": 67763,
      "identify causes": 51484,
      "knowledge make": 58060,
      "make difficult": 66659,
      "difficult use": 31671,
      "use available": 127914,
      "available knowledge": 11021,
      "user preference": 129019,
      "user feedback": 128984,
      "online forums": 81774,
      "forums users": 45321,
      "new tools": 79223,
      "analyze large": 5982,
      "large volumes": 61327,
      "topic models": 122550,
      "user opinions": 129017,
      "opinions easily": 82120,
      "decisionmaking process": 27952,
      "primary care": 90646,
      "reviews covering": 102603,
      "selection neural": 104808,
      "models dialog": 73063,
      "dialog response": 30583,
      "selection important": 104787,
      "step natural": 113281,
      "natural response": 76617,
      "conversational models": 23019,
      "mainly focuses": 66488,
      "learning response": 62967,
      "selection retrievalbased": 104823,
      "propose contextual": 92599,
      "reward function": 102672,
      "text online": 121154,
      "produce distributed": 91883,
      "representations dialog": 99593,
      "dialog context": 30554,
      "context responses": 22252,
      "contextual bandit": 22446,
      "bandit learning": 11412,
      "propose customized": 92610,
      "sampling method": 103602,
      "ubuntu dialogue": 126520,
      "dialogue corpus": 30653,
      "demonstrate significant": 28863,
      "significant performance": 108832,
      "gains proposed": 46405,
      "report encouraging": 98992,
      "selection performance": 104812,
      "training samples": 123826,
      "retrievalbased question": 102447,
      "study transfer": 114535,
      "aiming propose": 4493,
      "framework effectively": 45512,
      "effectively efficiently": 34805,
      "learned resourcerich": 62254,
      "resource poor": 100869,
      "domain specifically": 33652,
      "existing transfer": 40323,
      "target domains": 117611,
      "domains propose": 33834,
      "learn shared": 62148,
      "shared representations": 108007,
      "efficient effective": 35070,
      "method sentence": 69123,
      "model extensive": 71153,
      "identification natural": 51406,
      "compared competing": 19344,
      "competing models": 19619,
      "models transfer": 74238,
      "method help": 68870,
      "captured model": 15433,
      "bring significant": 14634,
      "sparse interpretable": 111227,
      "interpretable neural": 56247,
      "neural embeddings": 77902,
      "embeddings prediction": 35869,
      "success neural": 115105,
      "attributed ability": 10189,
      "expressive representations": 41774,
      "capture underlying": 15412,
      "data far": 25944,
      "novel variant": 80769,
      "generates highly": 47163,
      "highly efficient": 50320,
      "efficient interpretable": 35083,
      "embeddings beginning": 35570,
      "representations stateoftheart": 99907,
      "scale human": 103723,
      "evaluation report": 39361,
      "report resulting": 99033,
      "word2vec embeddings": 133671,
      "outperform existing": 82706,
      "existing popular": 40254,
      "embeddings diverse": 35665,
      "use dialogue": 127997,
      "humanmachine interaction": 51062,
      "increasingly prevalent": 53710,
      "prevalent paradigm": 90360,
      "growing number": 49178,
      "use conversation": 127965,
      "learned large": 62225,
      "biased offensive": 13859,
      "highlight potential": 50271,
      "potential ethical": 88552,
      "ethical issues": 38748,
      "implicit biases": 52004,
      "biases datadriven": 13868,
      "adversarial examples": 4117,
      "potential sources": 88586,
      "spur research": 112059,
      "units text": 127267,
      "work examine": 134498,
      "methods embeddings": 69466,
      "used deep": 128477,
      "effect performance": 34606,
      "tasks recurrent": 119445,
      "convolutional architectures": 23149,
      "better random": 13686,
      "random initialization": 95499,
      "especially considering": 38443,
      "speed learning": 111869,
      "learning hand": 62625,
      "significant difference": 108751,
      "various methods": 131131,
      "network use": 77465,
      "use space": 128288,
      "task support": 118768,
      "relations fact": 98174,
      "learn perform": 62110,
      "perform reasonably": 86054,
      "spam detection": 111114,
      "detection based": 29887,
      "learning widely": 63162,
      "learning analysis": 62330,
      "achieved high": 2637,
      "high degree": 50057,
      "detection important": 29969,
      "important application": 52098,
      "model related": 71885,
      "given attention": 47987,
      "order effectively": 82312,
      "effectively identify": 34818,
      "study characteristics": 114334,
      "explore novel": 41570,
      "characteristics textual": 16540,
      "textual semantics": 121734,
      "analysis detection": 5573,
      "mechanism based": 67952,
      "opinions paper": 82125,
      "model embedding": 71053,
      "pooling layer": 88041,
      "achieve accurate": 2414,
      "detection results": 30037,
      "dependencybased word": 29258,
      "brain activity": 14552,
      "evaluate different": 38819,
      "usefulness predicting": 128961,
      "activation patterns": 3100,
      "patterns associated": 85731,
      "models consider": 72962,
      "data popular": 26247,
      "popular neural": 88110,
      "neural distributional": 77898,
      "model reflects": 71881,
      "syntactic context": 116386,
      "cognitive plausibility": 18274,
      "understand improve": 126751,
      "methods interpreting": 69559,
      "imaging data": 51823,
      "models exhibit": 73189,
      "superior performance": 115685,
      "tasks consider": 119003,
      "model syntactically": 72130,
      "syntactically informed": 116514,
      "model gives": 71251,
      "performance predicting": 86610,
      "embeddings glove": 35734,
      "method gives": 68864,
      "words vectors": 134306,
      "error patterns": 38313,
      "support idea": 115974,
      "words suggest": 134249,
      "models account": 72652,
      "lead better": 61843,
      "associated words": 9615,
      "words learning": 134031,
      "translation history": 124850,
      "models generally": 73281,
      "sentences isolation": 106361,
      "documentlevel information": 33144,
      "information work": 55106,
      "augment nmt": 10262,
      "models lightweight": 73502,
      "representations translation": 99948,
      "models capability": 72873,
      "time experiments": 122028,
      "experiments multiple": 41027,
      "multiple domains": 75544,
      "domains different": 33762,
      "different topics": 31502,
      "negligible impact": 76976,
      "computational cost": 20366,
      "slot tagging": 110067,
      "understanding models": 126890,
      "models introduce": 73425,
      "information incorporate": 54678,
      "incorporate dialog": 53462,
      "networks encode": 77583,
      "form user": 45143,
      "user utterances": 129052,
      "real user": 96085,
      "user data": 128973,
      "microsoft cortana": 70059,
      "commercial personal": 18837,
      "personal assistant": 87113,
      "result showed": 101402,
      "showed significant": 108390,
      "tagging models": 117409,
      "using contextual": 129482,
      "information improved": 54672,
      "generated large": 47092,
      "data users": 26606,
      "providing meaningful": 94123,
      "business intelligence": 14975,
      "customers businesses": 25451,
      "able identify": 1651,
      "enables quantify": 36400,
      "using microsoft": 129883,
      "analysis develop": 5575,
      "learning classification": 62435,
      "contents sentiments": 21992,
      "tools analyzed": 122429,
      "factors social": 42900,
      "social influence": 110344,
      "traditional supervised": 122875,
      "supervised machine": 115787,
      "90 accuracy": 1365,
      "accuracy rate": 2255,
      "cognitive model": 18270,
      "model recent": 71860,
      "work attempted": 134385,
      "structure semantic": 113947,
      "task number": 118466,
      "search processes": 104340,
      "process language": 91517,
      "memory present": 68366,
      "model incrementally": 71333,
      "limited computational": 64220,
      "patterns human": 85748,
      "simple random": 109499,
      "perform thorough": 86093,
      "analyses showing": 5467,
      "showing combination": 108399,
      "combination structural": 18586,
      "features correlated": 43431,
      "patterns neural": 85763,
      "generation dynamic": 47379,
      "vocabularies study": 131863,
      "study response": 114503,
      "chatbots existing": 16671,
      "methods assume": 69321,
      "assume words": 9650,
      "noise causes": 80044,
      "high cost": 50053,
      "cost decoding": 24350,
      "decoding propose": 28129,
      "dynamic vocabulary": 34332,
      "decoding training": 28144,
      "training vocabulary": 123948,
      "jointly learned": 57352,
      "maximizing lower": 67523,
      "monte carlo": 74652,
      "carlo sampling": 15533,
      "method inference": 68907,
      "inference model": 54167,
      "irrelevant words": 56953,
      "words generation": 133972,
      "enjoys efficient": 37547,
      "efficient decoding": 35067,
      "results automatic": 101529,
      "terms response": 120376,
      "response quality": 101232,
      "time compared": 121995,
      "efficient baseline": 35062,
      "adversarial learning": 4130,
      "learning slot": 63031,
      "filling spoken": 44055,
      "representations slot": 99895,
      "task spoken": 118735,
      "slu models": 110119,
      "models domainspecific": 73107,
      "task domains": 118117,
      "adversarial training": 4169,
      "representations shared": 99888,
      "shared multiple": 107992,
      "domains model": 33820,
      "representations combined": 99547,
      "trained individual": 123159,
      "slu data": 110110,
      "data reduce": 26327,
      "reduce training": 97363,
      "domain experiments": 33524,
      "training helps": 123648,
      "helps learning": 49823,
      "models leading": 73476,
      "filling f1": 44040,
      "scores applying": 104152,
      "model helps": 71278,
      "achieving higher": 2953,
      "filling performance": 44051,
      "jointly optimized": 57373,
      "domainspecific models": 33909,
      "models lexical": 73501,
      "present compositional": 89410,
      "distributional analysis": 32693,
      "frobenius algebraic": 45896,
      "finite dimensional": 44533,
      "dimensional vector": 31754,
      "analysis relies": 5803,
      "type assignments": 126187,
      "reading using": 96038,
      "using extended": 129656,
      "lambek calculus": 58794,
      "calculus present": 15040,
      "allows single": 5188,
      "models augment": 72785,
      "augment data": 10255,
      "work highlights": 134553,
      "effectiveness combining": 34875,
      "patient data": 85699,
      "data relatively": 26331,
      "relatively large": 98395,
      "easily accessible": 34428,
      "deep natural": 28342,
      "text largescale": 121080,
      "analysis challenging": 5525,
      "initial work": 55229,
      "proposed automatic": 93232,
      "sentences written": 106549,
      "reports relevant": 99088,
      "relevant labels": 98564,
      "labels event": 58597,
      "main challenges": 66401,
      "high frequency": 50074,
      "terms presence": 120363,
      "incomplete sentences": 53424,
      "stateoftheart classification": 112608,
      "sentiment composition": 106724,
      "modeled simple": 72364,
      "simple heuristics": 109444,
      "capture true": 15411,
      "multiword phrases": 75948,
      "created dataset": 24661,
      "realvalued scores": 96144,
      "sentiment association": 106685,
      "using phrasal": 130012,
      "dataset analyze": 26734,
      "impact individual": 51872,
      "overall sentiment": 83258,
      "varies substantially": 130951,
      "learning promising": 62920,
      "automatic sentiment": 10652,
      "analysis product": 5775,
      "utmost importance": 130586,
      "huge number": 50724,
      "product paper": 92040,
      "related problems": 97889,
      "second problem": 104445,
      "generalization problem": 46793,
      "considered special": 21303,
      "identify novel": 51525,
      "novel question": 80695,
      "allow automatic": 5074,
      "called dual": 15057,
      "qa pair": 94510,
      "predict actual": 88870,
      "challenges addressed": 16125,
      "quantitatively qualitatively": 94893,
      "coverage accuracy": 24496,
      "accuracy compared": 2119,
      "satisfy needs": 103659,
      "paper identify": 83975,
      "novel qa": 80694,
      "qa corpus": 94500,
      "design neural": 29550,
      "network called": 77179,
      "network san": 77423,
      "questions model": 95330,
      "model leverages": 71456,
      "information perform": 54834,
      "perform semisupervised": 86061,
      "function high": 46024,
      "baselines hungarian": 12410,
      "fits data": 44586,
      "calculus graph": 15039,
      "dynamic process": 34321,
      "process guided": 91497,
      "novel methodology": 80638,
      "network specifically": 77442,
      "sentence matching": 105944,
      "model applies": 70671,
      "matching results": 67429,
      "baselines substantially": 12474,
      "replaced conventional": 98933,
      "statistical translation": 113173,
      "approach relying": 7854,
      "manual handcrafted": 67002,
      "network composed": 77202,
      "input language": 55353,
      "desired output": 29668,
      "sentence model": 105949,
      "architecture takes": 8756,
      "warm start": 132022,
      "problem end": 91027,
      "make following": 66674,
      "following contributions": 44967,
      "create parallel": 24634,
      "urdu language": 127841,
      "using bleu": 129418,
      "able correctly": 1625,
      "correctly predict": 24180,
      "sentences length": 106379,
      "achieving bleu": 2936,
      "shall serve": 107911,
      "serve baseline": 107288,
      "baseline work": 12341,
      "work domain": 134483,
      "domain neural": 33594,
      "using distributed": 129596,
      "representation avoiding": 99176,
      "retrievalbased conversation": 102440,
      "conversation retrievalbased": 22970,
      "conversation systems": 22975,
      "systems generally": 116917,
      "responses semantically": 101296,
      "similar identical": 109096,
      "given conversation": 48007,
      "conversation context": 22940,
      "context systems": 22289,
      "appropriate response": 8430,
      "mitigate problem": 70375,
      "hard negative": 49470,
      "negative mining": 76934,
      "mining approach": 70227,
      "resulting model": 101453,
      "model reduces": 71875,
      "terms average": 120278,
      "compared models": 19388,
      "approach hierarchical": 7602,
      "hierarchical text": 50013,
      "generation planning": 47544,
      "endtoend models": 36931,
      "challenging train": 16343,
      "entangled latent": 37689,
      "latent state": 61612,
      "state vectors": 112540,
      "introduce approach": 56380,
      "learning representations": 62960,
      "maximizing likelihood": 67522,
      "dialogue utterance": 30813,
      "latent sentence": 61607,
      "representations hierarchical": 99675,
      "approach increases": 7635,
      "achieved model": 2655,
      "improves effectiveness": 52974,
      "learning improve": 62639,
      "methods given": 69524,
      "text research": 121259,
      "cbow skipgram": 15849,
      "skipgram methods": 109985,
      "specific design": 111429,
      "distant domains": 32446,
      "way solving": 132131,
      "solving problems": 110648,
      "interested exploring": 56065,
      "specific aspects": 111407,
      "aspects product": 9404,
      "search specific": 104351,
      "specific needs": 111471,
      "needs specifically": 76899,
      "contribute novel": 22758,
      "abstracting specific": 1787,
      "approaches attentive": 8070,
      "attentive memory": 10116,
      "efficient machine": 35087,
      "answer based": 6517,
      "leveraging external": 63677,
      "natural conversation": 76246,
      "additional source": 3572,
      "source information": 110765,
      "information account": 54353,
      "utterances produced": 130662,
      "information conveyed": 54455,
      "conveyed user": 23125,
      "user conversation": 128972,
      "reading task": 96032,
      "task automated": 117907,
      "questions questions": 95348,
      "questions answered": 95271,
      "retrieval community": 102386,
      "setting paper": 107767,
      "key contribution": 57561,
      "speed important": 111866,
      "important requirement": 52225,
      "conversational turns": 23043,
      "datasets commonly": 27357,
      "achieves performance": 2830,
      "model ranking": 71846,
      "dialogue opendomain": 30716,
      "amazon alexa": 5264,
      "alexa prize": 4673,
      "time real": 122085,
      "humans popular": 51100,
      "ranker trained": 95654,
      "trained real": 123259,
      "problem train": 91266,
      "obtained competition": 81353,
      "emergence new": 36056,
      "social contexts": 110331,
      "forums social": 45319,
      "nonstandard language": 80294,
      "language form": 59053,
      "language focus": 59052,
      "focus linguistic": 44783,
      "largely neglected": 61343,
      "conduct largescale": 20883,
      "language increasingly": 59162,
      "increasingly used": 53715,
      "world online": 135040,
      "computationally analyzing": 20446,
      "phonological morphological": 87310,
      "syntactic properties": 116453,
      "patterns specific": 85781,
      "usage analyzing": 127855,
      "tens thousands": 120178,
      "reveals majority": 102519,
      "prevalent social": 90361,
      "social biases": 110326,
      "biases stereotypes": 13885,
      "prediction neural": 89087,
      "neural headline": 77926,
      "headline generation": 49596,
      "generation tasks": 47665,
      "model suffers": 72110,
      "important phrases": 52206,
      "phrases includes": 87431,
      "irrelevant entities": 56948,
      "token prediction": 122260,
      "prediction module": 89084,
      "method jointly": 68925,
      "estimates probability": 38644,
      "distributions source": 32769,
      "target vocabularies": 117740,
      "task additionally": 117846,
      "additionally method": 3613,
      "true alignments": 125845,
      "alignments words": 5024,
      "easier learn": 34416,
      "form learning": 45101,
      "distribution word": 32689,
      "words variable": 134301,
      "phonological level": 87309,
      "opposite direction": 82152,
      "phonological space": 87314,
      "space does": 110994,
      "opposite effect": 82153,
      "enriching lexical": 37566,
      "corpora manually": 23524,
      "kinds semantic": 57692,
      "available high": 11004,
      "high lexical": 50084,
      "lexical coverage": 63751,
      "combines domain": 18683,
      "domain specificity": 33653,
      "information distributional": 54501,
      "manually crafted": 67051,
      "lexical networks": 63787,
      "vocabulary terms": 131908,
      "information given": 54640,
      "representations obtain": 99786,
      "induced word": 54011,
      "ontology resulting": 81848,
      "semantic type": 105339,
      "judgments different": 57444,
      "method extrinsic": 68836,
      "knowledgebased word": 58258,
      "hybrid resource": 51193,
      "knowledge resources": 58154,
      "building sentiment": 14885,
      "sentiment corpus": 106726,
      "brazilian portuguese": 14572,
      "available social": 11104,
      "areas natural": 8894,
      "processing sentiment": 91792,
      "semantic characteristics": 105005,
      "motivates research": 74871,
      "research novel": 100573,
      "novel methods": 80639,
      "methods approaches": 69318,
      "classification high": 17221,
      "high demand": 50059,
      "domain sentences": 33636,
      "classes positive": 17070,
      "positive neutral": 88337,
      "annotators following": 6492,
      "reliability annotation": 98605,
      "baseline experiments": 12219,
      "description annotation": 29454,
      "use corpora": 127971,
      "controlled experiments": 22835,
      "experiments experiments": 40945,
      "data include": 26030,
      "large proportions": 61225,
      "descriptions addition": 29476,
      "corpora usually": 23615,
      "contexts need": 22410,
      "need annotated": 76779,
      "considerable time": 21264,
      "poor data": 88054,
      "data quality": 26308,
      "quality high": 94679,
      "costs work": 24404,
      "method annotation": 68637,
      "associations words": 9637,
      "intended facilitate": 55866,
      "facilitate design": 42763,
      "experiments produce": 41066,
      "email intent": 35337,
      "propose annotate": 92545,
      "argue approach": 8909,
      "good interannotator": 48482,
      "conversation domain": 22949,
      "domain adaptive": 33462,
      "neural bagofwords": 77850,
      "collection datasets": 18470,
      "datasets consisting": 27377,
      "rnns outperform": 102976,
      "outperform common": 82696,
      "speech act": 111641,
      "tasks experiment": 119117,
      "rnns learn": 102972,
      "useful representation": 128922,
      "personal names": 87121,
      "statistics based": 113193,
      "knowledge study": 58196,
      "datadriven analysis": 26655,
      "greater diversity": 49037,
      "patterns example": 85741,
      "hundreds unique": 51157,
      "respectively observed": 101153,
      "pronunciation lexicon": 92367,
      "stateoftheart speech": 112966,
      "recognition systems": 97019,
      "basic components": 12515,
      "model pronunciation": 71806,
      "lexicon language": 63890,
      "technical expertise": 119747,
      "lowresource domains": 65498,
      "techniques construct": 119855,
      "expert domain": 41218,
      "great demand": 49000,
      "despite having": 29691,
      "language terms": 60158,
      "terms standard": 120385,
      "publically available": 94278,
      "conversion tool": 23089,
      "form suitable": 45130,
      "use speech": 128291,
      "using lstmbased": 129844,
      "expert lexicon": 41226,
      "shows accuracy": 108545,
      "accuracy 64": 2069,
      "obtain word": 81336,
      "rate comparable": 95787,
      "sandhi splitting": 103626,
      "compound words": 20155,
      "process known": 91516,
      "compound word": 20154,
      "word splitting": 133585,
      "language highly": 59122,
      "highly challenging": 50295,
      "challenging identify": 16260,
      "identify location": 51519,
      "low accuracy": 65344,
      "multiple ways": 75745,
      "ways provide": 132171,
      "syntactically correct": 116509,
      "architecture called": 8632,
      "95 accuracy": 1419,
      "accuracy outperforming": 2226,
      "generalization capability": 46770,
      "capability deep": 15179,
      "model showing": 72007,
      "showing competitive": 108400,
      "results problem": 102055,
      "log data": 64978,
      "data challenging": 25725,
      "challenging practical": 16292,
      "research problem": 100587,
      "problem obtain": 91149,
      "obtain effective": 81278,
      "product titles": 92048,
      "make original": 66704,
      "traditional text": 122880,
      "text summarization": 121344,
      "summarization approaches": 115479,
      "approaches require": 8325,
      "novel multitask": 80659,
      "pointer networkbased": 87828,
      "networkbased sequencetosequence": 77493,
      "approach utilized": 8009,
      "utilized title": 130547,
      "extractive method": 42553,
      "encoderdecoder approach": 36590,
      "semantic embedding": 105039,
      "embedding original": 35465,
      "tasks attention": 118944,
      "extensive set": 41955,
      "experiments human": 40956,
      "online deployment": 81763,
      "demonstrate advantage": 28657,
      "proposed research": 93539,
      "english poetry": 37244,
      "poetry corpus": 87790,
      "neurocognitive poetics": 78742,
      "describes corpus": 29392,
      "gutenberg project": 49290,
      "fiction nonfiction": 43923,
      "narrative analysis": 76167,
      "used explore": 128538,
      "subcorpus gutenberg": 114657,
      "gutenberg english": 49287,
      "analysis significant": 5854,
      "research digital": 100470,
      "continuous neural": 22626,
      "global optimization": 48256,
      "optimization problems": 82205,
      "parameters adjusted": 84735,
      "function time": 46046,
      "obtain global": 81285,
      "summary known": 115644,
      "special cases": 111353,
      "correct answer": 24090,
      "analysis computational": 5540,
      "interactions roles": 56000,
      "important concepts": 52128,
      "concepts understanding": 20648,
      "understanding human": 126857,
      "distinct patterns": 32541,
      "active passive": 3119,
      "providing new": 94124,
      "individual group": 53910,
      "group different": 49135,
      "different combinations": 31045,
      "online collaborative": 81745,
      "collaborative learning": 18362,
      "worse learning": 135064,
      "learning outcomes": 62866,
      "validate novel": 130727,
      "linguistic techniques": 64565,
      "applied large": 7085,
      "mixedeffects modeling": 70409,
      "modeling used": 72583,
      "used assess": 128396,
      "assess validity": 9497,
      "patterns linguistic": 85758,
      "framework researchers": 45674,
      "researchers explore": 100689,
      "literary text": 64738,
      "text given": 121027,
      "rise new": 102868,
      "mt neural": 74976,
      "mt nmt": 74977,
      "assess translation": 9493,
      "text specifically": 121321,
      "specifically target": 111593,
      "target novels": 117682,
      "popular type": 88141,
      "text build": 120691,
      "translation direction": 124754,
      "paradigm mt": 84541,
      "mt pbsmt": 74982,
      "time train": 122126,
      "systems nmt": 117025,
      "nmt pbsmt": 79942,
      "amounts literary": 5349,
      "100 million": 165,
      "words evaluate": 133942,
      "widely known": 132548,
      "nmt results": 79956,
      "results 11": 101482,
      "11 relative": 226,
      "improvement points": 52742,
      "produced nmt": 91963,
      "speakers target": 111327,
      "quality translations": 94818,
      "professional human": 92078,
      "human translator": 50980,
      "partially inspired": 85305,
      "novel variational": 80770,
      "paper different": 83873,
      "latent random": 61595,
      "random variables": 95515,
      "way instead": 132096,
      "instead single": 55683,
      "variational autoencoder": 130909,
      "enables capture": 36379,
      "complex dependencies": 19807,
      "dependencies output": 29101,
      "translations different": 125460,
      "challenges performing": 16191,
      "chineseenglish englishgerman": 16849,
      "improvements conventional": 52835,
      "models contextual": 72979,
      "factorization machines": 42872,
      "classification existing": 17201,
      "achieved great": 2631,
      "classification typically": 17475,
      "explicitly capture": 41361,
      "word interaction": 133324,
      "poor results": 88062,
      "results finegrained": 101794,
      "level phrase": 63488,
      "phrase sentence": 87368,
      "possible approach": 88380,
      "recommender systems": 97097,
      "systems directly": 116854,
      "applicable task": 6829,
      "contexts word": 22435,
      "interaction context": 55941,
      "context position": 22219,
      "position information": 88288,
      "information information": 54685,
      "learning experimental": 62563,
      "methods documentlevel": 69456,
      "efficient text": 35119,
      "using treestructured": 130323,
      "principal component": 90670,
      "component analysis": 19965,
      "analysis novel": 5723,
      "novel text": 80750,
      "dimension reduction": 31745,
      "reduction technique": 97461,
      "proposed work": 93592,
      "work different": 134475,
      "different traditional": 31504,
      "reduces dimension": 97382,
      "tasks shown": 119499,
      "analysis pca": 5746,
      "results support": 102240,
      "data achieves": 25565,
      "stateoftheart recurrent": 112881,
      "rnn approach": 102900,
      "chinese dependency": 16759,
      "treebank web": 125651,
      "web text": 132264,
      "web 20": 132220,
      "revealing ones": 102511,
      "great source": 49027,
      "source tasks": 110850,
      "challenges current": 16144,
      "linguistic phenomenon": 64522,
      "understood context": 127023,
      "context especially": 22083,
      "great importance": 49005,
      "order promote": 82388,
      "promote research": 92319,
      "extracting relations": 42232,
      "events paper": 39590,
      "history computational": 50548,
      "computer science": 20488,
      "uses data": 129217,
      "historical research": 50533,
      "projects main": 92242,
      "target group": 117623,
      "key concepts": 57560,
      "term allowing": 120205,
      "inherent uncertainty": 55191,
      "detection benchmark": 29888,
      "text large": 121078,
      "highquality benchmark": 50372,
      "benchmark evaluation": 12821,
      "detection tools": 30084,
      "tools benchmark": 122441,
      "benchmark contains": 12745,
      "entities annotated": 37743,
      "annotated different": 6184,
      "clean text": 17704,
      "text taken": 121357,
      "taken wikipedia": 117520,
      "spoken data": 111972,
      "benchmark built": 12737,
      "crowd sourcing": 25097,
      "quality benchmark": 94605,
      "process guidelines": 91498,
      "guidelines used": 49266,
      "benchmark evaluating": 12820,
      "endtoend framework": 36902,
      "clear learn": 17726,
      "models morphological": 73583,
      "layers nmt": 61795,
      "nmt encoders": 79859,
      "systems parallel": 117049,
      "models extract": 73220,
      "training classifier": 123383,
      "performance classifier": 86205,
      "quality original": 94736,
      "original nmt": 82534,
      "quantitative analysis": 94862,
      "analysis yields": 5942,
      "yields interesting": 135423,
      "regarding representation": 97659,
      "learning nmt": 62851,
      "models instance": 73406,
      "instance higher": 55601,
      "higher layers": 50185,
      "layers better": 61765,
      "learning semantics": 62998,
      "layers tend": 61810,
      "better partofspeech": 13648,
      "observe little": 81201,
      "little effect": 64804,
      "representations especially": 99634,
      "especially higher": 38459,
      "higher quality": 50198,
      "continuous space": 22635,
      "models fall": 73235,
      "small context": 110143,
      "context sizes": 22272,
      "previously addressed": 90586,
      "addressed learning": 3783,
      "generalized representations": 46832,
      "alternative based": 5226,
      "using feedforward": 129673,
      "results improvements": 101845,
      "germanenglish englishgerman": 47933,
      "observed improvements": 81224,
      "used pos": 128687,
      "models modify": 73580,
      "bilingual corpus": 14029,
      "allows train": 5190,
      "train sequencetosequence": 123016,
      "model having": 71276,
      "having explicit": 49558,
      "reordering information": 98899,
      "information encoderdecoder": 54526,
      "relies solely": 98662,
      "handle long": 49391,
      "experiments did": 40905,
      "report summarizes": 99053,
      "exploratory study": 41513,
      "texts topics": 121637,
      "group people": 49140,
      "study investigated": 114414,
      "early modern": 34392,
      "available website": 11144,
      "university sheffield": 127355,
      "pipeline used": 87559,
      "used automatically": 128402,
      "automatically process": 10812,
      "network representation": 77408,
      "representation order": 99362,
      "entities texts": 37880,
      "ner task": 77088,
      "task classify": 117965,
      "text classes": 120706,
      "organization important": 82467,
      "important preprocessing": 52209,
      "step nlp": 113287,
      "tasks questionanswering": 119430,
      "summarization research": 115562,
      "research studies": 100630,
      "studies conducted": 114198,
      "english stateoftheart": 37293,
      "stateoftheart ner": 112772,
      "systems reached": 117095,
      "90 percent": 1369,
      "studies task": 114288,
      "ner dataset": 77036,
      "dataset train": 27240,
      "free research": 45764,
      "order construct": 82301,
      "standard ner": 112276,
      "ner datasets": 77037,
      "datasets constructed": 27379,
      "using news": 129966,
      "texts collected": 121475,
      "documents news": 33260,
      "news websites": 79398,
      "order provide": 82393,
      "english datasets": 37107,
      "linguistic rules": 64551,
      "novel sense": 80718,
      "temporal spatial": 120128,
      "existing wsd": 40343,
      "systems largely": 116972,
      "corpus specific": 24010,
      "specific word": 111517,
      "adapt existing": 3177,
      "algorithms identify": 4857,
      "available form": 10998,
      "methods identify": 69537,
      "time points": 122069,
      "judgment experiment": 57439,
      "performance approaches": 86147,
      "goaloriented chatbot": 48405,
      "dialog management": 30570,
      "learning goaloriented": 62618,
      "systems colloquially": 116787,
      "colloquially known": 18525,
      "known goal": 58298,
      "goal oriented": 48373,
      "oriented chatbots": 82487,
      "chatbots help": 16672,
      "help users": 49785,
      "users achieve": 129090,
      "achieve predefined": 2511,
      "predefined goal": 88824,
      "goal book": 48333,
      "book movie": 14408,
      "movie ticket": 74907,
      "ticket closed": 121957,
      "closed domain": 17843,
      "domain step": 33658,
      "step understand": 113307,
      "understand users": 126786,
      "users goal": 129124,
      "understanding techniques": 126993,
      "techniques goal": 119895,
      "goal known": 48360,
      "known bot": 58289,
      "bot manage": 14481,
      "manage dialogue": 66896,
      "dialogue achieve": 30625,
      "goal conducted": 48338,
      "conducted respect": 20940,
      "respect learnt": 101086,
      "learnt policy": 63250,
      "policy success": 87965,
      "success dialogue": 115073,
      "dialogue depends": 30660,
      "depends quality": 29289,
      "quality policy": 94753,
      "policy turn": 87970,
      "turn reliant": 125976,
      "reliant availability": 98634,
      "availability highquality": 10909,
      "highquality training": 50410,
      "data policy": 26243,
      "method instance": 68912,
      "instance deep": 55595,
      "specificity available": 111614,
      "typically low": 126441,
      "low allow": 65347,
      "allow training": 5093,
      "training good": 123642,
      "good dialogue": 48473,
      "dialogue policies": 30721,
      "introduce transfer": 56556,
      "method mitigate": 68963,
      "mitigate effects": 70363,
      "effects low": 34990,
      "low indomain": 65366,
      "availability transfer": 10928,
      "improves bots": 52955,
      "bots success": 14489,
      "rate 20": 95773,
      "20 relative": 520,
      "relative terms": 98379,
      "terms distant": 120308,
      "domains double": 33768,
      "double close": 33961,
      "close domains": 17811,
      "domains compared": 33745,
      "learning transfer": 63118,
      "learning chatbots": 62426,
      "chatbots learn": 16676,
      "learn policy": 62112,
      "policy 10": 87944,
      "10 times": 149,
      "faster finally": 43174,
      "finally transfer": 44241,
      "approach complementary": 7437,
      "complementary additional": 19733,
      "additional processing": 3559,
      "processing warmstarting": 91852,
      "warmstarting joint": 132026,
      "joint application": 57258,
      "application gives": 6852,
      "best outcomes": 13385,
      "knowledge multiple": 58070,
      "multiple distinct": 75542,
      "object features": 81051,
      "structure human": 113874,
      "human mind": 50917,
      "language represent": 60037,
      "knowledge semantic": 58167,
      "captured word": 15441,
      "meanings lexical": 67745,
      "represents words": 100078,
      "learned patterns": 62241,
      "language despite": 58949,
      "despite popularity": 29713,
      "fundamental concern": 46093,
      "overall similarity": 83259,
      "similarity human": 109239,
      "involve multiple": 56876,
      "features example": 43495,
      "similar size": 109146,
      "embeddings address": 35546,
      "issue introduce": 57000,
      "introduce powerful": 56524,
      "represent various": 99152,
      "various object": 131160,
      "object categories": 81048,
      "categories properties": 15746,
      "word cooccurrence": 132971,
      "cooccurrence statistics": 23233,
      "contextdependent meanings": 22353,
      "product description": 92031,
      "propose question": 93009,
      "selects sentence": 104866,
      "explore multiple": 41563,
      "encoding strategies": 36727,
      "attention layers": 9867,
      "yielding good": 135378,
      "ranking scores": 95685,
      "scores product": 104193,
      "convolutional autoencoder": 23152,
      "autoencoder text": 10408,
      "text article": 120659,
      "explore possible": 41574,
      "fixedlength representation": 44616,
      "representation possible": 99378,
      "achieved better": 2600,
      "networks proposed": 77713,
      "deep convolutional": 28221,
      "convolutional encoderdecoder": 23155,
      "residual connections": 100738,
      "layers encoder": 61773,
      "contains shared": 21839,
      "including arabic": 53262,
      "arabic chinese": 8495,
      "english analyses": 37067,
      "analyses conducted": 5444,
      "properties proposed": 92474,
      "model enhance": 71079,
      "representation outofvocabulary": 99364,
      "largest public": 61538,
      "public available": 94234,
      "corpus make": 23875,
      "feasible build": 43241,
      "build endtoend": 14763,
      "endtoend deep": 36883,
      "data challenge": 25723,
      "general pretrained": 46696,
      "vectors generated": 131439,
      "set address": 107352,
      "chen et": 16713,
      "enhanced lstm": 37511,
      "task utterance": 118836,
      "selection proposed": 104821,
      "method demonstrated": 68762,
      "improvement original": 52734,
      "original esim": 82520,
      "conversation corpus": 22942,
      "corpus addition": 23645,
      "understanding recurrent": 126940,
      "using memory": 129875,
      "network visualization": 77474,
      "visualization technique": 131828,
      "technique analyze": 119766,
      "recurrent state": 97260,
      "language acoustic": 58809,
      "acoustic models": 3020,
      "intermediate state": 56143,
      "state network": 112508,
      "models remains": 73922,
      "open challenge": 81882,
      "users understand": 129178,
      "grapheme sequence": 48919,
      "trains multiple": 123966,
      "multiple decoders": 75535,
      "predict prior": 88917,
      "user obtain": 129016,
      "memory behavior": 68292,
      "demonstrate methods": 28784,
      "extracting knowledge": 42217,
      "endtoend asr": 36874,
      "networks automatic": 77516,
      "languageindependent features": 60363,
      "features crosslingual": 43432,
      "crosslingual classification": 24933,
      "classification applications": 17116,
      "documents using": 33310,
      "using predefined": 130038,
      "approach performing": 7795,
      "performing text": 86977,
      "learning labeled": 62667,
      "labeled examples": 58458,
      "tasks difficult": 119063,
      "examples language": 39850,
      "learning examples": 62559,
      "examples languages": 39851,
      "called crosslingual": 15054,
      "crosslingual learning": 24969,
      "learning work": 63172,
      "approach solves": 7924,
      "categorization problem": 15768,
      "problem method": 91126,
      "generates training": 47188,
      "document set": 33078,
      "features using": 43778,
      "classifier classification": 17528,
      "classification stage": 17417,
      "stage generate": 112148,
      "unlabeled document": 127395,
      "document apply": 32951,
      "representation build": 99181,
      "utilize hierarchical": 130508,
      "ontology concept": 81840,
      "concept set": 20588,
      "preprocessing stage": 89333,
      "stage use": 112155,
      "languageindependent feature": 60362,
      "language collection": 58886,
      "used map": 128624,
      "method exploiting": 68828,
      "create virtual": 24651,
      "supporting documents": 116027,
      "documents languages": 33247,
      "languages tested": 60917,
      "tested method": 120576,
      "using wikipedia": 130376,
      "ontology commonly": 81838,
      "test collections": 120430,
      "collections crosslingual": 18502,
      "methods unified": 69822,
      "dialog framework": 30563,
      "framework conversational": 45476,
      "search propose": 104341,
      "propose unified": 93127,
      "search applications": 104298,
      "dialog interactions": 30567,
      "underlying data": 126679,
      "components required": 20046,
      "framework consists": 45467,
      "consists pipeline": 21489,
      "pipeline endtoend": 87537,
      "used semantically": 128745,
      "integrated framework": 55777,
      "users input": 129134,
      "improvement goal": 52715,
      "facilitate development": 42765,
      "development conversational": 30374,
      "systems identifying": 116942,
      "identifying components": 51590,
      "data adapted": 25570,
      "enduser applications": 36993,
      "applications demonstrate": 6913,
      "approach creating": 7466,
      "subwordlevel information": 115049,
      "vectors convolutional": 131422,
      "work tackle": 134839,
      "problem sentence": 91213,
      "detection applied": 29878,
      "combine convolutional": 18617,
      "embedding representations": 35483,
      "words morphology": 134065,
      "word represented": 133481,
      "models posterior": 73758,
      "architectures tested": 8851,
      "results general": 101807,
      "accuracy models": 2210,
      "models good": 73307,
      "lower scores": 65448,
      "f1 metric": 42662,
      "words encoded": 133937,
      "encoded embeddings": 36475,
      "behave like": 12643,
      "feasible use": 43244,
      "nmt widely": 80009,
      "improvements language": 52864,
      "systems generating": 116920,
      "translation remains": 125214,
      "challenges field": 16161,
      "interesting language": 56079,
      "overall meaning": 83237,
      "composed meanings": 20067,
      "important challenge": 52111,
      "challenge lack": 16044,
      "sets learning": 107683,
      "problem creating": 90988,
      "creating largescale": 24700,
      "set automatically": 107368,
      "includes language": 53255,
      "language direction": 58966,
      "targeted evaluation": 117766,
      "corpus sentences": 23988,
      "release data": 98445,
      "perform preliminary": 86048,
      "nmt experiments": 79864,
      "step better": 113253,
      "analyzing text": 6055,
      "method classifying": 68694,
      "used large": 128606,
      "dataset v10": 27265,
      "knearest neighbors": 57717,
      "neighbors knn": 76997,
      "knn model": 57726,
      "perceptron mlp": 85931,
      "features paper": 43644,
      "different evaluation": 31132,
      "classification data": 17169,
      "used research": 128729,
      "deep contextualized": 28214,
      "contextualized word": 22572,
      "new type": 79231,
      "models complex": 72942,
      "use syntax": 128310,
      "contexts model": 22409,
      "polysemy word": 88030,
      "deep bidirectional": 28202,
      "model bilm": 70780,
      "pretrained large": 90112,
      "added existing": 3358,
      "art challenging": 9038,
      "challenging nlp": 16285,
      "problems including": 91325,
      "including question": 53355,
      "answering textual": 6714,
      "entailment sentiment": 37675,
      "analysis showing": 5847,
      "downstream models": 34002,
      "models mix": 73571,
      "convolution networks": 23141,
      "human conversation": 50786,
      "crucial paper": 25157,
      "paper concentrate": 83769,
      "selection multiturn": 104807,
      "context previous": 22227,
      "capturing information": 15471,
      "rare keywords": 95739,
      "correct response": 24118,
      "long input": 65077,
      "sequences propose": 107133,
      "propose cross": 92608,
      "dataset largest": 26997,
      "based dialogue": 11645,
      "considerable improvements": 21251,
      "results open": 102014,
      "scientific text": 104024,
      "text evaluation": 120924,
      "extraction oie": 42416,
      "structured information": 114002,
      "including knowledge": 53307,
      "methods targeted": 69792,
      "independent evaluated": 53768,
      "evaluated primarily": 39001,
      "article evaluate": 9126,
      "scientific texts": 104025,
      "10 different": 129,
      "oie systems": 81636,
      "systems applying": 116733,
      "crowdsourcing approach": 25116,
      "significantly worse": 109056,
      "text encyclopedic": 120908,
      "encyclopedic text": 36791,
      "text provide": 121222,
      "analysis suggest": 5878,
      "reduce errors": 97326,
      "calculating similarity": 15034,
      "calculating semantic": 15032,
      "sentences long": 106385,
      "problem area": 90940,
      "analysis field": 5622,
      "role play": 103212,
      "research related": 100611,
      "related text": 97905,
      "issue incorporating": 56999,
      "calculate semantic": 15016,
      "method follows": 68846,
      "methodology applied": 69245,
      "variety domains": 130971,
      "methodology tested": 69267,
      "tested benchmark": 120562,
      "mean human": 67598,
      "dataset tested": 27236,
      "datasets gives": 27496,
      "gives highest": 48196,
      "value word": 130788,
      "similar models": 109113,
      "pearson correlation": 85836,
      "correlation coefficient": 24226,
      "deep transfer": 28426,
      "trained single": 123278,
      "dataset performance": 27093,
      "models rely": 73920,
      "size dataset": 109919,
      "dataset data": 26842,
      "presented training": 89801,
      "training work": 123954,
      "propose inductive": 92722,
      "inductive transfer": 54051,
      "method augment": 68650,
      "instances different": 55625,
      "nlp domain": 79606,
      "domain propose": 33619,
      "source dataset": 110740,
      "model representations": 71908,
      "representations instances": 99700,
      "learned retrieval": 62255,
      "retrieval relevant": 102427,
      "relevant source": 98585,
      "training target": 123902,
      "approach simultaneously": 7920,
      "exploits local": 41492,
      "approach shown": 7890,
      "major news": 66574,
      "news classification": 79316,
      "classification datasets": 17171,
      "datasets baseline": 27328,
      "experimental evaluations": 40535,
      "evaluations proposed": 39480,
      "approach reduces": 7848,
      "data significant": 26448,
      "significant margin": 108824,
      "dataset learning": 26999,
      "learning single": 63030,
      "corpus document": 23763,
      "detection detecting": 29921,
      "entire document": 37704,
      "intelligence ai": 55822,
      "widespread nlp": 132616,
      "extractive document": 42551,
      "news events": 79337,
      "events predicting": 39593,
      "scholarly articles": 103960,
      "articles important": 9193,
      "important problem": 52211,
      "level data": 63433,
      "evaluation automatic": 39124,
      "detection techniques": 30080,
      "classification framework": 17213,
      "gap present": 46472,
      "present resource": 89666,
      "news documents": 79334,
      "documents domains": 33223,
      "release annotated": 98432,
      "textual content": 121679,
      "content structure": 21953,
      "structure improve": 113876,
      "improve dialog": 52372,
      "messaging apps": 68525,
      "advances artificial": 3975,
      "requires understanding": 100329,
      "understanding data": 126824,
      "generated users": 47147,
      "users dialog": 129108,
      "dialog data": 30558,
      "different nature": 31282,
      "context temporal": 22295,
      "propose novelty": 92979,
      "novelty metric": 80790,
      "based text": 12113,
      "content information": 21892,
      "information related": 54911,
      "dialog structure": 30592,
      "switchboard dataset": 116284,
      "using evidence": 129646,
      "leads accurate": 61917,
      "opensource toolkit": 82032,
      "toolkit named": 122413,
      "scratch using": 104244,
      "training efficiency": 123599,
      "accelerates training": 1876,
      "various sizes": 131197,
      "knowledge implicit": 58004,
      "syntactically connected": 116508,
      "hard extract": 49460,
      "large numbers": 61194,
      "models implicit": 73357,
      "prediction simple": 89125,
      "cloze task": 17894,
      "scale allows": 103703,
      "model draws": 71030,
      "narrative coherence": 76168,
      "predictions model": 89181,
      "model superior": 72114,
      "performance synthetic": 86779,
      "synthetic natural": 116632,
      "decompositions word": 28163,
      "useful approach": 128860,
      "capture cooccurrence": 15284,
      "structures large": 114083,
      "corpora addition": 23402,
      "addition text": 3480,
      "associated individual": 9597,
      "like embedding": 64037,
      "naturally capture": 76639,
      "capture information": 15316,
      "cover new": 24487,
      "tensor decomposition": 120187,
      "model vector": 72304,
      "matrix model": 67479,
      "base embedding": 11453,
      "transformation matrix": 124265,
      "advantages approach": 4068,
      "approach data": 7469,
      "data efficiency": 25880,
      "efficiency interpretability": 35029,
      "embeddings compared": 35603,
      "standard approach": 112185,
      "using relevant": 130107,
      "subset data": 114828,
      "data related": 26329,
      "related methods": 97881,
      "methods furthermore": 69520,
      "algorithm datasets": 4712,
      "demonstrate used": 28896,
      "used address": 128381,
      "address natural": 3717,
      "natural questions": 76615,
      "code paper": 18127,
      "data generation": 25977,
      "generation natural": 47497,
      "data recently": 26326,
      "gained significant": 46372,
      "significant attention": 108726,
      "number approaches": 80841,
      "approaches generate": 8172,
      "language languages": 59236,
      "english work": 37352,
      "proposed generate": 93302,
      "research gap": 100523,
      "gap presenting": 46473,
      "portuguese language": 88199,
      "language evaluated": 59016,
      "experts nonexperts": 41244,
      "nonexperts results": 80203,
      "able generate": 1644,
      "generate text": 47032,
      "text similar": 121295,
      "generated humans": 47085,
      "offers opportunities": 81591,
      "improve natural": 52430,
      "semantically annotated": 105362,
      "annotated parallel": 6219,
      "sentences aligned": 106204,
      "representations order": 99791,
      "presupposition triggers": 89968,
      "formalism based": 45185,
      "computing precision": 20524,
      "tool evaluating": 122384,
      "accurate efficient": 2332,
      "study performed": 114467,
      "automatically changes": 10732,
      "changes meaning": 16388,
      "comparing meaning": 19505,
      "improving sentiment": 53163,
      "analysis arabic": 5500,
      "arabic language": 8519,
      "orthography dialects": 82605,
      "dialects makes": 30538,
      "makes sentiment": 66805,
      "arabic challenging": 8494,
      "challenging text": 16340,
      "short messages": 108221,
      "makes task": 66809,
      "task difficult": 118098,
      "years deep": 135247,
      "networks employed": 77582,
      "showed good": 108377,
      "results sentiment": 102155,
      "classification natural": 17299,
      "approach current": 7468,
      "powerful tool": 88679,
      "words contextual": 133883,
      "contextual text": 22502,
      "paper construct": 83786,
      "large arabic": 61034,
      "arabic corpus": 8499,
      "corpus obtained": 23913,
      "arab countries": 8486,
      "applying different": 7236,
      "networks different": 77568,
      "improved accuracy": 52592,
      "accuracy sentiment": 2272,
      "available arabic": 10949,
      "sentiment dataset": 106727,
      "dataset matching": 27014,
      "sentences hierarchical": 106338,
      "hierarchical sentence": 50004,
      "sentences identifying": 106346,
      "identifying relationship": 51618,
      "underlying natural": 126690,
      "prior research": 90727,
      "research proposed": 100595,
      "supervised deep": 115748,
      "learning schemes": 62984,
      "fail fully": 42959,
      "interactions paper": 55993,
      "representation components": 99192,
      "components different": 20025,
      "proposed sentence": 93547,
      "factorization technique": 42873,
      "technique leads": 119798,
      "invention new": 56688,
      "new unsupervised": 79236,
      "pair text": 83459,
      "text snippets": 121311,
      "optimal transport": 82174,
      "logical relationship": 65016,
      "models supervised": 74132,
      "training based": 123369,
      "based multiple": 11871,
      "microsoft research": 70060,
      "research paraphrase": 100579,
      "dataset extensive": 26919,
      "proposed hierarchical": 93307,
      "used significantly": 128761,
      "performance existing": 86349,
      "multiple supervised": 75714,
      "cnn long": 18008,
      "semeval2018 task": 105531,
      "task 11": 117811,
      "knowledge commonsense": 57829,
      "machine comprehension": 65746,
      "comprehension paper": 20207,
      "11 machine": 220,
      "comprehension using": 20240,
      "using commonsense": 129454,
      "knowledge use": 58224,
      "model interactions": 71357,
      "passage question": 85608,
      "question answers": 95127,
      "incorporate commonsense": 53453,
      "knowledge augment": 57759,
      "input relation": 55414,
      "relation embedding": 97972,
      "knowledge conceptnet": 57835,
      "conceptnet speer": 20602,
      "official test": 81605,
      "data code": 25732,
      "latent information": 61592,
      "processing area": 91624,
      "researchers main": 100697,
      "set words": 107639,
      "problem solved": 91236,
      "analysis achieve": 5484,
      "practical results": 88710,
      "applications instance": 6946,
      "features text": 43757,
      "words considering": 133877,
      "results deep": 101668,
      "high computational": 50044,
      "computational effort": 20377,
      "make model": 66698,
      "types applications": 126251,
      "models possible": 73757,
      "possible develop": 88396,
      "faster systems": 43192,
      "accuracy work": 2319,
      "work proposes": 134753,
      "proposes hybrid": 93601,
      "hybrid word": 51198,
      "suggestion model": 115362,
      "analysis considering": 5544,
      "completion challenge": 19792,
      "ancient chinese": 6090,
      "chinese limited": 16783,
      "corpora chinese": 23433,
      "chinese language": 16779,
      "reading sentences": 96031,
      "chinese paper": 16798,
      "model automatically": 70712,
      "corpora aligned": 23406,
      "sentencealigned corpora": 106135,
      "difficult train": 31668,
      "build sentence": 14806,
      "propose unsupervised": 93133,
      "unsupervised algorithm": 127599,
      "algorithm constructs": 4710,
      "using fact": 129665,
      "tokens based": 122299,
      "model copying": 70927,
      "copying mechanism": 23296,
      "local attention": 64904,
      "algorithm achieves": 4689,
      "score sentence": 104120,
      "relation argument": 97940,
      "argument extraction": 8941,
      "extraction work": 42541,
      "corpus query": 23956,
      "query entity": 94960,
      "model extract": 71160,
      "extract argument": 42068,
      "nonstandard entity": 80293,
      "type entities": 126196,
      "entities extracted": 37784,
      "standard named": 112272,
      "art corpus": 9043,
      "distantly supervised": 32489,
      "supervised dataset": 115747,
      "dataset based": 26757,
      "relations obtained": 98224,
      "develop compare": 30185,
      "range neural": 95589,
      "yielding large": 135379,
      "large improvements": 61107,
      "obtained neural": 81385,
      "neural question": 78640,
      "impact different": 51867,
      "architectures answer": 8781,
      "answer extraction": 6531,
      "systematically compared": 116693,
      "encoder based": 36498,
      "relative position": 98362,
      "position representations": 88291,
      "representations relying": 99852,
      "entirely attention": 37727,
      "mechanism transformer": 68042,
      "vaswani et": 131287,
      "contrast recurrent": 22712,
      "absolute position": 1746,
      "structure instead": 113885,
      "inputs work": 55499,
      "present alternative": 89363,
      "selfattention mechanism": 104883,
      "relative positions": 98366,
      "englishtogerman englishtofrench": 37446,
      "englishtofrench translation": 37444,
      "yields improvements": 135420,
      "improvements 13": 52789,
      "13 bleu": 282,
      "03 bleu": 16,
      "bleu absolute": 14275,
      "representations respectively": 99864,
      "representations yields": 100005,
      "efficient implementation": 35081,
      "selfattention mechanisms": 104889,
      "endtoend goaloriented": 36904,
      "goaloriented dialog": 48409,
      "advancements deep": 3969,
      "development endtoend": 30385,
      "endtoend trained": 36983,
      "systems systems": 117179,
      "systems achieve": 116711,
      "reallife scenarios": 96129,
      "address limitations": 3714,
      "positional information": 88303,
      "information fixed": 54618,
      "number possible": 80941,
      "response candidates": 101192,
      "positional encodings": 88302,
      "utterances furthermore": 130640,
      "furthermore using": 46221,
      "generate output": 46982,
      "candidates using": 15150,
      "using positional": 130034,
      "positional encoding": 88301,
      "better accuracies": 13498,
      "babi tasks": 11279,
      "network generating": 77268,
      "generating response": 47256,
      "computation time": 20354,
      "bengalienglish codemixed": 13041,
      "codemixed data": 18168,
      "phonetic based": 87290,
      "based lstm": 11817,
      "text remains": 121240,
      "remains challenging": 98790,
      "properties like": 92462,
      "present supervised": 89729,
      "level low": 63470,
      "character based": 16417,
      "models utilizing": 74314,
      "utilizing models": 130573,
      "models created": 72998,
      "created ensemble": 24666,
      "enhanced word": 37525,
      "bridging anaphora": 14603,
      "resolution current": 100757,
      "current models": 25300,
      "finegrained semantics": 44379,
      "semantics semantic": 105460,
      "resolving bridging": 100815,
      "requires knowledge": 100281,
      "instead semantic": 55682,
      "similarity information": 109242,
      "exploring syntactic": 41667,
      "demonstrate using": 28899,
      "isnotes corpus": 56967,
      "corpus furthermore": 23813,
      "furthermore achieve": 46142,
      "achieve substantial": 2577,
      "gain stateoftheart": 46354,
      "hou et": 50676,
      "modeling paradigm": 72502,
      "ability perform": 1558,
      "perform prediction": 86046,
      "models construct": 72973,
      "sequence probabilities": 107036,
      "parameters approach": 84736,
      "approach allows": 7348,
      "allows seamlessly": 5186,
      "items language": 57110,
      "tasks english": 119097,
      "english neural": 37219,
      "models utilize": 74313,
      "improve perplexity": 52496,
      "evaluation toolkit": 39426,
      "encompasses variety": 36740,
      "multiclass classification": 75051,
      "inference sentence": 54215,
      "similarity set": 109311,
      "consensus regarding": 21170,
      "appropriate evaluations": 8419,
      "preprocess datasets": 89318,
      "evaluate sentence": 38922,
      "sentence encoders": 105851,
      "way evaluating": 132076,
      "evaluating sentence": 39097,
      "representations achieving": 99490,
      "human parity": 50921,
      "automatic chinese": 10499,
      "english news": 37221,
      "rapid advances": 95709,
      "millions people": 70122,
      "people using": 85902,
      "today online": 122236,
      "applications order": 6981,
      "communicate language": 19017,
      "language barriers": 58856,
      "question naturally": 95187,
      "naturally arises": 76638,
      "human translations": 50979,
      "translations paper": 125483,
      "problem define": 91000,
      "accurately measure": 2385,
      "measure human": 67799,
      "parity translation": 84870,
      "translation measure": 124928,
      "measure quality": 67816,
      "translations widely": 125506,
      "task chinese": 117961,
      "latest neural": 61657,
      "translation reached": 125197,
      "stateoftheart translation": 113015,
      "quality human": 94680,
      "translations significantly": 125493,
      "significantly exceeds": 108911,
      "task word": 118851,
      "results shared": 102167,
      "induction wsi": 54044,
      "tasks conducted": 119001,
      "germanic languages": 47943,
      "disambiguation methods": 31971,
      "language shares": 60087,
      "languages rich": 60854,
      "participants asked": 85318,
      "senses provided": 105720,
      "datasets based": 27327,
      "different sense": 31409,
      "sense granularity": 105680,
      "datasets sampled": 27693,
      "corpus russian": 23978,
      "explanatory dictionary": 41303,
      "substantially outperform": 114904,
      "outperform competitive": 82697,
      "baselines previous": 12444,
      "previous years": 90583,
      "new lexical": 78993,
      "resource enriches": 100842,
      "resources semantic": 101042,
      "corpora features": 23488,
      "representations boost": 99524,
      "frame disambiguation": 45383,
      "disambiguation context": 31958,
      "development novel": 30408,
      "representations events": 99639,
      "wikidata wikipedia": 132634,
      "languages content": 60468,
      "domain wikipedia": 33695,
      "structured data": 113995,
      "architecture equipped": 8659,
      "equipped copy": 38241,
      "copy actions": 23285,
      "learns generate": 63212,
      "languages different": 60500,
      "arabic morphological": 8528,
      "language larger": 59238,
      "constructed language": 21664,
      "language known": 59231,
      "semantic natural": 105118,
      "platform paper": 87654,
      "research deep": 100460,
      "methods natural": 69631,
      "designed support": 29626,
      "researchers want": 100716,
      "build novel": 14797,
      "easily built": 34440,
      "built pytorch": 14935,
      "allowing dynamic": 5111,
      "computation graphs": 20349,
      "provides flexible": 94032,
      "working text": 134907,
      "framework makes": 45609,
      "reference implementations": 97523,
      "core semantic": 23332,
      "rajpurkar et": 95466,
      "doubt utility": 33973,
      "useful especially": 128883,
      "human translators": 50981,
      "quality improves": 94685,
      "longer single": 65162,
      "single gold": 109737,
      "hotel reviews": 50670,
      "reviews annotated": 102596,
      "aspectlevel sentiment": 9366,
      "research languages": 100544,
      "hindered lack": 50451,
      "resources research": 101037,
      "analysis focused": 5626,
      "focused unsupervised": 44882,
      "semisupervised approaches": 105592,
      "number resources": 80957,
      "reach performance": 95898,
      "performance supervised": 86775,
      "supervised approaches": 115724,
      "mind introduce": 70144,
      "introduce datasets": 56408,
      "datasets supervised": 27741,
      "provide highquality": 93844,
      "community researchers": 19098,
      "working languages": 134898,
      "expensive terms": 40434,
      "terms memory": 120347,
      "use lowresource": 128131,
      "lowresource devices": 65495,
      "method proposed": 69079,
      "embeddings preserving": 35874,
      "preserving semantic": 89938,
      "small memory": 110171,
      "autoencoder architecture": 10398,
      "allows reconstruct": 5183,
      "original vectors": 82561,
      "ones experimental": 81684,
      "analysis tasks": 5886,
      "leads loss": 61945,
      "size reduced": 109943,
      "using binary": 129416,
      "binary vectors": 14141,
      "faster using": 43198,
      "realvalued vectors": 96145,
      "learning provides": 62927,
      "methods applications": 69311,
      "study unsupervised": 114539,
      "native words": 76236,
      "language words": 60344,
      "key step": 57598,
      "problem unsupervised": 91279,
      "malayalam language": 66876,
      "key observation": 57590,
      "characters word": 16630,
      "word stem": 133587,
      "optimization method": 82195,
      "method relies": 69100,
      "distributions character": 32754,
      "iterative optimization": 57131,
      "evaluation illustrate": 39237,
      "illustrate method": 51740,
      "provides significant": 94074,
      "textual claims": 121675,
      "claims text": 17021,
      "text articles": 120660,
      "false claims": 43036,
      "fact fiction": 42823,
      "features tasks": 43754,
      "tasks main": 119283,
      "true false": 125848,
      "articles paper": 9205,
      "consists different": 21475,
      "different modules": 31267,
      "capturing various": 15497,
      "features includes": 43559,
      "retrieved knowledge": 102461,
      "architecture learns": 8684,
      "article experiments": 9128,
      "dataset reveals": 27172,
      "based linguistic": 11807,
      "wordlevel language": 133739,
      "modeling study": 72555,
      "study possibility": 114471,
      "representations encoded": 99626,
      "work extends": 134525,
      "extends recent": 41830,
      "modeling problem": 72515,
      "problem online": 91150,
      "continuously update": 22655,
      "identifying semantic": 51623,
      "semantic divergences": 105034,
      "correct translations": 24126,
      "translations semantically": 125490,
      "semantically equivalent": 105368,
      "parallel sentence": 84677,
      "pairs deep": 83511,
      "trained parallel": 123233,
      "annotation semantic": 6372,
      "model detects": 70981,
      "based surface": 12094,
      "matter neural": 67487,
      "dependency context": 29135,
      "context crosslingual": 22045,
      "determining word": 30157,
      "ability detect": 1505,
      "detect hypernymy": 29808,
      "event coreference": 39501,
      "bilingual word": 14069,
      "dependency contexts": 29136,
      "task compared": 117985,
      "lexical context": 63748,
      "approach robust": 7865,
      "showing promise": 108424,
      "using parser": 130007,
      "trained related": 123262,
      "loss performance": 65291,
      "challenging dataset": 16241,
      "task languages": 118340,
      "embeddings datasets": 35630,
      "paper studies": 84439,
      "studies problem": 114264,
      "extractive summarization": 42564,
      "summarization problem": 115550,
      "problem short": 91227,
      "text inputs": 121058,
      "propose featureenriched": 92676,
      "results framework": 101796,
      "baselines substantial": 12473,
      "summarization dataset": 115491,
      "learning general": 62609,
      "distributed sentence": 32626,
      "representations large": 99716,
      "success natural": 115101,
      "words trained": 134275,
      "amounts text": 5361,
      "representations typically": 99950,
      "purpose features": 94428,
      "words range": 134153,
      "representations sequences": 99884,
      "sentences remains": 106473,
      "problem recent": 91197,
      "techniques different": 119871,
      "training objectives": 123761,
      "objectives learn": 81138,
      "learn general": 62052,
      "effective multitask": 34715,
      "diverse training": 32858,
      "multiple training": 75733,
      "million sentences": 70107,
      "sentences extensive": 106310,
      "sharing single": 108140,
      "sentence encoder": 105849,
      "tasks leads": 119247,
      "leads consistent": 61931,
      "improvements previous": 52898,
      "present substantial": 89727,
      "learning lowresource": 62697,
      "settings using": 107844,
      "using learned": 129804,
      "generalpurpose representations": 46891,
      "fast neural": 43147,
      "framework integrated": 45579,
      "automatic differentiation": 10515,
      "design encoderdecoder": 29535,
      "training translation": 123938,
      "automatic normalization": 10619,
      "word variations": 133626,
      "variations codemixed": 130933,
      "codemixed social": 18181,
      "text social": 121312,
      "platforms twitter": 87673,
      "facebook popular": 42741,
      "popular multilingual": 88105,
      "multilingual societies": 75371,
      "south asian": 110963,
      "asian languages": 9289,
      "languages codemixed": 60449,
      "tasks codemixed": 118984,
      "data consist": 25777,
      "errors spelling": 38408,
      "spelling variations": 111916,
      "variations paper": 130938,
      "paper leverage": 84033,
      "leverage contextual": 63580,
      "property words": 92509,
      "spelling variation": 111915,
      "variation words": 130903,
      "share similar": 107961,
      "similar context": 109080,
      "context large": 22159,
      "noisy social": 80120,
      "text capture": 120694,
      "capture different": 15289,
      "manner using": 66963,
      "experiments reveal": 41121,
      "codemixed dataset": 18172,
      "stateoftheart partofspeech": 112802,
      "tasks emotions": 119088,
      "learning sentiment": 63006,
      "sentiment based": 106688,
      "representations resourcepoor": 99857,
      "siamese networks": 108653,
      "networks machine": 77653,
      "abundance resources": 1831,
      "siamese network": 108651,
      "architecture sentiment": 8742,
      "languages jointly": 60656,
      "training resourcerich": 123815,
      "resourcerich languages": 100926,
      "model consists": 70895,
      "consists twin": 21500,
      "twin bidirectional": 126088,
      "networks bilstm": 77525,
      "rnn shared": 102932,
      "shared parameters": 107999,
      "parameters joined": 84757,
      "joined contrastive": 57250,
      "contrastive loss": 22733,
      "function based": 45999,
      "similarity metric": 109266,
      "metric model": 69889,
      "language common": 58892,
      "common sentiment": 18925,
      "sentiment space": 106795,
      "using similarity": 130173,
      "individual sentiments": 53931,
      "sentiments model": 106832,
      "model projects": 71802,
      "projects sentences": 92243,
      "similar sentiment": 109144,
      "closer sentences": 17877,
      "different sentiment": 31415,
      "farther experiments": 43111,
      "largescale datasets": 61424,
      "datasets resourcerich": 27679,
      "spanish resourcepoor": 111167,
      "hindi telugu": 50473,
      "telugu reveal": 120073,
      "analysis approaches": 5499,
      "rules lexicon": 103416,
      "lexicon lists": 63893,
      "lists deep": 64715,
      "network representations": 77409,
      "attentive sequencetosequence": 10128,
      "west african": 132455,
      "african language": 4274,
      "language writing": 60351,
      "electronic texts": 35261,
      "texts limited": 121542,
      "provide morphological": 93876,
      "information crucial": 54462,
      "lexical disambiguation": 63757,
      "asr natural": 9438,
      "task experiment": 118170,
      "experiment different": 40462,
      "models process": 73805,
      "evaluation dataset": 39170,
      "pretrained models": 90133,
      "models datasets": 73015,
      "opensource project": 82025,
      "title generation": 122222,
      "generation ecommerce": 47380,
      "better search": 13716,
      "ecommerce websites": 34510,
      "set slot": 107584,
      "pairs given": 83556,
      "huge manual": 50723,
      "manual creation": 66985,
      "approaches depend": 8121,
      "depend heavily": 29049,
      "availability large": 10915,
      "language research": 60051,
      "research apply": 100407,
      "lowresourced languages": 65575,
      "leveraging transfer": 63709,
      "learning train": 63112,
      "data creating": 25808,
      "languages performance": 60798,
      "generation evaluated": 47390,
      "evaluated different": 38969,
      "fewshot text": 43917,
      "embeddings human": 35742,
      "human loop": 50909,
      "literature text": 64777,
      "problem given": 91069,
      "corpus labeled": 23851,
      "labeled documents": 58455,
      "classifier accurately": 17517,
      "unseen documents": 127520,
      "corpora documents": 23461,
      "documents classified": 33201,
      "web content": 132226,
      "labels applied": 58578,
      "aims make": 4548,
      "entire corpus": 37702,
      "unlabeled documents": 127396,
      "humanintheloop approach": 51036,
      "approach content": 7459,
      "automatically classified": 10733,
      "fewshot learning": 43904,
      "simple case": 109381,
      "measuring distance": 67909,
      "uses pretrained": 129265,
      "embeddings documents": 35666,
      "simple weighted": 109545,
      "weighted average": 132337,
      "embeddings tested": 35978,
      "accuracy approach": 2101,
      "approach existing": 7547,
      "existing labeled": 40151,
      "labeled datasets": 58453,
      "provide results": 93910,
      "results code": 101583,
      "reproducing results": 100091,
      "20 newsgroups": 515,
      "resource indian": 100849,
      "resource contains": 100835,
      "newly developed": 79269,
      "dictionary resource": 30891,
      "annotated gold": 6199,
      "standard corpus": 112217,
      "corpus consisting": 23717,
      "8483 verbs": 1321,
      "253 adverbs": 828,
      "according defined": 1972,
      "guidelines paper": 49264,
      "annotation procedure": 6355,
      "procedure present": 91395,
      "additionally discuss": 3604,
      "potential lexical": 88570,
      "senseannotated corpora": 105708,
      "corpora improving": 23501,
      "wsd tasks": 135174,
      "tasks telugu": 119551,
      "crowdsourced annotation": 25104,
      "classification adjectives": 17105,
      "short note": 108226,
      "note describes": 80385,
      "typelogical grammars": 126241,
      "fast implementation": 43138,
      "proof search": 92373,
      "deductive parsing": 28188,
      "universal decompositional": 127280,
      "decompositional semantics": 28160,
      "dataset yielding": 27284,
      "dataset date": 26846,
      "report model": 99012,
      "results extended": 101785,
      "dataset scalable": 27177,
      "conventional seq2seq": 22899,
      "sequences considering": 107117,
      "propose models": 92789,
      "model reinforcement": 71883,
      "based conventional": 11604,
      "develop evaluation": 30196,
      "popularly used": 88160,
      "used metrics": 128633,
      "used analyze": 128386,
      "fast effective": 43135,
      "translation examples": 124802,
      "process specifically": 91571,
      "sentence use": 106118,
      "pairs source": 83641,
      "source sides": 110826,
      "similar input": 109100,
      "aligned words": 4930,
      "words match": 134048,
      "sentences translation": 106529,
      "retrieved sentence": 102463,
      "based similarities": 12043,
      "used translate": 128830,
      "translate input": 124527,
      "improves nmt": 53000,
      "results bleu": 101557,
      "narrow domain": 76189,
      "domain translation": 33683,
      "increase translation": 53621,
      "retrievalbased method": 102444,
      "respect accuracy": 101073,
      "accuracy speed": 2282,
      "simplicity implementation": 109570,
      "digital text": 31727,
      "analysis todays": 5900,
      "bad news": 11354,
      "techniques identifying": 119903,
      "end user": 36835,
      "user work": 129058,
      "tested using": 120585,
      "driven approaches": 34149,
      "used annotating": 128389,
      "representation support": 99429,
      "classification deep": 17175,
      "make reliable": 66718,
      "showed training": 108393,
      "training accuracy": 123343,
      "accuracy 96": 2088,
      "test accuracy": 120419,
      "internal external": 56164,
      "multilingual parallel": 75323,
      "translation significant": 125256,
      "significant results": 108861,
      "results great": 101818,
      "problem lack": 91097,
      "absence parallel": 1719,
      "corpus languages": 23855,
      "suggests method": 115375,
      "corpus language": 23852,
      "pairs extracted": 83544,
      "extracted open": 42169,
      "set video": 107634,
      "titles attributes": 122225,
      "attributes like": 10200,
      "pairs desired": 83514,
      "desired language": 29666,
      "main problem": 66452,
      "ability make": 1540,
      "make context": 66638,
      "context based": 22019,
      "based parallel": 11922,
      "corpus filtering": 23801,
      "formal informal": 45162,
      "styles including": 114618,
      "advantage method": 4050,
      "make corpus": 66640,
      "corpus informal": 23840,
      "informal style": 54342,
      "generated corpus": 47065,
      "used realtime": 128715,
      "short time": 108266,
      "quality extremely": 94666,
      "equally important": 38230,
      "time possible": 122071,
      "costs paper": 24403,
      "floating point": 44675,
      "frustratingly easy": 45908,
      "creating accurate": 24693,
      "pretrained source": 90189,
      "source embeddings": 110752,
      "embeddings received": 35899,
      "attention lately": 9865,
      "shown produce": 108511,
      "produce accurate": 91872,
      "metaembeddings paper": 68536,
      "distinct word": 32549,
      "comparable better": 19129,
      "methods result": 69732,
      "spaces different": 111093,
      "different source": 31438,
      "grammar knowledge": 48643,
      "applied knowledge": 7080,
      "rely manual": 98719,
      "available applications": 10945,
      "enabling knowledge": 36410,
      "english corpus": 37098,
      "corpus additionally": 23649,
      "domainindependent information": 33714,
      "form summary": 45131,
      "lists word": 64722,
      "containing pairs": 21804,
      "pairs terms": 83645,
      "given semantic": 48113,
      "relation paper": 98061,
      "describes process": 29426,
      "special focus": 111362,
      "created far": 24668,
      "based random": 11968,
      "extracted word": 42196,
      "pairs results": 83630,
      "results necessary": 101981,
      "false negatives": 43041,
      "new patterns": 79072,
      "patterns context": 85734,
      "humor understanding": 51142,
      "contextual knowledge": 22476,
      "knowledge important": 58005,
      "discourse knowledge": 32064,
      "knowledge knowledge": 58029,
      "situational context": 109893,
      "discuss importance": 32258,
      "importance contextual": 52050,
      "knowledge understanding": 58219,
      "understanding humor": 126859,
      "process analyze": 91419,
      "effective tool": 34761,
      "knowledge includes": 58011,
      "knowledge english": 57897,
      "convey message": 23117,
      "way present": 132118,
      "linguistic cues": 64457,
      "context understanding": 22309,
      "problem community": 90967,
      "number positive": 80940,
      "proxy task": 94157,
      "learning user": 63149,
      "user embeddings": 128977,
      "embeddings user": 36007,
      "resulting embeddings": 101443,
      "identification common": 51371,
      "recognition using": 97038,
      "representations recent": 99840,
      "shown effectiveness": 108465,
      "features significantly": 43728,
      "significantly improving": 108962,
      "supervised ner": 115813,
      "study investigate": 114413,
      "representations additional": 99494,
      "systematically compare": 116692,
      "compare popular": 19282,
      "skipgram cbow": 109981,
      "cbow glove": 15847,
      "integrating word": 55806,
      "brown clustering": 14719,
      "concerning word": 20691,
      "second best": 104395,
      "brown clusters": 14720,
      "provides additional": 93996,
      "additional improvement": 3523,
      "nearly 10": 76732,
      "f1score baseline": 42714,
      "processing article": 91625,
      "model determining": 70984,
      "different syntactic": 31469,
      "underlying comprehension": 126677,
      "phrases model": 87440,
      "eeg signals": 34582,
      "signals recorded": 108704,
      "recorded reading": 97132,
      "distinctive linguistic": 32561,
      "linguistic processes": 64528,
      "selftraining using": 104965,
      "neural reinforcement": 78645,
      "learning selftraining": 62990,
      "learning leveraging": 62690,
      "leveraging raw": 63699,
      "texts enhancing": 121506,
      "enhancing model": 37537,
      "model performances": 71732,
      "methods depend": 69432,
      "heuristics model": 49886,
      "model confidence": 70882,
      "instance selection": 55610,
      "address challenges": 3661,
      "challenges propose": 16199,
      "selftraining strategy": 104962,
      "automatically based": 10726,
      "automatically learns": 10800,
      "learns optimal": 63224,
      "task extended": 118189,
      "results semeval2018": 102148,
      "relations scientific": 98250,
      "subtask 11": 114939,
      "subtask 12": 114940,
      "workshop paper": 135014,
      "al 2018": 4625,
      "technical details": 119744,
      "changes preprocessing": 16390,
      "postevaluation phase": 88494,
      "relations using": 98277,
      "using embeddings": 129622,
      "improved f1": 52604,
      "loan words": 64898,
      "frequently occurring": 45873,
      "posts analyzed": 88513,
      "dataset million": 27026,
      "posts written": 88528,
      "russianspeaking facebook": 103513,
      "words missing": 134055,
      "assumption words": 9666,
      "use recently": 128228,
      "true words": 125855,
      "words words": 134321,
      "texts russian": 121597,
      "russian national": 103495,
      "corpus result": 23975,
      "words composed": 133870,
      "identify various": 51575,
      "words large": 134027,
      "largest number": 61534,
      "serve starting": 107300,
      "learning joint": 62662,
      "joint semantic": 57320,
      "parsers multiple": 85033,
      "datasets target": 27745,
      "semantic formalisms": 105057,
      "drastically different": 34101,
      "different underlying": 31541,
      "latent structured": 61616,
      "baselines improvements": 12412,
      "parsing research": 85219,
      "focuses improving": 44903,
      "ambiguity inherent": 5283,
      "ambiguity important": 5282,
      "important error": 52156,
      "joint distribution": 57267,
      "define new": 28487,
      "dependency path": 29211,
      "analysis performance": 5747,
      "task second": 118668,
      "second demonstrate": 104400,
      "method parser": 69039,
      "use method": 128143,
      "extraction applications": 42263,
      "persons killed": 87154,
      "killed police": 57669,
      "framework utilize": 45736,
      "utilize large": 130513,
      "data addition": 25574,
      "predicted labels": 88961,
      "data select": 26381,
      "samples based": 103581,
      "augment training": 10268,
      "unlabeled labeled": 127399,
      "explore data": 41534,
      "highquality unlabeled": 50415,
      "unlabeled samples": 127407,
      "samples better": 103582,
      "specifically approach": 111527,
      "learn data": 62015,
      "labeled dataset": 58452,
      "clickbait detection": 17749,
      "generic text": 47811,
      "method obtain": 68994,
      "accurate text": 2368,
      "classification results": 17384,
      "description provided": 29473,
      "key components": 57558,
      "learning objectives": 62857,
      "research introduces": 100539,
      "similarity algorithms": 109192,
      "calculate similarity": 15017,
      "domain present": 33613,
      "algorithm integrating": 4756,
      "integrating domain": 55791,
      "domain corpus": 33492,
      "algorithm introduce": 4757,
      "representations facts": 99653,
      "associated texts": 9611,
      "datasets data": 27390,
      "facts related": 42922,
      "tackle challenging": 117288,
      "task introducing": 118309,
      "selection mechanism": 104798,
      "mechanism use": 68044,
      "multiinstance learning": 75145,
      "learning automatically": 62378,
      "automatically discover": 10760,
      "text pairs": 121161,
      "pairs used": 83660,
      "used enhance": 128517,
      "signal training": 108693,
      "objectives improve": 81137,
      "solely relies": 110533,
      "attention detecting": 9816,
      "interpreting neural": 56288,
      "medical treatment": 68230,
      "language changes": 58872,
      "patients cognitive": 85705,
      "potentially leading": 88617,
      "language samples": 60066,
      "new independent": 78954,
      "benchmark accuracy": 12728,
      "task importantly": 118277,
      "interpret neural": 56212,
      "learned linguistic": 62229,
      "techniques perform": 119948,
      "perform novel": 86038,
      "light limitations": 63993,
      "data endtoend": 25892,
      "tag parsing": 117338,
      "parsing neural": 85175,
      "highway connections": 50440,
      "parser jointly": 84966,
      "jointly performs": 57380,
      "global inference": 48241,
      "rich feature": 102743,
      "fundamental tradeoff": 46129,
      "systems demonstrate": 116837,
      "parser achieves": 84912,
      "parsing evaluation": 85109,
      "support claim": 115959,
      "require rich": 100195,
      "rich structural": 102791,
      "analysis sentences": 5836,
      "language proficiency": 59991,
      "levels description": 63530,
      "classification systems": 17424,
      "languages follow": 60593,
      "explore universal": 41603,
      "results preliminary": 102047,
      "experiments monolingual": 41022,
      "crosslingual multilingual": 24981,
      "languages german": 60609,
      "german czech": 47883,
      "italian results": 57101,
      "monolingual multilingual": 74598,
      "performance crosslingual": 86265,
      "monolingual classification": 74556,
      "ntuaslp semeval2018": 80819,
      "present deeplearning": 89435,
      "competed semeval2018": 19607,
      "task irony": 118315,
      "irony detection": 56938,
      "detection english": 29934,
      "english tweets": 37321,
      "tweets design": 126028,
      "independent models": 53775,
      "level order": 63483,
      "models augmented": 72786,
      "augmented selfattention": 10335,
      "mechanism order": 68023,
      "identify informative": 51510,
      "informative words": 55161,
      "words embedding": 133930,
      "embeddings pretrained": 35876,
      "collection 550": 18455,
      "550 million": 1116,
      "million english": 70097,
      "features lexicons": 43593,
      "external datasets": 41999,
      "information models": 54772,
      "data furthermore": 25964,
      "furthermore provide": 46210,
      "tokens attention": 122298,
      "attention layer": 9866,
      "interpret inner": 56209,
      "inner workings": 55277,
      "models ranked": 73852,
      "ranked 2nd": 95636,
      "results ranking": 102110,
      "domain dialogue": 33504,
      "competition amazon": 19621,
      "hybrid combining": 51177,
      "combining machine": 18732,
      "learning rule": 62980,
      "based approaches": 11516,
      "approaches discuss": 8130,
      "nlg knowledge": 79520,
      "sentences gapping": 106327,
      "relation arguments": 97941,
      "syntax representations": 116554,
      "sentences produced": 106455,
      "extraction typically": 42523,
      "typically designed": 126421,
      "information sentences": 54967,
      "structure paper": 113927,
      "dependencies graph": 29086,
      "graph representation": 48863,
      "additional nodes": 3553,
      "nodes edges": 80035,
      "gap demonstrate": 46449,
      "study swedish": 114529,
      "learning extract": 62575,
      "coherent summary": 18327,
      "coherence plays": 18306,
      "producing highquality": 92020,
      "document recent": 33060,
      "neural extractive": 77913,
      "summarization increasingly": 115520,
      "increasingly attractive": 53692,
      "extracting sentences": 42239,
      "coherent summaries": 18326,
      "summaries propose": 115453,
      "neural coherence": 77863,
      "syntactic coherence": 116380,
      "obviates need": 81485,
      "fashion using": 43124,
      "using unlabeled": 130340,
      "data empirical": 25886,
      "efficiently capture": 35139,
      "crosssentence coherence": 25069,
      "patterns using": 85793,
      "using combined": 129452,
      "output neural": 83101,
      "reward design": 102671,
      "train proposed": 122998,
      "extractive summarizer": 42569,
      "simultaneously experimental": 109675,
      "existing baselines": 40077,
      "performance term": 86791,
      "term rouge": 120242,
      "cnndaily mail": 18033,
      "mail dataset": 66390,
      "qualitative evaluation": 94561,
      "evaluation indicates": 39244,
      "summaries produced": 115452,
      "parsing multiple": 85165,
      "multiple sentences": 75688,
      "sentences improved": 106349,
      "approaches recognizing": 8312,
      "used parse": 128680,
      "parse input": 84883,
      "premises hypotheses": 89290,
      "logical formulas": 65010,
      "structure results": 113946,
      "results inconsistent": 101850,
      "predicate argument": 88847,
      "method extend": 68830,
      "extend existing": 41783,
      "existing logicbased": 40164,
      "systems method": 116999,
      "disentangled representations": 32350,
      "representations texts": 99932,
      "complementary aspects": 19739,
      "efficient model": 35093,
      "embeddings propose": 35886,
      "describing clinical": 29448,
      "clinical trials": 17795,
      "interventions outcomes": 56326,
      "learns representations": 63232,
      "representations encode": 99625,
      "used perform": 128685,
      "application experiments": 6851,
      "review corpora": 102538,
      "automatic stance": 10673,
      "stance detection": 112168,
      "agrees disagrees": 4413,
      "respect given": 101079,
      "evidence prediction": 39660,
      "paragraph level": 84568,
      "similarity matrix": 109254,
      "fake news": 43012,
      "news challenge": 79314,
      "challenge dataset": 16016,
      "identify extract": 51500,
      "knowledge crucial": 57846,
      "task recognizing": 118614,
      "method detecting": 68770,
      "solution relies": 110564,
      "representations experiments": 99648,
      "various paraphrases": 131161,
      "paraphrase databases": 84810,
      "paraphrases using": 84843,
      "rte tasks": 103342,
      "representations sentencelevel": 99879,
      "sentencelevel semantic": 106171,
      "using conversational": 129494,
      "conversational data": 22999,
      "method trains": 69196,
      "unsupervised model": 127670,
      "resulting sentence": 101470,
      "perform semantic": 86059,
      "benchmark semeval": 12855,
      "answering cqa": 6614,
      "question similarity": 95220,
      "combining conversational": 18716,
      "task extensive": 118191,
      "mixed systems": 70402,
      "systems tasks": 117189,
      "pathologies neural": 85684,
      "make interpretations": 66690,
      "way interpret": 132100,
      "model predictions": 71761,
      "highlight important": 50262,
      "words input": 134004,
      "interpretation methods": 56266,
      "understand limitations": 126756,
      "limitations methods": 64179,
      "use input": 128097,
      "pathological behaviors": 85682,
      "remaining words": 98785,
      "human experiments": 50854,
      "examples lack": 39849,
      "information support": 55021,
      "counterintuitive results": 24431,
      "draw connections": 34113,
      "confidence calibration": 20984,
      "trained maximum": 123197,
      "finetune models": 44410,
      "finetuned models": 44434,
      "models interpretable": 73422,
      "reduction accuracy": 97438,
      "accuracy loss": 2199,
      "contextualized representation": 22564,
      "tasks pretrained": 119394,
      "models lms": 73518,
      "brought significant": 14716,
      "fully leverage": 45965,
      "corpora capture": 23430,
      "specific task": 111496,
      "task parts": 118524,
      "heavy computation": 49665,
      "applications propose": 6993,
      "preserving useful": 89942,
      "information regard": 54909,
      "layers model": 61789,
      "different information": 31179,
      "method model": 68967,
      "dense connectivity": 29012,
      "training lms": 123689,
      "better robustness": 13712,
      "robustness experiments": 103098,
      "method direct": 68777,
      "transfer transfer": 124199,
      "embeddings semantic": 35925,
      "produce sentence": 91928,
      "typically evaluated": 126427,
      "transfer downstream": 124048,
      "work dedicated": 134458,
      "attention paper": 9963,
      "propose transfer": 93117,
      "learning setting": 63017,
      "transfer experiments": 124052,
      "standard text": 112323,
      "text similarity": 121296,
      "similarity datasets": 109217,
      "applying direct": 7237,
      "transfer existing": 124051,
      "encoders lead": 36655,
      "lead stateoftheart": 61869,
      "performance additionally": 86126,
      "additionally compare": 3599,
      "approaches transfer": 8380,
      "transfer sentence": 124185,
      "encoders semantic": 36662,
      "tasks showing": 119498,
      "greatly affects": 49044,
      "dataset unsupervised": 27253,
      "unsupervised natural": 127677,
      "generation denoising": 47360,
      "denoising autoencoders": 28998,
      "text structured": 121332,
      "tasks question": 119426,
      "systems domain": 116858,
      "supervision based": 115870,
      "text able": 120623,
      "able build": 1615,
      "build natural": 14789,
      "output use": 83144,
      "use denoising": 127991,
      "denoising autoencoder": 28995,
      "autoencoder reconstruct": 10407,
      "reconstruct sentence": 97110,
      "sentence introduce": 105909,
      "introduce noise": 56505,
      "noise training": 80070,
      "data resulting": 26357,
      "generate correct": 46919,
      "correct sentences": 24119,
      "given structured": 48139,
      "data dynamic": 25873,
      "systems selecting": 117129,
      "end introduce": 36811,
      "introduce dynamic": 56416,
      "learning embedding": 62534,
      "leads stateoftheart": 61964,
      "model class": 70826,
      "tasks subsequently": 119531,
      "shed new": 108156,
      "usage word": 127872,
      "embeddings nlp": 35833,
      "fact checking": 42816,
      "corpus reasonable": 23960,
      "potentially relevant": 88618,
      "sources news": 110913,
      "websites social": 132305,
      "making prediction": 66858,
      "able explain": 1636,
      "explain decision": 41255,
      "rationale extraction": 95840,
      "annotations corpus": 6418,
      "corpus implement": 23829,
      "parsing introduce": 85134,
      "task crosslingual": 118036,
      "parsing mapping": 85155,
      "language meaning": 59277,
      "based target": 12108,
      "designed allow": 29578,
      "systems target": 117185,
      "varying levels": 131263,
      "structural complexity": 113760,
      "shallow deep": 107916,
      "metric measure": 69886,
      "output reference": 83114,
      "representations endtoend": 99629,
      "endtoend model": 36928,
      "coreference evaluation": 23346,
      "modelling present": 72616,
      "novel effective": 80553,
      "effective technique": 34758,
      "deeper insights": 28444,
      "performance modern": 86542,
      "approaches nlp": 8258,
      "provide users": 93951,
      "final network": 44113,
      "network decision": 77217,
      "additional understanding": 3593,
      "work new": 134653,
      "type sentence": 126225,
      "embedding learned": 35426,
      "applied effectively": 7069,
      "deeper understanding": 28453,
      "data obtained": 26187,
      "embeddings train": 35982,
      "individual sentences": 53930,
      "predict location": 88899,
      "document form": 33010,
      "form distribution": 45082,
      "demonstrate embeddings": 28734,
      "embeddings combined": 35599,
      "combined simple": 18668,
      "heuristics used": 49887,
      "used achieve": 128376,
      "achieve performance": 2509,
      "performance competitive": 86244,
      "stateoftheart multiple": 112762,
      "multiple text": 75727,
      "tasks outperforming": 119356,
      "outperforming complex": 82805,
      "approaches additionally": 8049,
      "additionally demonstrate": 3600,
      "provide insights": 93859,
      "approach largescale": 7667,
      "domain classification": 33479,
      "intelligent personal": 55852,
      "personal digital": 87116,
      "digital assistants": 31716,
      "assistants ipdas": 9580,
      "application spoken": 6885,
      "understanding capabilities": 126807,
      "cover potentially": 24488,
      "thousands overlapping": 121926,
      "overlapping domains": 83334,
      "understanding task": 126982,
      "task finding": 118213,
      "finding best": 44266,
      "scale paper": 103745,
      "propose set": 93047,
      "efficient scalable": 35109,
      "models largescale": 73466,
      "stage performs": 112152,
      "additional contextual": 3501,
      "domains word": 33884,
      "classification technique": 17464,
      "technique report": 119812,
      "language applying": 58837,
      "augmentation methods": 10286,
      "methods specifically": 69768,
      "types noise": 126331,
      "noise adversarial": 80042,
      "adversarial noise": 4157,
      "apply constraints": 7168,
      "proposed data": 93247,
      "gain improvements": 46343,
      "words help": 133979,
      "paper specifically": 84436,
      "case neural": 15602,
      "generative story": 47759,
      "combines standard": 18698,
      "tokens sentence": 122332,
      "letters word": 63408,
      "word type": 133611,
      "structure word": 113984,
      "rnn generate": 102917,
      "embeddings naturally": 35825,
      "spelling token": 111912,
      "context comparing": 22032,
      "comparing baselines": 19501,
      "baselines including": 12414,
      "novel strong": 80738,
      "work establish": 134492,
      "establish stateoftheart": 38593,
      "results multiple": 101967,
      "incorporating knowledge": 53543,
      "endtoend taskoriented": 36976,
      "usually suffer": 130456,
      "suffer challenge": 115210,
      "challenge incorporating": 16042,
      "bases paper": 12505,
      "simple endtoend": 109425,
      "endtoend differentiable": 36892,
      "differentiable model": 31584,
      "neural generative": 77920,
      "multihop attention": 75136,
      "generation step": 47635,
      "mechanism helps": 67992,
      "addition model": 3451,
      "attain stateoftheart": 9721,
      "dialog datasets": 30560,
      "tweets universal": 126066,
      "dependencies using": 29123,
      "times larger": 122181,
      "kong et": 58342,
      "disagreements annotators": 31937,
      "consistent annotation": 21372,
      "computational efficiency": 20376,
      "efficiency propose": 35039,
      "method distill": 68781,
      "transitionbased parsers": 124513,
      "baseline outperforms": 12280,
      "parsers stateoftheart": 85047,
      "transfer semantic": 124184,
      "argument comprehension": 8937,
      "argument reasoning": 8950,
      "reasoning comprehension": 96237,
      "comprehension task": 20235,
      "understanding complex": 126815,
      "complex reasoning": 19867,
      "world knowledge": 135033,
      "knowledge focus": 57938,
      "complicated models": 19957,
      "models given": 73300,
      "small size": 110206,
      "dataset best": 26766,
      "bilstm encode": 14093,
      "encode input": 36434,
      "matching model": 67412,
      "achieves mean": 2817,
      "set accuracy": 107345,
      "significant gain": 108756,
      "gain best": 46335,
      "dataset provides": 27130,
      "position report": 88290,
      "report experiment": 98995,
      "model matches": 71506,
      "relatively low": 98404,
      "performance degradation": 86279,
      "model necessarily": 71562,
      "task exploiting": 118184,
      "translation graph": 124846,
      "networks semantic": 77755,
      "representations long": 99743,
      "potentially useful": 88621,
      "meaning preservation": 67664,
      "improving generalization": 53098,
      "methods work": 69856,
      "work incorporate": 134567,
      "structure source": 113959,
      "representations neural": 99772,
      "translation use": 125412,
      "use graph": 128075,
      "achieve improvements": 2492,
      "pair using": 83464,
      "using aspect": 129350,
      "extraction approaches": 42265,
      "user profiles": 129024,
      "reviews products": 102624,
      "rich information": 102747,
      "information users": 55079,
      "survey reviews": 116192,
      "review snippets": 102580,
      "certain aspect": 15930,
      "aspect identification": 9333,
      "extraction corpora": 42292,
      "corpora work": 23624,
      "sentences various": 106541,
      "judged human": 57431,
      "human evaluators": 50852,
      "baseline does": 12212,
      "second experiment": 104408,
      "distributions represent": 32768,
      "users reviews": 129166,
      "reviews written": 102641,
      "capture notions": 15351,
      "user preferences": 129020,
      "users generate": 129122,
      "different review": 31390,
      "detecting syntactic": 29853,
      "chinese present": 16805,
      "texts translated": 121639,
      "originally written": 82573,
      "using support": 130245,
      "machines svms": 66342,
      "corpus translation": 24048,
      "studies chinese": 114192,
      "constituent parse": 21542,
      "trees dependency": 125687,
      "dependency triples": 29249,
      "features lexical": 43592,
      "close results": 17831,
      "learning topic": 63110,
      "accurately distinguish": 2380,
      "distinguish translated": 32576,
      "translated original": 124554,
      "increased use": 53634,
      "nps vps": 80804,
      "bleu scheme": 14322,
      "scheme used": 103940,
      "annual conference": 6505,
      "conference machine": 20976,
      "translation wmt": 125426,
      "reference processing": 97527,
      "processing provide": 91778,
      "new tool": 79222,
      "distributional methods": 32710,
      "methods lexical": 69595,
      "applied successfully": 7122,
      "work questioned": 134760,
      "actually learn": 3153,
      "learn relation": 62121,
      "words specifically": 134234,
      "linear classifiers": 64334,
      "classifiers learn": 17616,
      "learn separate": 62147,
      "performance methods": 86529,
      "methods integrating": 69558,
      "evaluation different": 39184,
      "classifiers evaluation": 17610,
      "suitable evaluation": 115398,
      "evaluation setup": 39389,
      "setup task": 107858,
      "biases existing": 13872,
      "previous ones": 90441,
      "scheduled multitask": 103884,
      "learning syntax": 63081,
      "syntax translation": 116565,
      "results learning": 101888,
      "manner propose": 66957,
      "model begins": 70754,
      "focus translation": 44835,
      "achieve considerable": 2450,
      "large parallel": 61199,
      "wmt14 english": 132809,
      "inference propose": 54203,
      "propose process": 93002,
      "extent sentence": 41985,
      "use representations": 128246,
      "features train": 43762,
      "train natural": 122974,
      "classifier based": 17524,
      "existing semantic": 40276,
      "nmt encoder": 79857,
      "syntaxsemantics interface": 116590,
      "deficiencies existing": 28477,
      "process improved": 91507,
      "framework evaluating": 45529,
      "context global": 22118,
      "extraction extracting": 42337,
      "extracting entities": 42209,
      "text important": 121043,
      "task understanding": 118818,
      "corpora open": 23549,
      "relation tuples": 98078,
      "arguments predicate": 8991,
      "relation sentences": 98071,
      "sentences relation": 106471,
      "current open": 25319,
      "open systems": 81941,
      "focus modeling": 44789,
      "modeling local": 72464,
      "extract relation": 42104,
      "global statistics": 48272,
      "statistics large": 113202,
      "global structural": 48273,
      "facts external": 42912,
      "supervision new": 115902,
      "applied different": 7059,
      "extractions using": 42544,
      "using corpuslevel": 129508,
      "joint optimization": 57304,
      "measuring quality": 67927,
      "tuples extracted": 125949,
      "objective learning": 81091,
      "learning subtasks": 63067,
      "subtasks jointly": 114970,
      "produced subtask": 91971,
      "experiments realworld": 41106,
      "stateoftheart open": 112795,
      "common crawl": 18863,
      "mining parallel": 70254,
      "texts web": 121649,
      "web sites": 132258,
      "structure languages": 113891,
      "assumption propose": 9664,
      "based combination": 11583,
      "allows efficiently": 5149,
      "efficiently identify": 35147,
      "identify pairs": 51530,
      "parallel segments": 84676,
      "given web": 48172,
      "realworld data": 96156,
      "webcrawled data": 132279,
      "data improving": 26029,
      "improving coverage": 53078,
      "runtime complexity": 103479,
      "exact inference": 39717,
      "nonprojective transitionbased": 80281,
      "better coverage": 13555,
      "reduces time": 97400,
      "parsing hope": 85125,
      "systems better": 116759,
      "disambiguation underresourced": 31991,
      "word respect": 133482,
      "sense target": 105701,
      "model estimate": 71099,
      "similar word": 109171,
      "corresponding context": 24279,
      "problem architecture": 90939,
      "conduct evaluation": 20840,
      "outperforms sparse": 82988,
      "datasets according": 27291,
      "method sampling": 69114,
      "input string": 55446,
      "model contrast": 70910,
      "lstm demonstrate": 65615,
      "quality sample": 94782,
      "large state": 61279,
      "state spaces": 112516,
      "task 10": 117810,
      "vector pattern": 131336,
      "attributes paper": 10201,
      "attributes word": 10204,
      "embedding based": 35379,
      "semeval 2018": 105502,
      "2018 task": 688,
      "achieving f1": 2944,
      "identification dataset": 51377,
      "identification nli": 51410,
      "identifying authors": 51581,
      "language based": 58857,
      "based second": 12005,
      "student essays": 114138,
      "essays written": 38535,
      "european portuguese": 38771,
      "finegrained pos": 44372,
      "constituency parses": 21527,
      "nlp discuss": 79605,
      "discuss possible": 32276,
      "possible applications": 88379,
      "dataset present": 27101,
      "translation traditional": 125385,
      "traditional neural": 122851,
      "sentences learned": 106376,
      "efficient method": 35091,
      "method dynamically": 68787,
      "sample sentences": 103563,
      "sentences order": 106419,
      "order accelerate": 82264,
      "training approach": 123361,
      "weight assigned": 132322,
      "training costs": 123411,
      "nist chinesetoenglish": 79499,
      "chinesetoenglish wmt": 16875,
      "wmt englishtogerman": 132791,
      "training improve": 123654,
      "improve nmt": 52444,
      "nmt performance": 79944,
      "performance multitask": 86553,
      "allowing learning": 5113,
      "learning task": 63085,
      "task benefit": 117929,
      "learned paper": 62239,
      "task improving": 118282,
      "test case": 120425,
      "amr sdp": 5394,
      "ud parsing": 126531,
      "auxiliary tasks": 10892,
      "outofdomain settings": 82661,
      "syntaxbased nmt": 116582,
      "nmt explore": 79865,
      "explore strategies": 41591,
      "strategies incorporating": 113469,
      "target syntax": 117719,
      "syntax neural": 116545,
      "focus syntax": 44828,
      "containing multiple": 21803,
      "multiple sentence": 75686,
      "ensembles using": 37628,
      "especially effective": 38454,
      "effective long": 34699,
      "long representations": 65084,
      "representations like": 99739,
      "gives stateoftheart": 48208,
      "performance difficult": 86304,
      "crucial understanding": 25177,
      "analyses interviews": 5456,
      "problem investigate": 91092,
      "investigate problem": 56800,
      "suggesting significant": 115353,
      "work attempt": 134383,
      "contrast use": 22716,
      "methods particular": 69667,
      "particular word": 85465,
      "try identify": 125896,
      "identify differences": 51490,
      "sentiment expressed": 106740,
      "helps improve": 49817,
      "improve robustness": 52528,
      "existence problem": 40038,
      "conversational language": 23011,
      "generation rich": 47606,
      "rich content": 102732,
      "tourist information": 122708,
      "information typically": 55064,
      "attributes individual": 10199,
      "content generate": 21883,
      "hotel information": 50669,
      "information general": 54633,
      "domain rich": 33630,
      "rich complex": 102730,
      "complex content": 19804,
      "collecting data": 18445,
      "data inform": 26039,
      "features characterize": 43394,
      "data plan": 26239,
      "models generation": 73293,
      "utterances use": 130672,
      "automatic coding": 10502,
      "based associated": 11527,
      "number techniques": 80979,
      "techniques text": 119997,
      "text preprocessing": 121191,
      "compare effectiveness": 19242,
      "effectiveness different": 34883,
      "different classification": 31039,
      "ensemble decision": 37590,
      "stateoftheart technique": 113002,
      "single vector": 109811,
      "probing sentence": 90906,
      "training highquality": 123650,
      "poor understanding": 88063,
      "quality sentence": 94788,
      "tasks makes": 119287,
      "difficult infer": 31635,
      "probing tasks": 90911,
      "tasks designed": 119051,
      "capture simple": 15398,
      "simple linguistic": 109459,
      "features sentences": 43719,
      "generated different": 47073,
      "different encoders": 31126,
      "trained distinct": 123121,
      "transformation networks": 124269,
      "targetoriented sentiment": 117787,
      "classification aims": 17108,
      "aims classifying": 4512,
      "classifying sentiment": 17678,
      "sentiment polarities": 106769,
      "individual opinion": 53925,
      "opinion targets": 82105,
      "rnn attention": 102903,
      "attention good": 9848,
      "good fit": 48477,
      "characteristics task": 16538,
      "cnn perform": 18016,
      "perform classification": 85959,
      "model overcome": 71686,
      "overcome issues": 83288,
      "cnn layer": 18006,
      "rnn layer": 102921,
      "layers propose": 61801,
      "component generate": 19980,
      "preserving original": 89936,
      "layer experiments": 61717,
      "achieves new": 2818,
      "performance benchmarks": 86173,
      "benchmarks fast": 12903,
      "intelligent agents": 55843,
      "accurate models": 2344,
      "new natural": 79045,
      "language domains": 58985,
      "time data": 122009,
      "process propose": 91554,
      "reuse available": 102487,
      "popular commercial": 88087,
      "significantly increase": 108967,
      "accuracy low": 2200,
      "development accurate": 30366,
      "models data": 73009,
      "novel approaches": 80494,
      "experimental setting": 40728,
      "really useful": 96133,
      "advances state": 4020,
      "translation common": 124697,
      "common translation": 18941,
      "widely accepted": 132534,
      "mt approach": 74955,
      "set outperforms": 107520,
      "results nist": 101990,
      "nist openmt": 79506,
      "papers published": 84510,
      "linguistics artificial": 64604,
      "intelligence past": 55836,
      "11 years": 230,
      "standardized benchmark": 112346,
      "benchmark data": 12749,
      "data performance": 26232,
      "performance important": 86439,
      "team ranked": 119718,
      "datasets containing": 27381,
      "method employed": 68799,
      "representing contexts": 100044,
      "using offtheshelf": 129987,
      "offtheshelf pretrained": 81629,
      "pretrained distributional": 90024,
      "models vector": 74327,
      "ambiguous word": 5298,
      "result word": 101420,
      "evaluation downstream": 39194,
      "neural entity": 77911,
      "tasks conduct": 119000,
      "propose improvements": 92720,
      "extend model": 41787,
      "representation feature": 99238,
      "computation model": 20351,
      "hierarchical structured": 50007,
      "political parties": 87997,
      "particular issue": 85425,
      "spectrum paper": 111633,
      "propose twostage": 93125,
      "twostage model": 126162,
      "employ hierarchical": 36265,
      "hierarchical multitask": 49981,
      "deep model": 28338,
      "fine coarsegrained": 44321,
      "second step": 104457,
      "step perform": 113292,
      "perform posthoc": 86045,
      "positions using": 88313,
      "using probabilistic": 130058,
      "probabilistic soft": 90828,
      "soft logic": 110496,
      "outperforms stateofart": 82994,
      "languages multilingual": 60742,
      "multilingual dialogue": 75242,
      "evaluation adversarial": 39116,
      "learning automatic": 62377,
      "performance opendomain": 86587,
      "automatic dialogue": 10514,
      "evaluation existing": 39208,
      "methods mainly": 69608,
      "focus monolingual": 44791,
      "flexible transfer": 44665,
      "transfer different": 124045,
      "languages address": 60392,
      "issue propose": 57022,
      "multitask neural": 75883,
      "neural metric": 78142,
      "evaluation shared": 39394,
      "extraction languages": 42367,
      "languages evaluate": 60556,
      "experiments adversarial": 40762,
      "metric achieves": 69868,
      "high correlation": 50050,
      "human annotation": 50743,
      "performance monolingual": 86543,
      "monolingual ones": 74602,
      "various existing": 131092,
      "existing metrics": 40200,
      "metrics automatic": 69925,
      "automatic article": 10495,
      "article commenting": 9116,
      "dataset comments": 26801,
      "online articles": 81743,
      "improve user": 52581,
      "chatbots paper": 16680,
      "introduces largescale": 56617,
      "largescale chinese": 61403,
      "chinese dataset": 16754,
      "real comments": 96057,
      "human bias": 50769,
      "develop automatic": 30176,
      "broad set": 14682,
      "set popular": 107532,
      "referencebased metrics": 97548,
      "greatly improved": 49055,
      "discourseaware neural": 32106,
      "rewards reinforcement": 102685,
      "guide model": 49242,
      "generate long": 46967,
      "long coherent": 65063,
      "propose learn": 92743,
      "learn neural": 62099,
      "model crosssentence": 70944,
      "generator trained": 47783,
      "learned reward": 62256,
      "text models": 121128,
      "trained crossentropy": 123103,
      "using hierarchical": 129737,
      "rnns consider": 102964,
      "mechanism improves": 67993,
      "art task": 9096,
      "hierarchical bilstm": 49939,
      "embedding processes": 35473,
      "embeddings classify": 35595,
      "classify sentence": 17659,
      "faster train": 43193,
      "bilstm outperforms": 14102,
      "outperforms flat": 82906,
      "considers surrounding": 21345,
      "model broader": 70789,
      "broader discourse": 14701,
      "academic paper": 1859,
      "hierarchical convolutional": 49945,
      "academic papers": 1860,
      "paper order": 84060,
      "automatically determine": 10755,
      "build new": 14792,
      "network achieve": 77122,
      "baselines large": 12420,
      "dataset code": 26787,
      "automatic estimation": 10524,
      "spoken word": 112022,
      "number potential": 80942,
      "potential applications": 88536,
      "tools propose": 122467,
      "performance building": 86193,
      "qe machine": 94530,
      "experiments settings": 41129,
      "metric propose": 69894,
      "measures improve": 67871,
      "knowledgebased wsd": 58261,
      "performing tasks": 86974,
      "multilingual intent": 75261,
      "understand human": 126749,
      "data indomain": 26035,
      "indomain knowledge": 53969,
      "knowledge expanding": 57911,
      "user study": 129044,
      "study evaluate": 114370,
      "technology rapidly": 120045,
      "multilingual spoken": 75376,
      "systems enable": 116865,
      "language utterances": 60319,
      "translation human": 124852,
      "high potential": 50103,
      "potential human": 88559,
      "utterances high": 130644,
      "explore sentiment": 41586,
      "like happy": 64044,
      "patterns present": 85772,
      "supervised techniques": 115843,
      "techniques sentiment": 119982,
      "incorporates information": 53506,
      "information phrases": 54842,
      "speech sentiment": 111800,
      "association scores": 9628,
      "obtains accuracy": 81453,
      "far away": 43072,
      "know little": 57733,
      "use prior": 128213,
      "investigate role": 56811,
      "context lstm": 22170,
      "lstm lm": 65637,
      "ablation studies": 1584,
      "words shuffled": 134221,
      "datasets penn": 27616,
      "treebank wikitext2": 125652,
      "capable using": 15222,
      "tokens context": 122302,
      "context average": 22016,
      "context recent": 22242,
      "model highly": 71286,
      "highly sensitive": 50344,
      "sensitive order": 105742,
      "order words": 82429,
      "words recent": 134158,
      "longrange context": 65177,
      "grave et": 48990,
      "copy words": 23293,
      "neural lms": 77963,
      "light recent": 63998,
      "models unsupervised": 74272,
      "semantic frame": 105058,
      "use dependency": 127993,
      "corpus perform": 23928,
      "perform unsupervised": 86097,
      "dataset performing": 27094,
      "performing par": 86964,
      "translation sentence": 125238,
      "sentence translated": 106111,
      "sentences existing": 106298,
      "use correct": 127973,
      "sentences training": 106526,
      "translations sentence": 125491,
      "sentence share": 106070,
      "distinguish correct": 32568,
      "incorrect ones": 53578,
      "targets training": 117794,
      "stage order": 112151,
      "potentially correct": 88606,
      "appeared training": 6809,
      "translation dataset": 124734,
      "baselines bleu": 12365,
      "methods date": 69422,
      "date little": 27828,
      "methods realworld": 69706,
      "address present": 3726,
      "realworld texts": 96185,
      "algorithms neural": 4871,
      "perform best": 85952,
      "performance differences": 86294,
      "patterns observed": 85764,
      "low coherence": 65351,
      "texts domains": 121503,
      "underlying sentiment": 126700,
      "sentiment sentence": 106794,
      "sentence keeping": 105912,
      "keeping content": 57518,
      "data solve": 26466,
      "enables training": 36405,
      "unpaired data": 127485,
      "module evaluate": 74489,
      "review datasets": 102543,
      "datasets yelp": 27803,
      "especially proposed": 38494,
      "method substantially": 69168,
      "content preservation": 21925,
      "datasets respectively": 27683,
      "bilstm entity": 14095,
      "entity library": 37951,
      "contribution semeval": 22777,
      "task character": 117959,
      "character identification": 16434,
      "multiparty dialogues": 75483,
      "model key": 71392,
      "key innovation": 57582,
      "model finding": 71185,
      "finding potentially": 44281,
      "requires effective": 100266,
      "unbalanced data": 126606,
      "chinese corpus": 16751,
      "nontaskoriented dialogue": 80304,
      "presents largescale": 89858,
      "largescale corpus": 61413,
      "collected social": 18435,
      "annotate corpus": 6125,
      "according relevance": 1999,
      "relevance coherence": 98504,
      "various unsupervised": 131233,
      "models response": 73948,
      "confirm proposed": 21017,
      "proposed corpus": 93244,
      "sentences sequence": 106492,
      "simplify sentences": 109614,
      "network sequence": 77427,
      "model dubbed": 71031,
      "exploit linguistic": 41427,
      "sentences jointly": 106362,
      "pretrained finetuned": 90035,
      "data trained": 26569,
      "word substitutions": 133592,
      "remains difficult": 98794,
      "models promising": 73813,
      "promising text": 92310,
      "tasks style": 119530,
      "style transfer": 114601,
      "social power": 110448,
      "infer social": 54104,
      "context interactions": 22151,
      "understanding social": 126966,
      "structures affect": 114056,
      "affect way": 4228,
      "way interact": 132099,
      "great social": 49025,
      "social scientists": 110455,
      "fundamental questions": 46112,
      "questions human": 95315,
      "behavior computer": 12650,
      "build automatic": 14741,
      "paper employ": 83891,
      "nlp study": 79694,
      "context interaction": 22150,
      "specifically investigate": 111560,
      "captures important": 15446,
      "important distinctions": 52148,
      "learning write": 63175,
      "language output": 59727,
      "capture notion": 15350,
      "communicative goals": 19052,
      "grices maxims": 49087,
      "learning mixture": 62748,
      "mixture multiple": 70425,
      "guide decoding": 49234,
      "process human": 91500,
      "text generated": 120974,
      "preferred baselines": 89249,
      "significantly enhances": 108910,
      "enhances overall": 37531,
      "overall coherence": 83221,
      "relation classification": 97943,
      "classification different": 17185,
      "text entailment": 120914,
      "direct semantic": 31800,
      "extending traditional": 41821,
      "different existing": 31136,
      "models built": 72869,
      "lexical distributional": 63758,
      "vector features": 131306,
      "features proposed": 43674,
      "algorithm sequence": 4799,
      "problem semantic": 91212,
      "comparative analysis": 19201,
      "analysis multilingual": 5708,
      "relatedness using": 97932,
      "provides comparative": 94010,
      "stateoftheart distributional": 112643,
      "models dsms": 73112,
      "contrasting native": 22723,
      "languagespecific models": 60996,
      "improvement average": 52683,
      "spearman correlation": 111343,
      "correlation using": 24246,
      "stateoftheart machine": 112693,
      "translation approaches": 124645,
      "approaches results": 8329,
      "possible errors": 88399,
      "errors introduced": 38379,
      "translation languages": 124892,
      "english distributional": 37116,
      "model provided": 71824,
      "results consistently": 101632,
      "partially annotated": 85302,
      "annotated examples": 6192,
      "parsers neural": 85034,
      "advances word": 4026,
      "representations greatly": 99672,
      "need domain": 76801,
      "adaptation target": 3271,
      "syntactically similar": 116517,
      "similar source": 109149,
      "train parser": 122995,
      "domains provide": 33839,
      "simple way": 109543,
      "way adapt": 132054,
      "heldout set": 49688,
      "stateoftheart single": 112953,
      "absolute increase": 1742,
      "outside training": 83200,
      "easier models": 34419,
      "capture global": 15308,
      "global structures": 48276,
      "decomposable attention": 28150,
      "model word2vec": 72325,
      "tracking state": 122761,
      "state changes": 112486,
      "comprehension present": 20215,
      "genre text": 47832,
      "text describing": 120867,
      "dynamic world": 34335,
      "world new": 135039,
      "contain natural": 21749,
      "machinegenerated text": 66317,
      "entity states": 38106,
      "text challenging": 120702,
      "need inferred": 76823,
      "data achieve": 25558,
      "exploit alternative": 41404,
      "alternative mechanisms": 5234,
      "state prediction": 112510,
      "prediction particular": 89100,
      "using lstm": 129841,
      "lstm input": 65628,
      "span prediction": 111128,
      "prediction new": 89090,
      "new models": 79036,
      "models available": 72797,
      "available community": 10963,
      "spanbased constituency": 111142,
      "parsing achieved": 85063,
      "achieved competitive": 2608,
      "extremely simple": 42609,
      "al 2017a": 4624,
      "art accuracy": 9034,
      "cubic time": 25202,
      "longer sentences": 65160,
      "applications sentence": 7013,
      "sentence boundaries": 105779,
      "endtoend discourse": 36893,
      "discourse parsing": 32071,
      "joint sentence": 57321,
      "constituency parser": 21525,
      "runs time": 103476,
      "pruning compared": 94165,
      "lineartime parser": 64404,
      "parser substantially": 84999,
      "sentences penn": 106442,
      "parsing achieves": 85064,
      "highest f1": 50225,
      "f1 accuracy": 42647,
      "model endtoend": 71075,
      "endtoend systems": 36974,
      "encoder contextualized": 36503,
      "architecture argument": 8618,
      "task semeval": 118677,
      "2018 simple": 685,
      "consisting parts": 21461,
      "built set": 14939,
      "set given": 107453,
      "plausible model": 87684,
      "vectors pretrained": 131463,
      "form transfer": 45137,
      "learning help": 62628,
      "help mitigate": 49750,
      "lack training": 58762,
      "lstms trained": 65717,
      "models achieving": 72706,
      "set 60": 107343,
      "advanced methods": 3948,
      "recently obtained": 96725,
      "obtained stateofthe": 81415,
      "art machine": 9055,
      "systems popular": 117067,
      "availability data": 10903,
      "data lowresourced": 26101,
      "lowresourced language": 65574,
      "field lack": 43961,
      "attempt build": 9733,
      "build nmt": 14795,
      "methods reduce": 69711,
      "systems addition": 116721,
      "addition proposed": 3467,
      "effective word": 34782,
      "segmentation vietnamese": 104651,
      "vietnamese texts": 131698,
      "texts alleviate": 121453,
      "based universal": 12144,
      "representations sentence": 99876,
      "capture wide": 15420,
      "range information": 95571,
      "based character": 11567,
      "word ngrams": 133370,
      "ngrams paper": 79467,
      "train sentence": 123013,
      "using smallscale": 130192,
      "evaluation sentence": 39383,
      "data tasks": 26545,
      "tasks improve": 119169,
      "improve automatic": 52338,
      "translation experimental": 124809,
      "performance sentence": 86698,
      "features sentence": 43717,
      "comparison semantic": 19572,
      "inspired propose": 55577,
      "employing multiple": 36317,
      "embeddings evaluating": 35701,
      "word multiple": 133369,
      "novel sentence": 80719,
      "comparison method": 19554,
      "strong performances": 113698,
      "tasks measure": 119289,
      "dataset semeval": 27182,
      "methods textual": 69801,
      "use handcrafted": 128080,
      "alignment features": 4960,
      "features ngram": 43628,
      "dependency features": 29140,
      "features does": 43462,
      "require pretrained": 100188,
      "endtoend approach": 36870,
      "form content": 45075,
      "text typically": 121382,
      "relying features": 98759,
      "grammatical function": 48704,
      "function paper": 46038,
      "networks makes": 77655,
      "makes decisions": 66769,
      "corpus neural": 23898,
      "model substantially": 72101,
      "improves strong": 53050,
      "baselines data": 12379,
      "data models": 26137,
      "models publicly": 73842,
      "rise neural": 102866,
      "networks particularly": 77695,
      "common models": 18895,
      "initial word": 55228,
      "word encodings": 133260,
      "composed recurrent": 20070,
      "representation learned": 99280,
      "learned pretrained": 62243,
      "consider context": 21192,
      "word subsequent": 133591,
      "recurrent layers": 97180,
      "word subword": 133593,
      "subword information": 115020,
      "investigate models": 56783,
      "sentencelevel context": 106149,
      "character wordbased": 16481,
      "wordbased representations": 133700,
      "learns combine": 63198,
      "states present": 113052,
      "partofspeech morphological": 85529,
      "performance number": 86580,
      "representations crosslingual": 99579,
      "crosslingual information": 24960,
      "important challenging": 52113,
      "task especially": 118152,
      "especially low": 38468,
      "resource scenarios": 100877,
      "propose training": 93116,
      "powerful technique": 88678,
      "enables neural": 36394,
      "representations robust": 99867,
      "robust noise": 103055,
      "yielding better": 135376,
      "networks sentence": 77756,
      "pair modeling": 83444,
      "critical nlp": 24820,
      "tasks paraphrase": 119367,
      "identification semantic": 51435,
      "inference stateoftheart": 54226,
      "tasks rely": 119453,
      "rely pretrained": 98730,
      "sentencelevel semantics": 106172,
      "works attempted": 134923,
      "really need": 96132,
      "models effective": 73116,
      "effective tasks": 34757,
      "single sentence": 109793,
      "input including": 55350,
      "including language": 53308,
      "modeling machine": 72470,
      "experiments subword": 41157,
      "results social": 102197,
      "media datasets": 68103,
      "datasets competitive": 27365,
      "results news": 101989,
      "news data": 79326,
      "data paraphrase": 26220,
      "variation neural": 130895,
      "generators natural": 47787,
      "control style": 22819,
      "date work": 27832,
      "taskoriented neural": 118899,
      "primarily focused": 90640,
      "semantic fidelity": 105055,
      "contexts difficult": 22388,
      "different sequencetosequence": 31420,
      "models carefully": 72886,
      "content style": 21954,
      "control semantic": 22818,
      "explicit model": 41335,
      "model adds": 70636,
      "context vector": 22319,
      "parameters input": 84756,
      "state encoder": 112492,
      "encoder time": 36581,
      "time step": 122112,
      "supervision training": 115926,
      "large learning": 61131,
      "networks learning": 77641,
      "learning phrase": 62888,
      "words obtain": 134087,
      "rotten tomatoes": 103286,
      "tomatoes dataset": 122359,
      "keeping sentence": 57521,
      "embeddings fixed": 35726,
      "usergenerated texts": 129078,
      "information online": 54814,
      "make decision": 66655,
      "users read": 129164,
      "predict ratings": 88921,
      "sentimental words": 106822,
      "rating prediction": 95819,
      "high information": 50078,
      "information gain": 54628,
      "efficient compared": 35065,
      "compared words": 19492,
      "addition explore": 3428,
      "explore best": 41525,
      "features predicting": 43662,
      "ratings reviews": 95829,
      "kind semantic": 57682,
      "types exploit": 126286,
      "head word": 49579,
      "languages sentiment": 60862,
      "analysis lowresource": 5684,
      "corpora estimate": 23471,
      "highperforming models": 50362,
      "translation bilingual": 124675,
      "amounts parallel": 5353,
      "sufficiently capture": 115292,
      "capture sentiment": 15390,
      "sentiment information": 106746,
      "information introduce": 54700,
      "introduce bilingual": 56388,
      "annotated sentiment": 6231,
      "embeddings language": 35772,
      "language perform": 59807,
      "language combinations": 58888,
      "spanish catalan": 111148,
      "sentencelevel crosslingual": 106152,
      "methods experimental": 69487,
      "experimental setups": 40731,
      "complementary information": 19746,
      "translation analysis": 124637,
      "analysis resulting": 5810,
      "resourcepoor target": 100919,
      "machinetranslated data": 66348,
      "bootstrapping natural": 14460,
      "use case": 127933,
      "voicecontrolled device": 131928,
      "cost time": 24375,
      "time needed": 122055,
      "corpus new": 23900,
      "language having": 59117,
      "having large": 49560,
      "order utterances": 82427,
      "improve nlu": 52443,
      "nlu performance": 79812,
      "performance languagespecific": 86486,
      "postprocessing methods": 88508,
      "methods investigated": 69563,
      "investigated methods": 56839,
      "nlu task": 79814,
      "task translating": 118807,
      "10 millions": 141,
      "millions training": 70124,
      "german results": 47911,
      "large improvement": 61106,
      "improvement using": 52782,
      "using mt": 129904,
      "inhouse data": 55203,
      "manual effort": 66989,
      "results does": 101737,
      "describes participation": 29422,
      "group text": 49143,
      "word provided": 133425,
      "task instance": 118302,
      "instance word": 55612,
      "word occurs": 133373,
      "proposed evaluation": 93272,
      "datasets varying": 27790,
      "varying complexity": 131254,
      "wikipedia web": 132687,
      "experiments positive": 41062,
      "contexts represented": 22423,
      "stateoftheart production": 112873,
      "systems team": 117190,
      "datasets best": 27333,
      "participating teams": 85359,
      "dyadic interactions": 34288,
      "schemes automatically": 103947,
      "target speaker": 117714,
      "target speakers": 117715,
      "control degree": 22803,
      "investigate degree": 56740,
      "crosslingual document": 24940,
      "document classifier": 32970,
      "resources language": 100994,
      "language transferring": 60183,
      "language additional": 58814,
      "additional resources": 3567,
      "resources approaches": 100946,
      "approaches proposed": 8295,
      "proposed literature": 93324,
      "covers languages": 24556,
      "published works": 94376,
      "transfer english": 124050,
      "prior distributions": 90700,
      "distributions differ": 32756,
      "differ significantly": 30932,
      "languages argue": 60409,
      "corpus balanced": 23676,
      "russian japanese": 103492,
      "japanese chinese": 57184,
      "language transfer": 60181,
      "directions using": 31855,
      "embeddings respectively": 35915,
      "framework evaluate": 45527,
      "evaluate crosslingual": 38814,
      "hope foster": 50627,
      "important area": 52102,
      "implementation paper": 51949,
      "describes submissions": 29434,
      "workshop neural": 135010,
      "university edinburgh": 127349,
      "adam mickiewicz": 3167,
      "mickiewicz university": 70033,
      "focus efficient": 44756,
      "deeplearning model": 28458,
      "fast inference": 43139,
      "inference engine": 54138,
      "engine neural": 37032,
      "translation improve": 124857,
      "performance efficient": 86331,
      "submissions using": 114756,
      "drawn attention": 34131,
      "promising translation": 92313,
      "recently studies": 96770,
      "studies indicate": 114239,
      "using phrase": 130013,
      "main idea": 66427,
      "nmt make": 79890,
      "correct predictions": 24116,
      "sentence construct": 105804,
      "word set": 133548,
      "set matching": 107489,
      "translated target": 124566,
      "words nmt": 134081,
      "word using": 133625,
      "translation probability": 125142,
      "finally integrate": 44198,
      "nmt improve": 79874,
      "remarkable improvements": 98833,
      "attentionbased nmt": 10093,
      "japanese predicate": 57195,
      "words owing": 134100,
      "computational costs": 20373,
      "softmax function": 110507,
      "increase vocabulary": 53622,
      "varieties research": 130960,
      "research focus": 100517,
      "various forms": 131101,
      "forms verbs": 45265,
      "methods considered": 69395,
      "embedded vector": 35365,
      "using methods": 129879,
      "methods demonstrate": 69429,
      "corpus nmt": 23904,
      "models output": 73692,
      "output words": 83152,
      "set furthermore": 107449,
      "englishtojapanese translation": 37457,
      "order source": 82408,
      "significantly influences": 108970,
      "effectively address": 34786,
      "require manual": 100174,
      "manual feature": 66998,
      "feature design": 43262,
      "making language": 66849,
      "features raw": 43686,
      "inputs experiments": 55483,
      "quality stateoftheart": 94799,
      "training nlp": 123747,
      "training sequencetosequence": 123843,
      "models features": 73244,
      "benchmarks machine": 12917,
      "translation speech": 125282,
      "recognition tasks": 97027,
      "building blocks": 14826,
      "models solve": 74066,
      "range tasks": 95609,
      "including neural": 53336,
      "translation automatic": 124657,
      "hypernym discovery": 51216,
      "discovery term": 32150,
      "task aims": 117864,
      "aims discover": 4520,
      "input concepts": 55309,
      "predefined vocabulary": 88838,
      "concerned task": 20688,
      "task empirically": 118138,
      "empirically study": 36243,
      "study various": 114547,
      "various neural": 131149,
      "models build": 72867,
      "latent space": 61608,
      "phrases evaluated": 87422,
      "models include": 73375,
      "network gated": 77262,
      "unit recurrent": 127209,
      "including word": 53406,
      "sense embedding": 105675,
      "embedding better": 35382,
      "performance generating": 86417,
      "largescale knowledge": 61448,
      "graphs provide": 48969,
      "vast amounts": 131272,
      "amounts structured": 5355,
      "short textual": 108264,
      "entities lack": 37810,
      "network generates": 77267,
      "generates short": 47175,
      "description entity": 29460,
      "embeddings dynamic": 35672,
      "dynamic context": 34297,
      "generated sequence": 47121,
      "discern relevant": 32011,
      "relevant information": 98554,
      "relationship text": 98310,
      "mining knowledge": 70238,
      "discovery data": 32135,
      "analysis process": 5772,
      "analysis confirms": 5542,
      "reduce computational": 97318,
      "vocabulary set": 131901,
      "learning encoderdecoder": 62543,
      "method selecting": 69116,
      "cooccurrence information": 23227,
      "capture using": 15416,
      "algorithm apply": 4697,
      "apply proposed": 7207,
      "method tasks": 69180,
      "translation grammatical": 124843,
      "method english": 68806,
      "english grammatical": 37162,
      "points higher": 87859,
      "distilling knowledge": 32528,
      "modeled structured": 72365,
      "search problem": 104337,
      "different initialization": 31181,
      "learning match": 62709,
      "use ensemble": 128029,
      "distillation effectively": 32504,
      "improve single": 52541,
      "single models": 109766,
      "performance final": 86382,
      "achieves improvements": 2809,
      "baselines outperforms": 12438,
      "models previous": 73794,
      "parser abstract": 84907,
      "representations learns": 99735,
      "tree representations": 125611,
      "representations compositional": 99555,
      "compositional structure": 20134,
      "structure amr": 113806,
      "graph allows": 48756,
      "use standard": 128293,
      "tree parsing": 125605,
      "linguistically principled": 64596,
      "decoding algorithms": 28073,
      "outperform strong": 82761,
      "learning global": 62616,
      "global interpretability": 48243,
      "extraction particular": 42427,
      "entity classification": 37913,
      "combines benefits": 18678,
      "benefits traditional": 13023,
      "use limited": 128125,
      "extraction patterns": 42429,
      "robust learning": 103043,
      "algorithm iteratively": 4759,
      "embeddings multiword": 35819,
      "multiword entities": 75942,
      "datasets conll2003": 27373,
      "conll2003 ontonotes": 21097,
      "embeddings approach": 35555,
      "approach outputs": 7781,
      "model consisting": 70894,
      "ranking patterns": 95679,
      "embedding given": 35412,
      "given class": 47992,
      "performs close": 86989,
      "produce interpretable": 91903,
      "models small": 74061,
      "cooperative game": 23257,
      "new feature": 78916,
      "pitch accent": 87572,
      "speakers gender": 111317,
      "different functions": 31158,
      "strategy task": 113542,
      "paraphrase plagiarism": 84830,
      "identification paraphrase": 51419,
      "task given": 118244,
      "accordingly paper": 2015,
      "relatedness given": 97922,
      "given texts": 48157,
      "information external": 54557,
      "external resource": 42037,
      "words resulting": 134176,
      "training supervised": 123893,
      "supervised classifier": 115741,
      "detecting paraphrase": 29846,
      "proposed metrics": 93420,
      "addition results": 3472,
      "competitive stateofthe": 19691,
      "art methods": 9059,
      "effective solution": 34747,
      "language games": 59059,
      "private information": 90785,
      "ii generate": 51708,
      "reason pragmatically": 96203,
      "strategy propose": 113533,
      "importance capturing": 52047,
      "using crowdsourcing": 129524,
      "indonesian conversational": 53990,
      "conversational texts": 23041,
      "texts despite": 121495,
      "despite long": 29707,
      "processing community": 91639,
      "texts texts": 121635,
      "variations increase": 130936,
      "number oov": 80929,
      "poses difficulty": 88263,
      "evidence effectiveness": 39642,
      "labeling models": 58512,
      "models character": 72898,
      "character models": 16450,
      "f1 points": 42668,
      "improvement high": 52716,
      "15 f1": 344,
      "models robust": 73966,
      "robust high": 103039,
      "oov rate": 81866,
      "semeval 2019": 105504,
      "2019 shared": 708,
      "task ucca": 118817,
      "submit systems": 114763,
      "given success": 48141,
      "parsing shared": 85236,
      "task significant": 118713,
      "parsing particular": 85188,
      "particular semantic": 85449,
      "furthermore existing": 46170,
      "applications semantic": 7011,
      "semantic evaluation": 105045,
      "greatly benefit": 49048,
      "benefit better": 12961,
      "better automatic": 13513,
      "competition website": 19628,
      "define semantic": 28491,
      "texts multiple": 121556,
      "systems date": 116832,
      "capture relations": 15367,
      "relations target": 98264,
      "target relation": 117695,
      "relation time": 98075,
      "approach jointly": 7654,
      "multiple relations": 75665,
      "separate tasks": 106863,
      "tasks does": 119071,
      "outperform singletask": 82745,
      "approach traditional": 7976,
      "performance relation": 86667,
      "dataset does": 26881,
      "studies identified": 114236,
      "anderson et": 6095,
      "processing linguistic": 91699,
      "shared semantic": 108008,
      "space semantic": 111058,
      "used variety": 128838,
      "variety linguistic": 130989,
      "tasks claim": 118976,
      "current studies": 25373,
      "content representations": 21935,
      "designed solve": 29623,
      "fail distinguish": 42956,
      "solve different": 110595,
      "tasks conclude": 118998,
      "neural representation": 78650,
      "training semisupervised": 123834,
      "performance japanese": 86473,
      "increase size": 53616,
      "large raw": 61233,
      "raw corpora": 95849,
      "novel japanese": 80606,
      "analysis model": 5698,
      "based semisupervised": 12014,
      "training raw": 123804,
      "existing stateoftheart": 40290,
      "annotations dataset": 6421,
      "dataset construction": 26823,
      "direct assessment": 31783,
      "assign scores": 9546,
      "labels proposed": 58633,
      "leads increased": 61941,
      "dataset creation": 26838,
      "understanding meanings": 126887,
      "feedback understanding": 43838,
      "task providing": 118587,
      "multilingual environment": 75253,
      "feedback secondly": 43834,
      "classification used": 17481,
      "task capturing": 117953,
      "linear classifier": 64333,
      "close tasks": 17837,
      "tasks high": 119157,
      "domain suggestion": 33664,
      "problem definition": 91001,
      "task suggestion": 118763,
      "domain applications": 33469,
      "human perception": 50926,
      "datasets task": 27746,
      "used manually": 128623,
      "present annotation": 89369,
      "annotation study": 6378,
      "study based": 114328,
      "based observations": 11909,
      "observations propose": 81181,
      "task definition": 118053,
      "study provide": 114493,
      "available labeled": 11022,
      "difference natural": 30943,
      "language corpora": 58921,
      "corpora observed": 23544,
      "software systems": 110523,
      "systems known": 116964,
      "corpora does": 23463,
      "languages does": 60514,
      "reading writing": 96039,
      "writing code": 135090,
      "support argument": 115957,
      "studies set": 114278,
      "learner corpora": 62291,
      "used humans": 128573,
      "present challenge": 89402,
      "approach introduced": 7647,
      "challenge sets": 16101,
      "difficult machines": 31643,
      "translate correctly": 124525,
      "sentences designed": 106275,
      "test ability": 120418,
      "ability systems": 1571,
      "systems correctly": 116820,
      "correctly translate": 24182,
      "words prepositions": 134134,
      "set testing": 107607,
      "google translate": 48524,
      "different dates": 31081,
      "january 2018": 57181,
      "resulting data": 101439,
      "data publicly": 26305,
      "domain adversarial": 33465,
      "propose domain": 92631,
      "recognition problem": 96970,
      "problem order": 91151,
      "order reduce": 82396,
      "labeled source": 58469,
      "data standard": 26499,
      "learning objective": 62853,
      "objective encourage": 81072,
      "character error": 16431,
      "baseline trained": 12325,
      "data benefit": 25695,
      "used combination": 128431,
      "combination training": 18590,
      "training automatic": 123367,
      "automatic transcriptions": 10703,
      "findings second": 44312,
      "second workshop": 104467,
      "generation document": 47374,
      "describes findings": 29405,
      "held concert": 49681,
      "concert annual": 20695,
      "association computational": 9619,
      "2018 summarize": 687,
      "summarize research": 115609,
      "research trends": 100653,
      "trends papers": 125751,
      "papers presented": 84508,
      "presented proceedings": 89794,
      "particular linguistic": 85429,
      "adaptation data": 3213,
      "resources analysis": 100941,
      "analysis models": 5702,
      "models second": 73982,
      "second results": 104450,
      "participants tasked": 85327,
      "tasked creating": 118876,
      "translation taskspecific": 125370,
      "taskspecific attention": 119619,
      "translation addresses": 124628,
      "addresses task": 3805,
      "multiple source": 75695,
      "technique improving": 119792,
      "quality sequencetosequence": 94789,
      "multilingual translation": 75395,
      "approach seeks": 7870,
      "model particular": 71714,
      "attention provides": 9983,
      "provides consistent": 94015,
      "consistent gains": 21377,
      "quality possible": 94757,
      "possible translation": 88439,
      "directions compared": 31836,
      "parameters shared": 84785,
      "extreme lowresource": 42594,
      "zeroshot translation": 135534,
      "directions model": 31847,
      "taskspecific representation": 119639,
      "classification resource": 17381,
      "poor languages": 88059,
      "limited dependence": 64229,
      "enhance text": 37495,
      "resource availability": 100829,
      "bilstm network": 14100,
      "network shared": 77432,
      "sentences common": 106245,
      "common space": 18929,
      "similarity assigned": 109199,
      "tags model": 117480,
      "evaluated model": 38991,
      "analysis emoji": 5590,
      "emoji prediction": 36088,
      "various useful": 131236,
      "statistics corpus": 113194,
      "corpus given": 23820,
      "form represent": 45119,
      "morphological feature": 74700,
      "forms present": 45257,
      "approach employs": 7520,
      "model prior": 71789,
      "prior distribution": 90698,
      "rare ones": 95740,
      "task report": 118636,
      "natural question": 76614,
      "work languages": 134600,
      "crosslinguistic comparison": 25054,
      "comparison language": 19551,
      "asked predict": 9303,
      "information conduct": 54436,
      "conduct study": 20895,
      "21 languages": 761,
      "languages demonstrating": 60492,
      "languages textual": 60919,
      "ngram lstm": 79441,
      "capturing interactions": 15472,
      "new japanese": 78965,
      "models integrate": 73409,
      "integrate label": 55756,
      "label prediction": 58406,
      "prediction information": 89069,
      "input layers": 55356,
      "standard deep": 112229,
      "birnn model": 14229,
      "model models": 71525,
      "using mechanisms": 129874,
      "pooling attention": 88040,
      "directly capture": 31865,
      "capture potential": 15361,
      "distance experiments": 32417,
      "accuracy specifically": 2279,
      "art overall": 9074,
      "benchmark corpus": 12747,
      "solving task": 110650,
      "requires detecting": 100264,
      "detecting recurring": 29848,
      "events discourse": 39575,
      "context applications": 22007,
      "summarization dialogue": 115495,
      "systems create": 116822,
      "new datasets": 78871,
      "task derived": 118068,
      "treebank annotated": 125625,
      "annotated english": 6189,
      "english gigaword": 37159,
      "corpora novel": 23543,
      "baseline recurrent": 12292,
      "additional trainable": 3585,
      "trainable parameters": 123057,
      "mechanism demonstrate": 67966,
      "model statistically": 72088,
      "outperforms number": 82927,
      "number baselines": 80846,
      "model sequencetosequence": 71987,
      "learning taskoriented": 63088,
      "dialogue dialogue": 30665,
      "pipeline models": 87547,
      "models taskoriented": 74161,
      "explicit modeling": 41336,
      "dialogue states": 30756,
      "domainspecific knowledge": 33904,
      "dialogue history": 30687,
      "history response": 50567,
      "response current": 101194,
      "current turn": 25383,
      "explicit knowledge": 41328,
      "framework leverages": 45600,
      "leverages advantages": 63642,
      "advantages classic": 4070,
      "framework models": 45615,
      "use representation": 128245,
      "representation query": 99390,
      "multidomain taskoriented": 75097,
      "shows framework": 108579,
      "sequencetosequence based": 107156,
      "based baseline": 11544,
      "models automatic": 72789,
      "gate model": 46500,
      "usually trained": 130461,
      "bilingual sentence": 14058,
      "translate sentence": 124535,
      "sentence time": 106105,
      "intersentence information": 56308,
      "information make": 54746,
      "inconsistent translations": 53444,
      "neighboring sentences": 76994,
      "order handle": 82333,
      "handle issue": 49383,
      "propose intersentence": 92728,
      "encoder encode": 36523,
      "adjacent sentences": 3851,
      "way proposed": 132124,
      "sentences neural": 106410,
      "baseline second": 12305,
      "approach accurate": 7293,
      "accurate prediction": 2353,
      "fundamental building": 46085,
      "building block": 14824,
      "knowledge gaps": 57946,
      "applying approach": 7230,
      "achieved highest": 2640,
      "score evaluation": 104070,
      "metrics datasets": 69948,
      "2018 shared": 676,
      "modeling model": 72477,
      "model discuss": 71009,
      "production environment": 92056,
      "evaluation unsupervised": 39431,
      "compositional representations": 20128,
      "representations evaluated": 99637,
      "evaluated various": 39024,
      "various compositional": 131064,
      "models bagofwords": 72801,
      "bagofwords representations": 11381,
      "rnnbased models": 102948,
      "unsupervised evaluation": 127634,
      "evaluation benchmarks": 39135,
      "benchmarks results": 12935,
      "vector averaging": 131297,
      "models benchmarks": 72828,
      "models useful": 74294,
      "useful certain": 128867,
      "certain classification": 15934,
      "identify aspects": 51477,
      "aspects meaning": 9394,
      "characteristics various": 16543,
      "models explain": 73206,
      "analysis case": 5523,
      "study recent": 114498,
      "field consistently": 43937,
      "released code": 98476,
      "settings learning": 107817,
      "complementary methods": 19748,
      "methods perform": 69670,
      "perform largescale": 86023,
      "different english": 31127,
      "datasets reflecting": 27664,
      "reproduction experiments": 100093,
      "experiments consider": 40854,
      "variety datasets": 130968,
      "methods published": 69703,
      "aid understanding": 4436,
      "bags words": 11383,
      "analytics based": 5949,
      "requires deeper": 100262,
      "deeper insight": 28443,
      "documents created": 33214,
      "requires richer": 100319,
      "used required": 128727,
      "required human": 100216,
      "effort construct": 35163,
      "corpora resulting": 23577,
      "provide practical": 93895,
      "new class": 78830,
      "techniques textual": 119998,
      "lexical contextual": 63749,
      "information predicting": 54857,
      "state level": 112503,
      "received significant": 96375,
      "previous literature": 90420,
      "state governments": 112498,
      "bills year": 14087,
      "level analysis": 63418,
      "analysis received": 5794,
      "factors affect": 42880,
      "utilize lexical": 130515,
      "lexical content": 63747,
      "features build": 43385,
      "predictive models": 89205,
      "models allowing": 72745,
      "allowing comparison": 5109,
      "factors important": 42891,
      "process furthermore": 91493,
      "achieving average": 2927,
      "average improvement": 11195,
      "improvement accuracy": 52679,
      "using conditional": 129470,
      "efficient word": 35135,
      "languages word": 60970,
      "script cursive": 104252,
      "regardless word": 97669,
      "word boundary": 132938,
      "uses conditional": 129207,
      "learns predict": 63226,
      "nonjoiner zwnj": 80218,
      "boundary identification": 14521,
      "identification tasks": 51450,
      "tasks code": 118981,
      "corpus publicly": 23951,
      "available make": 11040,
      "make results": 66719,
      "results reproducible": 102132,
      "logical representations": 65017,
      "multilingual datasets": 75236,
      "recent research": 96506,
      "multilingual setup": 75369,
      "monolingual semantic": 74610,
      "language leveraging": 59250,
      "data annotated": 25602,
      "languages remains": 60842,
      "study learning": 114424,
      "used improving": 128585,
      "existing monolingual": 40213,
      "parsers incorporate": 85028,
      "able yield": 1705,
      "yield improved": 135341,
      "improved semantic": 52639,
      "geoquery dataset": 47874,
      "user product": 129021,
      "attention sentiment": 10004,
      "classification use": 17480,
      "userproduct information": 129084,
      "important especially": 52158,
      "number reviews": 80959,
      "reviews limited": 102613,
      "limited current": 64226,
      "models deal": 73016,
      "present hybrid": 89512,
      "contains modules": 21831,
      "fast word": 43156,
      "short long": 108216,
      "range dependency": 95558,
      "aware attention": 11258,
      "mechanism considers": 67960,
      "encoded word": 36487,
      "vectors constructed": 131418,
      "used original": 128675,
      "sufficient information": 115280,
      "selective gate": 104855,
      "experiments terms": 41174,
      "better compared": 13542,
      "datasets despite": 27422,
      "having complexity": 49554,
      "importantly model": 52298,
      "humor detection": 51138,
      "detection englishhindi": 29937,
      "englishhindi codemixed": 37401,
      "content corpus": 21864,
      "corpus baseline": 23678,
      "generated data": 47067,
      "data social": 26463,
      "sites led": 109877,
      "linguistics past": 64622,
      "past decade": 85638,
      "automatic humor": 10564,
      "understanding text": 126998,
      "text required": 121257,
      "required detect": 100214,
      "humor makes": 51139,
      "multilingual speakers": 75373,
      "posting social": 88498,
      "media called": 68086,
      "called codemixing": 15051,
      "field linguistic": 43962,
      "develop techniques": 30242,
      "codemixed tweets": 18190,
      "paper analyze": 83732,
      "analyze task": 6010,
      "corpus containing": 23721,
      "tagged words": 117352,
      "words tweets": 134283,
      "language tags": 60143,
      "corpus provide": 23947,
      "baseline classification": 12200,
      "humorous nonhumorous": 51144,
      "toolkit paper": 122416,
      "toolkit neural": 122414,
      "models crf": 73002,
      "crf inference": 24748,
      "inference layer": 54160,
      "custom model": 25434,
      "model structure": 72092,
      "flexible neural": 44661,
      "neural feature": 77915,
      "analysis correlation": 5551,
      "correlation syntax": 24245,
      "active area": 3103,
      "research linguistics": 100548,
      "typically studied": 126462,
      "controlled conditions": 22833,
      "successfully used": 115196,
      "parsing shallow": 85235,
      "close gap": 17817,
      "gap investigating": 46459,
      "functions using": 46082,
      "using linear": 129813,
      "linear mixed": 64352,
      "mixed effects": 70395,
      "effects models": 34995,
      "corpus able": 23638,
      "try optimize": 125899,
      "factual content": 42930,
      "obtain significant": 81320,
      "function compared": 46009,
      "negative effect": 76924,
      "linguistic probing": 64527,
      "methods challenging": 69367,
      "comprehensive evaluations": 20260,
      "evaluations different": 39459,
      "techniques past": 119946,
      "especially development": 38449,
      "work perform": 134668,
      "using wide": 130372,
      "tasks simple": 119502,
      "bagofwords recently": 11379,
      "model deep": 70957,
      "embeddings proved": 35892,
      "tasks compared": 118990,
      "compared sentence": 19438,
      "universal encoder": 127305,
      "encoder perform": 36553,
      "perform consistently": 85975,
      "skipgram word2vec": 109996,
      "word2vec recent": 133684,
      "words distributed": 133920,
      "gained popularity": 46369,
      "various areas": 131043,
      "capture syntactic": 15405,
      "explicit supervision": 41352,
      "respect propose": 101100,
      "structure training": 113976,
      "large gains": 61096,
      "original test": 82552,
      "method convolutional": 68745,
      "understanding slot": 126956,
      "filling important": 44043,
      "users intent": 129135,
      "assigning semantic": 9559,
      "semantic concept": 105014,
      "concept word": 20592,
      "combines convolutional": 18680,
      "word training": 133606,
      "dataset feature": 26929,
      "computational results": 20426,
      "bidirectional sequential": 13976,
      "empirically investigate": 36234,
      "training demonstrate": 123578,
      "domain mismatch": 33582,
      "mismatch training": 70305,
      "training deployment": 123579,
      "conduct series": 20893,
      "series simulation": 107280,
      "simulation experiments": 109648,
      "test effectiveness": 120448,
      "based feedback": 11717,
      "categorization semantic": 15769,
      "express relationships": 41706,
      "number linguistic": 80908,
      "definitions available": 28533,
      "make useful": 66749,
      "useful semantic": 128929,
      "analysis subset": 5877,
      "dictionary definition": 30876,
      "identifying patterns": 51608,
      "development information": 30391,
      "extraction frameworks": 42343,
      "building knowledge": 14849,
      "graph natural": 48835,
      "interpretable text": 56253,
      "entailment recognition": 37669,
      "source knowledge": 110769,
      "method provide": 69082,
      "automatically building": 10728,
      "trained classifier": 123087,
      "classifier automatically": 17523,
      "automatically labeling": 10794,
      "graph built": 48766,
      "according methodology": 1988,
      "recognition approach": 96808,
      "paths graph": 85689,
      "provide clear": 93773,
      "classification hidden": 17220,
      "goal detect": 48347,
      "using hidden": 129734,
      "allows capture": 5130,
      "level linguistic": 63469,
      "features computed": 43417,
      "include syntactic": 53235,
      "statistical word": 113176,
      "proposed evaluated": 93271,
      "corpus obtain": 23911,
      "approaches offer": 8264,
      "adapt word": 3200,
      "model learned": 71435,
      "general written": 46734,
      "texts data": 121490,
      "data spoken": 26498,
      "reviews model": 102619,
      "model dynamics": 71036,
      "approach interpretation": 7645,
      "tasks increasingly": 119197,
      "electronic form": 35249,
      "agents work": 4330,
      "work area": 134378,
      "focused classifying": 44850,
      "coherence models": 18305,
      "multiple turns": 75739,
      "challenge stateoftheart": 16104,
      "stateoftheart dialogue": 112639,
      "automatically learning": 10799,
      "distribution entities": 32652,
      "entities multiple": 37823,
      "automatic summaries": 10677,
      "extensions model": 41848,
      "crucial aspect": 25131,
      "dialogue structure": 30758,
      "representation dialogue": 99210,
      "structure conversation": 113828,
      "standard tasks": 112317,
      "assessment different": 9522,
      "different dialogue": 31095,
      "dialogue datasets": 30657,
      "modelling dialogue": 72603,
      "densely connected": 29025,
      "characterlevel features": 16565,
      "step achieving": 113243,
      "alleviate need": 5040,
      "need human": 76817,
      "human labor": 50895,
      "features methods": 43607,
      "methods utilize": 69840,
      "utilize neural": 130520,
      "architectures convolutional": 8796,
      "proposed shown": 93550,
      "shown great": 108476,
      "great results": 49021,
      "results cnn": 101582,
      "using densely": 129552,
      "network automatically": 77153,
      "language task": 60149,
      "shows robustness": 108623,
      "rnnbased methods": 102946,
      "methods evaluating": 69483,
      "method sequence": 69127,
      "tasks slot": 119505,
      "tagging namedentity": 117422,
      "accuracy slot": 2277,
      "tagging pos": 117437,
      "modeling multiturn": 72484,
      "multiturn conversation": 75911,
      "challenge building": 16003,
      "building intelligent": 14848,
      "intelligent dialogue": 55848,
      "systems work": 117241,
      "response matching": 101221,
      "matching multiturn": 67420,
      "simply concatenates": 109624,
      "utterances ignoring": 130646,
      "interactions previous": 55996,
      "utterances context": 130630,
      "context modeling": 22180,
      "paper formulate": 83960,
      "aggregation model": 4364,
      "model form": 71204,
      "context representation": 22248,
      "selfmatching attention": 104926,
      "vital information": 131848,
      "information utterance": 55082,
      "utterance model": 130604,
      "matches response": 67388,
      "final matching": 44107,
      "score obtained": 104099,
      "obtained attentive": 81345,
      "aggregation experimental": 4358,
      "work addresses": 134359,
      "addresses challenges": 3795,
      "challenges arising": 16134,
      "entities textual": 37881,
      "cost data": 24348,
      "model accuracy": 70529,
      "selecting appropriate": 104744,
      "appropriate evaluation": 8418,
      "overall quality": 83252,
      "quality annotation": 94593,
      "annotation present": 6354,
      "framework integrates": 45580,
      "set expansion": 107438,
      "reduce annotation": 97313,
      "annotation cost": 6293,
      "data provide": 26300,
      "provide online": 93884,
      "allows rapid": 5181,
      "annotation subsequent": 6381,
      "evaluate framework": 38838,
      "framework publicly": 45664,
      "available datasets": 10976,
      "drastically reduces": 34106,
      "average 85": 11178,
      "robust performance": 103060,
      "words modeling": 134057,
      "words work": 134323,
      "modeling propose": 72518,
      "propose possible": 92991,
      "possible solution": 88432,
      "evaluation analysis": 39118,
      "leads performance": 61948,
      "contextual language": 22477,
      "provide high": 93843,
      "accuracy variety": 2315,
      "framework contextual": 45474,
      "adaptation provides": 3255,
      "different possible": 31343,
      "goaloriented conversational": 48407,
      "agents natural": 4324,
      "nongoal oriented": 80209,
      "data partitioned": 26225,
      "using topic": 130297,
      "topic labels": 122536,
      "topic classifier": 122502,
      "classifier obtain": 17563,
      "relative wer": 98383,
      "wer improvement": 132449,
      "decoding strategy": 28138,
      "decoding framework": 28094,
      "model 15": 70497,
      "15 relative": 351,
      "recognizing named": 97074,
      "parsing chart": 85083,
      "constraints specify": 21613,
      "neural tagger": 78696,
      "tagger predicts": 117360,
      "pruning techniques": 94172,
      "speedup orders": 111886,
      "textual corpora": 121683,
      "algorithm word": 4814,
      "different words": 31576,
      "used similar": 128762,
      "actions objects": 3092,
      "based factors": 11713,
      "race gender": 95424,
      "similar documents": 109085,
      "quantitative evaluations": 94868,
      "evaluations technique": 39488,
      "humanities social": 51042,
      "features continuous": 43425,
      "systems conduct": 116801,
      "user models": 129014,
      "models decisions": 73018,
      "limited ability": 64200,
      "model fast": 71174,
      "using rnns": 130122,
      "data applied": 25611,
      "applied make": 7091,
      "just limited": 57467,
      "making predictions": 66859,
      "traditional acoustic": 122789,
      "acoustic features": 3004,
      "features generally": 43522,
      "perform worse": 86099,
      "outperform previously": 82730,
      "baselines joint": 12418,
      "detection dynamic": 29931,
      "classification spoken": 17416,
      "systems previous": 117073,
      "detectors trained": 30101,
      "trained separately": 123271,
      "indomain ind": 53968,
      "detection given": 29959,
      "evaluation scores": 39380,
      "model domain": 71025,
      "accuracy evaluating": 2154,
      "evaluating domain": 39051,
      "utterances large": 130653,
      "efficient approach": 35057,
      "approach encoding": 7525,
      "context spoken": 22278,
      "language user": 60306,
      "utterances semantic": 130666,
      "semantic frames": 105059,
      "frames making": 45400,
      "context prior": 22228,
      "encode context": 36422,
      "multiple utterances": 75741,
      "utterances dialogue": 130634,
      "dialogue turn": 30808,
      "resulting significant": 101473,
      "tradeoffs accuracy": 122783,
      "downstream components": 33996,
      "components like": 20031,
      "state tracker": 112518,
      "tracker dst": 122744,
      "track dialogue": 122727,
      "slu specifically": 110122,
      "rnn based": 102904,
      "subtasks shared": 114981,
      "speech asr": 111649,
      "asr usually": 9454,
      "does predict": 33378,
      "lack punctuation": 58739,
      "causes problems": 15841,
      "problems result": 91370,
      "overcome limitations": 83292,
      "models bidirectional": 72850,
      "memory blstm": 68296,
      "annotation experiments": 6311,
      "using sequence": 130156,
      "alignment algorithm": 4943,
      "algorithm neural": 4774,
      "networks trained": 77798,
      "common web": 18948,
      "web crawl": 132228,
      "glove embedding": 48300,
      "word time": 133603,
      "cnns yield": 18047,
      "words time": 134268,
      "time pretrained": 122075,
      "language variety": 60325,
      "variety translation": 131029,
      "translation trained": 125388,
      "trained translate": 123317,
      "standard national": 112274,
      "national varieties": 76215,
      "varieties language": 130956,
      "brazilian european": 14570,
      "method phrasebased": 69052,
      "translation report": 125215,
      "report performance": 99023,
      "02 bleu": 10,
      "carried human": 15541,
      "evaluation experiment": 39210,
      "output produced": 83109,
      "comparison statistical": 19578,
      "efficient robust": 35107,
      "robust word": 103083,
      "tool supports": 122400,
      "nlp practitioners": 79666,
      "common research": 18917,
      "research problems": 100589,
      "script using": 104261,
      "evaluating seven": 39100,
      "large word": 61329,
      "task involving": 118314,
      "raw sentence": 95856,
      "tasks reach": 119433,
      "reach goal": 95890,
      "intents dialogue": 55922,
      "dialogue work": 30816,
      "work small": 134808,
      "annotation data": 6296,
      "nlu module": 79811,
      "requiring heavy": 100341,
      "slots filled": 110088,
      "set initial": 107466,
      "initial training": 55227,
      "training sentences": 123837,
      "sentences classifying": 106238,
      "utterances generated": 130641,
      "dialogue development": 30664,
      "development framework": 30388,
      "accurate semantic": 2362,
      "parsing syntactic": 85251,
      "sentence semantic": 106063,
      "aim capture": 4450,
      "parser dozat": 84942,
      "dozat manning": 34068,
      "manning 2017": 66968,
      "beating previous": 12620,
      "complex stateoftheart": 19881,
      "developed resource": 30299,
      "sense annotated": 105651,
      "primary secondary": 90656,
      "relatively recent": 98410,
      "large scope": 61266,
      "relevant features": 98550,
      "analysis different": 5576,
      "classifiers applied": 17599,
      "results experiment": 101777,
      "resource effective": 100839,
      "effective using": 34773,
      "languages major": 60715,
      "major contribution": 66549,
      "contribution work": 22782,
      "annotations manually": 6444,
      "corpus consists": 23718,
      "annotation statistics": 6375,
      "translation rulebased": 125226,
      "data efficient": 25882,
      "making appropriate": 66820,
      "appropriate languages": 8423,
      "extensive training": 41960,
      "define language": 28486,
      "language rules": 60065,
      "meaning individual": 67641,
      "time representation": 122088,
      "concepts terms": 20644,
      "time proposed": 122081,
      "proposed natural": 93504,
      "generation parsing": 47536,
      "translation strategies": 125296,
      "example data": 39777,
      "success nlp": 115109,
      "tasks difficulty": 119064,
      "recurrent structure": 97261,
      "takes time": 117538,
      "ability obtain": 1555,
      "highlevel information": 50246,
      "multiple layers": 75602,
      "extra parameters": 42062,
      "use linear": 128126,
      "times fast": 122171,
      "standard rnns": 112298,
      "rnns faster": 102969,
      "longer sequences": 65161,
      "sequences experiments": 107122,
      "analysis datasets": 5558,
      "inflectional morphological": 54293,
      "morphological systems": 74739,
      "quantify linguistic": 94853,
      "complexity different": 19908,
      "languages morphological": 60736,
      "paradigm size": 84549,
      "languages inflectional": 60644,
      "jointly predict": 57381,
      "variational approximation": 130908,
      "large morphological": 61161,
      "cnn lstm": 18009,
      "model arabic": 70679,
      "analysis deep": 5560,
      "challenging large": 16268,
      "range application": 95539,
      "features long": 43598,
      "learning sequential": 63015,
      "approaches reported": 8322,
      "areas image": 8890,
      "image processing": 51789,
      "messages twitter": 68519,
      "twitter challenging": 126096,
      "task complexity": 117994,
      "language sentiment": 60081,
      "tasks arabic": 118935,
      "arabic rich": 8545,
      "tools arabic": 122433,
      "limited research": 64271,
      "area paper": 8878,
      "investigate benefits": 56724,
      "cnns lstms": 18038,
      "obtained improved": 81379,
      "accuracy arabic": 2102,
      "datasets additionally": 27299,
      "arabic words": 8569,
      "enhancing lexical": 37536,
      "resource using": 100887,
      "interface tool": 56103,
      "web words": 132272,
      "works shown": 134977,
      "shown importance": 108481,
      "wordlevel annotations": 133727,
      "annotations sentiment": 6463,
      "aim analyze": 4445,
      "analyze importance": 5978,
      "performing task": 86973,
      "task sentiment": 118691,
      "analysis explain": 5614,
      "task known": 118330,
      "development advanced": 30367,
      "interactivepredictive translation": 56041,
      "continuous learning": 22624,
      "version popular": 131618,
      "models deployed": 73042,
      "deployed used": 29312,
      "given high": 48041,
      "different problems": 31361,
      "image video": 51801,
      "video captioning": 131655,
      "marginal distributions": 67204,
      "essential tool": 38571,
      "techniques developed": 119869,
      "developed purpose": 30295,
      "used smoothing": 128766,
      "variants including": 130875,
      "widely considered": 132541,
      "smoothing method": 110269,
      "method preserves": 69068,
      "distributions original": 32766,
      "original model": 82531,
      "overcome propose": 83299,
      "advantages previous": 4086,
      "achieve results": 2523,
      "modelling task": 72627,
      "design use": 29571,
      "use graphbased": 128077,
      "framework toolkit": 45716,
      "released opensource": 98490,
      "opensource python": 82028,
      "python software": 94487,
      "independent easily": 53766,
      "easily extensible": 34452,
      "firstorder graphbased": 44568,
      "encoders decoders": 36639,
      "word analogies": 132914,
      "linear transformations": 64384,
      "new intrinsic": 78962,
      "including english": 53291,
      "stateoftheart monolingual": 112756,
      "shared space": 108013,
      "using dictionaries": 129565,
      "word translations": 133610,
      "translations compare": 125458,
      "tested linear": 120575,
      "achieve average": 2416,
      "monolingual bilingual": 74555,
      "bilingual multilingual": 14049,
      "respectively lowresource": 101148,
      "domainadversarial learning": 33706,
      "learning deep": 62486,
      "techniques recently": 119964,
      "shown successful": 108534,
      "successful natural": 115162,
      "systems require": 117112,
      "explores use": 41657,
      "avoid overfitting": 11239,
      "overfitting training": 83320,
      "training domain": 123592,
      "domain invariant": 33551,
      "invariant features": 56681,
      "complex neural": 19846,
      "networks lowresource": 77650,
      "zeroresource settings": 135485,
      "new target": 79193,
      "languages monolingual": 60734,
      "vectors directly": 131427,
      "final performance": 44116,
      "pretrained multilingual": 90159,
      "focus effectively": 44755,
      "integrating information": 55795,
      "concepts words": 20653,
      "lower dimensional": 65429,
      "dimensional space": 31751,
      "semantics broad": 105399,
      "broad context": 14669,
      "understanding investigate": 126870,
      "core nlp": 23329,
      "detection classification": 29896,
      "classification automatic": 17129,
      "targeted sentiment": 117769,
      "corpus open": 23916,
      "open corpus": 81887,
      "corpus eec": 23767,
      "corpus contemporary": 23724,
      "lexicon research": 63904,
      "architecture predict": 8728,
      "power relations": 88653,
      "relations language": 98196,
      "analysis reveal": 5815,
      "reveal underlying": 102506,
      "relations exist": 98162,
      "interaction prior": 55962,
      "nlp shown": 79688,
      "shown promise": 108512,
      "automatically predicting": 10810,
      "using nlp": 129972,
      "social interactions": 110346,
      "interactions remains": 55999,
      "architecture captures": 8636,
      "manifestations power": 66921,
      "way order": 132113,
      "infer direction": 54097,
      "direction power": 31824,
      "power pairs": 88646,
      "pairs participants": 83604,
      "participants email": 85319,
      "obtain accuracy": 81254,
      "task apply": 117884,
      "based entire": 11684,
      "entire set": 37715,
      "stateoftheart techniques": 113003,
      "amounts unlabeled": 5364,
      "embeddings yield": 36040,
      "yield superior": 135363,
      "superior results": 115695,
      "subsequent training": 114819,
      "training stages": 123879,
      "quality embeddings": 94640,
      "embeddings highly": 35741,
      "highly dependent": 50311,
      "extracts information": 42579,
      "domain portability": 33611,
      "requiring retraining": 100355,
      "indomain corpora": 53956,
      "corpora achieve": 23397,
      "achieve robustness": 2529,
      "robustness work": 103137,
      "present multitask": 89564,
      "multitask paradigm": 75887,
      "unsupervised contextual": 127615,
      "unsupervised domain": 127626,
      "generated process": 47109,
      "process increases": 91510,
      "tasks statistical": 119522,
      "nlu models": 79810,
      "present main": 89541,
      "main techniques": 66469,
      "feature hashing": 43288,
      "techniques complementary": 119854,
      "complementary existing": 19745,
      "existing model": 40201,
      "l1 regularization": 58371,
      "scale nlu": 103743,
      "reduction memory": 97450,
      "original models": 82532,
      "models minimal": 73567,
      "pca based": 85825,
      "reduction method": 97451,
      "problem called": 90957,
      "paper theoretical": 84473,
      "extension previous": 41838,
      "unlike conventional": 127428,
      "wordtovector embedding": 134340,
      "sequence level": 107006,
      "able facilitate": 1639,
      "results dense": 101712,
      "preprocessed data": 89320,
      "nmt neural": 79927,
      "machine translationnmt": 66309,
      "performance unfortunately": 86822,
      "present studies": 89722,
      "studies mainly": 114251,
      "measure ability": 67778,
      "ability nmt": 1553,
      "systems dealing": 116833,
      "propose quantitative": 93008,
      "automatically evaluate": 10764,
      "translation evaluate": 124797,
      "evaluate metrics": 38860,
      "metrics comparing": 69938,
      "scores human": 104177,
      "reveal strong": 102505,
      "indepth analyses": 53793,
      "bleu proposed": 14317,
      "significance metrics": 108714,
      "division labor": 32920,
      "perspective recent": 87171,
      "recent debates": 96445,
      "theory mind": 121821,
      "speakers listeners": 111322,
      "achieve successful": 2580,
      "successful communication": 115157,
      "model augmenting": 70708,
      "series simulations": 107281,
      "costs benefits": 24397,
      "experiments testing": 41176,
      "presence absence": 89347,
      "task speakers": 118727,
      "utterances used": 130673,
      "errors time": 38414,
      "time taken": 122120,
      "contextually appropriate": 22587,
      "framework provide": 45662,
      "processing constraints": 91646,
      "adoption neural": 3906,
      "stateoftheart wmt": 113027,
      "evaluated automatic": 38952,
      "automatic metric": 10597,
      "metric bleu": 69872,
      "model inference": 71340,
      "production deployment": 92054,
      "require expensive": 100132,
      "expensive human": 40422,
      "quality evaluation": 94657,
      "framework automatic": 45432,
      "evaluation translation": 39427,
      "novel bidirectional": 80506,
      "bidirectional transformer": 13977,
      "named neural": 76131,
      "model pretrained": 71772,
      "based prior": 11952,
      "data subsequently": 26515,
      "features fed": 43515,
      "simple bilstm": 109380,
      "evaluation experimental": 39211,
      "performance quality": 86644,
      "areas sentiment": 8901,
      "analysis opinion": 5729,
      "networks gained": 77606,
      "attention performance": 9969,
      "work applied": 134372,
      "achieves improved": 2807,
      "datasets compared": 27362,
      "highway networks": 50441,
      "batch normalization": 12556,
      "existing pretrained": 40255,
      "cnns used": 18046,
      "core task": 23333,
      "utterances conversation": 130631,
      "memory read": 68370,
      "capture variability": 15417,
      "natural conversations": 76247,
      "empirically compare": 36219,
      "approaches various": 8399,
      "conversational datasets": 23002,
      "consistently achieves": 21409,
      "qualitative evaluations": 94562,
      "autoencoding variational": 10421,
      "deep generative": 28234,
      "jointly shared": 57389,
      "shared latent": 107988,
      "networks perform": 77696,
      "training using": 123944,
      "amortised variational": 5324,
      "variational inference": 130925,
      "joint modelling": 57296,
      "efficient approximation": 35059,
      "maximum posteriori": 67547,
      "translation scenarios": 125229,
      "indomain training": 53985,
      "training learning": 123685,
      "modelling standard": 72625,
      "presents extension": 89849,
      "lowresource parallel": 65552,
      "corpus collected": 23695,
      "endangered language": 36839,
      "language griko": 59111,
      "computational research": 20423,
      "minutes speech": 70280,
      "transcribed translated": 123987,
      "speechtotranslation alignments": 111860,
      "alignments corpus": 5010,
      "morphosyntactic tags": 74818,
      "applying automatic": 7232,
      "unit discovery": 127202,
      "collected cleaned": 18406,
      "cleaned processed": 17708,
      "processed illustrate": 91598,
      "illustrate use": 51746,
      "use zeroresource": 128371,
      "alignment unsupervised": 4999,
      "discovery dataset": 32136,
      "dataset available": 26753,
      "computational language": 20383,
      "language documentation": 58978,
      "documentation experiments": 33121,
      "people organizations": 85881,
      "names people": 76150,
      "organization location": 82468,
      "research challenge": 100440,
      "computational techniques": 20437,
      "techniques identify": 119902,
      "identify entities": 51496,
      "tools algorithms": 122426,
      "improve nlp": 52441,
      "corpus proposed": 23943,
      "relies common": 98649,
      "recent word": 96564,
      "embeddings models": 35812,
      "models named": 73599,
      "entities used": 37890,
      "cluster similar": 17929,
      "similar entities": 109089,
      "types human": 126295,
      "measured using": 67841,
      "results metrics": 101930,
      "metrics demonstrate": 69949,
      "years field": 135266,
      "models survey": 74139,
      "provides brief": 94007,
      "brief introduction": 14612,
      "core linguistic": 23327,
      "processing issues": 91687,
      "applications computational": 6905,
      "provided recommendations": 93980,
      "recommendations future": 97092,
      "loss neural": 65284,
      "limitedsize vocabulary": 64306,
      "novel nmt": 80676,
      "nmt loss": 79886,
      "function encourages": 46012,
      "decoder generate": 28012,
      "generate words": 47044,
      "helps decoder": 49813,
      "best candidates": 13314,
      "sets proposed": 107698,
      "method showed": 69129,
      "improvements standard": 52910,
      "baseline datasets": 12208,
      "datasets especially": 27452,
      "1000 words": 178,
      "interactive neural": 56024,
      "data streams": 26508,
      "study application": 114321,
      "translation main": 124924,
      "stream source": 113551,
      "supervised human": 115758,
      "agent user": 4315,
      "data useful": 26604,
      "adapting neural": 3316,
      "information attention": 54382,
      "pipeline allows": 87522,
      "reduce effort": 97324,
      "quality translation": 94817,
      "outperforms classical": 82862,
      "approaches large": 8207,
      "task combining": 117973,
      "english subtasks": 37297,
      "subtasks semeval": 114980,
      "semeval 2016": 105498,
      "2016 task": 620,
      "department computer": 29037,
      "pattern recognition": 85718,
      "recognition human": 96887,
      "instances using": 55640,
      "pairs semantic": 83634,
      "words knowledge": 134020,
      "approach obtained": 7745,
      "compared task": 19474,
      "task participants": 118516,
      "gender bias": 46583,
      "bias neural": 13825,
      "systems reflect": 117105,
      "biases training": 13886,
      "data define": 25821,
      "quantify gender": 94849,
      "tasks empirical": 119089,
      "neural coreference": 77879,
      "rnnbased language": 102944,
      "trained benchmark": 123078,
      "bias models": 13820,
      "mitigate bias": 70359,
      "methodology corpus": 69253,
      "corpus augmentation": 23669,
      "genderneutral words": 46625,
      "words empirically": 133933,
      "bias preserving": 13834,
      "preserving accuracy": 89933,
      "explore space": 41588,
      "mitigation strategies": 70391,
      "embedding debiasing": 35392,
      "methods effectively": 69462,
      "original data": 82509,
      "corpus mining": 23883,
      "approach parallel": 7789,
      "embeddings embedding": 35678,
      "trained produce": 123246,
      "produce similar": 91934,
      "similar representations": 109136,
      "pairs translations": 83654,
      "novel training": 80753,
      "hard negatives": 49472,
      "consisting sentences": 21462,
      "sentences translations": 106530,
      "assessing machine": 9514,
      "trained gold": 123155,
      "pairs sentence": 83635,
      "corpus sentence": 23986,
      "achieve parallel": 2508,
      "parallel document": 84664,
      "document matching": 33036,
      "matching accuracy": 67392,
      "accuracy comparable": 2118,
      "computationally intensive": 20457,
      "data able": 25551,
      "able train": 1696,
      "trained original": 123227,
      "neural speech": 78685,
      "translation development": 124748,
      "systems improved": 116947,
      "improved significantly": 52641,
      "systems able": 116707,
      "sentences target": 106518,
      "complete sentences": 19769,
      "generate complete": 46912,
      "complete sentence": 19768,
      "sentence nmt": 105959,
      "systems adapted": 116720,
      "available furthermore": 10999,
      "performance original": 86588,
      "original training": 82556,
      "able reduce": 1682,
      "reduce number": 97343,
      "following approaches": 44962,
      "lexical meaning": 63782,
      "modern times": 74424,
      "ontology language": 81846,
      "framework proposes": 45661,
      "structures paper": 114095,
      "speakers languages": 111321,
      "following annotation": 44959,
      "type distribution": 126195,
      "distribution different": 32649,
      "compared using": 19485,
      "embedding used": 35523,
      "resource work": 100888,
      "text encoding": 120905,
      "fixedsize vector": 44623,
      "useful building": 128865,
      "building nlp": 14873,
      "systems including": 116950,
      "significantly reduce": 109045,
      "parameters improve": 84754,
      "accuracy finally": 2165,
      "representations created": 99578,
      "created model": 24675,
      "endtoend speech": 36966,
      "recognition existing": 96878,
      "typically built": 126414,
      "level known": 63464,
      "known dialog": 58297,
      "sentences speakers": 106501,
      "processing long": 91701,
      "recent progress": 96495,
      "progress endtoend": 92147,
      "information acoustic": 54360,
      "acoustic language": 3008,
      "natural dialog": 76249,
      "improve recognition": 52516,
      "recognition accuracy": 96801,
      "model explicitly": 71140,
      "explicitly uses": 41399,
      "information endtoend": 54529,
      "context trained": 22303,
      "manner evaluate": 66944,
      "switchboard conversational": 116280,
      "outperforms comparable": 82865,
      "recognition wordlevel": 97045,
      "neural temporal": 78701,
      "classification unsupervised": 17478,
      "unsupervised pretrained": 127695,
      "used effectively": 128506,
      "processing leverage": 91694,
      "leverage unlabeled": 63630,
      "data embeddings": 25884,
      "fixed word": 44610,
      "representations taskspecific": 99929,
      "taskspecific classification": 119621,
      "task loss": 118361,
      "model ensure": 71085,
      "representations contain": 99563,
      "features learned": 43590,
      "learned supervised": 62262,
      "supervised loss": 115786,
      "loss component": 65259,
      "learned unsupervised": 62275,
      "unsupervised loss": 127657,
      "component evaluate": 19977,
      "task temporal": 118781,
      "extraction clinical": 42281,
      "clinical records": 17785,
      "continued training": 22606,
      "training embeddings": 123604,
      "unsupervised objective": 127685,
      "objective task": 81120,
      "task objective": 118468,
      "better taskspecific": 13739,
      "using generaldomain": 129698,
      "partofspeech tagger": 85543,
      "adversarial domain": 4111,
      "learning source": 63039,
      "different target": 31475,
      "sufficient annotated": 115272,
      "data source": 26470,
      "domain limited": 33572,
      "limited labeled": 64244,
      "domain effectively": 33515,
      "effectively utilize": 34861,
      "utilize existing": 130506,
      "source domains": 110751,
      "issue domain": 56994,
      "trained source": 123288,
      "finetuned small": 44436,
      "set target": 107599,
      "existing knowledge": 40146,
      "domain adapt": 33426,
      "related domain": 97856,
      "small indomain": 110159,
      "data exploiting": 25932,
      "representations chinese": 99543,
      "using multichannel": 129905,
      "multichannel convolutional": 75042,
      "network effective": 77232,
      "effective representation": 34737,
      "text critical": 120843,
      "tasks particular": 119371,
      "particular task": 85456,
      "effect representations": 34609,
      "analysis proposing": 5781,
      "representation experimental": 99232,
      "generally improves": 46861,
      "outperform conventional": 82700,
      "model fasttext": 71175,
      "analysis pos": 5755,
      "extraction background": 42269,
      "extraction biomedical": 42273,
      "support knowledge": 115976,
      "knowledge capture": 57821,
      "task syntactic": 118769,
      "information valuable": 55083,
      "highest performance": 50228,
      "perform empirical": 85987,
      "study comparing": 114337,
      "comparing stateoftheart": 19520,
      "traditional featurebased": 122819,
      "parsing benchmark": 85079,
      "biomedical corpora": 14169,
      "work making": 134628,
      "making comparisons": 66831,
      "context specifically": 22276,
      "general neural": 46684,
      "models downstream": 73109,
      "downstream application": 33986,
      "performance does": 86311,
      "extraction performance": 42431,
      "detailed empirical": 29768,
      "models pos": 73754,
      "task availability": 117919,
      "chinese translation": 16830,
      "translation large": 124893,
      "modern chinese": 74382,
      "lack largescale": 58729,
      "based characteristics": 11569,
      "information achieves": 54359,
      "method create": 68748,
      "new largescale": 78986,
      "pairs best": 83494,
      "knowledge large": 58035,
      "dataset furthermore": 26946,
      "analyzed compared": 6020,
      "various nmt": 131158,
      "dataset provided": 27128,
      "baseline task": 12320,
      "task fake": 118202,
      "sentence detection": 105821,
      "training task": 123904,
      "trained language": 123168,
      "results sentencelevel": 102154,
      "sentencelevel tasks": 106177,
      "detection new": 30004,
      "new training": 79225,
      "generate fake": 46939,
      "sentences source": 106499,
      "effective detecting": 34661,
      "detecting fake": 29834,
      "sentences binary": 106231,
      "bilstm encoder": 14094,
      "encoder trained": 36582,
      "trained task": 123306,
      "task strong": 118744,
      "trains faster": 123963,
      "20 hours": 504,
      "using smaller": 130191,
      "shows learned": 108594,
      "selection multilingual": 104806,
      "interesting challenge": 56068,
      "introduce multilingual": 56461,
      "response input": 101218,
      "systems utilize": 117228,
      "language data": 58932,
      "present knowledge": 89523,
      "transfer methods": 124151,
      "methods conversational": 69408,
      "systems evaluate": 116877,
      "methods create": 69411,
      "new multilingual": 79040,
      "conversation dataset": 22946,
      "experiments dataset": 40865,
      "effectiveness methods": 34912,
      "systems suffer": 117171,
      "fundamental shortcoming": 46114,
      "shortcoming sequence": 108278,
      "errors early": 38365,
      "inputs model": 55486,
      "model quickly": 71839,
      "generation address": 47295,
      "regularization method": 97751,
      "method nmt": 68991,
      "nmt decoders": 79852,
      "achieved introducing": 2650,
      "kullbackleibler divergence": 58360,
      "regularization terms": 97760,
      "training objective": 123753,
      "objective reduce": 81110,
      "output probabilities": 83108,
      "addition employ": 3425,
      "employ joint": 36266,
      "training strategy": 123888,
      "improve interactive": 52397,
      "live video": 64852,
      "comment generation": 18799,
      "live comments": 64848,
      "comments paper": 18819,
      "task live": 118359,
      "live comment": 64847,
      "new form": 78921,
      "form comments": 45072,
      "comments videos": 18828,
      "relevant video": 98599,
      "users work": 129184,
      "work construct": 134436,
      "generate humanlike": 46954,
      "users comments": 129101,
      "model constructed": 70901,
      "dataset experimental": 26909,
      "outperform baselines": 82693,
      "baselines modeling": 12430,
      "modeling semantics": 72540,
      "gated graph": 46506,
      "graph neural": 48840,
      "networks knowledge": 77636,
      "base question": 11475,
      "approaches knowledge": 8204,
      "answering based": 6604,
      "semantic parses": 105140,
      "work largely": 134603,
      "largely focused": 61338,
      "focused selecting": 44874,
      "relations question": 98244,
      "semantic parse": 105132,
      "connections entities": 21146,
      "encode graph": 36431,
      "graph networks": 48839,
      "networks outperform": 77684,
      "models explicitly": 73208,
      "approach successfully": 7944,
      "process complex": 91442,
      "automatic referencebased": 10642,
      "referencebased evaluation": 97546,
      "pronoun translation": 92344,
      "test suite": 120540,
      "issues limit": 57057,
      "limit performance": 64156,
      "performance automated": 86157,
      "test suites": 120543,
      "semimarkov conditional": 105580,
      "characterlevel models": 16574,
      "characterlevel model": 16573,
      "effectiveness multilingual": 34917,
      "multilingual settings": 75367,
      "field neural": 43972,
      "networks character": 77532,
      "segment representation": 104564,
      "matches stateoftheart": 67389,
      "baselines various": 12489,
      "various languages": 131117,
      "noisy dataset": 80099,
      "dataset publicly": 27133,
      "use small": 128281,
      "belong different": 12714,
      "distribution types": 32686,
      "types tokens": 126374,
      "different morphological": 31269,
      "morphological word": 74757,
      "internal structure": 56176,
      "structure words": 113985,
      "translation deep": 124740,
      "data deployed": 25829,
      "results recently": 102118,
      "processing power": 91773,
      "significant boost": 108731,
      "boost terms": 14437,
      "terms efficiency": 120316,
      "accuracy paper": 2228,
      "paper look": 84034,
      "signal processing": 108690,
      "deep recurrent": 28402,
      "planning neural": 87640,
      "sentences missing": 106397,
      "missing current": 70321,
      "current language": 25290,
      "planning phase": 87641,
      "structure output": 113926,
      "words conditioned": 133874,
      "learned capture": 62205,
      "structure target": 113968,
      "codes design": 18202,
      "tags target": 117485,
      "performance generally": 86415,
      "generally improved": 46860,
      "different structures": 31458,
      "common phenomenon": 18908,
      "frequent senses": 45862,
      "constructing knowledge": 21678,
      "information importance": 54667,
      "senses paper": 105718,
      "explore way": 41615,
      "based textual": 12117,
      "descriptions multiple": 29488,
      "multiple senses": 75685,
      "expression online": 41741,
      "relationship different": 98289,
      "models experiment": 73196,
      "precision score": 88804,
      "mean average": 67595,
      "big margin": 13995,
      "adversarial neural": 4154,
      "study crosslingual": 114343,
      "language adversarial": 58818,
      "previously shown": 90622,
      "effective training": 34762,
      "crosslingual sentence": 25008,
      "sentence classifiers": 105788,
      "enable effective": 36351,
      "effective transfer": 34764,
      "types adversarial": 126245,
      "training tasks": 123908,
      "tasks dependency": 119047,
      "crosslingual performance": 24991,
      "trained baseline": 123077,
      "baseline exploiting": 12221,
      "exploiting deep": 41463,
      "learning persian": 62885,
      "persian sentiment": 87084,
      "rise social": 102870,
      "express opinions": 41704,
      "analysis automatically": 5509,
      "sentiment positive": 106779,
      "particular aspect": 85381,
      "topic product": 122562,
      "emerged powerful": 36052,
      "powerful machine": 88668,
      "technique tackle": 119816,
      "growing demand": 49171,
      "accurate sentiment": 2364,
      "analysis limited": 5680,
      "work deep": 134459,
      "developed applied": 30253,
      "applied novel": 7103,
      "shallow multilayer": 107924,
      "simulation results": 109649,
      "enhanced performance": 37514,
      "performance deep": 86276,
      "learning stateoftheart": 63055,
      "using spectral": 130205,
      "spectral clustering": 111630,
      "wellstudied problem": 132436,
      "data statistical": 26503,
      "processing features": 91671,
      "features solving": 43736,
      "solving various": 110653,
      "answering named": 6648,
      "used technique": 128805,
      "field image": 43954,
      "processing speech": 91798,
      "context nlp": 22200,
      "specifically method": 111568,
      "general word": 46730,
      "word lexicon": 133349,
      "classical nlp": 17092,
      "labeling dependency": 58495,
      "parsing compare": 85085,
      "clustering methods": 17948,
      "methods addition": 69295,
      "showing model": 108415,
      "captures information": 15447,
      "unstructured textual": 127588,
      "words emerge": 133932,
      "leading great": 61893,
      "minimum semantic": 70214,
      "units human": 127245,
      "manual construction": 66984,
      "timeconsuming laborintensive": 122146,
      "automatically predict": 10808,
      "predict lexical": 88898,
      "based descriptions": 11637,
      "weakly ordered": 132193,
      "novel soft": 80732,
      "soft loss": 110497,
      "problem experiments": 91044,
      "sememe knowledge": 105487,
      "base hownet": 11460,
      "model beats": 70753,
      "baselines significantly": 12465,
      "subset test": 114837,
      "topic natural": 122554,
      "essential generate": 38551,
      "achieve enhanced": 2454,
      "performance various": 86840,
      "good performances": 48492,
      "enhanced using": 37524,
      "simple sentence": 109516,
      "representation method": 99325,
      "paper inspired": 83983,
      "inspired human": 55570,
      "propose following": 92683,
      "semantic coherence": 105011,
      "method similar": 69151,
      "similar sentences": 109143,
      "close embedding": 17815,
      "identification datasets": 51379,
      "learning graph": 62622,
      "measures present": 67888,
      "structural measures": 113780,
      "measures pairwise": 67884,
      "pairwise node": 83678,
      "node similarities": 80023,
      "representations nodes": 99785,
      "dense space": 29018,
      "userdefined graph": 129064,
      "graph distance": 48795,
      "shortest path": 108303,
      "path distance": 85669,
      "distance distance": 32413,
      "measures information": 67872,
      "information graph": 54646,
      "structure account": 113804,
      "account evaluation": 2024,
      "evaluation proposed": 39338,
      "measures approach": 67852,
      "yields competitive": 135405,
      "outperforming strong": 82833,
      "graph embedding": 48797,
      "embedding baselines": 35380,
      "baselines model": 12429,
      "model computationally": 70875,
      "systems translate": 117206,
      "translate words": 124549,
      "data humans": 26009,
      "words novel": 134085,
      "novel morphological": 80652,
      "morphological variants": 74755,
      "following observation": 44975,
      "general methods": 46678,
      "use subword": 128305,
      "languages varying": 60967,
      "yields consistent": 135408,
      "consistent bleu": 21373,
      "bleu gains": 14284,
      "average 20": 11176,
      "languages especially": 60553,
      "lowresource scenarios": 65558,
      "syntactic hierarchy": 116416,
      "embeddings aims": 35547,
      "aims map": 4550,
      "sense words": 105706,
      "space order": 111035,
      "embeddings domain": 35668,
      "specific data": 111426,
      "data helps": 26000,
      "express concepts": 41697,
      "relevant use": 98598,
      "knowledge embeddings": 57888,
      "propose graph": 92706,
      "graph based": 48764,
      "based embedding": 11670,
      "results shown": 102177,
      "algorithm improves": 4753,
      "building good": 14843,
      "compare neural": 19263,
      "systems traditional": 117200,
      "corpora including": 23502,
      "investigate importance": 56766,
      "special preprocessing": 111366,
      "script presented": 104259,
      "presented results": 89795,
      "based test": 12112,
      "2005 2012": 551,
      "points compared": 87852,
      "tuning model": 125936,
      "quality corpus": 94621,
      "corpus like": 23865,
      "gives substantial": 48210,
      "competitive traditional": 19699,
      "slu model": 110118,
      "labeling new": 58518,
      "corpus target": 24032,
      "corpora consist": 23439,
      "semantic labels": 105092,
      "differences paper": 30966,
      "transferring task": 124245,
      "indomain parallel": 53976,
      "adaptation general": 3225,
      "importantly use": 52302,
      "rl finetune": 102889,
      "receive higher": 96353,
      "rewards evaluate": 102683,
      "approach chinese": 7418,
      "slu systems": 110123,
      "systems experimental": 116890,
      "results generated": 101810,
      "generated english": 47077,
      "slot f1": 110036,
      "accuracy domain": 2140,
      "demonstrates effectiveness": 28942,
      "accuracy relatively": 2259,
      "models investigate": 73427,
      "investigate neural": 56788,
      "interaction lexical": 55950,
      "information task": 55031,
      "task event": 118162,
      "study build": 114330,
      "clauseembedding verbs": 17694,
      "use dataset": 127979,
      "dataset make": 27011,
      "behavior current": 12652,
      "systems showing": 117141,
      "systematic errors": 116671,
      "nonnative speakers": 80247,
      "data especially": 25904,
      "errors make": 38382,
      "make robust": 66721,
      "combination automatic": 18549,
      "errors present": 38399,
      "set spanish": 107587,
      "correction corpus": 24145,
      "corpus allows": 23654,
      "real grammatical": 96065,
      "impressive performance": 52318,
      "improvement new": 52732,
      "postprocessing techniques": 88511,
      "techniques called": 119848,
      "principal components": 90674,
      "performance apply": 86143,
      "proposed postprocessing": 93521,
      "graph clustering": 48768,
      "clustering applications": 17936,
      "applications sense": 7012,
      "induction present": 54029,
      "present detailed": 89443,
      "detailed theoretical": 29787,
      "theoretical computational": 121792,
      "computational analysis": 20358,
      "widely applicable": 132537,
      "clustering discover": 17942,
      "discover clusters": 32114,
      "shows competitive": 108565,
      "applications unsupervised": 7032,
      "induction distributional": 54022,
      "generic applied": 47790,
      "networks linguistic": 77645,
      "data influence": 26038,
      "embedding stability": 35506,
      "stability word": 112107,
      "set recently": 107557,
      "raised concerns": 95459,
      "compare word": 19317,
      "different sizes": 31432,
      "strategies used": 113494,
      "training procedures": 123792,
      "par skipgram": 84521,
      "usage internet": 127862,
      "structured format": 114001,
      "parliamentary data": 84873,
      "provide dataset": 93797,
      "stance classification": 112167,
      "identifying speaker": 51626,
      "thorough manual": 121891,
      "developed annotation": 30251,
      "experiments automatic": 40785,
      "automated classification": 10433,
      "novel english": 80560,
      "news news": 79351,
      "news agencies": 79283,
      "websites world": 132308,
      "processing new": 91723,
      "domains textual": 33874,
      "textual processing": 121725,
      "complex structure": 19882,
      "structure news": 113922,
      "persian paper": 87083,
      "analyze data": 5961,
      "extract entities": 42079,
      "tagging reducing": 117441,
      "reducing gender": 97417,
      "bias abusive": 13790,
      "abusive language": 1844,
      "detection abusive": 29863,
      "detection models": 29994,
      "tend problem": 120159,
      "identity words": 51639,
      "imbalanced training": 51830,
      "trained existing": 123141,
      "model bias": 70775,
      "robust practical": 103061,
      "practical use": 88720,
      "use work": 128369,
      "work measure": 134631,
      "measure gender": 67798,
      "gender biases": 46592,
      "biases models": 13877,
      "language datasets": 58937,
      "datasets analyzing": 27310,
      "analyzing effect": 6040,
      "effect different": 34591,
      "different pretrained": 31348,
      "bias mitigation": 13818,
      "mitigation methods": 70390,
      "effectively reduce": 34842,
      "reduce gender": 97329,
      "scenarios learning": 103856,
      "sentiment modification": 106762,
      "sentiment input": 106747,
      "aligned sentences": 4919,
      "sentences content": 106261,
      "different sentiments": 31418,
      "data hard": 25997,
      "extract sentiment": 42115,
      "independent content": 53762,
      "sentiment unsupervised": 106813,
      "way previous": 132120,
      "strong cues": 113664,
      "friendly propose": 45890,
      "appropriate sentiment": 8437,
      "information learned": 54723,
      "according specific": 2003,
      "specific context": 111420,
      "context experiments": 22095,
      "examine methods": 39751,
      "augmentation textbased": 10312,
      "problem derive": 91004,
      "existing augmentation": 40070,
      "simple data": 109393,
      "augmentation strategy": 10307,
      "randomly replacing": 95531,
      "replacing words": 98953,
      "words source": 134228,
      "different scales": 31398,
      "achieving better": 2934,
      "comparable performances": 19164,
      "performances strong": 86899,
      "strong alternatives": 113619,
      "alternatives word": 5255,
      "word dropout": 133001,
      "sennrich et": 105644,
      "spoken english": 111985,
      "feature scoring": 43305,
      "nonnative english": 80241,
      "involved using": 56886,
      "existing automatic": 40071,
      "asr engine": 9427,
      "engine convert": 37028,
      "speech text": 111824,
      "pattern results": 85719,
      "results highly": 101827,
      "distinct types": 32548,
      "parameters like": 84766,
      "difficult word": 31672,
      "lexical density": 63756,
      "improving results": 53157,
      "string kernels": 113597,
      "arabic dialect": 8503,
      "identification native": 51403,
      "set samples": 107571,
      "approach simple": 7919,
      "selftraining method": 104958,
      "test samples": 120488,
      "confidence scores": 20997,
      "added training": 3368,
      "groundtruth labels": 49130,
      "added test": 3367,
      "instead use": 55691,
      "classifier training": 17588,
      "set report": 107564,
      "report significantly": 99045,
      "accuracy rates": 2257,
      "classification arabic": 17121,
      "identification deep": 51380,
      "deep probabilistic": 28394,
      "unifying framework": 127151,
      "learning emerged": 62539,
      "tasks superior": 119538,
      "learning applicability": 62333,
      "limited reliance": 64270,
      "examples difficult": 39825,
      "labeling functions": 58501,
      "generate noisy": 46977,
      "noisy examples": 80102,
      "examples unlabeled": 39893,
      "strengths limitations": 113575,
      "supervision endtoend": 115881,
      "endtoend modeling": 36930,
      "inference learning": 54161,
      "models label": 73445,
      "decisions latent": 27961,
      "knowledge relations": 58139,
      "using weighted": 130370,
      "end task": 36832,
      "supervision using": 115929,
      "em framework": 35335,
      "supervision methods": 115898,
      "novel combination": 80510,
      "rich domain": 102741,
      "domain linguistic": 33575,
      "knowledge experiments": 57914,
      "experiments biomedical": 40810,
      "demonstrate promise": 28826,
      "promise approach": 92257,
      "approach identifying": 7614,
      "identifying domain": 51595,
      "inevitably fail": 54082,
      "commonly treated": 18965,
      "outofdomain classification": 82647,
      "present comparison": 89408,
      "various baselines": 131050,
      "baselines used": 12486,
      "used solve": 128769,
      "solve propose": 110614,
      "downstream semantic": 34024,
      "using global": 129708,
      "properties semantic": 92480,
      "semantic graphs": 105067,
      "wordnet resources": 133778,
      "layers local": 61788,
      "individual relations": 53928,
      "hypernymy meronymy": 51223,
      "understanding words": 127018,
      "properties entire": 92450,
      "paper combine": 83763,
      "graph models": 48833,
      "novel extension": 80567,
      "graph model": 48831,
      "demonstrate global": 28749,
      "modeling improves": 72446,
      "yielding new": 135380,
      "dataset challenging": 26782,
      "link prediction": 64637,
      "characteristic wellformed": 16517,
      "semantic ontologies": 105129,
      "utterancelevel semantic": 130622,
      "generation generating": 47416,
      "generating semantically": 47259,
      "challenge dialogue": 16024,
      "different conventional": 31064,
      "conventional text": 22905,
      "mapping inputs": 67134,
      "inputs responses": 55494,
      "responses conversations": 101262,
      "inputs outputs": 55488,
      "learn dependency": 62020,
      "model contains": 70907,
      "mapping module": 67139,
      "representations inputs": 99698,
      "module learns": 74497,
      "representations experimental": 99644,
      "responses high": 101274,
      "coherence fluency": 18300,
      "models code": 72915,
      "major research": 66585,
      "like speech": 64096,
      "speech tagging": 111817,
      "summarization evaluation": 115501,
      "evaluation standard": 39404,
      "fscore classification": 45910,
      "evaluating automatic": 39038,
      "detection systems": 30071,
      "reliable metric": 98618,
      "extracting sentiment": 42240,
      "corpus including": 23836,
      "relations document": 98141,
      "annotated sentiments": 6233,
      "mentioned entities": 68435,
      "relations entities": 98146,
      "task experimented": 118178,
      "svm random": 116245,
      "models conventionally": 72986,
      "trained minimizing": 123200,
      "grammatical sentences": 48721,
      "sentences demonstrate": 106273,
      "method aims": 68625,
      "good bad": 48461,
      "widely applied": 132538,
      "text compared": 120804,
      "wer reduction": 132451,
      "10 bleu": 124,
      "unit language": 127206,
      "modeling contextual": 72404,
      "success task": 115131,
      "enables learning": 36391,
      "space generalization": 111006,
      "generalization power": 46791,
      "interactions including": 55986,
      "results wordlevel": 102343,
      "recent stateoftheart": 96517,
      "stateoftheart language": 112682,
      "13 points": 287,
      "points learning": 87862,
      "parameters similar": 84786,
      "number model": 80912,
      "exploit different": 41410,
      "behavior language": 12662,
      "representation paper": 99365,
      "method adversarial": 68623,
      "independent vectors": 53782,
      "specific aspect": 111406,
      "method case": 68690,
      "method capable": 68687,
      "sentence learning": 105920,
      "representation style": 99424,
      "style sentence": 114593,
      "furthermore evaluate": 46167,
      "meaning embeddings": 67634,
      "languages previous": 60811,
      "transitionbased neural": 124509,
      "languages lead": 60681,
      "present evaluation": 89477,
      "27 different": 850,
      "different parameter": 31317,
      "sharing strategies": 108141,
      "languages representing": 60844,
      "pairs related": 83624,
      "classifier parameters": 17569,
      "word andor": 132922,
      "andor character": 6098,
      "character lstm": 16449,
      "varies based": 130947,
      "result propose": 101395,
      "model linguistically": 71465,
      "obtains significant": 81474,
      "unrelated languages": 127502,
      "sharing parameters": 108137,
      "parameters does": 84743,
      "does help": 33356,
      "tags dependency": 117469,
      "parsing provide": 85204,
      "information important": 54668,
      "models fact": 73230,
      "produces large": 92001,
      "baseline using": 12334,
      "embeddings combining": 35600,
      "words frequency": 133964,
      "order systematically": 82416,
      "systematically investigate": 116697,
      "investigate techniques": 56818,
      "quality word": 94828,
      "openclass words": 81954,
      "help disambiguate": 49713,
      "large character": 61045,
      "embedding sizes": 35493,
      "character sets": 16467,
      "using noisy": 129973,
      "noisy channel": 80091,
      "channel model": 16406,
      "fluent sentences": 44719,
      "model scores": 71951,
      "model reranking": 71914,
      "reranking process": 100377,
      "improves stateoftheart": 53048,
      "networks recent": 77722,
      "years natural": 135274,
      "representations various": 99970,
      "learn input": 62070,
      "input representation": 55415,
      "representations derived": 99589,
      "derived output": 29356,
      "preexisting systems": 89234,
      "network acnn": 77124,
      "layer capture": 61704,
      "speech experiments": 111686,
      "task increase": 118288,
      "result task": 101411,
      "language commands": 58891,
      "provides rich": 94069,
      "language environment": 59008,
      "second article": 104387,
      "choose correct": 16925,
      "capture various": 15419,
      "various phenomena": 131168,
      "relational reasoning": 98097,
      "article implemented": 9135,
      "phenomena present": 87243,
      "seqtoseq models": 106911,
      "models incremental": 73392,
      "dialogue investigate": 30696,
      "synthetic dataset": 116625,
      "attention using": 10031,
      "using visualisation": 130362,
      "visualisation diagnostic": 131818,
      "diagnostic classifiers": 30509,
      "incrementally built": 53744,
      "built model": 14927,
      "models develop": 73059,
      "data appears": 25609,
      "model create": 70937,
      "attention patterns": 9968,
      "reliable benchmark": 98611,
      "infrequent word": 55175,
      "models rare": 73855,
      "recently enjoyed": 96687,
      "enjoyed surge": 37544,
      "effective handling": 34686,
      "words play": 134123,
      "comparison techniques": 19582,
      "paper existing": 83907,
      "existing benchmark": 40080,
      "annotations limited": 6442,
      "limited vocabulary": 64299,
      "solid comparison": 110541,
      "framework order": 45637,
      "order evaluation": 82321,
      "evaluation gap": 39218,
      "gap propose": 46474,
      "highly reliable": 50340,
      "challenging benchmark": 16228,
      "representation techniques": 99434,
      "experiments best": 40807,
      "best mainstream": 13362,
      "embeddings millions": 35810,
      "millions words": 70128,
      "dataset annotation": 26739,
      "universal dependency": 127297,
      "general transitionbased": 46723,
      "parser paper": 84980,
      "experiments applying": 40776,
      "conll 2018": 21083,
      "neural transitionbased": 78718,
      "parser use": 85003,
      "enhanced dependencies": 37504,
      "ud trees": 126536,
      "learning code": 62441,
      "empirical success": 36208,
      "approaches recently": 8310,
      "weighted finite": 132348,
      "leading new": 61898,
      "insights work": 55548,
      "work recurrent": 134763,
      "networks share": 77763,
      "formally defining": 45204,
      "fresh view": 45885,
      "modeling text": 72562,
      "approach designing": 7483,
      "attention human": 9856,
      "human rationales": 50943,
      "models successful": 74121,
      "continuous attention": 22611,
      "central hypothesis": 15899,
      "general domains": 46654,
      "resourcerich domains": 100922,
      "domains lowresource": 33810,
      "lowresource ones": 65548,
      "ones model": 81697,
      "domaininvariant representation": 33717,
      "results validate": 102306,
      "approach delivers": 7475,
      "gains stateoftheart": 46409,
      "average error": 11186,
      "reduction benchmark": 97440,
      "release corpus": 98443,
      "human editor": 50795,
      "single contiguous": 109718,
      "collected data": 18414,
      "language generated": 59064,
      "observe standard": 81211,
      "corpora models": 23531,
      "encode different": 36426,
      "trained raw": 123257,
      "text release": 121237,
      "systems wmt18": 117238,
      "submission wmt18": 114740,
      "wmt18 news": 132844,
      "combines neural": 18695,
      "small consistent": 110142,
      "gains strong": 46410,
      "strong transformer": 113724,
      "split rephrase": 111943,
      "task breaking": 117945,
      "sentence shorter": 106071,
      "shorter ones": 108295,
      "convey meaning": 23115,
      "task mining": 118393,
      "split examples": 111941,
      "corpus introduced": 23846,
      "narayan et": 76161,
      "benchmark task": 12862,
      "task incorporating": 118287,
      "produces model": 92003,
      "better predictions": 13674,
      "points prior": 87871,
      "prior best": 90693,
      "models social": 74063,
      "media language": 68123,
      "approaches far": 8162,
      "words way": 134312,
      "higher rate": 50200,
      "tasks effectively": 119080,
      "approach different": 7492,
      "predictive tasks": 89208,
      "tasks spanning": 119510,
      "adaptation significantly": 3267,
      "stateoftheart incorporating": 112674,
      "embeddings new": 35832,
      "morphological phonological": 74719,
      "subword representations": 115030,
      "languages making": 60718,
      "generalization new": 46787,
      "lessresourced languages": 63402,
      "languages challenging": 60436,
      "approaches improving": 8191,
      "languages adapting": 60388,
      "using linguistically": 129819,
      "motivated subword": 74862,
      "corpora bilingual": 23428,
      "gain performance": 46348,
      "performance previous": 86618,
      "methods relying": 69723,
      "effectiveness approaches": 34873,
      "recognition languages": 96901,
      "resource languages": 100858,
      "experiments machine": 40993,
      "translation exploiting": 124816,
      "monolingual setting": 74616,
      "setting achieve": 107729,
      "new applications": 78783,
      "applications methods": 6967,
      "domain expert": 33525,
      "explore extent": 41545,
      "expressed text": 41725,
      "text collection": 120797,
      "expressed natural": 41718,
      "corpus create": 23735,
      "tasks derived": 119050,
      "derived existing": 29348,
      "introduce realistic": 56525,
      "study new": 114454,
      "model built": 70795,
      "built natural": 14929,
      "language entailment": 59007,
      "entailment data": 37657,
      "ones produced": 81705,
      "work provides": 134757,
      "key challenges": 57546,
      "script generation": 104254,
      "define knowledge": 28485,
      "challenges learning": 16172,
      "hierarchical nature": 49985,
      "knowledge example": 57909,
      "type information": 126205,
      "autoencoder model": 10404,
      "space defined": 110989,
      "categorical variables": 15724,
      "vector quantization": 131337,
      "associated latent": 9599,
      "latent hierarchy": 61591,
      "embeddings given": 35733,
      "setting model": 107758,
      "model effectively": 71044,
      "effectively encodes": 34807,
      "recent language": 96467,
      "method standard": 69162,
      "tasks allowing": 118926,
      "achieve substantially": 2579,
      "substantially lower": 114903,
      "lower perplexity": 65445,
      "perplexity scores": 87072,
      "scores compared": 104161,
      "neural text": 78705,
      "generation including": 47437,
      "quite successful": 95406,
      "multiple references": 75661,
      "available references": 11089,
      "multiple human": 75574,
      "human references": 50954,
      "importantly propose": 52299,
      "algorithm generate": 4744,
      "approaches lead": 8210,
      "baselines machine": 12426,
      "strategy proposing": 113535,
      "strategy generating": 113519,
      "brand new": 14565,
      "trains neural": 123967,
      "models endtoend": 73152,
      "endtoend nature": 36936,
      "nature makes": 76659,
      "makes hard": 66786,
      "strategy reinforcement": 113537,
      "solutions paper": 110580,
      "propose modular": 92791,
      "modular approach": 74464,
      "strategy using": 113544,
      "learning reinforcement": 62951,
      "produce diverse": 91885,
      "utterances test": 130669,
      "test approach": 120421,
      "approach recently": 7842,
      "based real": 11971,
      "shows systems": 108639,
      "higher task": 50211,
      "model explainable": 71135,
      "explainable neural": 41267,
      "operations allow": 82068,
      "markers contrast": 67231,
      "modern neural": 74402,
      "explicit word": 41358,
      "practical machine": 88705,
      "improves explainability": 52979,
      "transformer architecture": 124291,
      "identifying sentiment": 51624,
      "novel setting": 80726,
      "continuous sentiment": 22633,
      "approach sentiment": 7878,
      "narrative time": 76182,
      "time provide": 122082,
      "gender associated": 46581,
      "preferences different": 89244,
      "respect previous": 101095,
      "areas research": 8899,
      "research distant": 100476,
      "tagging introduce": 117394,
      "crosslingual neural": 24985,
      "neural partofspeech": 78629,
      "model exploits": 71142,
      "framework approach": 45427,
      "effective resulting": 34741,
      "resulting new": 101457,
      "art access": 9033,
      "gold annotated": 48429,
      "minimal resources": 70175,
      "resources languages": 100995,
      "languages annotated": 60402,
      "resources unsupervised": 101060,
      "unsupervised transfer": 127742,
      "transfer natural": 124161,
      "models namedentity": 73603,
      "languages appealing": 60403,
      "differences words": 30982,
      "languages make": 60716,
      "make challenging": 66630,
      "items languages": 57111,
      "method finds": 68841,
      "based bilingual": 11557,
      "use selfattention": 128259,
      "stateoftheart competitive": 112614,
      "ner performance": 77064,
      "performance commonly": 86218,
      "tested languages": 120573,
      "resource requirements": 100872,
      "past approaches": 85634,
      "approaches evaluate": 8147,
      "challenges applying": 16131,
      "applying methods": 7255,
      "term dependency": 120206,
      "embeddings kind": 35769,
      "major limitations": 66572,
      "common important": 18885,
      "important understanding": 52287,
      "mainly effective": 66483,
      "theoretical foundation": 121794,
      "words enables": 133935,
      "necessity using": 76771,
      "furthermore approach": 46145,
      "systematic evaluation": 116672,
      "shows effectiveness": 108571,
      "comparison state": 19575,
      "qa datasets": 94503,
      "component various": 20012,
      "tasks goal": 119150,
      "map input": 67107,
      "different lengths": 31226,
      "different alphabets": 30988,
      "used sequencetosequence": 128754,
      "mechanism learn": 68007,
      "generation output": 47526,
      "hard monotonic": 49468,
      "monotonic attention": 74643,
      "attention used": 10029,
      "used hard": 128570,
      "xu et": 135225,
      "exponential number": 41678,
      "hard attention": 49452,
      "model compare": 70852,
      "soft hard": 110492,
      "exact algorithm": 39714,
      "translation englishfrench": 124792,
      "analysis error": 5597,
      "pronouns longstanding": 92357,
      "longstanding challenge": 65192,
      "study performance": 114466,
      "rulebased statistical": 103373,
      "based extensive": 11707,
      "suite enables": 115414,
      "light difficulties": 63988,
      "difficulties task": 31689,
      "systems exhibit": 116882,
      "exhibit significant": 40004,
      "lack awareness": 58686,
      "recent transformerbased": 96552,
      "crosssentence context": 25070,
      "shows promising": 108611,
      "considerable room": 21258,
      "crosssentence dependencies": 25071,
      "structures introduce": 114079,
      "avoid expensive": 11235,
      "improve strong": 52552,
      "semantics frame": 105417,
      "semantics coreference": 105403,
      "resolution achieving": 100750,
      "model interpretable": 71363,
      "processing greatly": 91675,
      "introduction attention": 56654,
      "standard attention": 112194,
      "limited interpretability": 64242,
      "inference steps": 54227,
      "model constructs": 70905,
      "incremental representations": 53739,
      "input easily": 55326,
      "easily interpretable": 34458,
      "interpretable manner": 56242,
      "performance close": 86208,
      "embeddings target": 35971,
      "classifiers neural": 17620,
      "models leads": 73477,
      "leads faster": 61932,
      "training better": 123376,
      "quality given": 94678,
      "parameters particular": 84777,
      "propose structureaware": 93089,
      "layer captures": 61705,
      "captures semantic": 15453,
      "output space": 83124,
      "words joint": 134018,
      "model generalized": 71229,
      "generalized form": 46827,
      "allows learning": 5168,
      "model shares": 72003,
      "allows better": 5128,
      "better leverage": 13618,
      "leverage prior": 63616,
      "datasets shows": 27709,
      "method strong": 69164,
      "encoderdecoder baselines": 36595,
      "baselines trained": 12483,
      "extracting keywords": 42216,
      "survey data": 116166,
      "basis research": 12547,
      "data generally": 25973,
      "answers multiplechoice": 6734,
      "multiplechoice questions": 75757,
      "data allow": 25592,
      "allow new": 5088,
      "valuable source": 130770,
      "essential linguistic": 38555,
      "linguistic nature": 64513,
      "social entities": 110337,
      "present computational": 89416,
      "responses openended": 101287,
      "openended questions": 81988,
      "extract keywords": 42093,
      "need perform": 76846,
      "systems performs": 117065,
      "performs task": 87034,
      "potential biases": 88540,
      "data science": 26377,
      "using multiscale": 129922,
      "cues multiple": 25215,
      "multiple modalities": 75616,
      "cues separate": 25217,
      "rnn architecture": 102901,
      "used incorporate": 128586,
      "gaze features": 46561,
      "informal texts": 54344,
      "alternative uses": 5249,
      "uses neural": 129254,
      "neural parsers": 78627,
      "tree model": 125602,
      "parsing converts": 85088,
      "converts natural": 23108,
      "language utterance": 60318,
      "machine interpretable": 65761,
      "interpretable meaning": 56243,
      "joint representation": 57316,
      "neural component": 77865,
      "component model": 19986,
      "inference extensive": 54142,
      "dataset languages": 26995,
      "performance languages": 86485,
      "representation contextual": 99197,
      "estimation task": 38667,
      "wordlevel quality": 133747,
      "consists taking": 21498,
      "taking source": 117552,
      "translation predicting": 125127,
      "words output": 134098,
      "output correct": 83063,
      "correct wrong": 24130,
      "effectively encode": 34806,
      "encode local": 36437,
      "tags languages": 117479,
      "languages second": 60857,
      "integrate local": 55761,
      "feedforward recurrent": 43853,
      "sentence making": 105938,
      "achieves strong": 2909,
      "dual conditional": 34228,
      "conditional crossentropy": 20753,
      "crossentropy filtering": 24912,
      "filtering noisy": 44085,
      "noisy parallel": 80114,
      "introduce dual": 56415,
      "data sentence": 26394,
      "corpus compute": 23710,
      "scores results": 104200,
      "higher bleu": 50166,
      "scores models": 104187,
      "method context": 68740,
      "task parallel": 118512,
      "achieve overall": 2507,
      "investigative journalism": 56869,
      "introduce advanced": 56378,
      "pipeline automatically": 87529,
      "collections unstructured": 18506,
      "data purpose": 26307,
      "serves new": 107316,
      "input processor": 55404,
      "german news": 47904,
      "containing unknown": 21810,
      "based automatic": 11536,
      "extraction entities": 42323,
      "documents contrast": 33213,
      "following major": 44972,
      "composition multiple": 20100,
      "multiple stateoftheart": 75703,
      "support multilingual": 115978,
      "40 languages": 1006,
      "entities various": 37894,
      "customer support": 25448,
      "agents chatbots": 4320,
      "advances neural": 4002,
      "models transformer": 74241,
      "applied various": 7140,
      "research directions": 100473,
      "cases feasible": 15645,
      "retrieval techniques": 102433,
      "model ii": 71298,
      "transformer experiments": 124324,
      "experiments twitter": 41188,
      "terms semantics": 120379,
      "word overlap": 133389,
      "text foreign": 120959,
      "language evaluation": 59018,
      "evaluation usefulness": 39433,
      "questions language": 95324,
      "certain words": 15964,
      "words removed": 134169,
      "using machinetranslated": 129851,
      "machinetranslated text": 66350,
      "translations multiple": 125476,
      "multiple mt": 75620,
      "document context": 32979,
      "main findings": 66422,
      "clearly identify": 17738,
      "vary widely": 131250,
      "systems hard": 116928,
      "support use": 115995,
      "conventional method": 22888,
      "text length": 121086,
      "contextual neural": 22483,
      "works neural": 134964,
      "translation translating": 125395,
      "conversations explore": 23055,
      "explore neural": 41566,
      "exploit source": 41441,
      "source targetside": 110848,
      "task introduce": 118307,
      "datasets extracted": 27480,
      "europarl v7": 38765,
      "bleu manual": 14290,
      "translation noisy": 125065,
      "noisy text": 80125,
      "text noisy": 121147,
      "modern machine": 74395,
      "systems growing": 116926,
      "growing research": 49181,
      "systems publicly": 117089,
      "noisy inputs": 80107,
      "work resorted": 134784,
      "created datasets": 24662,
      "dataset machine": 27005,
      "consisting noisy": 21457,
      "noisy comments": 80095,
      "professionally sourced": 92087,
      "sourced translations": 110873,
      "translations english": 125461,
      "japanese french": 57190,
      "comments english": 18812,
      "qualitatively quantitatively": 94580,
      "quantitatively examine": 94890,
      "included dataset": 53245,
      "demonstrate existing": 28739,
      "performing adaptation": 86942,
      "data indicates": 26034,
      "dataset provide": 27126,
      "handling noisy": 49417,
      "text mt": 121132,
      "augmentation neural": 10292,
      "generalization ability": 46761,
      "ability models": 1545,
      "investigate data": 56737,
      "different standard": 31448,
      "augmentation techniques": 10311,
      "techniques method": 119931,
      "synthesized data": 116607,
      "network machine": 77308,
      "comprehension model": 20197,
      "approaches perspectives": 8281,
      "training applied": 123360,
      "applied target": 7123,
      "model inputs": 71348,
      "embeddings control": 35618,
      "adversarial perturbations": 4158,
      "original target": 82549,
      "variables training": 130860,
      "training improves": 123657,
      "improves robustness": 53040,
      "robustness generalization": 103100,
      "generalization model": 46780,
      "second propose": 104446,
      "propose multilayer": 92797,
      "multilayer attention": 75168,
      "interaction question": 55964,
      "question passage": 95196,
      "reasonable representation": 96212,
      "understanding model": 126889,
      "combining contributions": 18714,
      "contributions enhance": 22787,
      "enhance diversity": 37466,
      "information extracting": 54561,
      "extracting ability": 42199,
      "model time": 72170,
      "models improving": 73373,
      "understanding causal": 126808,
      "mental health": 68409,
      "manual identification": 67003,
      "personal writing": 87125,
      "alternative expensive": 5229,
      "expensive manual": 40424,
      "door new": 33951,
      "climate change": 17758,
      "explore automating": 41523,
      "building discourse": 14838,
      "presenting novel": 89809,
      "novel subtasks": 80739,
      "causality detection": 15816,
      "identification identifying": 51389,
      "achieve strong": 2573,
      "tasks different": 119060,
      "finally explore": 44185,
      "explore applications": 41519,
      "word sentiment": 133545,
      "sentiment change": 106689,
      "used causal": 128419,
      "responses neural": 101285,
      "models distributional": 73098,
      "constraints neural": 21605,
      "tend generate": 120150,
      "generate safe": 47004,
      "generic responses": 47808,
      "decoding objectives": 28116,
      "tasks diverse": 119069,
      "incorporating information": 53542,
      "form distributional": 45083,
      "constraints generated": 21599,
      "help generate": 49729,
      "generate content": 46918,
      "responses based": 101256,
      "arora et": 9019,
      "2016 evaluate": 606,
      "approach variety": 8012,
      "competitive baselines": 19639,
      "baselines using": 12487,
      "approach generates": 7587,
      "mechanism called": 67955,
      "predict text": 88945,
      "data represented": 26341,
      "accurate representation": 2357,
      "representation nodes": 99359,
      "neighborhood information": 76990,
      "information better": 54398,
      "better reflect": 13691,
      "method presented": 69067,
      "presented used": 89802,
      "data graphs": 25987,
      "method apply": 68640,
      "translation morphology": 124975,
      "comparison analysis": 19525,
      "nmt lowresource": 79887,
      "difficult data": 31617,
      "used help": 128571,
      "help reduce": 49771,
      "reduce sparsity": 97359,
      "compared bpe": 19341,
      "improvements lowresource": 52869,
      "lowresource data": 65491,
      "translate languages": 124529,
      "multiway parallel": 75940,
      "cases using": 15671,
      "performs best": 86983,
      "words best": 133837,
      "semantics use": 105477,
      "generation training": 47684,
      "component conversational": 19972,
      "process producing": 91553,
      "quality responses": 94776,
      "domains knowledge": 33795,
      "knowledge little": 58056,
      "neural generators": 77924,
      "experiments models": 41021,
      "responses different": 101264,
      "different style": 31459,
      "input meaning": 55362,
      "trained achieve": 123065,
      "target produce": 117693,
      "produce outputs": 91916,
      "carefully evaluate": 15523,
      "training style": 123889,
      "learned models": 62231,
      "models simply": 74053,
      "parameters produce": 84779,
      "segmentation morphological": 104603,
      "context entire": 22079,
      "entire sentence": 37711,
      "propose structured": 93090,
      "prediction framework": 89064,
      "model adopt": 70638,
      "adopt approaches": 3890,
      "approaches generally": 8171,
      "mcdonald et": 67581,
      "percentage improvement": 85921,
      "sequential labelling": 107232,
      "plays important": 87730,
      "drawing conclusions": 34127,
      "results nlp": 101993,
      "nlp papers": 79658,
      "statistical tests": 113172,
      "common tasks": 18933,
      "attention dialogue": 9818,
      "modeling spoken": 72549,
      "slu essential": 110113,
      "contexts provide": 22422,
      "provide informative": 93853,
      "cues better": 25210,
      "contextual slu": 22499,
      "paid attention": 83414,
      "attention related": 9990,
      "content history": 21887,
      "history utterances": 50573,
      "recent utterances": 96560,
      "utterances important": 130648,
      "important recent": 52223,
      "recent ones": 96485,
      "timeaware attention": 122142,
      "attention function": 9843,
      "based content": 11600,
      "roles contexts": 103230,
      "demonstrates remarkable": 28960,
      "dialogue contexts": 30651,
      "contexts experiments": 22392,
      "benchmark dialogue": 12814,
      "challenge dstc4": 16028,
      "dstc4 dataset": 34216,
      "contextual understanding": 22507,
      "understanding performance": 126918,
      "sentence generation": 105888,
      "prediction major": 89075,
      "generation large": 47456,
      "size equal": 109922,
      "improve efficiency": 52378,
      "prediction method": 89077,
      "method predicts": 69062,
      "input generate": 55344,
      "used supervised": 128794,
      "supervised reinforcement": 115824,
      "learning steps": 63057,
      "captioning datasets": 15258,
      "datasets method": 27564,
      "gpu memory": 48580,
      "equal better": 38222,
      "better baselines": 13518,
      "personalized dialogue": 87143,
      "current dialogue": 25276,
      "zhang et": 135537,
      "2018 showed": 684,
      "engagement level": 37015,
      "endtoend dialogue": 36890,
      "models increases": 73389,
      "providing personalized": 94129,
      "dataset used": 27255,
      "limited size": 64278,
      "different personas": 31328,
      "dataset providing": 27131,
      "performance endtoend": 86339,
      "coverage dataset": 24499,
      "dataset finetuning": 26938,
      "finetuning model": 44478,
      "achieving stateoftheart": 2984,
      "models universal": 74271,
      "parsing pipeline": 85193,
      "consisting components": 21449,
      "second predicts": 104443,
      "partof speech": 85522,
      "speech tags": 111818,
      "predicts dependency": 89215,
      "tags instead": 117478,
      "training single": 123866,
      "multiple treebanks": 75738,
      "language closely": 58880,
      "greatly reducing": 49064,
      "models official": 73660,
      "ranked 7th": 95641,
      "obtained best": 81350,
      "overall word": 83270,
      "segmentation universal": 104646,
      "universal pos": 127321,
      "features data": 43435,
      "augmentation spoken": 10303,
      "generation data": 47351,
      "main obstacles": 66448,
      "slu datasets": 110111,
      "datasets recent": 27661,
      "text generative": 121022,
      "models variational": 74318,
      "autoencoder vae": 10410,
      "generating plausible": 47247,
      "natural sentences": 76620,
      "novel generative": 80582,
      "architecture leverages": 8685,
      "generative power": 47749,
      "fully annotated": 45941,
      "utterances experiments": 130638,
      "experiments existing": 40943,
      "existing slu": 40286,
      "trained additional": 123066,
      "additional synthetic": 3579,
      "synthetic examples": 116627,
      "examples achieve": 39805,
      "approach helps": 7601,
      "helps alleviate": 49809,
      "scarcity issue": 103805,
      "slu task": 110124,
      "task datasets": 118048,
      "various slu": 131198,
      "considers single": 21343,
      "task partofspeech": 118522,
      "languages work": 60974,
      "require manually": 100177,
      "tasks investigated": 119225,
      "linguistic insights": 64495,
      "insights gained": 55533,
      "models textual": 74188,
      "sentence like": 105937,
      "demographic groups": 28649,
      "model higherorder": 71284,
      "rates compared": 95817,
      "enable new": 36360,
      "rely discourse": 98687,
      "discourse understanding": 32097,
      "quantitative text": 94886,
      "grounded word": 49116,
      "words images": 133988,
      "embeddings word2vec": 36037,
      "learned embeddings": 62213,
      "real number": 96073,
      "learn grounded": 62059,
      "embeddings representing": 35913,
      "learned model": 62230,
      "indicates model": 53862,
      "model interpretation": 71364,
      "methods explain": 69492,
      "input feature": 55336,
      "networks robust": 77751,
      "measure model": 67807,
      "model uncertainty": 72241,
      "importance input": 52061,
      "behavior neural": 12665,
      "algorithm provides": 4790,
      "provides robust": 94071,
      "use generate": 128066,
      "align human": 4900,
      "interpretation method": 56265,
      "annotation artifacts": 6278,
      "artifacts neural": 9242,
      "learn sentence": 62143,
      "responses taskoriented": 101300,
      "multiple propositions": 75652,
      "use sentence": 128265,
      "best option": 13384,
      "content related": 21931,
      "combines multiple": 18694,
      "planning surface": 87644,
      "endtoend learning": 36911,
      "shown neural": 108496,
      "content single": 21947,
      "training systematically": 123899,
      "corpora exhibit": 23478,
      "particular sentence": 85450,
      "compare models": 19260,
      "provide explicit": 93822,
      "training models": 123713,
      "operations generalize": 82070,
      "parseme shared": 84902,
      "multiword expression": 75943,
      "adapted task": 3294,
      "expression mwe": 41740,
      "mwe identification": 75992,
      "identification employ": 51384,
      "employ neural": 36272,
      "crf layer": 24750,
      "participated open": 85338,
      "open track": 81942,
      "track parseme": 122732,
      "use pretrained": 128203,
      "embeddings outperformed": 35847,
      "participating systems": 85356,
      "performance unseen": 86823,
      "unseen data": 127519,
      "improvements evaluation": 52851,
      "induction recent": 54031,
      "recent attempts": 96430,
      "induction systems": 54035,
      "2018 modern": 665,
      "shown accurate": 108441,
      "accurate early": 2331,
      "complexity grows": 19913,
      "work instead": 134573,
      "johnson et": 57245,
      "increasing accuracy": 53657,
      "accuracy resulting": 2264,
      "chinese german": 16772,
      "new inference": 78955,
      "technique able": 119760,
      "able produce": 1674,
      "competitively stateoftheart": 19703,
      "greedy search": 49074,
      "probabilistic ngram": 90825,
      "ngram matching": 79444,
      "trained wordlevel": 123340,
      "loss using": 65305,
      "using teacher": 130270,
      "teacher forcing": 119694,
      "evaluates translation": 39031,
      "suffers exposure": 115256,
      "exposure bias": 41693,
      "mitigate problems": 70377,
      "high variance": 50145,
      "based probabilistic": 11955,
      "addition method": 3449,
      "search training": 104359,
      "training uses": 123943,
      "uses predicted": 129264,
      "inference alleviate": 54114,
      "problem exposure": 91047,
      "improvement 15": 52668,
      "tagging lemmatization": 117397,
      "lemmatization using": 63342,
      "lstmbased neural": 65685,
      "additionally generate": 3608,
      "network jointly": 77292,
      "lemmas partofspeech": 63332,
      "segmentation tokenization": 104643,
      "demonstrate viability": 28907,
      "viability proposed": 131639,
      "multitask architecture": 75810,
      "performance remains": 86670,
      "stateoftheart multilingual": 112760,
      "extractive reading": 42558,
      "translation despite": 124745,
      "comprehension rc": 20223,
      "progress limited": 92158,
      "rc training": 95876,
      "data english": 25893,
      "leverages existing": 63645,
      "existing rc": 40262,
      "rc model": 95871,
      "model translate": 72233,
      "language obtain": 59723,
      "answer using": 6590,
      "language finally": 59046,
      "corresponding answer": 24275,
      "original language": 82529,
      "attention scores": 10000,
      "scores nmt": 104189,
      "create evaluation": 24613,
      "rc data": 95869,
      "data nonenglish": 26179,
      "nonenglish languages": 80196,
      "languages japanese": 60654,
      "method experimental": 68823,
      "baseline stateoftheart": 12314,
      "evaluated common": 38962,
      "common dataset": 18865,
      "models english": 73153,
      "rarely explored": 95759,
      "indonesian paper": 53997,
      "explored various": 41638,
      "including rulebased": 53362,
      "crf neural": 24754,
      "score achieved": 104051,
      "achieved recurrent": 2678,
      "standard future": 112243,
      "work release": 134771,
      "dataset split": 27214,
      "neural latent": 77958,
      "relations vector": 98278,
      "space capturing": 110977,
      "capturing semantic": 15485,
      "words vector": 134304,
      "tasks promising": 119403,
      "patternbased approach": 85723,
      "patternbased models": 85725,
      "models experimental": 73197,
      "results measuring": 101911,
      "measuring relational": 67928,
      "similarity demonstrate": 109221,
      "addition combined": 3405,
      "additional semantic": 3568,
      "semantic relational": 105207,
      "filling missing": 44046,
      "recognizing lexical": 97070,
      "relations recognizing": 98246,
      "pairs important": 83560,
      "task applications": 117882,
      "mainstream approaches": 66498,
      "approaches task": 8369,
      "task exploit": 118183,
      "reflect semantic": 97610,
      "pairs method": 83583,
      "law states": 61687,
      "rarely paper": 95761,
      "methods neural": 69636,
      "problem proposed": 91185,
      "cooccur corpus": 23221,
      "corpus extract": 23797,
      "capturing relational": 15484,
      "pairs experimental": 83539,
      "previous neural": 90435,
      "focused problem": 44873,
      "inferences including": 54252,
      "models hmms": 73338,
      "framework supports": 45706,
      "previous clustering": 90399,
      "develop algorithm": 30170,
      "structure parameter": 113929,
      "based expectation": 11696,
      "expectation maximization": 40382,
      "maximization evaluate": 67506,
      "evaluate number": 38879,
      "number natural": 80918,
      "results algorithm": 101502,
      "alignment problem": 4981,
      "problem stateoftheart": 91247,
      "multihead attention": 75122,
      "based transformer": 12127,
      "architecture demonstrate": 8643,
      "transformer models": 124346,
      "models improved": 73368,
      "attention component": 9805,
      "used compute": 128442,
      "weights use": 132384,
      "performance study": 86763,
      "effect adding": 34584,
      "task user": 118828,
      "user wants": 129057,
      "38 bleu": 979,
      "translation factor": 124822,
      "carry experiments": 15552,
      "task assessing": 117900,
      "composition sentence": 20104,
      "sentence vector": 106123,
      "representations produced": 99827,
      "sentence composition": 105795,
      "composition models": 20099,
      "method address": 68619,
      "challenge developing": 16022,
      "tasks directly": 119065,
      "enable creation": 36350,
      "annotated sentence": 6228,
      "meeting specified": 68255,
      "lexical constraints": 63746,
      "applying method": 7254,
      "information embeddings": 54520,
      "embeddings number": 35837,
      "number existing": 80885,
      "models method": 73563,
      "method able": 68579,
      "implications results": 51995,
      "results respect": 102133,
      "systems capturing": 116769,
      "used experiments": 128533,
      "generation evaluating": 47391,
      "task current": 118038,
      "current automatic": 25262,
      "model measure": 71509,
      "human written": 50992,
      "written sentences": 135147,
      "sentences fail": 106314,
      "evaluation taking": 39413,
      "advantage model": 4051,
      "model utilize": 72287,
      "finegrained semantic": 44377,
      "meanings word": 67751,
      "specific contexts": 111421,
      "representation help": 99249,
      "dependency words": 29253,
      "build largescale": 14779,
      "results dataset": 101656,
      "outperforms competitive": 82867,
      "embedding textual": 35517,
      "relation text": 98074,
      "hypothesis recent": 51272,
      "methods achieved": 69285,
      "interactions premisehypothesis": 55995,
      "similarity evaluation": 109230,
      "premise hypothesis": 89285,
      "hypothesis paper": 51269,
      "way enhance": 132074,
      "enhance existing": 37468,
      "entailment algorithms": 37652,
      "algorithms using": 4888,
      "embeddings experimental": 35708,
      "embeddings significantly": 35937,
      "entailment models": 37667,
      "model 21": 70499,
      "accuracy improvement": 2181,
      "improvement prior": 52751,
      "left right": 63299,
      "2018 recently": 675,
      "action sequences": 3079,
      "strong generalization": 113678,
      "generalization abilities": 46760,
      "recurrent sequencetosequence": 97259,
      "suggested models": 115344,
      "lack ability": 58676,
      "closer look": 17874,
      "does capture": 33333,
      "propose complementary": 92588,
      "dataset requires": 27159,
      "closely aligned": 17849,
      "learning extractive": 62576,
      "summarisation biomedical": 115467,
      "biomedical data": 14170,
      "text summarisation": 121343,
      "biomedical publications": 14202,
      "pressing need": 89957,
      "plethora information": 87752,
      "impact supervised": 51894,
      "multidocument summaries": 75077,
      "summaries given": 115442,
      "particular compare": 85393,
      "compare classification": 19229,
      "classification regression": 17372,
      "regression approaches": 97692,
      "querybased extractive": 94998,
      "extractive summarisation": 42563,
      "bioasq challenge": 14148,
      "tackled problem": 117317,
      "training classification": 123382,
      "simple annotation": 109359,
      "annotation approach": 6277,
      "endtoend language": 36910,
      "systems making": 116995,
      "types sentences": 126357,
      "automatically label": 10791,
      "experiments neural": 41035,
      "neural generator": 77923,
      "training different": 123586,
      "controlled training": 22848,
      "style generated": 114579,
      "generated utterances": 47151,
      "utterances using": 130674,
      "used existing": 128530,
      "existing large": 40157,
      "corpus showing": 23995,
      "vary terms": 131248,
      "semantic quality": 105191,
      "memory dialogue": 68310,
      "generation traditional": 47681,
      "traditional generative": 122821,
      "information generating": 54637,
      "generating specific": 47264,
      "certain query": 15953,
      "generation informative": 47440,
      "utterances recently": 130663,
      "researchers attempted": 100678,
      "information gap": 54630,
      "exploiting information": 41467,
      "techniques generating": 119894,
      "dialogues retrieved": 30842,
      "entire training": 37721,
      "data considered": 25774,
      "performance paper": 86594,
      "framework exploits": 45537,
      "query response": 94984,
      "generation experimental": 47398,
      "results approaches": 101521,
      "approaches significantly": 8341,
      "improve diversity": 52374,
      "diversity informativeness": 32880,
      "informativeness generated": 55165,
      "risk factor": 102876,
      "patient health": 85700,
      "health records": 49627,
      "critically important": 24846,
      "important difficult": 52142,
      "clinical narratives": 17778,
      "electronic health": 35250,
      "records ehrs": 97136,
      "range topics": 95611,
      "robust interpretable": 103041,
      "topic extraction": 122515,
      "extraction component": 42284,
      "created data": 24660,
      "data pipeline": 26237,
      "using document": 129602,
      "metrics perform": 69990,
      "perform topic": 86094,
      "ehr data": 35217,
      "readmission risk": 96042,
      "extraction model": 42390,
      "model identify": 71295,
      "written news": 135140,
      "depending time": 29275,
      "time training": 122127,
      "text pretrained": 121198,
      "sliding window": 110013,
      "fast flexible": 43137,
      "learn follow": 62051,
      "learning follow": 62588,
      "human instructions": 50866,
      "task particularly": 118520,
      "employed language": 36302,
      "handful examples": 49365,
      "manually engineered": 67058,
      "engineered features": 37041,
      "features provide": 43678,
      "strong inductive": 113683,
      "make learning": 66694,
      "knowledge acquired": 57741,
      "procedure slow": 91401,
      "learning stage": 63050,
      "learns general": 63211,
      "structure task": 113969,
      "task fast": 118204,
      "fast online": 43148,
      "online adaptation": 81741,
      "language new": 59716,
      "model adapts": 70626,
      "new vocabulary": 79245,
      "human speakers": 50965,
      "speakers language": 111320,
      "language usage": 60298,
      "artificial training": 9267,
      "network make": 77311,
      "automatically acquired": 10711,
      "unsupervised sense": 127713,
      "extraction present": 42441,
      "method extracting": 68834,
      "sets synonyms": 107715,
      "evaluation gold": 39227,
      "method successfully": 69170,
      "hearst patterns": 49644,
      "analysis properties": 5777,
      "investigate type": 56820,
      "general information": 46663,
      "order using": 82425,
      "does contain": 33337,
      "structure level": 113898,
      "higher order": 50191,
      "information transfer": 55054,
      "transfer multitask": 124159,
      "interpretation paper": 56272,
      "learning challenging": 62424,
      "experiments indepth": 40963,
      "indepth error": 53797,
      "analysis transfer": 5907,
      "parameter initialization": 84715,
      "help neural": 49757,
      "neural classification": 77859,
      "demonstrate dual": 28703,
      "relations set": 98254,
      "exploited improve": 41452,
      "overall accuracy": 83210,
      "accuracy neural": 2214,
      "neural classifier": 77861,
      "relations better": 98116,
      "existing encoderdecoder": 40125,
      "effectively modeling": 34832,
      "modeling coherence": 72394,
      "output diversity": 83069,
      "introduce measure": 56453,
      "embedding similarity": 35491,
      "similarity dialogue": 109222,
      "generated response": 47114,
      "filter training": 44076,
      "corpora based": 23422,
      "based measure": 11834,
      "lexically diverse": 63858,
      "response generator": 101215,
      "generator using": 47784,
      "opensubtitles corpus": 82036,
      "improvement competitive": 52697,
      "competitive neural": 19657,
      "filtering task": 44088,
      "presents nicts": 89876,
      "nicts participation": 79478,
      "participation wmt18": 85369,
      "wmt18 shared": 132847,
      "words germanenglish": 133973,
      "corpus crawled": 23734,
      "using clean": 129443,
      "task designed": 118071,
      "systems empirical": 116862,
      "results nmt": 101996,
      "achieve promising": 2515,
      "models latent": 73468,
      "generate generic": 46947,
      "propose latent": 92742,
      "seq2seq neural": 106906,
      "make training": 66740,
      "training easier": 123596,
      "neural topic": 78710,
      "sentence build": 105782,
      "global topic": 48278,
      "representation learnt": 99314,
      "contribute better": 22746,
      "models languages": 73456,
      "generate diverse": 46931,
      "diverse interesting": 32818,
      "subjective human": 114693,
      "overall preferred": 83251,
      "representations nlp": 99781,
      "interpretable representations": 56251,
      "gumbel softmax": 49276,
      "representations outperform": 99792,
      "outperform naive": 82720,
      "corpus search": 23981,
      "search terms": 104356,
      "terms target": 120389,
      "highlight differences": 50258,
      "dataset document": 26879,
      "conversations paper": 23063,
      "conversations define": 23053,
      "conversations average": 23049,
      "provide relevant": 93906,
      "chat history": 16657,
      "provide source": 93924,
      "architectures provide": 8840,
      "provide benchmark": 93766,
      "benchmark performance": 12847,
      "fluent responses": 44717,
      "type classification": 126190,
      "classification capabilities": 17140,
      "based type": 12143,
      "discourse functions": 32057,
      "depends heavily": 29286,
      "contexts propose": 22421,
      "representations predicting": 99811,
      "model read": 71854,
      "modeling context": 72403,
      "humanlevel performance": 51050,
      "generating english": 47216,
      "acquisition information": 3047,
      "extraction existing": 42331,
      "transliteration generation": 125523,
      "generation require": 47600,
      "easier task": 34424,
      "present bootstrapping": 89397,
      "improve generation": 52391,
      "generation used": 47690,
      "evaluate transliteration": 38938,
      "generation performance": 47541,
      "crosslingual candidate": 24932,
      "generation entity": 47389,
      "entity linking": 37953,
      "typical downstream": 126398,
      "evaluation approach": 39120,
      "languages written": 60985,
      "building conversation": 14829,
      "systems existing": 116884,
      "sequence utterances": 107095,
      "utterances responses": 130665,
      "resulted development": 101427,
      "sequencetosequence generation": 107160,
      "given sequence": 48120,
      "generate response": 47001,
      "way humans": 132089,
      "humans converse": 51074,
      "opposed simply": 82148,
      "simply relying": 109626,
      "previous sequence": 90461,
      "produce utterances": 91945,
      "topic facilitate": 122516,
      "mimic human": 70132,
      "human process": 50935,
      "models pure": 73846,
      "models ignore": 73351,
      "information background": 54391,
      "knowledge required": 58149,
      "prediction based": 89037,
      "models predict": 73764,
      "predict appropriate": 88874,
      "conversational ai": 22991,
      "paper surveys": 84458,
      "developed years": 30326,
      "progress challenges": 92143,
      "using specific": 130204,
      "specific systems": 111494,
      "systems models": 117004,
      "models case": 72889,
      "crossview training": 25090,
      "training unsupervised": 123941,
      "accuracy supervised": 2295,
      "learn taskspecific": 62163,
      "taskspecific labeled": 119630,
      "main training": 66473,
      "training phase": 123780,
      "representations bilstm": 99522,
      "encoder using": 36586,
      "labeled unlabeled": 58475,
      "data labeled": 26064,
      "standard supervised": 112313,
      "sentence match": 105943,
      "modules model": 74526,
      "model share": 72000,
      "intermediate representations": 56140,
      "representations turn": 99949,
      "improves model": 52994,
      "particularly effective": 85481,
      "learning evaluate": 62555,
      "embeddings previous": 35881,
      "sparse representations": 111234,
      "learned existing": 62217,
      "existing dense": 40109,
      "dense embeddings": 29013,
      "recognizable feature": 97049,
      "transfer idea": 124061,
      "explore approaches": 41521,
      "sparse representation": 111233,
      "methods observe": 69646,
      "increase interpretability": 53601,
      "interpretability compared": 56221,
      "compared dense": 19356,
      "scene descriptions": 103876,
      "coco dataset": 18066,
      "monolingual sentence": 74612,
      "matching text": 67437,
      "work improves": 134563,
      "alignment text": 4993,
      "specifically text": 111596,
      "wikipedia introduce": 132659,
      "introduce convolutional": 56397,
      "network structure": 77446,
      "structure model": 113906,
      "model similarity": 72041,
      "semisupervised way": 105634,
      "knowledgebased method": 58252,
      "graph information": 48821,
      "information aggregation": 54371,
      "practical utility": 88723,
      "utility natural": 130485,
      "world common": 135022,
      "sentence extracting": 105868,
      "works modeling": 134963,
      "sequential modeling": 107238,
      "low efficiency": 65360,
      "jointly extract": 57342,
      "multiple event": 75559,
      "event triggers": 39553,
      "triggers arguments": 125792,
      "introducing syntactic": 56651,
      "enhance information": 37471,
      "graph convolution": 48776,
      "model graph": 71265,
      "information experiment": 54548,
      "framework achieves": 45408,
      "fast simple": 43152,
      "despite known": 29699,
      "memory computational": 68303,
      "unleash power": 127421,
      "coding schemes": 18245,
      "reduce vocabulary": 97366,
      "encoding mechanisms": 36710,
      "memory consumption": 68307,
      "cider score": 16978,
      "04 bleu": 24,
      "respectively achieving": 101120,
      "englishtogerman task": 37448,
      "predictive embeddings": 89202,
      "hate speech": 49530,
      "speech detection": 111671,
      "detection twitter": 30087,
      "approach classifying": 7421,
      "online hate": 81776,
      "racist sexist": 95431,
      "able predict": 1671,
      "occurrence hate": 81510,
      "speech commonly": 111660,
      "datasets models": 27577,
      "match outperform": 67361,
      "outperform state": 82747,
      "f1 performance": 42667,
      "datasets using": 27782,
      "using significantly": 130170,
      "minimal feature": 70159,
      "largescale multidomain": 61469,
      "dialogue modelling": 30711,
      "learning major": 62703,
      "dialogue research": 30732,
      "scale data": 103709,
      "available address": 10937,
      "dataset multiwoz": 27046,
      "conversations spanning": 23070,
      "domains topics": 33875,
      "magnitude larger": 66384,
      "larger previous": 61377,
      "dataset labelled": 26992,
      "dialogue belief": 30644,
      "belief states": 12691,
      "states dialogue": 113038,
      "description data": 29458,
      "collection procedure": 18484,
      "summary data": 115635,
      "annotators secondly": 6499,
      "set benchmark": 107377,
      "benchmark results": 12854,
      "tracking dialogue": 122753,
      "sets baseline": 107651,
      "future studies": 46314,
      "study empirically": 114367,
      "referring expressions": 97575,
      "specifically study": 111591,
      "study ability": 114304,
      "choosing right": 16931,
      "human accuracy": 50736,
      "produced generation": 91957,
      "algorithm especially": 4727,
      "novel natural": 80661,
      "sentences fluency": 106321,
      "sentences propose": 106456,
      "sentence editing": 105830,
      "editing networks": 34549,
      "generated recurrent": 47111,
      "lexical gap": 63766,
      "gap source": 46480,
      "vector sentence": 131368,
      "previous step": 90486,
      "experiments 10": 40752,
      "sequences extracted": 107123,
      "yelp review": 135321,
      "baselines text": 12481,
      "discuss directions": 32247,
      "opportunities future": 82142,
      "research text": 100639,
      "basis propose": 12546,
      "simple robust": 109508,
      "representations inspired": 99699,
      "word surrounding": 133597,
      "based aspects": 11526,
      "relatedness word": 97933,
      "representations approach": 99503,
      "parameters efficient": 84745,
      "efficient inference": 35082,
      "inference performance": 54191,
      "shows superior": 108635,
      "competitive approaches": 19636,
      "approaches relying": 8320,
      "amounts labelled": 5346,
      "hindienglish codeswitching": 50487,
      "codeswitching speech": 18240,
      "corpus codeswitching": 23694,
      "languages sentence": 60860,
      "global phenomenon": 48260,
      "phenomenon multilingual": 87253,
      "multilingual communities": 75217,
      "research increasing": 100534,
      "increasing demand": 53669,
      "codeswitching automatic": 18226,
      "corpus highly": 23823,
      "training systems": 123900,
      "systems limited": 116981,
      "available work": 11149,
      "present efforts": 89463,
      "efforts building": 35191,
      "codeswitching asr": 18225,
      "speech database": 111668,
      "applied speech": 7119,
      "asr language": 9432,
      "identification language": 51393,
      "paper mainly": 84036,
      "results asr": 101523,
      "asr task": 9449,
      "data expansion": 25917,
      "understanding spoken": 126971,
      "systems widely": 117236,
      "slu consists": 110109,
      "nlu model": 79809,
      "shared different": 107973,
      "models need": 73619,
      "data train": 26565,
      "introduce efficient": 56418,
      "method expand": 68822,
      "limited indomain": 64240,
      "detect informative": 29810,
      "using ngrams": 129971,
      "intent label": 55905,
      "train new": 122989,
      "rate cer": 95786,
      "traditional data": 122809,
      "ones based": 81679,
      "e2e nlg": 34361,
      "nlg challenge": 79513,
      "challenge paper": 16077,
      "experimental setup": 40730,
      "task endtoend": 118146,
      "endtoend e2e": 36895,
      "recent endtoend": 96460,
      "systems promising": 117081,
      "reduce need": 97341,
      "syntactic complexity": 116381,
      "diverse discourse": 32808,
      "62 systems": 1165,
      "covering wide": 24552,
      "approaches including": 8192,
      "models seq2seq": 74007,
      "winograd schema": 132727,
      "schema challenge": 103894,
      "challenge wsc": 16120,
      "reasoning task": 96314,
      "complex forms": 19812,
      "inference knowledge": 54155,
      "uses knowledge": 129236,
      "engine extracts": 37029,
      "resolution approach": 100754,
      "improves f1": 52980,
      "best represents": 13430,
      "approach competitive": 7436,
      "choice plausible": 16896,
      "plausible alternatives": 87679,
      "copa task": 23269,
      "transitionbased model": 124507,
      "model nested": 71565,
      "mention recognition": 68427,
      "common entity": 18872,
      "entity mentions": 37972,
      "contain mentions": 21746,
      "bottomup manner": 14502,
      "action sequence": 3078,
      "length based": 63353,
      "based stacklstm": 12065,
      "efficiently effectively": 35140,
      "component capture": 19971,
      "patterns model": 85760,
      "results ace": 101491,
      "datasets showing": 27707,
      "showing effectiveness": 108403,
      "recognition work": 97046,
      "overlapping entity": 83335,
      "datasets model": 27569,
      "built new": 14931,
      "representation able": 99160,
      "models maintaining": 73546,
      "complexity inference": 19920,
      "inference present": 54196,
      "present theoretical": 89741,
      "representation better": 99180,
      "better alternative": 13507,
      "alternative representations": 5242,
      "representational power": 99477,
      "datasets annotated": 27311,
      "understanding current": 126823,
      "local syntactic": 64938,
      "information fail": 54611,
      "use highlevel": 128087,
      "building effective": 14839,
      "network capable": 77180,
      "capable representing": 15218,
      "leveraging text": 63708,
      "text structure": 121331,
      "structure long": 113903,
      "long document": 65070,
      "critical information": 24813,
      "information generally": 54634,
      "encoding neural": 36715,
      "build hierarchical": 14771,
      "right wrong": 102837,
      "multiple right": 75674,
      "text sequence": 121286,
      "scope detection": 104037,
      "chinese corpora": 16750,
      "languages learned": 60683,
      "language trained": 60178,
      "develop neural": 30216,
      "learn crosslingual": 62014,
      "embeddings universal": 35998,
      "dependencies english": 29079,
      "english test": 37306,
      "showing work": 108437,
      "work surprisingly": 134837,
      "monolingual settings": 74617,
      "settings crosslingual": 107792,
      "task direct": 118100,
      "tag information": 117336,
      "using regular": 130099,
      "patterns natural": 85761,
      "data need": 26159,
      "tagging data": 117380,
      "need way": 76868,
      "way increase": 132094,
      "increase precision": 53611,
      "data scalable": 26369,
      "discourse structured": 32093,
      "framework generating": 45554,
      "language description": 58945,
      "data tables": 26534,
      "problem comes": 90965,
      "datatotext natural": 27815,
      "nlg systems": 79527,
      "endtoend statistical": 36973,
      "learn limited": 62082,
      "limited taskspecific": 64289,
      "exhibit limited": 40002,
      "limited scalability": 64274,
      "data relies": 26335,
      "offtheshelf nlp": 81626,
      "easily adaptable": 34430,
      "coherent fluent": 18317,
      "fluent adequate": 44704,
      "modules experiments": 74523,
      "datatotext approaches": 27811,
      "approaches demonstrate": 8120,
      "demonstrate robustness": 28861,
      "popular datasets": 88088,
      "covering diverse": 24538,
      "diverse data": 32803,
      "data types": 26586,
      "types knowledge": 126303,
      "search machine": 104326,
      "comprehension perspective": 20214,
      "attention enables": 9826,
      "users interact": 129137,
      "engine natural": 37031,
      "efficient manner": 35090,
      "understanding traditional": 127000,
      "tracking query": 122760,
      "query understanding": 94992,
      "different challenging": 31033,
      "challenging diverse": 16245,
      "complex intentions": 19821,
      "intentions work": 55919,
      "work define": 134460,
      "update internal": 127793,
      "propose self": 93028,
      "self attention": 104869,
      "network handle": 77272,
      "handle task": 49403,
      "dataset suggest": 27224,
      "suggest proposed": 115330,
      "exact match": 39718,
      "match accuracy": 67355,
      "accuracy f1": 2160,
      "showing potential": 108422,
      "comprehension like": 20189,
      "systems spontaneous": 117153,
      "intended meaning": 55868,
      "able develop": 1630,
      "able generalise": 1642,
      "generalise diverse": 46742,
      "robustness paper": 103121,
      "switchboard dialogue": 116288,
      "swda corpus": 116258,
      "corpus present": 23935,
      "10 percentage": 144,
      "simpler architecture": 109552,
      "test models": 120474,
      "dataset additional": 26725,
      "goaloriented dialogues": 48416,
      "light types": 64000,
      "sequential neural": 107241,
      "case gender": 15586,
      "rnns efficient": 102967,
      "challenge data": 16015,
      "processing morphologically": 91716,
      "modeling morphological": 72479,
      "dependencies improve": 29088,
      "various experiments": 131095,
      "interpretable features": 56239,
      "models crosslingual": 73005,
      "modern period": 74410,
      "develop unsupervised": 30248,
      "character segmentation": 16462,
      "cluster sequences": 17928,
      "models empirical": 73137,
      "korean question": 58347,
      "core issue": 23325,
      "difficult extract": 31629,
      "content automatically": 21860,
      "challenging languages": 16267,
      "parse sentence": 84885,
      "neural paraphrasing": 78624,
      "paraphrasing systems": 84849,
      "introduced recently": 56596,
      "structured annotation": 113990,
      "argument mining": 8947,
      "scheme dataset": 103916,
      "dataset expected": 26906,
      "help machines": 49748,
      "machines understand": 66344,
      "graph knowledge": 48826,
      "composed entities": 20065,
      "relationship entities": 98293,
      "application scenarios": 6879,
      "usually large": 130432,
      "retrieval natural": 102413,
      "general method": 46676,
      "external user": 42049,
      "embedding techniques": 35514,
      "external textual": 42046,
      "data entities": 25898,
      "attribute values": 10187,
      "work applying": 134375,
      "methods apply": 69315,
      "important attributes": 52107,
      "previous generated": 90414,
      "text new": 121145,
      "generates customized": 47158,
      "informative messages": 55143,
      "learning jointly": 62663,
      "pronouns shared": 92361,
      "reconstruction mechanism": 97121,
      "pronouns frequently": 92352,
      "frequently omitted": 45874,
      "omitted prodrop": 81655,
      "chinese generally": 16768,
      "generally leading": 46866,
      "leading significant": 61904,
      "challenges respect": 16205,
      "respect production": 101097,
      "production complete": 92051,
      "complete translations": 19774,
      "wang et": 132007,
      "2018 proposed": 674,
      "novel reconstructionbased": 80699,
      "reconstructionbased approach": 97125,
      "approach alleviating": 7345,
      "pronoun dp": 92338,
      "dp translation": 34074,
      "translation problems": 125144,
      "problems neural": 91348,
      "work improve": 134562,
      "employ shared": 36279,
      "better exploit": 13582,
      "decoder representations": 28050,
      "representations second": 99869,
      "learn translate": 62171,
      "errors propagated": 38401,
      "dp prediction": 34073,
      "language representation": 60038,
      "representation machine": 99318,
      "translation universal": 125402,
      "thanks new": 121771,
      "based combining": 11585,
      "variational autoencoders": 130918,
      "introducing interlingual": 56636,
      "interlingual loss": 56121,
      "loss additional": 65252,
      "objective adding": 81061,
      "adding forcing": 3378,
      "forcing interlingual": 45019,
      "loss able": 65246,
      "train multiple": 122969,
      "multiple encoders": 75549,
      "decoders language": 28065,
      "language sharing": 60088,
      "universal representation": 127324,
      "representation final": 99240,
      "decoding latent": 28099,
      "task shows": 118712,
      "architecture capable": 8633,
      "representation simultaneously": 99412,
      "universal morphology": 127313,
      "morphology universal": 74808,
      "morphology unimorph": 74807,
      "annotated text": 6245,
      "token level": 122257,
      "level corpus": 63432,
      "built different": 14918,
      "homograph disambiguation": 50616,
      "present deterministic": 89451,
      "dependencies v2": 29124,
      "unimorph schema": 127165,
      "validate approach": 130714,
      "paucity data": 85798,
      "present critical": 89424,
      "critical evaluation": 24811,
      "text representations": 121252,
      "large performance": 61206,
      "gains various": 46413,
      "tasks alleviating": 118925,
      "need manual": 76833,
      "linguistic understanding": 64570,
      "harder interpret": 49487,
      "approaches provide": 8300,
      "reach better": 95888,
      "interpretability performance": 56229,
      "performance present": 86612,
      "framework studying": 45702,
      "text context": 120819,
      "features combining": 43408,
      "flexible extensible": 44658,
      "learning curve": 62479,
      "super characters": 115668,
      "method named": 68978,
      "problem image": 91076,
      "texts images": 121525,
      "cnn models": 18014,
      "extracted automatically": 42141,
      "need explicit": 76805,
      "media corpus": 68096,
      "characters method": 16615,
      "classification topic": 17469,
      "contents different": 21986,
      "japanese korean": 57192,
      "english contextual": 37096,
      "contextual topic": 22505,
      "identifying topics": 51632,
      "unsupervised topic": 127739,
      "context dialog": 22057,
      "dialog act": 30544,
      "incorporating context": 53522,
      "relative gains": 98355,
      "accuracy 35": 2063,
      "detection recall": 30033,
      "span multiple": 111126,
      "dialog evaluation": 30561,
      "predict user": 88951,
      "keywords used": 57650,
      "used guide": 128569,
      "datadriven text": 26682,
      "text rewriting": 121270,
      "rewriting propose": 102696,
      "simple unsupervised": 109538,
      "representative different": 100010,
      "text styles": 121339,
      "papers approach": 84501,
      "instead relies": 55678,
      "documents sentences": 33289,
      "evaluation text": 39422,
      "normal simple": 80329,
      "pseudoparallel sentences": 94195,
      "sentences extracted": 106313,
      "supplement existing": 115937,
      "data lead": 26080,
      "corpora quality": 23566,
      "obtained web": 81431,
      "negatively affect": 76962,
      "affect quality": 4223,
      "problems data": 91313,
      "systems identify": 116941,
      "sentences input": 106356,
      "input corpora": 55313,
      "latency using": 61569,
      "useful scenarios": 128928,
      "framework suitable": 45704,
      "fullsentence translation": 45934,
      "implicitly learns": 52033,
      "single translation": 109805,
      "model framework": 71206,
      "framework present": 45655,
      "waitk policy": 131991,
      "trained generate": 123151,
      "strategy achieves": 113501,
      "low latency": 65370,
      "informal text": 54343,
      "based existing": 11695,
      "corpus nus": 23907,
      "nus sms": 81033,
      "sms corpus": 110275,
      "sms messages": 110276,
      "graphical models": 48930,
      "empirical evaluations": 36162,
      "evaluations new": 39477,
      "yielded similar": 135371,
      "similar accuracy": 109072,
      "significantly lower": 108974,
      "learning recognize": 62947,
      "entities paper": 37833,
      "focuses study": 44914,
      "entities motivated": 37822,
      "representation jointly": 99270,
      "jointly encode": 57340,
      "overlap compare": 83329,
      "introduce notion": 56508,
      "difficulty level": 31704,
      "advantages model": 4083,
      "model previous": 71783,
      "linearchain crfs": 64394,
      "results evaluated": 101768,
      "applications play": 6988,
      "computational tools": 20441,
      "aid process": 4435,
      "process translation": 91582,
      "benefit translation": 12995,
      "translation small": 125267,
      "7000 languages": 1213,
      "translators work": 125517,
      "digital world": 31729,
      "proposes framework": 93600,
      "new theory": 79221,
      "phraselevel translation": 87405,
      "datasets baselines": 27329,
      "goldstandard datasets": 48453,
      "corpus automatically": 23672,
      "annotated wikipedia": 6262,
      "corpus news": 23901,
      "evaluate popular": 38894,
      "recognition models": 96910,
      "datasets release": 27668,
      "100 200": 158,
      "glove word": 48304,
      "trained collection": 123093,
      "news blogs": 79310,
      "fundamental nlp": 46102,
      "applications despite": 6914,
      "despite remarkable": 29726,
      "remarkable results": 98845,
      "supervised settings": 115837,
      "task lowresource": 118363,
      "approaches leverage": 8214,
      "leverage machine": 63600,
      "translation techniques": 125371,
      "techniques translate": 120003,
      "richresource language": 102817,
      "language approaches": 58839,
      "industrial applications": 54057,
      "applications spoken": 7019,
      "efficiency required": 35043,
      "propose multilingual": 92799,
      "framework tackle": 45708,
      "language spanish": 60108,
      "rich annotation": 102726,
      "data rich": 26363,
      "english approach": 37068,
      "approach extended": 7558,
      "multilingual encoder": 75251,
      "encoder pretrained": 36554,
      "task incorporate": 118286,
      "nature shared": 76667,
      "encoder sentence": 36573,
      "multiple representations": 75670,
      "representations different": 99595,
      "target translation": 117734,
      "used ensemble": 128518,
      "model improve": 71306,
      "evaluation demonstrate": 39174,
      "demonstrate superiority": 28879,
      "superiority method": 115702,
      "baseline approach": 12188,
      "approach consistently": 7447,
      "inference reasoning": 54207,
      "pairs words": 83667,
      "problems paper": 91354,
      "implicitly represent": 52035,
      "knowledge relationships": 58140,
      "embeddings computed": 35606,
      "function word": 46052,
      "crosssentence attention": 25068,
      "layer existing": 61714,
      "inference models": 54168,
      "models bidaf": 72849,
      "replacing existing": 98948,
      "embeddings experiments": 35710,
      "recently released": 96749,
      "adversarial squad": 4165,
      "squad datasets": 112077,
      "bilingual contextual": 14023,
      "dataset evaluating": 26898,
      "similarity bcws": 109202,
      "pairs corresponding": 83504,
      "annotated human": 6202,
      "higher consistency": 50170,
      "consistency compared": 21355,
      "compared similar": 19440,
      "similar datasets": 109082,
      "datasets establish": 27453,
      "establish baselines": 38581,
      "bilingual embedding": 14035,
      "benchmark experiments": 12822,
      "crosslingual sense": 25007,
      "provided dataset": 93961,
      "understanding multilingual": 126893,
      "mentions mention": 68458,
      "capable recognizing": 15216,
      "used capture": 128417,
      "inference performed": 54192,
      "model recognizing": 71867,
      "analysis standard": 5872,
      "requires lexical": 100292,
      "words times": 134270,
      "shared properties": 108004,
      "generalpurpose lexical": 46886,
      "framework make": 45607,
      "dataset extracted": 26922,
      "reference corpus": 97509,
      "scale language": 103729,
      "modeling lm": 72463,
      "plays key": 87736,
      "role important": 103176,
      "applications speech": 7017,
      "community data": 19070,
      "like data": 64031,
      "words modern": 134060,
      "methods far": 69508,
      "far goal": 43082,
      "goal various": 48399,
      "words character": 133851,
      "number unique": 80991,
      "unique words": 127197,
      "words types": 134284,
      "batch size": 12557,
      "embedding matrix": 35438,
      "datasets scale": 27694,
      "scale number": 103744,
      "time speeds": 122109,
      "accuracy training": 2306,
      "dataset taking": 27230,
      "increase training": 53620,
      "learning discriminate": 62506,
      "information neural": 54799,
      "problem study": 91252,
      "pays special": 85818,
      "argue exist": 8917,
      "information global": 54641,
      "affect translations": 4226,
      "sentence specific": 106088,
      "specific words": 111519,
      "accordingly propose": 2016,
      "learns jointly": 63214,
      "trained dataset": 123108,
      "dataset derived": 26863,
      "derived original": 29355,
      "original parallel": 82536,
      "external labeled": 42022,
      "annotation experimental": 6308,
      "various realworld": 131180,
      "realworld scenarios": 96176,
      "scenarios language": 103855,
      "pairs neural": 83590,
      "better incorporate": 13605,
      "incorporate external": 53470,
      "information noisy": 54807,
      "noisy conditions": 80097,
      "convolution recurrent": 23146,
      "sentences uses": 106537,
      "consider global": 21201,
      "sentences combination": 106243,
      "best stateoftheart": 13455,
      "multilevel memory": 75192,
      "task oriented": 118483,
      "oriented dialogs": 82493,
      "endtoend task": 36975,
      "oriented dialog": 82491,
      "work makes": 134627,
      "use triples": 128343,
      "represent knowledge": 99115,
      "knowledge combines": 57827,
      "results memory": 101913,
      "size makes": 109931,
      "makes reasoning": 66802,
      "memory design": 68309,
      "strong assumptions": 113621,
      "existing architectures": 40067,
      "used modeling": 128636,
      "results instead": 101865,
      "instead using": 55692,
      "results introduce": 101867,
      "novel multilevel": 80656,
      "memory architecture": 68286,
      "query corresponding": 94953,
      "results multilevel": 101964,
      "conduct detailed": 20834,
      "detailed experiments": 29774,
      "available task": 11122,
      "models report": 73925,
      "entity f1": 37938,
      "words article": 133826,
      "present design": 89441,
      "design implementation": 29543,
      "developed detect": 30270,
      "language document": 58976,
      "web interface": 132237,
      "collects new": 18511,
      "account context": 2020,
      "addition commonly": 3408,
      "information textual": 55040,
      "automatically determines": 10757,
      "topics text": 122657,
      "text containing": 120816,
      "overview developed": 83363,
      "existing entailment": 40127,
      "datasets mainly": 27555,
      "pose problems": 88254,
      "attention grammar": 9849,
      "grammar word": 48666,
      "examples different": 39823,
      "order change": 82291,
      "based synthetic": 12105,
      "natural entailment": 76254,
      "entailment examples": 37664,
      "examples snli": 39879,
      "change meaning": 16366,
      "messages difficult": 68507,
      "task harder": 118251,
      "large manually": 61136,
      "datasets created": 27386,
      "16 times": 380,
      "larger previously": 61378,
      "previously released": 90617,
      "released datasets": 98480,
      "datasets combined": 27355,
      "context use": 22311,
      "work particular": 134667,
      "conversations widely": 23077,
      "manuallyannotated data": 67088,
      "develop robust": 30229,
      "robust datadriven": 103027,
      "help advance": 49695,
      "improving context": 53077,
      "multimodal dialogue": 75425,
      "generation multimodal": 47494,
      "fashion domain": 43116,
      "domain introduce": 33550,
      "introduce multimodal": 56462,
      "recurrent encoderdecoder": 97174,
      "model extension": 71152,
      "baselines terms": 12477,
      "textbased similarity": 121428,
      "vision language": 131761,
      "models performing": 73740,
      "analysis systems": 5883,
      "systems output": 117040,
      "based conversational": 11607,
      "systems conversational": 116816,
      "increasingly popular": 53706,
      "modern society": 74414,
      "users challenging": 129099,
      "demonstrate deep": 28697,
      "agent called": 4297,
      "generating relevant": 47253,
      "happiness users": 49444,
      "team wins": 119724,
      "agent needs": 4308,
      "needs detect": 76889,
      "sensitive topics": 105751,
      "behavior users": 12669,
      "significant research": 108858,
      "research challenges": 100441,
      "world users": 135053,
      "till date": 121973,
      "multitask deep": 75814,
      "coding propose": 18244,
      "propose methodology": 92775,
      "methodology estimating": 69257,
      "human behaviors": 50766,
      "learning paradigms": 62874,
      "data human": 26007,
      "human interactions": 50869,
      "labels relevant": 58635,
      "datasets specifically": 27728,
      "single multiple": 109771,
      "approaches incorporating": 8194,
      "context demonstrate": 22049,
      "discuss insights": 32260,
      "insights models": 55541,
      "better internal": 13609,
      "words sequence": 134204,
      "models recently": 73881,
      "tasks gap": 119140,
      "shallow simple": 107931,
      "research learning": 100545,
      "representations address": 99495,
      "questions investigate": 95320,
      "representations conduct": 99558,
      "different stateoftheart": 31449,
      "stateoftheart convolutional": 112620,
      "convolutional models": 23161,
      "investigate advantages": 56713,
      "words composing": 133871,
      "composing characters": 20080,
      "limited supervised": 64285,
      "model sequence": 71985,
      "labeling datasets": 58494,
      "including named": 53329,
      "recognition partofspeech": 96960,
      "syntactic chunking": 116376,
      "indepth analysis": 53794,
      "models obtains": 73658,
      "obtains new": 81468,
      "relying external": 98757,
      "resources machine": 101008,
      "vietnamese english": 131676,
      "networks state": 77778,
      "english vietnamese": 37334,
      "shortage parallel": 108270,
      "hyperparameter search": 51228,
      "search present": 104336,
      "present practical": 89645,
      "practical challenges": 88703,
      "paper highlights": 83971,
      "efforts improving": 35199,
      "corpus date": 23745,
      "scores experiments": 104173,
      "experiments provide": 41088,
      "employing different": 36315,
      "models lowresource": 73532,
      "spoken dialogs": 111977,
      "cooperative competitive": 23256,
      "cooperative games": 23258,
      "given knowledge": 48049,
      "knowledge high": 58002,
      "entrainment control": 38154,
      "graphbased framework": 48899,
      "framework information": 45575,
      "modern information": 74392,
      "local dependencies": 64915,
      "improve predictions": 52498,
      "predictions paper": 89184,
      "textual units": 121752,
      "units words": 127271,
      "propagates information": 92387,
      "nodes graph": 80036,
      "graph convolutions": 48790,
      "improve wordlevel": 52587,
      "predictions evaluation": 89166,
      "extraction shows": 42487,
      "stateoftheart sequence": 112949,
      "model significant": 72017,
      "appropriate word": 8444,
      "digitized texts": 31734,
      "texts language": 121537,
      "language agglutinative": 58821,
      "korean language": 58346,
      "language problem": 59829,
      "propose realtime": 93010,
      "automatic word": 10708,
      "propose datadriven": 92613,
      "corpus various": 24062,
      "qualitative comparison": 94557,
      "comparison text": 19583,
      "network selfattention": 77424,
      "attracted increasing": 10150,
      "flexibility modeling": 44650,
      "modeling dependencies": 72414,
      "multiheaded attention": 75133,
      "different representation": 31380,
      "representation subspaces": 99425,
      "subspaces different": 114846,
      "different positions": 31341,
      "2017 work": 645,
      "novel convolutional": 80516,
      "capture neighboring": 15348,
      "model interaction": 71356,
      "interaction multiple": 55956,
      "multiple attention": 75497,
      "attention heads": 9851,
      "results wmt14": 102337,
      "wmt14 englishtogerman": 132821,
      "englishtogerman translation": 37449,
      "transformer baseline": 124306,
      "comparing previous": 19513,
      "work model": 134635,
      "new parameters": 79071,
      "multilingual task": 75380,
      "utterance interpretation": 130602,
      "ai systems": 4426,
      "identify user": 51573,
      "user intents": 129001,
      "intents corresponding": 55921,
      "corresponding slots": 24306,
      "language train": 60177,
      "models largely": 73465,
      "multilingual training": 75385,
      "different crosslingual": 31070,
      "translating training": 124610,
      "crosslingual pretrained": 24993,
      "embeddings novel": 35836,
      "translation encoder": 124786,
      "given training": 48163,
      "examples target": 39885,
      "data lowresource": 26099,
      "settings multilingual": 107822,
      "multilingual contextual": 75219,
      "representations better": 99521,
      "static embeddings": 113063,
      "embeddings compare": 35602,
      "compare crosslingual": 19233,
      "resources form": 100978,
      "form contextual": 45076,
      "elmo representations": 35324,
      "given just": 48048,
      "just small": 57473,
      "amounts target": 5357,
      "highlights need": 50284,
      "natural disasters": 76251,
      "like able": 64017,
      "coherent narratives": 18322,
      "make task": 66738,
      "multiple documents": 75543,
      "documents generated": 33234,
      "exploiting wikipedia": 41488,
      "challenging inference": 16262,
      "novel challenging": 80509,
      "simple bagofwords": 109372,
      "falls short": 43034,
      "short task": 108246,
      "task necessary": 118445,
      "generating texts": 47278,
      "concepttotext generation": 20655,
      "generation typically": 47687,
      "typically employs": 126425,
      "pipeline architecture": 87528,
      "architecture leads": 8682,
      "leads suboptimal": 61969,
      "greedily select": 49066,
      "select important": 104700,
      "important facts": 52162,
      "words express": 133953,
      "space limited": 111015,
      "use fewer": 128053,
      "fewer words": 43893,
      "words report": 134172,
      "sentence aggregation": 105765,
      "resulting texts": 101475,
      "available state": 11111,
      "art natural": 9065,
      "programming model": 92133,
      "model unlike": 72247,
      "unlike pipeline": 127438,
      "architectures jointly": 8811,
      "jointly considers": 57338,
      "choices available": 16914,
      "local decisions": 64914,
      "results summarizing": 102238,
      "summarizing structured": 115630,
      "information limited": 54730,
      "limited space": 64283,
      "model considers": 70888,
      "limited form": 64236,
      "generation avoids": 47322,
      "texts need": 121560,
      "compared pipeline": 19410,
      "pipeline systems": 87556,
      "perceived quality": 85913,
      "generated texts": 47139,
      "countbased word": 24420,
      "trained purely": 123254,
      "contribution unsupervised": 22781,
      "experiments 25": 40755,
      "500 sentences": 1082,
      "dialogue summarization": 30760,
      "summarization tasks": 115577,
      "tasks meeting": 119291,
      "meeting summarization": 68257,
      "dialogue text": 30806,
      "text input": 121056,
      "concise description": 20700,
      "short description": 108206,
      "main topic": 66471,
      "topic conversation": 122505,
      "clear picture": 17728,
      "conversation based": 22937,
      "dataset build": 26772,
      "attentive model": 10119,
      "interaction utterances": 55975,
      "utterances different": 130635,
      "different speakers": 31445,
      "compared baselines": 19333,
      "multilingual model": 75280,
      "personality traits": 87133,
      "traits text": 123977,
      "spanish dutch": 111152,
      "words having": 133978,
      "similar semantic": 109141,
      "meaning different": 67632,
      "languages necessarily": 60758,
      "necessarily correspond": 76746,
      "words correlate": 133890,
      "multilingual vector": 75399,
      "embeddings training": 35987,
      "related training": 97909,
      "training features": 123634,
      "english lowresource": 37201,
      "languages better": 60429,
      "better multilingual": 13632,
      "multilingual embeddings": 75250,
      "embeddings achieve": 35540,
      "average fscore": 11192,
      "monolingual model": 74596,
      "using cnn": 129445,
      "regression tasks": 97717,
      "better classification": 13536,
      "results evaluating": 101770,
      "model separate": 71981,
      "phrases local": 87439,
      "global contexts": 48232,
      "emerging entities": 36067,
      "documents web": 33313,
      "context help": 22120,
      "context important": 22129,
      "important machines": 52184,
      "problem answer": 90933,
      "undertake task": 127034,
      "task describing": 118069,
      "describing given": 29451,
      "contexts solve": 22428,
      "description model": 29468,
      "context encoders": 22076,
      "definition generation": 28523,
      "2018 model": 664,
      "dataset newly": 27055,
      "wikipedia demonstrate": 132651,
      "longform text": 65170,
      "task previous": 118558,
      "coherence cohesion": 18293,
      "model equipped": 71090,
      "provide feedback": 93827,
      "feedback signals": 43835,
      "efficient variant": 35130,
      "sequence training": 107089,
      "proposed eliminate": 93263,
      "eliminate need": 35296,
      "need training": 76866,
      "training separate": 123839,
      "showing improvements": 108409,
      "recurrent attentionbased": 97162,
      "models recent": 73872,
      "years witnessed": 135312,
      "driven neural": 34155,
      "comes substantial": 18784,
      "cost performance": 24369,
      "inference latency": 54158,
      "devices paper": 30459,
      "make observation": 66703,
      "compare stateoftheart": 19296,
      "standard benchmarks": 112208,
      "dialogue conducted": 30646,
      "conducted based": 20911,
      "based given": 11739,
      "hu et": 50707,
      "2014 work": 580,
      "study largescale": 114423,
      "architectures datasets": 8797,
      "set neural": 107504,
      "architectures using": 8857,
      "representations considering": 99560,
      "various ways": 131237,
      "components test": 20048,
      "models collect": 72919,
      "dataset grounded": 26958,
      "humanhuman conversations": 51028,
      "speakers asked": 111308,
      "key factor": 57567,
      "guo et": 49282,
      "al 2019": 4640,
      "2019 dataset": 695,
      "efficacy approach": 35010,
      "particular obtain": 85436,
      "task best": 117933,
      "model par": 71699,
      "ability neural": 1550,
      "hypernymy hyponymy": 51222,
      "relationships word": 98338,
      "predominant approach": 89228,
      "approach generally": 7585,
      "generally involves": 46863,
      "limited quantity": 64266,
      "systems article": 116737,
      "article propose": 9150,
      "advantage knowledge": 4048,
      "sense tags": 105700,
      "disambiguate words": 31942,
      "words lexical": 134033,
      "method leads": 68933,
      "leads state": 61961,
      "results wsd": 102348,
      "wsd evaluation": 135170,
      "time size": 122103,
      "size models": 109936,
      "method combined": 68702,
      "tagged training": 117351,
      "systems support": 117175,
      "including social": 53375,
      "train systems": 123036,
      "systems produce": 117079,
      "engaging responses": 37022,
      "unannotated data": 126586,
      "dataset real": 27144,
      "real conversations": 96059,
      "conversations collected": 23051,
      "2017 alexa": 626,
      "challenge developed": 16021,
      "developed neural": 30290,
      "responses user": 101301,
      "likely lead": 64140,
      "engaging conversations": 37018,
      "user ratings": 129031,
      "larger amounts": 61353,
      "using conversation": 129493,
      "make data": 66641,
      "learning ranking": 62935,
      "ranking short": 95688,
      "short social": 108236,
      "user queries": 129025,
      "queries using": 94943,
      "simple wordlevel": 109548,
      "architecture augmented": 8622,
      "attentionbased mechanisms": 10079,
      "tokens extensive": 122308,
      "datasets trec": 27767,
      "effectiveness existing": 34890,
      "exploit diverse": 41411,
      "attentionbased matching": 10078,
      "models shared": 74020,
      "community support": 19102,
      "support future": 115970,
      "mapping instructions": 67135,
      "map natural": 67110,
      "language instructions": 59212,
      "search commonly": 104304,
      "using beam": 129384,
      "partial programs": 85298,
      "finding good": 44274,
      "difficult work": 31674,
      "uses target": 129290,
      "world state": 135050,
      "known training": 58328,
      "network predicts": 77393,
      "expected reward": 40401,
      "dataset algorithm": 26730,
      "dramatically improves": 34095,
      "performance domains": 86314,
      "standard beam": 112200,
      "baselines neural": 12433,
      "task representations": 118637,
      "supervision model": 115899,
      "model agnostic": 70648,
      "transferring model": 124241,
      "time effort": 122021,
      "effort paper": 35180,
      "framework easily": 45510,
      "easily effectively": 34448,
      "effectively transferring": 34855,
      "languages framework": 60594,
      "representations form": 99659,
      "form weak": 45146,
      "task agnostic": 117858,
      "data loss": 26096,
      "defined task": 28503,
      "framework transferring": 45722,
      "tests models": 120610,
      "number strong": 80972,
      "results rely": 102125,
      "complex approaches": 19799,
      "resources data": 100961,
      "data additionally": 25578,
      "framework proposed": 45660,
      "semantically rich": 105375,
      "meaningful representations": 67725,
      "representations languages": 99715,
      "despite lack": 29700,
      "wizard wikipedia": 132761,
      "dialogue intelligent": 30691,
      "use knowledge": 128104,
      "popular sequence": 88127,
      "typically generate": 126435,
      "generic utterances": 47813,
      "weights model": 132380,
      "mapping input": 67133,
      "input utterances": 55468,
      "output employing": 83072,
      "knowledge context": 57840,
      "knowledge far": 57934,
      "proved difficult": 93711,
      "difficult lack": 31638,
      "lack supervised": 58754,
      "learning benchmark": 62402,
      "end collect": 36796,
      "collect release": 18390,
      "release large": 98454,
      "retrieved wikipedia": 102464,
      "architectures capable": 8789,
      "natural responses": 76618,
      "able conduct": 1623,
      "new benchmark": 78811,
      "learning languageindependent": 62678,
      "model fusion": 71215,
      "work explores": 134520,
      "better adaptation": 13502,
      "external language": 42023,
      "framework transfer": 45721,
      "build languageindependent": 14777,
      "architecture shared": 8748,
      "external lm": 42033,
      "lm integrated": 64865,
      "s2s model": 103515,
      "effectively incorporate": 34823,
      "context target": 22291,
      "language investigate": 59225,
      "investigate various": 56827,
      "seed models": 104505,
      "iarpa babel": 51300,
      "improves performances": 53023,
      "simple transfer": 109534,
      "reduces performance": 97393,
      "performance gap": 86407,
      "hybrid systems": 51196,
      "built evaluation": 14922,
      "year paper": 135238,
      "years especially": 135259,
      "classification component": 17151,
      "improve runtime": 52534,
      "trained convolutional": 123099,
      "networks combined": 77538,
      "combined traditional": 18671,
      "traditional evaluation": 122815,
      "evaluation designed": 39178,
      "network endtoend": 77243,
      "endtoend performance": 36952,
      "achieved rank": 2675,
      "systems participating": 117051,
      "participating task": 85358,
      "embed sentences": 35349,
      "using attentive": 129358,
      "trees sentence": 125710,
      "effective feature": 34673,
      "deep learningbased": 28325,
      "learningbased nlp": 63186,
      "structures existing": 114074,
      "explicit mechanism": 41333,
      "words dynamically": 133927,
      "according importance": 1982,
      "importance task": 52086,
      "specifically construct": 111534,
      "construct latent": 21638,
      "tree sentence": 125612,
      "sentence proposed": 106009,
      "important words": 52294,
      "embedding propose": 35475,
      "methods sentence": 69740,
      "tasks data": 119026,
      "feature decay": 43258,
      "decay algorithms": 27911,
      "target data": 117596,
      "applied neural": 7099,
      "model retrieving": 71930,
      "use training": 128337,
      "select data": 104696,
      "set document": 107421,
      "document translated": 33099,
      "using sourceside": 130201,
      "sentences selected": 106486,
      "given testset": 48154,
      "domain corpora": 33490,
      "corpora subtitle": 23594,
      "contain parallel": 21751,
      "inaccurate translations": 53193,
      "general idea": 46662,
      "selection approach": 104765,
      "algorithms fda": 4853,
      "data selected": 26382,
      "selected using": 104739,
      "using test": 130284,
      "set source": 107586,
      "findings reveal": 44311,
      "reveal models": 102500,
      "combination outputs": 18573,
      "set obtain": 107516,
      "obtain statistically": 81330,
      "points strong": 87877,
      "sourceside information": 110940,
      "endtoend approaches": 36871,
      "simplifying training": 109617,
      "data perform": 26230,
      "perform large": 86021,
      "tasks aim": 118924,
      "broader range": 14702,
      "explore potential": 41575,
      "use endtoend": 128027,
      "methods small": 69761,
      "smaller datasets": 110232,
      "expanding vocabulary": 40366,
      "examples new": 39860,
      "learning results": 62969,
      "competitive strong": 19696,
      "baseline small": 12312,
      "speech commands": 111657,
      "commands dataset": 18793,
      "problem simple": 91232,
      "simple strategy": 109522,
      "strategy achieved": 113500,
      "accuracy new": 2216,
      "new keywords": 78968,
      "prediction reading": 89114,
      "solely syntactic": 110535,
      "syntactic cues": 116389,
      "cues make": 25213,
      "model implicit": 71301,
      "pointer networks": 87829,
      "shows good": 108581,
      "performance argument": 86150,
      "task nominal": 118462,
      "relevant knowledge": 98562,
      "modeling response": 72535,
      "conversation critical": 22943,
      "make good": 66678,
      "good use": 48508,
      "use relevant": 128242,
      "attributes entities": 10195,
      "distinguish uses": 32580,
      "proper entities": 92413,
      "validate proposed": 130729,
      "models long": 73526,
      "connections recent": 21147,
      "model sequential": 71988,
      "difficulty capturing": 31695,
      "capturing longterm": 15480,
      "longterm dependencies": 65200,
      "dependencies work": 29127,
      "work tried": 134863,
      "tried alleviate": 125774,
      "learn directly": 62022,
      "information training": 55052,
      "novel reinforcement": 80705,
      "dependency relationship": 29223,
      "model computes": 70878,
      "transition functions": 124490,
      "advantage rnns": 4059,
      "entire sentences": 37713,
      "number prediction": 80943,
      "experiment proposed": 40486,
      "accuracy nearly": 2213,
      "lowresource natural": 65540,
      "recent deep": 96446,
      "utilize knowledge": 130512,
      "lowresource setting": 65560,
      "data crucial": 25813,
      "model tackle": 72133,
      "nlg problem": 79525,
      "dataset integrate": 26982,
      "encoderdecoder generator": 36598,
      "procedure experiments": 91391,
      "showed proposed": 108385,
      "having sufficient": 49571,
      "sufficient training": 115289,
      "dataset strong": 27220,
      "strong ability": 113617,
      "language disambiguation": 58969,
      "large portion": 61211,
      "syntactic characteristics": 116375,
      "language especially": 59013,
      "headfinal languages": 49588,
      "importance identifying": 52059,
      "identifying speakers": 51627,
      "speakers intention": 111318,
      "paper suggests": 84452,
      "main point": 66449,
      "given utterance": 48169,
      "statement question": 112547,
      "intuitive understanding": 56670,
      "speech validate": 111847,
      "utility test": 130491,
      "speech recognizers": 111791,
      "number customers": 80862,
      "attract customers": 10139,
      "limited screen": 64276,
      "previous researchers": 90452,
      "researchers mainly": 100698,
      "mainly consider": 66480,
      "consider textual": 21234,
      "view training": 131712,
      "propose multimodal": 92800,
      "adversarial network": 4150,
      "image information": 51787,
      "information original": 54817,
      "generation reinforcement": 47590,
      "learning personalized": 62886,
      "systems consider": 116806,
      "conversation content": 22939,
      "unsolved issues": 127563,
      "model attempt": 70696,
      "profile model": 92099,
      "distributed embeddings": 32612,
      "similar users": 109166,
      "preference model": 89241,
      "captures user": 15458,
      "handle ambiguity": 49374,
      "models combined": 72924,
      "qualitative performance": 94568,
      "outperforms approaches": 82838,
      "approaches terms": 8373,
      "terms task": 120390,
      "task completion": 117992,
      "deep ensemble": 28231,
      "ensemble framework": 37596,
      "framework fake": 45545,
      "news detection": 79329,
      "misinformation detection": 70297,
      "social fabrics": 110339,
      "information increasing": 54682,
      "increasing rapidly": 53682,
      "sources including": 110894,
      "media feeds": 68113,
      "feeds news": 43857,
      "blogs online": 14374,
      "online newspapers": 81786,
      "newspapers paper": 79413,
      "various deep": 131075,
      "classifying predefined": 17673,
      "develop models": 30214,
      "representations obtained": 99787,
      "models fed": 73245,
      "final classification": 44099,
      "results overall": 102022,
      "model opinion": 71612,
      "opinion target": 82103,
      "target extraction": 117622,
      "extraction target": 42501,
      "target sentiment": 117704,
      "targetbased sentiment": 117757,
      "works usually": 134990,
      "usually studied": 130455,
      "hinders practical": 50458,
      "use paper": 128187,
      "aims solve": 4575,
      "complete task": 19771,
      "analysis endtoend": 5595,
      "novel unified": 80761,
      "unified tagging": 127130,
      "tagging scheme": 117443,
      "stacked recurrent": 112129,
      "upper predicts": 127822,
      "predicts unified": 89225,
      "produce final": 91889,
      "output results": 83115,
      "network improve": 77280,
      "task explore": 118186,
      "dependency propose": 29218,
      "propose explicitly": 92667,
      "polarities propose": 87902,
      "gate mechanism": 46499,
      "mechanism models": 68014,
      "models relation": 73905,
      "relation features": 98037,
      "achieves consistently": 2780,
      "results domain": 101738,
      "domain agnostic": 33467,
      "specificity prediction": 111618,
      "level sentence": 63501,
      "information discourse": 54496,
      "useful downstream": 128880,
      "systems predict": 117069,
      "coarse labels": 18052,
      "labels binary": 58583,
      "tailored specific": 117501,
      "domains news": 33825,
      "work generalize": 134543,
      "domains labeled": 33796,
      "prediction specifically": 89126,
      "designed output": 29615,
      "distribution labels": 32663,
      "generalizes different": 46837,
      "stateoftheart trained": 113010,
      "trained news": 123222,
      "deterministic algorithm": 30159,
      "work bridging": 134401,
      "resolution poesio": 100776,
      "poesio et": 87782,
      "al 2004": 4593,
      "2004 hou": 546,
      "al 2013b": 4607,
      "calculate word": 15019,
      "nps head": 80803,
      "fully capture": 45950,
      "paper create": 83790,
      "vectors combining": 131415,
      "knowledge resource": 58152,
      "deterministic approach": 30160,
      "head noun": 49577,
      "compared best": 19339,
      "additionally improve": 3610,
      "results bridging": 101558,
      "combining simple": 18750,
      "ii neural": 51712,
      "neural based": 77851,
      "controversial nature": 22865,
      "fact language": 42828,
      "understanding use": 127003,
      "does hold": 33357,
      "focus specific": 44822,
      "approach relies": 7853,
      "relies recurrent": 98656,
      "networks order": 77683,
      "introduced bias": 56574,
      "advantages neural": 4085,
      "word lexicons": 133350,
      "lexicons handcrafted": 63926,
      "able distinguish": 1632,
      "biased statements": 13860,
      "significantly outperforming": 108998,
      "outperforming baseline": 82798,
      "models improvement": 73370,
      "finally release": 44229,
      "largest corpus": 61530,
      "jointly identifying": 57345,
      "identifying opinion": 51607,
      "features opinion": 43639,
      "mainly involves": 66490,
      "relations opinion": 98226,
      "opinion expressions": 82087,
      "expressions related": 41762,
      "information previous": 54867,
      "previous researches": 90453,
      "typically handled": 126436,
      "extraction results": 42471,
      "results complexity": 101616,
      "increased paper": 53630,
      "elements specifically": 35283,
      "new terms": 79215,
      "ensure accuracy": 37636,
      "based fuzzy": 11729,
      "algorithm effectively": 4721,
      "identify main": 51520,
      "simultaneously outperform": 109686,
      "help select": 49781,
      "interesting observations": 56082,
      "iwslt 2018": 57162,
      "basque english": 12553,
      "english low": 37199,
      "resource mt": 100865,
      "morphologicallyrich language": 74778,
      "challenge neural": 16064,
      "usually achieve": 130404,
      "performance trained": 86807,
      "data accordingly": 25557,
      "authentic data": 10355,
      "backtranslated data": 11323,
      "data create": 25805,
      "sentences trained": 106525,
      "sentences close": 106239,
      "set model": 107496,
      "model finetuned": 71190,
      "multitask approach": 75809,
      "effort devoted": 35171,
      "evaluate multitask": 38874,
      "nlp downstream": 79608,
      "lack understanding": 58767,
      "settings multitask": 107823,
      "learning significant": 63025,
      "trained multitask": 123209,
      "learning setup": 63019,
      "selected semantic": 104731,
      "fashion introduce": 43119,
      "complex tasks": 19888,
      "recognition entity": 96867,
      "features external": 43503,
      "external nlp": 42035,
      "tools like": 122454,
      "training supervision": 123895,
      "set shared": 107578,
      "model layers": 71421,
      "represent complex": 99100,
      "information mining": 54765,
      "terms referring": 120371,
      "work rank": 134761,
      "query term": 94990,
      "problem twophase": 91273,
      "twophase task": 126154,
      "pairs followed": 83548,
      "followed organizing": 44949,
      "organizing pairs": 82481,
      "approaches fail": 8161,
      "fail model": 42968,
      "holistic semantics": 50601,
      "semantics set": 105468,
      "suffer error": 115218,
      "framework named": 45620,
      "efficiently generates": 35145,
      "using example": 129649,
      "consists novel": 21486,
      "classifier jointly": 17553,
      "learns represent": 63230,
      "permutation invariant": 87053,
      "include new": 53231,
      "algorithm enumerates": 4726,
      "classifier detect": 17537,
      "detect entity": 29803,
      "sets using": 107724,
      "sentiment induction": 106745,
      "online communities": 81748,
      "ways text": 132174,
      "users sentiment": 129168,
      "differences representations": 30969,
      "social group": 110341,
      "communities high": 19060,
      "vice versa": 131648,
      "vary according": 131241,
      "communityspecific sentiment": 19106,
      "sentiment useful": 106815,
      "useful indicator": 128895,
      "indicator words": 53879,
      "words social": 134226,
      "social meaning": 110347,
      "values especially": 130791,
      "especially context": 38444,
      "social platforms": 110446,
      "settings different": 107797,
      "missing existing": 70325,
      "sense knowledge": 105689,
      "neural multitask": 78216,
      "analysis given": 5634,
      "contains relevant": 21834,
      "model validates": 72293,
      "cnn outperforms": 18015,
      "tasks multitask": 119317,
      "setup demonstrate": 107851,
      "additional performance": 3555,
      "cue words": 25205,
      "longer conversations": 65155,
      "interesting challenging": 56070,
      "challenging tasks": 16335,
      "intelligence research": 55837,
      "efforts dedicated": 35195,
      "building dialogue": 14836,
      "systems shed": 117139,
      "modeling conversation": 72406,
      "common people": 18907,
      "people talk": 85894,
      "aspects conversation": 9377,
      "topics coherent": 122617,
      "demonstrates necessity": 28950,
      "dialogue flow": 30672,
      "select adaptive": 104688,
      "terms effectiveness": 120315,
      "model longterm": 71475,
      "paper experiments": 83909,
      "experiments reallife": 41105,
      "language wide": 60339,
      "public largescale": 94262,
      "work introduces": 134583,
      "randomly assigned": 95525,
      "documents compiled": 33204,
      "dataset contained": 26824,
      "set average": 107371,
      "information logistic": 54735,
      "classifier fit": 17544,
      "documents average": 33190,
      "utilize intrinsic": 130511,
      "correlation multiple": 24241,
      "train supervised": 123031,
      "effective summarizing": 34755,
      "diversitypromoting objective": 32902,
      "function neural": 46035,
      "estimation mle": 38655,
      "trains models": 123965,
      "new objective": 79061,
      "token frequency": 122253,
      "classes larger": 17063,
      "encourages model": 36768,
      "loss model": 65282,
      "model establishes": 71096,
      "score maintaining": 104090,
      "maintaining good": 66515,
      "score comparable": 104061,
      "networks sequence": 77759,
      "recently large": 96703,
      "number neural": 80922,
      "neural mechanisms": 78138,
      "mechanisms models": 68064,
      "transformer model": 124339,
      "networks gnns": 77614,
      "complementary strengths": 19753,
      "propose contextualized": 92601,
      "dynamically construct": 34342,
      "sentence leverage": 105936,
      "leverage rich": 63622,
      "dependencies particular": 29106,
      "providing better": 94104,
      "better interpretability": 13610,
      "improve lowresource": 52407,
      "tagging natural": 117424,
      "learning revolution": 62971,
      "shifted focus": 108183,
      "handcrafted symbolic": 49356,
      "adequate representations": 3834,
      "learned automatically": 62201,
      "automatically corpora": 10746,
      "working lowresource": 134899,
      "corpora additional": 23403,
      "typically rely": 126453,
      "shown clear": 108452,
      "combining best": 18708,
      "improvements depend": 52841,
      "coverage quality": 24521,
      "paper seeks": 84417,
      "gap providing": 46476,
      "thorough analysis": 121877,
      "resources crosslingual": 100960,
      "tagging neural": 117429,
      "study improving": 114404,
      "pose challenges": 88250,
      "mt tasks": 74999,
      "language lrl": 59262,
      "lrl pairs": 65581,
      "pairs language": 83569,
      "corpora exist": 23479,
      "exist work": 40033,
      "instance learning": 55603,
      "learning set": 63016,
      "used language": 128602,
      "work character": 134410,
      "synchronic diachronic": 116335,
      "level nmt": 63481,
      "task characterize": 117960,
      "typical errors": 126399,
      "errors method": 38384,
      "improves bleu": 52953,
      "generalize languages": 46811,
      "languages applying": 60406,
      "pairs work": 83668,
      "work seen": 134791,
      "step process": 113296,
      "creating effective": 24698,
      "resourceconstrained languages": 100891,
      "languages iii": 60625,
      "enhanced semantic": 37518,
      "characterlevel tasks": 16586,
      "tasks hierarchical": 119156,
      "years sequencetosequence": 135300,
      "challenges especially": 16154,
      "especially neural": 38487,
      "present hierarchical": 89510,
      "hierarchical deep": 49951,
      "proposed network": 93505,
      "shorter sequences": 108297,
      "network long": 77306,
      "short sentences": 108233,
      "based sequencetosequence": 12028,
      "network concatenated": 77204,
      "experiments shows": 41139,
      "achieve superior": 2581,
      "evaluation understudy": 39429,
      "style words": 114611,
      "networks strategy": 77781,
      "sampling training": 103617,
      "retrievalbased dialogue": 102441,
      "approach quality": 7831,
      "quality improvement": 94683,
      "automated dialogue": 10436,
      "negative samples": 76947,
      "according distribution": 1975,
      "randomly chosen": 95526,
      "original results": 82539,
      "case using": 15632,
      "knowledge distributed": 57874,
      "embeddings independent": 35760,
      "information plays": 54845,
      "chinese lexical": 16782,
      "ontology based": 81836,
      "disambiguation corpus": 31960,
      "pieces knowledge": 87499,
      "exploit hierarchical": 41417,
      "hierarchical information": 49969,
      "information tackle": 55026,
      "tackle data": 117289,
      "problem instance": 91086,
      "representation morphemes": 99345,
      "evaluation validate": 39438,
      "syntagmatic relations": 116528,
      "new source": 79142,
      "artificial neural": 9259,
      "labeled grammatical": 58459,
      "grammatical acceptability": 48688,
      "different verbal": 31555,
      "verbal argument": 131530,
      "converging evidence": 22933,
      "dataset investigate": 26987,
      "investigate syntactic": 56816,
      "suggesting representations": 115351,
      "word sentencelevel": 133543,
      "models information": 73401,
      "present word": 89765,
      "embeddings passed": 35858,
      "embeddings crosslingual": 35627,
      "crosslingual approaches": 24929,
      "resolution dialogue": 100759,
      "slotfilling paradigm": 110077,
      "paradigm user": 84551,
      "user refer": 129033,
      "refer slots": 97499,
      "slots context": 110084,
      "context conversation": 22039,
      "conversation goal": 22952,
      "goal contextual": 48342,
      "understanding resolve": 126943,
      "resolve referring": 100805,
      "expressions appropriate": 41747,
      "appropriate slots": 8438,
      "scalable multidomain": 103695,
      "resolving references": 100819,
      "trivial task": 125831,
      "task large": 118341,
      "language main": 59266,
      "main focus": 66423,
      "way alleviate": 132056,
      "language crosslingual": 58928,
      "setup assume": 107850,
      "assume access": 9645,
      "access annotated": 1902,
      "resources trained": 101058,
      "language little": 59256,
      "little annotated": 64788,
      "setting experiments": 107747,
      "furthermore combined": 46152,
      "live data": 64849,
      "systems dialog": 116849,
      "ranking used": 95694,
      "used rank": 128712,
      "attention focused": 9841,
      "systems real": 117096,
      "errors paper": 38398,
      "propose contextaware": 92598,
      "matching scores": 67432,
      "scores candidate": 104159,
      "current dialog": 25275,
      "context combines": 22029,
      "distribution candidates": 32642,
      "reranking using": 100380,
      "models handcrafted": 73317,
      "proposed endtoend": 93267,
      "improving robustness": 53158,
      "robustness neural": 103112,
      "dialog models": 30574,
      "models lack": 73446,
      "lack robustness": 58743,
      "user input": 128996,
      "input leads": 55357,
      "dialog behavior": 30553,
      "production environments": 92057,
      "data access": 25556,
      "examples paper": 39862,
      "problem robustness": 91209,
      "robustness systems": 103134,
      "seen unseen": 104552,
      "dataset studying": 27222,
      "babi dialog": 11278,
      "dialog task": 30606,
      "dataset original": 27075,
      "74 f1score": 1237,
      "furthermore introduce": 46183,
      "accuracy original": 2225,
      "dataset outperforming": 27077,
      "document structure": 33082,
      "discovery problem": 32147,
      "problem finding": 91059,
      "isa relationship": 56961,
      "new context": 78838,
      "context type": 22307,
      "measure differentiate": 67790,
      "types semantic": 126354,
      "based hierarchical": 11757,
      "position terms": 88294,
      "definition text": 28529,
      "measure quantifies": 67817,
      "multiple attributes": 75501,
      "sequential inference": 107226,
      "selection paper": 104811,
      "presents endtoend": 89846,
      "selection model": 104803,
      "model track": 72175,
      "technology challenges": 120041,
      "challenges dstc7": 16153,
      "correct utterance": 24127,
      "given partial": 48079,
      "enhanced sequential": 37519,
      "esim model": 38424,
      "model following": 71200,
      "challenge outofvocabulary": 16073,
      "words second": 134181,
      "second attentive": 104389,
      "attentive hierarchical": 10111,
      "recurrent encoder": 97172,
      "encoder ahre": 36492,
      "sentences hierarchically": 106339,
      "generate descriptive": 46924,
      "descriptive representations": 29504,
      "pooling method": 88043,
      "pooling used": 88050,
      "used instead": 128595,
      "simple combination": 109385,
      "average pooling": 11204,
      "layer added": 61699,
      "emphasize importance": 36140,
      "utterance context": 130592,
      "context response": 22251,
      "released evaluation": 98481,
      "ubuntu dataset": 126517,
      "2018 paper": 666,
      "presents description": 89837,
      "indic languages": 53829,
      "task 2018": 117821,
      "explore statistical": 41590,
      "pairs present": 83607,
      "approaches context": 8110,
      "models highest": 73331,
      "metrics english": 69955,
      "english telugu": 37305,
      "hindi bengali": 50460,
      "image editing": 51781,
      "editing natural": 34547,
      "task modifying": 118405,
      "editing program": 34552,
      "program using": 92117,
      "language written": 60352,
      "collected crowdsourcing": 18412,
      "framework composed": 45461,
      "actions entities": 3091,
      "process complete": 91441,
      "annotation corpus": 6290,
      "experimented different": 40745,
      "bidirectional lstmcrf": 13958,
      "models best": 72842,
      "editing actions": 34543,
      "realtime applications": 96137,
      "applications users": 7035,
      "users daily": 129107,
      "support systems": 115993,
      "systems just": 116960,
      "particular text": 85458,
      "form text": 45134,
      "processed information": 91601,
      "task regarding": 118620,
      "number methods": 80911,
      "pipelines paper": 87568,
      "investigate propose": 56803,
      "step address": 113245,
      "address word": 3779,
      "german language": 47901,
      "language approach": 58838,
      "methods conclude": 69390,
      "sophisticated nlp": 110674,
      "performance exploring": 86365,
      "importance context": 52049,
      "neural ner": 78226,
      "ner models": 77055,
      "nlu systems": 79813,
      "systems taskoriented": 117187,
      "systems slot": 117146,
      "using gazetteers": 129695,
      "domain ontology": 33601,
      "features endtoend": 43482,
      "recent state": 96515,
      "short noisy": 108225,
      "previous utterance": 90515,
      "utterance dialogue": 130594,
      "methods evaluated": 69482,
      "public english": 94258,
      "datasets belonging": 27330,
      "domains respectively": 33851,
      "additional evaluation": 3511,
      "experiments adding": 40760,
      "automatically translated": 10835,
      "versions english": 131622,
      "english dataset": 37106,
      "embeddings popular": 35865,
      "unsupervised embeddings": 127632,
      "embeddings inadequate": 35752,
      "embeddings lead": 35778,
      "lead suboptimal": 61870,
      "suboptimal performance": 114790,
      "performance propose": 86625,
      "novel learning": 80620,
      "general nlp": 46687,
      "embeddings ensure": 35696,
      "method various": 69219,
      "tasks consistent": 119006,
      "consistent improvement": 21381,
      "autoencoder neural": 10406,
      "models conditional": 72953,
      "conditional text": 20789,
      "generation context": 47344,
      "encoderdecoder nmt": 36617,
      "paradigm introducing": 84537,
      "model features": 71176,
      "coattention mechanism": 18064,
      "parikh et": 84862,
      "inference network": 54173,
      "additional challenges": 3497,
      "discrete nature": 32171,
      "approaches mitigate": 8234,
      "mitigate issue": 70370,
      "baseline presented": 12288,
      "present exploration": 89485,
      "learned latent": 62228,
      "capable capturing": 15194,
      "utilizes latent": 130554,
      "model scalable": 71947,
      "interactive speech": 56030,
      "systems time": 117197,
      "existing training": 40321,
      "longer sufficient": 65163,
      "user interactions": 129004,
      "grammar training": 48661,
      "use constrained": 127957,
      "evaluated approach": 38950,
      "rate 15": 95772,
      "available application": 10944,
      "recognition significant": 96999,
      "practical problem": 88708,
      "spurred greater": 112066,
      "online marketplaces": 81781,
      "want identify": 132016,
      "extract important": 42087,
      "mobile phone": 70469,
      "phone models": 87273,
      "semisupervised methods": 105615,
      "poorly domains": 88066,
      "domains research": 33850,
      "addresses challenge": 3794,
      "algorithm identifying": 4750,
      "facebook posts": 42742,
      "algorithm consists": 4709,
      "fields crf": 43992,
      "model partofspeech": 71716,
      "set simple": 107582,
      "candidate names": 15113,
      "remove spurious": 98877,
      "using clustering": 129444,
      "maximumlikelihood estimation": 67554,
      "sequence tasks": 107086,
      "target token": 117732,
      "prediction probabilities": 89106,
      "sequence tokens": 107088,
      "probabilities tokens": 90852,
      "results low": 101894,
      "generated sequences": 47122,
      "sequences paper": 107130,
      "tokens target": 122338,
      "adjust weights": 3875,
      "token sentence": 122265,
      "summarization task": 115574,
      "tasks achieves": 118916,
      "scores wmt14": 104220,
      "englishgerman wmt17": 37397,
      "wmt17 chineseenglish": 132838,
      "task generalpurpose": 118235,
      "significant portion": 108847,
      "documentary linguists": 33118,
      "linguists work": 64630,
      "work advances": 134361,
      "advances natural": 3998,
      "processing help": 91676,
      "training material": 123702,
      "questions remain": 95355,
      "human involvement": 50874,
      "project attempt": 92211,
      "documentation process": 33122,
      "process use": 91583,
      "nlp technology": 79781,
      "based recent": 11972,
      "massively multilingual": 67341,
      "data current": 25814,
      "current progress": 25327,
      "progress fronts": 92152,
      "automatic phoneme": 10622,
      "problems corpus": 91311,
      "hypotheses based": 51250,
      "size 150": 109910,
      "corpus constructed": 23719,
      "real text": 96080,
      "using crowd": 129522,
      "believe corpus": 12696,
      "naturallanguage inference": 76633,
      "inference systems": 54230,
      "measuring similarity": 67933,
      "readers interests": 95959,
      "procedures used": 91408,
      "used computational": 128441,
      "encountered nlp": 36744,
      "sentences compared": 106249,
      "sql patterns": 112069,
      "unseen questions": 127535,
      "scenario approach": 103819,
      "led improved": 63277,
      "understanding building": 126806,
      "networkbased natural": 77489,
      "collect training": 18393,
      "generation synthetic": 47651,
      "quick way": 95386,
      "collect data": 18376,
      "data variety": 26616,
      "real natural": 96071,
      "challenge using": 16117,
      "real data": 96061,
      "understanding evaluate": 126836,
      "test datasets": 120443,
      "subtitles movies": 114995,
      "dataset multiple": 27043,
      "multiple topic": 75731,
      "problems themes": 91378,
      "population structure": 88168,
      "structure conversations": 113829,
      "considered application": 21286,
      "asr error": 9428,
      "noise speaker": 80068,
      "types speech": 126362,
      "speech disfluencies": 111678,
      "requires composition": 100252,
      "strategy introduced": 113524,
      "introduced paper": 56595,
      "survey large": 116176,
      "available given": 11003,
      "given time": 48160,
      "time period": 122067,
      "containing data": 21784,
      "data analyzed": 25600,
      "consider methods": 21215,
      "local feature": 64919,
      "type feature": 126201,
      "methods novel": 69645,
      "high density": 50060,
      "precisely detected": 88786,
      "accuracy improved": 2180,
      "hypotheses generated": 51254,
      "basic systems": 12535,
      "provided corpus": 93960,
      "proposed strategy": 93558,
      "supervised domain": 115751,
      "mechanism utilizes": 68048,
      "softmax attention": 110505,
      "supervised attention": 115728,
      "attention information": 9860,
      "analysis seen": 5833,
      "seen progress": 104537,
      "past decades": 85639,
      "rnns cnns": 102963,
      "task recently": 118612,
      "recently new": 96724,
      "networks selfattention": 77754,
      "selfattention networks": 104895,
      "networks sans": 77753,
      "utilizes attention": 130551,
      "basic building": 12513,
      "recurrence convolutions": 97154,
      "analysis demonstrate": 5561,
      "accuracy datasets": 2124,
      "model characteristics": 70819,
      "characteristics training": 16541,
      "sequence position": 107029,
      "analysis study": 5876,
      "corpus encyclopedic": 23771,
      "texts article": 121458,
      "presents automatic": 89822,
      "analysis evaluated": 5604,
      "evaluated corpus": 38965,
      "corpus french": 23811,
      "texts annotated": 121456,
      "annotated according": 6136,
      "labeling model": 58511,
      "frame identification": 45384,
      "purpose study": 94440,
      "data point": 26240,
      "rnns implicitly": 102971,
      "representations recurrent": 99843,
      "representations symbolic": 99922,
      "structures sequences": 114105,
      "hypothesis rnns": 51275,
      "words test": 134262,
      "hypothesis introduce": 51266,
      "treebased rnn": 125671,
      "trained encode": 123126,
      "sentences largely": 106372,
      "marginal improvements": 67205,
      "representations standard": 99906,
      "sequence representations": 107043,
      "time existing": 122027,
      "structural representations": 113792,
      "modeling framework": 72434,
      "framework built": 45449,
      "providing simple": 94132,
      "model components": 70870,
      "components using": 20051,
      "models inference": 73399,
      "report experience": 98994,
      "novel modeling": 80648,
      "industrial scale": 54059,
      "scale multiple": 103739,
      "conversation contain": 22938,
      "proposed multiple": 93502,
      "based cosine": 11616,
      "entire conversation": 37701,
      "selected words": 104742,
      "successive words": 115200,
      "words considered": 133876,
      "results results": 102136,
      "obtained proposed": 81399,
      "outperform results": 82738,
      "obtained support": 81419,
      "audio document": 10229,
      "humanhuman conversation": 51026,
      "analysis speech": 5868,
      "individual neurons": 53924,
      "nlp interpretability": 79624,
      "interpretability remains": 56232,
      "challenge previous": 16085,
      "learn representation": 62126,
      "representation level": 99315,
      "individual dimensions": 53906,
      "methods linguistic": 69600,
      "correlation analysis": 24221,
      "based supervised": 12088,
      "supervised method": 115799,
      "analysis unsupervised": 5923,
      "effectiveness techniques": 34966,
      "networks performance": 77697,
      "aim address": 4442,
      "address following": 3680,
      "following questions": 44981,
      "distributed different": 32611,
      "properties models": 92466,
      "models ii": 73353,
      "properties iii": 92455,
      "important neurons": 52195,
      "overall task": 83262,
      "task code": 117968,
      "dalvi et": 25523,
      "neurons neural": 78748,
      "toolkit provides": 122419,
      "respect model": 101091,
      "user visualize": 129056,
      "measure effect": 67794,
      "effect model": 34602,
      "model test": 72158,
      "analysis potential": 5758,
      "various research": 131185,
      "architectural choices": 8600,
      "choices model": 16915,
      "dialog learning": 30568,
      "systems key": 116961,
      "key idea": 57574,
      "space clusters": 110980,
      "utterance conversation": 130593,
      "model relies": 71897,
      "model methodology": 71513,
      "applied endtoend": 7070,
      "dialog technology": 30609,
      "challenges dstc6": 16152,
      "candidates order": 15141,
      "results technique": 102262,
      "technique proposed": 119811,
      "paper achieves": 83697,
      "correct candidate": 24095,
      "networks identifying": 77622,
      "coherence features": 18299,
      "german corpus": 47882,
      "showed method": 108380,
      "method achieved": 68587,
      "achieved best": 2597,
      "equal error": 38223,
      "including translation": 53401,
      "paper generation": 83967,
      "generation best": 47326,
      "language low": 59259,
      "model vietnamese": 72306,
      "joint vietnamese": 57331,
      "segmentation partofspeech": 104610,
      "bist graphbased": 14232,
      "parser kiperwasser": 84967,
      "kiperwasser goldberg": 57700,
      "vietnamese benchmark": 131669,
      "results joint": 101869,
      "competitive performances": 19672,
      "parsing selfattention": 85223,
      "unsupervised pretraining": 127696,
      "benefits pretraining": 13018,
      "fasttext elmo": 43202,
      "elmo bert": 35310,
      "english bert": 37078,
      "bert outperforms": 13196,
      "model capacity": 70805,
      "fasttext embeddings": 43203,
      "embeddings pretraining": 35880,
      "tested large": 120574,
      "model sizes": 72058,
      "million parameters": 70102,
      "expensive train": 40436,
      "train separate": 123014,
      "separate models": 106856,
      "language address": 58816,
      "address shortcoming": 3764,
      "joint multilingual": 57299,
      "multilingual pretraining": 75341,
      "pretraining finetuning": 90258,
      "languages final": 60585,
      "size compared": 109914,
      "32 relative": 932,
      "relative error": 98351,
      "finetuning gives": 44461,
      "way benefit": 132060,
      "benefit larger": 12977,
      "larger datasets": 61361,
      "finally demonstrate": 44164,
      "demonstrate new": 28808,
      "918 f1": 1394,
      "literature review": 64771,
      "review machine": 102554,
      "past couple": 85636,
      "couple years": 24456,
      "exchange information": 39945,
      "different regions": 31375,
      "regional languages": 97679,
      "translation poses": 125124,
      "numerous challenges": 81015,
      "given languages": 48056,
      "numerous methods": 81020,
      "proposed past": 93519,
      "aim improving": 4462,
      "study robustness": 114507,
      "measuring performance": 67922,
      "languages literature": 60701,
      "approaches particular": 8277,
      "objects relations": 81150,
      "statistical spoken": 113167,
      "multidomain dialogue": 75090,
      "dialogue structures": 30759,
      "relations work": 98285,
      "novel dialogue": 80545,
      "relations multiple": 98212,
      "entities type": 37883,
      "prototype implementation": 93685,
      "trained policy": 123239,
      "policy using": 87974,
      "baseline furthermore": 12226,
      "modelling relations": 72619,
      "capable processing": 15213,
      "relations present": 98237,
      "present user": 89759,
      "input learns": 55358,
      "vocabulary reduction": 131897,
      "work systematically": 134838,
      "models extensively": 73216,
      "extensively compare": 41964,
      "choice vocabulary": 16911,
      "extremely sparse": 42611,
      "tables text": 117268,
      "applies text": 7155,
      "analyze customer": 5960,
      "reviews automatically": 102597,
      "star rating": 112391,
      "platform users": 87661,
      "users share": 129169,
      "insights strengths": 55546,
      "extracted sentence": 42179,
      "nlp library": 79628,
      "associated aspects": 9593,
      "aspects based": 9375,
      "words fed": 133957,
      "sentiment score": 106791,
      "performance sequential": 86706,
      "attentionbased network": 10083,
      "challenge track": 16114,
      "multiturn context": 75910,
      "systems ranked": 117094,
      "datasets challenge": 27342,
      "challenge focused": 16035,
      "networks explicitly": 77595,
      "different turns": 31521,
      "turns utterances": 125997,
      "sequential matching": 107236,
      "based chain": 11566,
      "multiturn response": 75922,
      "demonstrate potentials": 28820,
      "approaches fully": 8167,
      "fully exploited": 45957,
      "challenge proposed": 16092,
      "including stateoftheart": 53382,
      "models achieves": 72698,
      "performances largescale": 86883,
      "largescale public": 61499,
      "selection benchmark": 104771,
      "datasets language": 27537,
      "language representations": 60048,
      "induce distributed": 54001,
      "words similar": 134222,
      "words end": 133938,
      "learn distributed": 62028,
      "corpus translated": 24047,
      "translated english": 124552,
      "source languages": 110788,
      "languages just": 60660,
      "just like": 57466,
      "thorny problem": 121875,
      "problem separate": 91216,
      "investigate correlations": 56736,
      "causal relationships": 15813,
      "structural similarity": 113796,
      "languages structural": 60899,
      "similarity correlate": 109214,
      "evaluation previous": 39333,
      "case nlp": 15603,
      "sentence rewriting": 106054,
      "parsing major": 85152,
      "mismatch problem": 70304,
      "ontology paper": 81847,
      "propose sentence": 93036,
      "rewriting based": 102693,
      "sentence new": 105958,
      "form structure": 45128,
      "methods common": 69380,
      "common types": 18943,
      "templatebased method": 120081,
      "parser benchmark": 84924,
      "results outperforms": 102020,
      "outperforms base": 82839,
      "gain f1": 46341,
      "analysis pretrained": 5766,
      "pretrained sentence": 90182,
      "acceptability judgments": 1891,
      "evaluating grammatical": 39060,
      "grammatical knowledge": 48712,
      "knowledge pretrained": 58115,
      "phenomena introduce": 87227,
      "new analysis": 78771,
      "analysis dataset": 5557,
      "set corpus": 107406,
      "corpus linguistic": 23867,
      "linguistic acceptability": 64418,
      "acceptability cola": 1889,
      "set investigate": 107469,
      "pretrained encoders": 90033,
      "encoders bert": 36634,
      "bert devlin": 13095,
      "devlin et": 30468,
      "2018 gpt": 662,
      "gpt radford": 48554,
      "radford et": 95433,
      "models strong": 74102,
      "structures like": 114086,
      "book read": 14413,
      "dependencies like": 29093,
      "challenge models": 16055,
      "models bert": 72831,
      "bert gpt": 13136,
      "nearhuman performance": 76729,
      "acceptability classification": 1888,
      "classification overall": 17317,
      "technology challenge": 120039,
      "challenges dstc": 16151,
      "shared datasets": 107971,
      "datasets explore": 27477,
      "problem building": 90956,
      "building dialog": 14834,
      "recently endtoend": 96686,
      "modeling approaches": 72376,
      "dialog tasks": 30607,
      "developing technologies": 30363,
      "audio visual": 10243,
      "visual scene": 131807,
      "dialog paper": 30577,
      "paper summarizes": 84454,
      "descriptions different": 29479,
      "provided datasets": 93962,
      "datasets overall": 27607,
      "overall trends": 83265,
      "results track": 102276,
      "participants achieved": 85311,
      "stateoftheart endtoend": 112654,
      "quantitative reasoning": 94882,
      "intelligent natural": 55850,
      "handle present": 49401,
      "performance published": 86642,
      "nli models": 79546,
      "achieve absolute": 2411,
      "improvement majorityclass": 52721,
      "majorityclass baseline": 66614,
      "implicitly learn": 52030,
      "establish new": 38584,
      "nli model": 79545,
      "achieves success": 2916,
      "numerical reasoning": 81008,
      "verbal reasoning": 131538,
      "reasoning capabilities": 96230,
      "framework support": 45705,
      "learning compositional": 62449,
      "flexible ways": 44667,
      "recent improvements": 96465,
      "improvements machine": 52871,
      "algorithms require": 4879,
      "require vast": 100205,
      "struggle generalize": 114125,
      "generalize new": 46813,
      "human abilities": 50733,
      "learn use": 62182,
      "examples fewshot": 39835,
      "learning successfully": 63072,
      "complex ways": 19893,
      "additional experiments": 3513,
      "people make": 85879,
      "solving tasks": 110652,
      "mutual exclusivity": 75975,
      "onetoone mappings": 81731,
      "cognitive modeling": 18271,
      "humanlike language": 51055,
      "learning capabilities": 62417,
      "autonomous vehicles": 10853,
      "understanding passenger": 126916,
      "passenger intents": 85622,
      "developing contextual": 30337,
      "contextual dialogue": 22453,
      "responsible handling": 101311,
      "amie automatedvehicle": 5319,
      "automatedvehicle multimodal": 10483,
      "multimodal incabin": 75431,
      "incabin experience": 53207,
      "support various": 115998,
      "dataset multiturn": 27045,
      "various recent": 131183,
      "based techniques": 12109,
      "scenarios experimental": 103846,
      "achieved f1score": 2624,
      "f1score 091": 42706,
      "utterancelevel intent": 130621,
      "intent recognition": 55907,
      "extraction models": 42393,
      "analyzing word": 6058,
      "changes time": 16393,
      "point identify": 87803,
      "unique properties": 127192,
      "presented work": 89804,
      "exploration generating": 41501,
      "pairs generated": 83553,
      "glove embeddings": 48301,
      "algorithm detecting": 4718,
      "small dataset": 110147,
      "work left": 134609,
      "space sentence": 111059,
      "transition matrix": 124491,
      "topic field": 122519,
      "sentence key": 105914,
      "classification document": 17190,
      "summarization various": 115592,
      "sts tasks": 114133,
      "measure degree": 67788,
      "semantic preservation": 105183,
      "yielded stateoftheart": 135372,
      "method practical": 69059,
      "applied sentence": 7113,
      "achieve robust": 2528,
      "examples improving": 39846,
      "improving sequencetosequence": 53165,
      "learning optimal": 62861,
      "models commonly": 72929,
      "commonly trained": 18964,
      "mle training": 70444,
      "word given": 133298,
      "focuses modeling": 44909,
      "syntactic patterns": 116447,
      "longrange semantic": 65184,
      "structure present": 113935,
      "alleviate issues": 5038,
      "based optimal": 11914,
      "preservation semantic": 89917,
      "distribution extensive": 32654,
      "utility proposed": 130489,
      "improvements wide": 52938,
      "translation abstractive": 124614,
      "abstractive text": 1823,
      "summarization image": 115516,
      "captioning crosslingual": 15257,
      "model pretraining": 71779,
      "pretraining recent": 90311,
      "studies demonstrated": 114206,
      "generative pretraining": 47753,
      "approach multiple": 7720,
      "languages effectiveness": 60521,
      "effectiveness crosslingual": 34879,
      "crosslingual pretraining": 24995,
      "pretraining propose": 90308,
      "unsupervised relies": 127702,
      "new crosslingual": 78849,
      "results crosslingual": 101648,
      "art absolute": 9032,
      "absolute gain": 1730,
      "accuracy unsupervised": 2312,
      "bleu wmt16": 14345,
      "wmt16 germanenglish": 132832,
      "obtain new": 81302,
      "outperforming previous": 82816,
      "best approach": 13305,
      "approach bleu": 7395,
      "code pretrained": 18132,
      "semisupervised text": 105628,
      "method semisupervised": 69122,
      "embeddings outperforms": 35848,
      "outperforms classifier": 82863,
      "randomly initialized": 95529,
      "embeddings empirically": 35684,
      "empirically observed": 36239,
      "method builds": 68683,
      "form model": 45105,
      "model ensemble": 71083,
      "differently using": 31599,
      "using random": 130078,
      "focus different": 44752,
      "different predictions": 31344,
      "data following": 25954,
      "selftraining framework": 104956,
      "framework use": 45730,
      "cotraining framework": 24409,
      "automated essay": 10437,
      "essay scoring": 38531,
      "current stateofart": 25347,
      "essay score": 38530,
      "adversarial samples": 4161,
      "problem develop": 91009,
      "experiments compare": 40834,
      "robustness models": 103111,
      "surpasses baselines": 116103,
      "adding adversarial": 3371,
      "original datasets": 82511,
      "great extent": 49003,
      "analysis lexical": 5679,
      "having multiple": 49563,
      "senses text": 105721,
      "semantic task": 105325,
      "task particular": 118519,
      "appropriate given": 8420,
      "refers identification": 97582,
      "meaning polysemous": 67660,
      "computational algorithms": 20356,
      "algorithms language": 4861,
      "official language": 81600,
      "language india": 59167,
      "work prepared": 134673,
      "new senses": 79125,
      "hindi corpus": 50461,
      "words furthermore": 133967,
      "analyzed novel": 6027,
      "word evaluation": 133276,
      "favorable results": 43215,
      "evaluating stateoftheart": 39103,
      "endtoend natural": 36934,
      "provides comprehensive": 94014,
      "analysis shared": 5841,
      "avenues future": 11167,
      "based results": 11995,
      "task aimed": 117862,
      "aimed assess": 4478,
      "assess recent": 9492,
      "endtoend nlg": 36949,
      "systems generate": 116919,
      "generate complex": 46914,
      "learning datasets": 62484,
      "introducing novel": 56649,
      "great potential": 49010,
      "wordoverlap metrics": 133788,
      "vanilla seq2seq": 130821,
      "strong semantic": 113709,
      "mechanism applied": 67947,
      "applied decoding": 7056,
      "models outperformed": 73688,
      "terms overall": 120354,
      "quality complexity": 94614,
      "diversity outputs": 32888,
      "number recent": 80951,
      "systems chatbots": 116778,
      "combination transfer": 18591,
      "based training": 12124,
      "training scheme": 123829,
      "finetuning performed": 44490,
      "finetuned model": 44433,
      "strong improvements": 113681,
      "improvements current": 52837,
      "like memory": 64065,
      "memory augmented": 68289,
      "personachat dataset": 87110,
      "dataset conversational": 26831,
      "conversational intelligence": 23009,
      "intelligence challenge": 55828,
      "challenge approach": 15995,
      "approach obtains": 7746,
      "45 absolute": 1033,
      "46 absolute": 1040,
      "review sentiment": 102575,
      "carry sentiment": 15558,
      "custom dictionaries": 25433,
      "endtoend conversation": 36881,
      "work track": 134855,
      "aims evaluate": 4526,
      "factual texts": 42938,
      "models widely": 74339,
      "used endtoend": 128513,
      "endtoend generative": 36903,
      "generative conversation": 47718,
      "responses previous": 101290,
      "aims promote": 4562,
      "promote diversity": 92317,
      "follows twostage": 44993,
      "twostage pipeline": 126163,
      "generate multiple": 46971,
      "multiple responses": 75673,
      "variational generative": 130924,
      "retrieval based": 102383,
      "based retrieval": 11996,
      "model rank": 71843,
      "response training": 101247,
      "ranking process": 95683,
      "according official": 1991,
      "official evaluation": 81597,
      "diversity metrics": 32883,
      "metrics entropy": 69956,
      "meteor metrics": 68570,
      "pay attention": 85810,
      "lightweight dynamic": 64011,
      "dynamic convolutions": 34299,
      "mechanism build": 67954,
      "build generative": 14768,
      "step paper": 113291,
      "perform competitively": 85968,
      "selfattention results": 104902,
      "simpler efficient": 109556,
      "efficient selfattention": 35110,
      "kernels based": 57536,
      "based solely": 12057,
      "order determine": 82309,
      "operations required": 82074,
      "approach scales": 7867,
      "scales linearly": 103770,
      "selfattention models": 104891,
      "models wmt14": 74346,
      "englishgerman test": 37385,
      "bleu training": 14342,
      "classification explore": 17207,
      "explore various": 41614,
      "computing sentence": 20529,
      "representations pretrained": 99815,
      "surprisingly little": 116142,
      "make important": 66684,
      "classification evaluation": 17200,
      "scratch paper": 104243,
      "neural pipeline": 78631,
      "input performs": 55391,
      "performs tasks": 87035,
      "tasks required": 119463,
      "submission achieved": 114725,
      "2nd 1st": 887,
      "3rd official": 996,
      "systems lowresource": 116986,
      "metrics large": 69978,
      "different model": 31262,
      "studies endtoend": 114217,
      "current conversational": 25269,
      "systems attracted": 116741,
      "develop dialogue": 30190,
      "automatic medical": 10590,
      "medical diagnosis": 68201,
      "collect additional": 18372,
      "topic transition": 122579,
      "question understanding": 95234,
      "understanding automatic": 126798,
      "medical knowledge": 68209,
      "relations existing": 98163,
      "li et": 63953,
      "datadriven learning": 26664,
      "learning able": 62302,
      "transition dialogue": 124487,
      "generation novel": 47520,
      "deep qnetwork": 28400,
      "topic transitions": 122581,
      "beats stateoftheart": 12622,
      "accuracy superiority": 2294,
      "retaining original": 102362,
      "data patients": 26228,
      "sharing hierarchical": 108130,
      "hierarchical hidden": 49967,
      "context adaptive": 21999,
      "users utterances": 129183,
      "utterances corresponding": 130632,
      "slots values": 110093,
      "available addition": 10935,
      "addition semantic": 3473,
      "data contain": 25787,
      "framework based": 45435,
      "used present": 128692,
      "phonetic transcriptions": 87299,
      "results positive": 102043,
      "positive effects": 88319,
      "representations evaluation": 99638,
      "evaluation experiments": 39213,
      "heldout data": 49686,
      "high accuracies": 50031,
      "amounts training": 5362,
      "easy data": 34475,
      "boosting performance": 14446,
      "synonym replacement": 116348,
      "replacement random": 98941,
      "deletion text": 28597,
      "particularly strong": 85497,
      "results smaller": 102195,
      "datasets training": 27765,
      "using 50": 129313,
      "set achieved": 107347,
      "achieved accuracy": 2592,
      "performed extensive": 86922,
      "studies suggest": 114285,
      "heads layer": 49607,
      "provide superior": 93936,
      "compared simpler": 19442,
      "models determining": 73058,
      "source context": 110736,
      "context relevant": 22247,
      "model open": 71609,
      "extension transformer": 41840,
      "architecture makes": 8691,
      "use hidden": 128084,
      "information predict": 54856,
      "predict word": 88952,
      "trained bilingual": 123081,
      "optimize attention": 82223,
      "word resulting": 133485,
      "naive approach": 76011,
      "transformer attention": 124295,
      "generalization models": 46781,
      "models generic": 73298,
      "semantic framework": 105060,
      "arguments use": 8997,
      "framework construct": 45469,
      "covering entirety": 24539,
      "entirety universal": 37737,
      "english web": 37335,
      "web treebank": 132266,
      "treebank use": 125649,
      "elmo word": 35326,
      "multilingual transfer": 75390,
      "transfer nlp": 124166,
      "languages applied": 60404,
      "applied lowresource": 7090,
      "lowresource target": 65568,
      "language prior": 59828,
      "used single": 128764,
      "single source": 109799,
      "model carefully": 70814,
      "problem poor": 91164,
      "transfer particularly": 124175,
      "particularly distant": 85478,
      "propose techniques": 93112,
      "zeroshot fewshot": 135500,
      "learning respectively": 62966,
      "evaluating named": 39079,
      "including standard": 53381,
      "ensembling unsupervised": 37634,
      "model human": 71290,
      "human acceptability": 50735,
      "extractive sentence": 42560,
      "compression recent": 20321,
      "rely parallel": 98724,
      "particular application": 85380,
      "used support": 128798,
      "compression task": 20324,
      "task release": 118626,
      "release model": 98458,
      "dataset future": 26947,
      "analogical reasoning": 5409,
      "new theoretical": 79220,
      "embedding proposed": 35477,
      "reasoning refers": 96304,
      "basic concept": 12516,
      "use techniques": 128321,
      "data distributed": 25859,
      "platforms review": 87672,
      "review conversational": 102537,
      "comprehension inspired": 20186,
      "comprehension crc": 20174,
      "studies novel": 114257,
      "task leveraging": 118351,
      "answer multiturn": 6544,
      "dataset propose": 27112,
      "model bert": 70761,
      "bert pretraining": 13208,
      "pretraining domainspecific": 90254,
      "performance end": 86337,
      "approach highly": 7605,
      "datasets lowresource": 27552,
      "lowresource machine": 65532,
      "nepalienglish sinhalaenglish": 77015,
      "majority language": 66603,
      "world considered": 135023,
      "considered lowresource": 21295,
      "little parallel": 64822,
      "technical challenges": 119742,
      "learning limited": 62692,
      "difficult evaluate": 31625,
      "trained lowresource": 123191,
      "pairs lack": 83568,
      "freely publicly": 45790,
      "available benchmarks": 10952,
      "benchmarks work": 12942,
      "based sentences": 12019,
      "outofdomain parallel": 82656,
      "data freely": 25959,
      "process collect": 91437,
      "report baseline": 98978,
      "baseline performance": 12283,
      "using learning": 129806,
      "fully supervised": 45971,
      "supervised weakly": 115855,
      "lowresource mt": 65535,
      "code reproduce": 18149,
      "reproduce experiments": 100081,
      "experiments available": 40788,
      "learned vector": 62280,
      "semantics tasks": 105472,
      "models far": 73238,
      "far failed": 43078,
      "investigate approaches": 56718,
      "approaches jointly": 8202,
      "jointly train": 57392,
      "train word": 123050,
      "linked entities": 64644,
      "distinct approaches": 32531,
      "training stateoftheart": 123881,
      "versions corpus": 131621,
      "node embeddings": 80020,
      "cooccurrence graph": 23226,
      "embeddings classical": 35594,
      "classical word": 17094,
      "embeddings variety": 36013,
      "variety word": 131031,
      "investigate performance": 56793,
      "training popular": 123782,
      "models annotated": 72753,
      "acceptable performance": 1894,
      "performance common": 86217,
      "common test": 18937,
      "test cases": 120426,
      "results discuss": 101731,
      "method addresses": 68621,
      "analysis problems": 5771,
      "text images": 121042,
      "models classify": 72907,
      "classify sentiment": 17661,
      "straightforward apply": 113429,
      "latin languages": 61664,
      "model designed": 70973,
      "level applying": 63423,
      "model classify": 70831,
      "sentence applied": 105772,
      "dataset obtained": 27062,
      "accuracy gain": 2168,
      "gain compared": 46339,
      "method multimodal": 68976,
      "multimodal data": 75422,
      "data unstructured": 26596,
      "unstructured natural": 127575,
      "comprehension answer": 20164,
      "answer reranking": 6571,
      "new ideas": 78952,
      "networks transformer": 77801,
      "key ingredients": 57581,
      "stateoftheart dialog": 112638,
      "generate meaningful": 46969,
      "unseen situation": 127542,
      "data build": 25711,
      "build reliable": 14803,
      "reliable model": 98619,
      "answers different": 6727,
      "question context": 95140,
      "context train": 22302,
      "reranking framework": 100371,
      "framework improve": 45566,
      "semantics individual": 105425,
      "individual models": 53919,
      "coherent topic": 18330,
      "searching information": 104371,
      "information human": 54658,
      "spots relevant": 112044,
      "challenge present": 16082,
      "documents coherent": 33202,
      "learns latent": 63217,
      "text segment": 121275,
      "segment document": 104561,
      "topic shifts": 122576,
      "labeled sections": 58466,
      "distinct domains": 32534,
      "classification 30": 17097,
      "topics english": 122626,
      "stateoftheart cnn": 112609,
      "segmentation leveraging": 104597,
      "treebanks parsing": 125664,
      "data argument": 25620,
      "problem parsing": 91161,
      "morphologicallyrich languages": 74779,
      "occurs frequently": 81521,
      "evaluate stateoftheart": 38929,
      "parsing new": 85176,
      "trees sentences": 125711,
      "movie scripts": 74905,
      "posts hindi": 88518,
      "applied conversational": 7052,
      "alternative word": 5252,
      "structures treebank": 114112,
      "representations contextual": 99566,
      "tell story": 120069,
      "story words": 113421,
      "detailed mathematical": 29780,
      "algorithms does": 4844,
      "does focus": 33352,
      "translation question": 125190,
      "ideas presented": 51351,
      "papers authors": 84502,
      "reading document": 96019,
      "general understanding": 46725,
      "understanding word": 127017,
      "known word": 58331,
      "problems solve": 91374,
      "changed time": 16379,
      "discussion recent": 32325,
      "recent advance": 96388,
      "language similarities": 60094,
      "years novel": 135283,
      "accurate language": 2339,
      "effectively models": 34833,
      "share lexical": 107954,
      "idea representing": 51337,
      "lower dimension": 65428,
      "newly discovered": 79270,
      "vocabulary existing": 131874,
      "existing languages": 40156,
      "investigating effect": 56855,
      "segmentation methods": 104600,
      "based sentiment": 12020,
      "segmentation approaches": 104575,
      "segmentation deep": 104583,
      "model segmentation": 71956,
      "focuses preprocessing": 44911,
      "subword tokenization": 115039,
      "focuses evaluation": 44897,
      "rnn model": 102926,
      "models suffers": 74129,
      "suffers generating": 115258,
      "greatly reduce": 49060,
      "incoherent phrases": 53413,
      "languages linguistic": 60700,
      "linguistic differences": 64466,
      "gap research": 46477,
      "applied approach": 7045,
      "performance translation": 86815,
      "translation evaluated": 124798,
      "score metric": 104092,
      "translation bleu": 124679,
      "approach performed": 7793,
      "commonly seen": 18963,
      "slightly higher": 110022,
      "higher score": 50206,
      "fast robust": 43150,
      "biomedical natural": 14197,
      "processing despite": 91661,
      "processing statistical": 91801,
      "models processing": 73806,
      "text perform": 121176,
      "domain shift": 33637,
      "biomedical clinical": 14166,
      "spacy library": 111111,
      "models released": 73915,
      "neural named": 78218,
      "ner important": 77048,
      "task nlp": 118460,
      "nlp challenging": 79585,
      "conversational domain": 23006,
      "texts available": 121462,
      "making supervised": 66867,
      "required previous": 100224,
      "biases transfer": 13888,
      "explore transfer": 41599,
      "data generic": 25979,
      "generic domain": 47796,
      "training limited": 123688,
      "data report": 26338,
      "learning variants": 63155,
      "pretrained lm": 90120,
      "lm finetuning": 64862,
      "sentences yielding": 106550,
      "yielding absolute": 135375,
      "points test": 87878,
      "test f1": 120455,
      "furthermore pretrained": 46203,
      "lm encodes": 64860,
      "partofspeech information": 85527,
      "information strong": 55006,
      "classification knowledge": 17242,
      "knowledge powered": 58111,
      "powered attention": 88661,
      "documents short": 33292,
      "information poses": 54850,
      "poses great": 88265,
      "great challenge": 48995,
      "challenge classification": 16006,
      "knowledge external": 57924,
      "enhance semantic": 37491,
      "representation short": 99409,
      "information kind": 54708,
      "knowledge incorporate": 58013,
      "purpose measuring": 94432,
      "measuring importance": 67913,
      "importance knowledge": 52062,
      "knowledge introduce": 58023,
      "introduce attention": 56382,
      "mechanisms propose": 68067,
      "text st": 121327,
      "text help": 121033,
      "information unlike": 55069,
      "approaches model": 8235,
      "like human": 64046,
      "data machines": 26107,
      "pays attention": 85815,
      "attention important": 9857,
      "important knowledge": 52176,
      "knowledge conduct": 57838,
      "tasks experimental": 119118,
      "studies model": 114254,
      "attention improving": 9859,
      "multilingual sentence": 75360,
      "embedding using": 35524,
      "additive margin": 3642,
      "margin softmax": 67200,
      "learn multilingual": 62094,
      "use pairs": 128186,
      "embeddings constructed": 35610,
      "task document": 118113,
      "achieve 97": 2410,
      "pairs lastly": 83572,
      "mining task": 70262,
      "task learned": 118347,
      "compared current": 19354,
      "stateoftheart level": 112691,
      "cognitive models": 18272,
      "models naturalistic": 73616,
      "knowledge categories": 57823,
      "categories exist": 15736,
      "allow efficiently": 5081,
      "efficiently represent": 35155,
      "tested data": 120565,
      "experiments involving": 40968,
      "artificially restricted": 9274,
      "college students": 18513,
      "rich noisy": 102775,
      "environment humans": 38185,
      "humans live": 51088,
      "languages cultures": 60477,
      "present bayesian": 89394,
      "designed jointly": 29604,
      "text allows": 120638,
      "allows evaluate": 5151,
      "performance large": 86487,
      "richly structured": 102813,
      "illustrates potential": 51750,
      "advances computational": 3978,
      "computational modeling": 20409,
      "science research": 103983,
      "novel representation": 80707,
      "approach inspired": 7640,
      "used image": 128577,
      "image representation": 51794,
      "representation works": 99475,
      "order learn": 82352,
      "document associated": 32953,
      "range recent": 95594,
      "methods demonstrating": 69430,
      "considerable improvement": 21250,
      "review data": 102541,
      "approach code": 7422,
      "gender prediction": 46607,
      "task gender": 118232,
      "prediction present": 89104,
      "detection dutch": 29930,
      "models cnn": 72913,
      "data preprocessing": 26262,
      "models combination": 72921,
      "combination word": 18596,
      "traditional nonneural": 122857,
      "previous tasks": 90509,
      "results described": 101715,
      "described previous": 29380,
      "models performed": 73738,
      "weighted ensemble": 132344,
      "models final": 73247,
      "language technologies": 60154,
      "evaluations tasks": 39487,
      "entity discovery": 37924,
      "discovery linking": 32140,
      "linking edl": 64655,
      "analysis adversarial": 5491,
      "adversarial settings": 4164,
      "detected using": 29826,
      "conducting extensive": 20963,
      "extensive literature": 41943,
      "review existing": 102548,
      "existing empirical": 40124,
      "empirical work": 36213,
      "work demonstrate": 134461,
      "features indicative": 43562,
      "fail generalize": 42961,
      "semantic domains": 105037,
      "domains suggest": 33864,
      "classifying texts": 17681,
      "potentially deceptive": 88607,
      "discuss forms": 32252,
      "style obfuscation": 114586,
      "surveying literature": 116210,
      "author identification": 10365,
      "current style": 25375,
      "style transformation": 114607,
      "transformation methods": 124267,
      "fail achieve": 42947,
      "faithfulness original": 43009,
      "propose future": 92692,
      "method reduce": 69095,
      "highquality translation": 50414,
      "preserve meaning": 89925,
      "meaning sense": 67686,
      "sense sentences": 105699,
      "sentence language": 105918,
      "language stateoftheart": 60122,
      "meaning space": 67696,
      "space different": 110992,
      "states world": 113058,
      "je suis": 57223,
      "language cases": 58869,
      "cases results": 15665,
      "results important": 101840,
      "bayesian models": 12584,
      "models informative": 73402,
      "informative utterance": 55159,
      "translation terms": 125372,
      "pretrained neural": 90171,
      "model method": 71512,
      "method increases": 68901,
      "quality measured": 94715,
      "measured bleu": 67836,
      "capturing meaning": 15481,
      "long challenging": 65062,
      "linear combinations": 64337,
      "combinations word": 18603,
      "conduct semantic": 20891,
      "units phrases": 127264,
      "sentences documents": 106286,
      "language instance": 59210,
      "levels semantic": 63564,
      "complexvalued vector": 19950,
      "proposed implement": 93310,
      "implement framework": 51933,
      "framework text": 45713,
      "task evaluation": 118158,
      "results benchmarking": 101546,
      "framework works": 45742,
      "informationtheoretic approach": 55123,
      "chinese gigaword": 16773,
      "gigaword corpus": 47964,
      "information bits": 54400,
      "classes nouns": 17067,
      "nouns adjectives": 80439,
      "pretrained convolutional": 90016,
      "convolutional decoders": 23153,
      "framework compare": 45457,
      "variety existing": 130979,
      "existing novel": 40244,
      "keyword generation": 57636,
      "suggestion important": 115356,
      "sponsored search": 112028,
      "search improve": 104320,
      "improve online": 52447,
      "common challenges": 18861,
      "keywords difficult": 57645,
      "difficult discover": 31622,
      "search result": 104346,
      "result pages": 101389,
      "suggest novel": 115324,
      "generative neural": 47748,
      "networks keyword": 77635,
      "sequence input": 106962,
      "generate set": 47013,
      "set keywords": 107470,
      "relevant input": 98560,
      "domain constraint": 33487,
      "utilize domainspecific": 130505,
      "domainspecific information": 33903,
      "shows generative": 108580,
      "improve coverage": 52364,
      "clickthrough rate": 17753,
      "distillation multilingual": 32514,
      "training online": 123762,
      "translation usually": 125421,
      "counterpart using": 24438,
      "using individual": 129750,
      "pair language": 83436,
      "diversity model": 32884,
      "limitations paper": 64183,
      "approach boost": 7396,
      "boost accuracy": 14429,
      "accuracy multilingual": 2211,
      "data match": 26120,
      "models simultaneously": 74056,
      "simultaneously knowledge": 109681,
      "distillation experiments": 32506,
      "particularly model": 85489,
      "accuracy individual": 2189,
      "adapt unseen": 3198,
      "unseen classes": 127518,
      "challenging scenarios": 16310,
      "studies used": 114295,
      "used metalearning": 128629,
      "queries compared": 94915,
      "support set": 115990,
      "class support": 17049,
      "compare new": 19264,
      "induction network": 54027,
      "dynamic routing": 34323,
      "routing algorithm": 103323,
      "way model": 132108,
      "able induce": 1658,
      "better evaluate": 13576,
      "classification dataset": 17170,
      "dataset chinese": 26784,
      "graph attention": 48759,
      "attention networks": 9948,
      "proposed text": 93570,
      "tasks existing": 119112,
      "existing deep": 40107,
      "deep models": 28339,
      "designed semantic": 29622,
      "matching pair": 67422,
      "texts paraphrase": 121574,
      "identification question": 51427,
      "task relevance": 118630,
      "partially fact": 85303,
      "essential characteristics": 38545,
      "models specifically": 74078,
      "specifically methods": 111569,
      "fail handle": 42965,
      "underlying structural": 126703,
      "text long": 121094,
      "critical problems": 24823,
      "problems like": 91336,
      "querydocument matching": 95002,
      "matching information": 67407,
      "undirected graph": 127046,
      "graph propose": 48858,
      "propose multiresolution": 92804,
      "network gcn": 77264,
      "text snippet": 121310,
      "document attention": 32954,
      "mechanisms applied": 68054,
      "deep matching": 28333,
      "evaluating text": 39107,
      "building meaningful": 14862,
      "representations challenging": 99538,
      "representations perform": 99804,
      "static word": 113066,
      "meaning shift": 67690,
      "information performance": 54835,
      "performance far": 86377,
      "tasks related": 119448,
      "composition effects": 20094,
      "research aiming": 100401,
      "aiming improve": 4490,
      "constituent parsers": 21544,
      "models constituent": 72970,
      "faster accurate": 43160,
      "high error": 50069,
      "large label": 61121,
      "label sets": 58414,
      "learn predict": 62116,
      "mitigate issues": 70371,
      "auxiliary losses": 10880,
      "combining techniques": 18753,
      "surpass performance": 116092,
      "performance sequence": 86704,
      "parsers english": 85020,
      "chinese penn": 16801,
      "penn treebanks": 85867,
      "improvements board": 52821,
      "including new": 53340,
      "optimization neural": 82198,
      "problem making": 91121,
      "making efficient": 66840,
      "efficient use": 35127,
      "use heterogeneous": 128083,
      "nmt specifically": 79963,
      "specifically given": 111557,
      "sentencelevel feature": 106160,
      "examples training": 39892,
      "training curriculum": 123415,
      "curriculum framework": 25421,
      "arbitrary number": 8587,
      "number times": 80981,
      "knowledge design": 57857,
      "learning learn": 62684,
      "learn automatically": 61992,
      "jointly nmt": 57371,
      "single training": 109804,
      "run approach": 103460,
      "wmt englishtofrench": 132790,
      "match performance": 67362,
      "chinese using": 16832,
      "learning shown": 63023,
      "scarcity data": 103804,
      "especially languages": 38463,
      "english propose": 37250,
      "lexicosyntactic features": 63940,
      "data apply": 25612,
      "translationbased baselines": 125443,
      "extraction datasets": 42295,
      "datasets digital": 27433,
      "humanities domain": 51041,
      "domain evaluation": 33522,
      "research manually": 100554,
      "manually create": 67052,
      "create highquality": 24617,
      "highquality datasets": 50377,
      "evaluation language": 39247,
      "models step": 74101,
      "fantasy novel": 43063,
      "novel book": 80507,
      "book series": 14415,
      "task types": 118815,
      "various popular": 131170,
      "glove fasttext": 48302,
      "suitability word": 115391,
      "models specific": 74076,
      "specific relation": 111484,
      "comparably small": 19197,
      "corpus sizes": 24000,
      "term frequencies": 120214,
      "task difficulty": 118099,
      "datasets underlying": 27771,
      "underlying word": 126709,
      "extended new": 41807,
      "datasets tasks": 27748,
      "results transferred": 102283,
      "princeton wordnet": 90668,
      "important resources": 52235,
      "available english": 10991,
      "languages expensive": 60567,
      "manual process": 67012,
      "process beneficial": 91430,
      "highquality automatic": 50371,
      "approach support": 7950,
      "wordnet new": 133774,
      "need translate": 76867,
      "including low": 53317,
      "low frequency": 65364,
      "current machine": 25296,
      "experiment shows": 40510,
      "shows significant": 108624,
      "translation contextual": 124709,
      "semeval2019 task": 105543,
      "french discuss": 45805,
      "yielded improvements": 135366,
      "baseline languages": 12237,
      "settings results": 107836,
      "tasks website": 119601,
      "spoken dialogues": 111981,
      "research development": 100467,
      "framework inspired": 45577,
      "humanhuman dialogue": 51031,
      "characteristics spoken": 16536,
      "spoken interactions": 111987,
      "bidirectional attention": 13899,
      "80 f1": 1289,
      "score heldout": 104076,
      "heldout test": 49689,
      "ability automatically": 1487,
      "healthcare domain": 49632,
      "exploiting limited": 41468,
      "demonstrate feasibility": 28743,
      "effective extraction": 34672,
      "extraction retrieval": 42472,
      "retrieval comprehension": 102388,
      "symptom checking": 116331,
      "spoken conversations": 111970,
      "approach select": 7872,
      "recent advanced": 96389,
      "advanced embedding": 3943,
      "models selecting": 73986,
      "aka word": 4586,
      "best fit": 13335,
      "task nontrivial": 118463,
      "extraction provide": 42454,
      "provide method": 93873,
      "space explore": 111003,
      "approach pretrained": 7814,
      "models vietnamese": 74331,
      "models suitable": 74131,
      "pretrained embedding": 90026,
      "task utilize": 118835,
      "apply approach": 7158,
      "embedding selection": 35486,
      "selection helps": 104785,
      "helps users": 49834,
      "select suitable": 104716,
      "embeddings addition": 35543,
      "approach facilitate": 7567,
      "tasks source": 119507,
      "cues using": 25220,
      "unit recognition": 127208,
      "recognition analyze": 96805,
      "acoustic patterns": 3021,
      "analysis spoken": 5869,
      "fusion visual": 46254,
      "based matching": 11830,
      "applying neural": 7263,
      "models level": 73497,
      "individual posts": 53927,
      "approaches identify": 8183,
      "sequential nature": 107240,
      "nature social": 76668,
      "detection problem": 30026,
      "problem social": 91234,
      "adopt neural": 3895,
      "chronological order": 16947,
      "posts specifically": 88524,
      "specifically establish": 111546,
      "establish neural": 38583,
      "neural baseline": 77852,
      "lowresource syntactic": 65567,
      "transfer unsupervised": 124205,
      "method dependency": 68764,
      "parsing takes": 85256,
      "account problem": 2035,
      "data commonly": 25751,
      "data transfer": 26579,
      "bible corpus": 13893,
      "treebanks multiple": 125663,
      "source treebanks": 110860,
      "treebanks training": 125666,
      "language improves": 59156,
      "european language": 38769,
      "dependencies corpus": 29072,
      "corpus achieve": 23640,
      "languages achieve": 60382,
      "method content": 68739,
      "representations syntactic": 99924,
      "analysis plays": 5752,
      "role semantic": 103216,
      "ongoing debate": 81733,
      "comparative studies": 19210,
      "studies syntactic": 114287,
      "hinders development": 50455,
      "parsing methods": 85157,
      "long tail": 65135,
      "approaches finally": 8163,
      "finally proposed": 44225,
      "methodology used": 69268,
      "used finegrained": 128555,
      "finegrained evaluation": 44356,
      "highlighting challenges": 50278,
      "benefit downstream": 12964,
      "applications syntactic": 7023,
      "studied context": 114154,
      "medical records": 68221,
      "spans personal": 111207,
      "fully investigated": 45963,
      "end define": 36802,
      "define task": 28494,
      "present pipeline": 89643,
      "asr ner": 9440,
      "transcript text": 123993,
      "novel metric": 80640,
      "evaluation benchmark": 39133,
      "benchmark consisting": 12744,
      "large labeled": 61123,
      "switchboard fisher": 116289,
      "audio datasets": 10228,
      "set probabilistic": 107544,
      "probabilistic models": 90823,
      "markov chains": 67255,
      "used identify": 128574,
      "model internal": 71360,
      "employed identify": 36299,
      "results comparison": 101610,
      "approach pretraining": 7815,
      "pretraining bidirectional": 90237,
      "understanding problems": 126928,
      "problems model": 91344,
      "text experiments": 120936,
      "demonstrate large": 28767,
      "benchmarks consistent": 12892,
      "introduced bert": 56572,
      "bert model": 13166,
      "analysis number": 5724,
      "number factors": 80889,
      "factors contribute": 42884,
      "contribute effective": 22750,
      "effective pretraining": 34732,
      "including data": 53282,
      "size model": 109935,
      "describes machine": 29408,
      "englishhindi translation": 37411,
      "corpus provided": 23949,
      "provided organizers": 93978,
      "architectures train": 8852,
      "train additional": 122900,
      "analysis comparison": 5535,
      "results systems": 102250,
      "salient differences": 103548,
      "differences systems": 30978,
      "implements number": 51985,
      "number tools": 80983,
      "tools analysis": 122428,
      "generation particular": 47537,
      "types words": 126388,
      "advanced features": 3944,
      "features use": 43771,
      "linguistic labels": 64504,
      "models aims": 72738,
      "aims easily": 4522,
      "types analysis": 126246,
      "study open": 114460,
      "domain question": 33621,
      "answering current": 6615,
      "approaches natural": 8246,
      "focus domainspecific": 44754,
      "domainspecific taskoriented": 33917,
      "slot types": 110070,
      "types usually": 126385,
      "considering previous": 21329,
      "context furthermore": 22112,
      "furthermore approaches": 46146,
      "examples available": 39811,
      "domains work": 33885,
      "explores feasibility": 41647,
      "dialog applications": 30552,
      "opendomain question": 81976,
      "qa based": 94495,
      "opendomain qa": 81974,
      "generation quality": 47577,
      "quality experiment": 94659,
      "types second": 126353,
      "second perform": 104440,
      "benchmark model": 12841,
      "model popular": 71750,
      "experiment using": 40518,
      "additional input": 3528,
      "quality experiments": 94663,
      "learning statistical": 63056,
      "nlg models": 79522,
      "bayesian approach": 12578,
      "uses language": 129237,
      "language speakers": 60111,
      "theories language": 121811,
      "able account": 1591,
      "expressions present": 41761,
      "production model": 92058,
      "framework treats": 45724,
      "relax assumption": 98422,
      "capture large": 15324,
      "complex scenes": 19872,
      "features increase": 43561,
      "parsing pointer": 85194,
      "transitionbased algorithm": 124499,
      "sentences left": 106377,
      "length input": 63360,
      "similarly recent": 109348,
      "ma et": 65734,
      "framework given": 45556,
      "position sentence": 88293,
      "transition sequence": 124495,
      "sequence length": 107003,
      "parser runs": 84994,
      "twice fast": 126086,
      "achieving best": 2928,
      "accuracy date": 2126,
      "date english": 27825,
      "english ptb": 37253,
      "ptb dataset": 94224,
      "fullysupervised singlemodel": 45990,
      "topdown transition": 122493,
      "recurrent attention": 97161,
      "networks dialogue": 77567,
      "topic growing": 122527,
      "goal task": 48389,
      "task select": 118670,
      "challenge dstc7": 16030,
      "track based": 122724,
      "characteristics natural": 16530,
      "knowledge grounding": 57997,
      "candidates proposed": 15142,
      "generalized different": 46826,
      "datasets settings": 27702,
      "learning multilevel": 62811,
      "recurrent transformer": 97263,
      "transformer increasing": 124327,
      "increasing research": 53684,
      "generation emerging": 47383,
      "formulating task": 45292,
      "task selecting": 118671,
      "selecting sentences": 104758,
      "following recent": 44983,
      "success transformer": 115137,
      "based multihead": 11861,
      "attention called": 9802,
      "recurrent model": 97181,
      "transformer proposed": 124371,
      "capability proposed": 15185,
      "model modeling": 71524,
      "representations despite": 99592,
      "despite growing": 29690,
      "models introduced": 73426,
      "lack standardized": 58746,
      "help community": 49705,
      "performance design": 86288,
      "substantial computational": 114854,
      "resources researchers": 101038,
      "researchers access": 100675,
      "access recent": 1921,
      "development nlp": 30407,
      "nlp use": 79790,
      "tasks test": 119552,
      "existing studies": 40301,
      "exploring linguistic": 41664,
      "information encoded": 54523,
      "morphologically poor": 74763,
      "address introduce": 3684,
      "tasks case": 118972,
      "case marking": 15599,
      "24 languages": 801,
      "creation evaluation": 24716,
      "multilingual setting": 75366,
      "setting present": 107771,
      "experiments diverse": 40918,
      "diverse multilingual": 32824,
      "models relate": 73904,
      "probing task": 90907,
      "performance diverse": 86307,
      "tasks postagging": 119387,
      "postagging dependency": 88459,
      "labeling named": 58514,
      "high positive": 50101,
      "positive correlation": 88316,
      "correlation downstream": 24227,
      "tasks especially": 119101,
      "explore word": 41617,
      "blackbox neural": 14255,
      "models linguistic": 73511,
      "high scores": 50137,
      "datasets leads": 27545,
      "leads poor": 61950,
      "case lowresource": 15594,
      "simple text": 109531,
      "text augmentation": 120671,
      "tree fragments": 125592,
      "dependencies project": 29107,
      "characterlevel sequence": 16585,
      "improvements models": 52876,
      "data majority": 26111,
      "knowledgegrounded response": 58277,
      "generation achieved": 47291,
      "using handcrafted": 129723,
      "task corpus": 118028,
      "approaches unable": 8388,
      "unable generate": 126569,
      "paper attempts": 83746,
      "generating diverse": 47215,
      "diverse informative": 32816,
      "informative responses": 55151,
      "federated learning": 43813,
      "purpose expanding": 94427,
      "virtual keyboard": 131745,
      "keyboard smartphones": 57615,
      "exporting sensitive": 41685,
      "sensitive text": 105750,
      "approach settings": 7888,
      "dataset popular": 27096,
      "popular social": 88128,
      "using federated": 129672,
      "words setting": 134209,
      "setting demonstrate": 107739,
      "learn meaningful": 62090,
      "entropy loss": 38165,
      "building extending": 14841,
      "thesaurus paper": 121841,
      "new semiautomatically": 79124,
      "seamlessly integrated": 104290,
      "systems methodology": 117000,
      "corpora languages": 23513,
      "languages involved": 60651,
      "usage examples": 127858,
      "extraction new": 42411,
      "candidate terms": 15126,
      "accessible webbased": 1933,
      "webbased application": 132276,
      "information term": 55035,
      "similar terms": 109158,
      "specialized corpora": 111385,
      "terms central": 120290,
      "terms help": 120333,
      "providing competitive": 94106,
      "ecommerce platform": 34506,
      "successful product": 115169,
      "generation attracted": 47314,
      "industrial communities": 54058,
      "communities existing": 19059,
      "methods performance": 69673,
      "performance limited": 86497,
      "limited paper": 64258,
      "explore new": 41568,
      "new way": 79246,
      "way generate": 132087,
      "combining power": 18744,
      "power neural": 88641,
      "context ecommerce": 22071,
      "framework transformer": 45723,
      "transformer sequence": 124377,
      "using selfattention": 130137,
      "variety important": 130983,
      "important factors": 52161,
      "including product": 53353,
      "product aspects": 92025,
      "base experiments": 11456,
      "various metrics": 131132,
      "improvement 97": 52678,
      "anecdotal evidence": 6102,
      "largest online": 61535,
      "discontinuous constituency": 32027,
      "stack data": 112117,
      "uses set": 129277,
      "possible construct": 88394,
      "transitions sentence": 124517,
      "parsing strategy": 85247,
      "based assumption": 11528,
      "oracle new": 82256,
      "new transition": 79228,
      "oracle parser": 82257,
      "parser obtains": 84975,
      "german discontinuous": 47887,
      "ranking selecting": 95687,
      "better predict": 13672,
      "human needs": 50920,
      "understand sentiments": 126776,
      "way explain": 132081,
      "sentiment text": 106806,
      "text humans": 121038,
      "good understanding": 48507,
      "situations described": 109898,
      "described natural": 29377,
      "extract rank": 42103,
      "filter select": 44074,
      "relation paths": 98063,
      "paths commonsense": 85687,
      "efficiently integrate": 35149,
      "acquired knowledge": 3038,
      "paths neural": 85690,
      "knowledge using": 58229,
      "gated attention": 46503,
      "assess models": 9482,
      "published dataset": 94351,
      "integrating knowledge": 55796,
      "boosts performance": 14451,
      "establishes new": 38609,
      "interpretability learned": 56224,
      "learned attention": 62200,
      "attention map": 9876,
      "encoded knowledge": 36477,
      "inherently difficult": 55193,
      "greatest challenges": 49041,
      "reduce translation": 97365,
      "supervised sequence": 115834,
      "struggle translating": 114128,
      "word experimental": 133278,
      "extensible toolkit": 41833,
      "modeling toolkit": 72567,
      "allows researchers": 5184,
      "researchers developers": 100685,
      "developers train": 30330,
      "train custom": 122915,
      "custom models": 25435,
      "translation summarization": 125301,
      "summarization language": 115523,
      "distributed training": 32629,
      "multiple gpus": 75570,
      "modern gpus": 74389,
      "demo video": 28642,
      "representations propose": 99829,
      "propose generative": 92703,
      "sentence uses": 106120,
      "syntax sentence": 116561,
      "represent semantics": 99137,
      "better disentanglement": 13564,
      "syntactic representations": 116458,
      "wordorder information": 133786,
      "models popular": 73753,
      "popular pretrained": 88118,
      "embeddings standard": 35951,
      "standard semantic": 112300,
      "tasks empirically": 119091,
      "empirically model": 36238,
      "future time": 46317,
      "aspect important": 9335,
      "determining natural": 30150,
      "concepts paper": 20630,
      "entailment dataset": 37658,
      "range recently": 95596,
      "proposed nlp": 93511,
      "perform inference": 86014,
      "inference temporal": 54235,
      "models encode": 73146,
      "encode substantial": 36457,
      "information relating": 54912,
      "inferences require": 54253,
      "require reasoning": 100192,
      "reasoning semantic": 96307,
      "pragmatically informative": 88760,
      "informative text": 55157,
      "generation improve": 47433,
      "output text": 83136,
      "correctly identify": 24177,
      "original input": 82527,
      "text describes": 120866,
      "describes approaches": 29386,
      "approaches widely": 8402,
      "learning received": 62942,
      "performance strong": 86756,
      "strong existing": 113673,
      "generation structured": 47640,
      "structured meaning": 114014,
      "classification scientific": 17387,
      "use methods": 128145,
      "methods comparing": 69383,
      "comparing results": 19515,
      "results critical": 101647,
      "automated analysis": 10428,
      "propose structural": 93088,
      "effective classification": 34643,
      "stateoftheart existing": 112659,
      "acl anthology": 2995,
      "increase f1": 53598,
      "addition introduce": 3441,
      "multiple scientific": 75678,
      "scientific domains": 104005,
      "compared existing": 19364,
      "datasets code": 27347,
      "informative contexts": 55133,
      "learning highquality": 62634,
      "highquality embeddings": 50381,
      "embeddings rare": 35897,
      "problem sparse": 91243,
      "standard algorithm": 112183,
      "form used": 45142,
      "compute embeddings": 20465,
      "given access": 47980,
      "access words": 1926,
      "contexts learns": 22407,
      "learns attend": 63195,
      "embedding evaluation": 35402,
      "work attentive": 134387,
      "errors equal": 38366,
      "critical component": 24806,
      "detection model": 29992,
      "model despite": 70978,
      "despite years": 29745,
      "years speech": 135302,
      "recognition research": 96989,
      "asr accuracy": 9423,
      "generated asr": 47050,
      "asr errors": 9430,
      "investigate influence": 56768,
      "affect detection": 4214,
      "features syntactic": 43751,
      "representation speech": 99418,
      "performance multimodal": 86551,
      "mobile phones": 70470,
      "evaluation process": 39336,
      "process paper": 91544,
      "demonstrate algorithm": 28661,
      "deploy model": 29302,
      "realworld online": 96173,
      "effectively boost": 34790,
      "learning spoken": 63046,
      "understanding typically": 127002,
      "trained annotated": 123068,
      "data costly": 25800,
      "aiming reduce": 4494,
      "reduce data": 97321,
      "weight transfer": 132330,
      "transfer approach": 124036,
      "slu framework": 110117,
      "separate training": 106864,
      "different weight": 31563,
      "transfer settings": 124186,
      "settings work": 107846,
      "structure similar": 113955,
      "works including": 134953,
      "build single": 14808,
      "model solving": 72068,
      "multitask models": 75880,
      "compared singletask": 19443,
      "singletask model": 109846,
      "sharing knowledge": 108132,
      "addition work": 3486,
      "work employ": 134488,
      "nary relation": 76201,
      "multiscale representation": 75774,
      "focus binary": 44740,
      "relations expressed": 98167,
      "expressed single": 41721,
      "highvalue domains": 50438,
      "nary relations": 76203,
      "relations great": 98180,
      "far apart": 43068,
      "document existing": 33003,
      "crosssentence relation": 25078,
      "consecutive sentences": 21165,
      "severely limits": 107885,
      "multiscale neural": 75773,
      "extraction combines": 42282,
      "learned various": 62279,
      "spans document": 111199,
      "weak signals": 132181,
      "document multiscale": 33043,
      "noisy labels": 80109,
      "labels distant": 58591,
      "supervision experiments": 115887,
      "reading approach": 95971,
      "methods alternative": 69306,
      "weighting schemes": 132367,
      "elmo embeddings": 35311,
      "embeddings peters": 35860,
      "peters et": 87208,
      "community recent": 19094,
      "use embeddings": 128025,
      "nlp architectures": 79578,
      "embeddings like": 35789,
      "vectors token": 131478,
      "al proposed": 4655,
      "proposed learn": 93320,
      "weighting scheme": 132366,
      "necessarily yield": 76748,
      "achieve best": 2418,
      "best possible": 13409,
      "possible performance": 88420,
      "published language": 94357,
      "decreases performance": 28175,
      "complexity language": 19921,
      "training speedup": 123877,
      "mining paper": 70253,
      "model architecture": 70682,
      "architecture submitted": 8753,
      "submitted semeval2019": 114770,
      "task competition": 117989,
      "mining online": 70249,
      "crossdomain suggestion": 24904,
      "proposed recurrent": 93533,
      "bilstm layers": 14097,
      "layers selfattention": 61808,
      "encode words": 36466,
      "using elmo": 129619,
      "different setups": 31426,
      "function best": 46003,
      "evaluation score": 39379,
      "official results": 81603,
      "subtasks respectively": 114978,
      "graph paths": 48855,
      "textual definitions": 121688,
      "method mapping": 68955,
      "framing task": 45752,
      "task sequencetosequence": 118702,
      "sequencetosequence problem": 107198,
      "state input": 112501,
      "text decoder": 120860,
      "decoder directly": 28003,
      "target node": 117681,
      "mapping systems": 67151,
      "model outputs": 71684,
      "structured predictions": 114031,
      "present proofofconcept": 89652,
      "proofofconcept experiment": 92375,
      "context introduce": 22152,
      "contextually relevant": 22588,
      "information entity": 54536,
      "barack obama": 11427,
      "created automatically": 24655,
      "particular news": 85435,
      "use crowdsourcing": 127976,
      "contextual relevance": 22489,
      "generation approaches": 47311,
      "approaches baselines": 8084,
      "baselines dataset": 12380,
      "large room": 61239,
      "improvement terms": 52776,
      "relevant facts": 98548,
      "20 improvement": 506,
      "generating appropriate": 47199,
      "sufficient accurate": 115271,
      "generation conduct": 47340,
      "promising directions": 92272,
      "outlier detection": 82627,
      "improved data": 52601,
      "quality diversity": 94632,
      "systems corpus": 116819,
      "informative samples": 55152,
      "datasets guiding": 27500,
      "attention nlp": 9958,
      "nlp particularly": 79660,
      "corpus short": 23992,
      "pipeline built": 87532,
      "unique data": 127184,
      "data samples": 26367,
      "samples experiments": 103584,
      "effective finding": 34677,
      "errors data": 38363,
      "highly diverse": 50317,
      "diverse corpora": 32801,
      "corpora turn": 23606,
      "slotfilling models": 110076,
      "approaches ability": 8038,
      "ability control": 1499,
      "candidate responses": 15119,
      "unacceptable responses": 126573,
      "best response": 13431,
      "semantic correctness": 105023,
      "datasets grammatical": 27498,
      "distribution errors": 32653,
      "generated datadriven": 47068,
      "systems explore": 116898,
      "sensitive distribution": 105737,
      "data distributions": 25861,
      "significantly source": 109051,
      "domain possible": 33612,
      "possible achieve": 88375,
      "dataset general": 26949,
      "introduce general": 56428,
      "tasks share": 119496,
      "span representations": 111130,
      "dynamically constructed": 34343,
      "graphs graphs": 48956,
      "entity spans": 38105,
      "relation types": 98082,
      "coreference relation": 23358,
      "iteratively refine": 57144,
      "refine span": 97587,
      "interaction tasks": 55971,
      "tasks shared": 119497,
      "multiple information": 75576,
      "tasks multiple": 119314,
      "reflecting different": 97620,
      "domains observe": 33826,
      "score improvement": 104085,
      "stringtotree neural": 113608,
      "translation known": 124886,
      "difficult parallelize": 31652,
      "discrete operations": 32172,
      "problem utilizing": 91284,
      "utilizing state": 130582,
      "regard different": 97650,
      "significantly faster": 108912,
      "faster compared": 43166,
      "undergo process": 126666,
      "global structure": 48275,
      "entity properties": 37988,
      "structured neural": 114019,
      "model tracks": 72176,
      "continuous representation": 22627,
      "contain relevant": 21754,
      "relevant state": 98588,
      "state information": 112500,
      "discrete state": 32177,
      "constraints entity": 21597,
      "states time": 113057,
      "propara dataset": 92403,
      "autoencoder unsupervised": 10409,
      "unsupervised abstractive": 127594,
      "abstractive sentence": 1806,
      "compression neural": 20316,
      "models currently": 73008,
      "currently dominant": 25403,
      "dominant approach": 33928,
      "sequence discrete": 106932,
      "discrete latent": 32166,
      "sequences input": 107127,
      "sentences respectively": 106480,
      "compressed sentence": 20299,
      "latent word": 61640,
      "forces model": 45017,
      "input pretrained": 55398,
      "model acting": 70621,
      "prior latent": 90715,
      "compressed sentences": 20300,
      "gradientbased optimization": 48610,
      "achieving promising": 2970,
      "results unsupervised": 102294,
      "unsupervised sentence": 127717,
      "compression benchmark": 20308,
      "models nlm": 73644,
      "learning dense": 62493,
      "dense representation": 29016,
      "representation rare": 99392,
      "words challenging": 133846,
      "probability estimates": 90860,
      "representations rare": 99837,
      "words pretrained": 134137,
      "improve probability": 52505,
      "probability estimation": 90861,
      "estimation performance": 38660,
      "embedding matrices": 35437,
      "hypotheses output": 51256,
      "reduces word": 97403,
      "improves recognition": 53036,
      "16 absolute": 375,
      "absolute compared": 1725,
      "construction evaluation": 21686,
      "essential ingredient": 38553,
      "quantification semantic": 94837,
      "evaluate ability": 38781,
      "created english": 24665,
      "thai language": 121767,
      "work create": 134447,
      "datasets datasets": 27392,
      "pairs total": 83648,
      "different characteristics": 31034,
      "characteristics terms": 16539,
      "domain coverage": 33493,
      "notion similarity": 80410,
      "features help": 43534,
      "help gain": 49728,
      "evaluated word": 39026,
      "baseline evaluations": 12215,
      "models identify": 73349,
      "available nlp": 11055,
      "community online": 19083,
      "human speech": 50966,
      "speech transcription": 111827,
      "transcription errors": 123997,
      "contexts associated": 22378,
      "errors transcription": 38416,
      "shedding light": 108159,
      "light human": 63991,
      "speech phenomena": 111722,
      "transcripts results": 124013,
      "draws inspiration": 34139,
      "inspiration traditional": 55558,
      "generation techniques": 47676,
      "content say": 21938,
      "building successful": 14891,
      "successful encoderdecoder": 115158,
      "encoderdecoder paradigm": 36618,
      "content representation": 21934,
      "representation given": 99248,
      "text produce": 121215,
      "produce output": 91915,
      "text training": 121371,
      "decoder evaluate": 28007,
      "model abstractive": 70522,
      "performance outperforms": 86589,
      "comparable baselines": 19127,
      "language difference": 58962,
      "selection present": 104816,
      "apply methods": 7187,
      "addressing problem": 3815,
      "selecting relevant": 104755,
      "data general": 25971,
      "method incrementally": 68902,
      "efficiently model": 35152,
      "selection approaches": 104766,
      "used time": 128813,
      "indomain vocabulary": 53987,
      "data methods": 26127,
      "new approaches": 78799,
      "better systems": 13737,
      "viable alternative": 131641,
      "additionally able": 3597,
      "smt models": 110286,
      "closely matches": 17859,
      "corpus frequency": 23812,
      "construction grammar": 21688,
      "generalization paper": 46789,
      "minimum description": 70205,
      "length paradigm": 63371,
      "paradigm experiments": 84536,
      "embeddings develop": 35647,
      "crosslingual alignment": 24925,
      "alignment approaches": 4946,
      "sequential encoderdecoder": 107224,
      "applied models": 7094,
      "models joint": 73432,
      "mapping using": 67154,
      "text guide": 121032,
      "alignment results": 4986,
      "scalable approach": 103687,
      "embeddings observe": 35838,
      "observe better": 81190,
      "models intrinsic": 73423,
      "evaluations particularly": 39478,
      "sets parallel": 107693,
      "articles using": 9224,
      "using event": 129645,
      "attacks natural": 9714,
      "wikidata knowledge": 132630,
      "produce semantic": 91926,
      "keyword based": 57634,
      "based search": 12004,
      "belonging specific": 12722,
      "automatically inferred": 10790,
      "summary evaluation": 115638,
      "evaluation considered": 39158,
      "pyramid method": 94476,
      "researchers resorted": 100708,
      "pairwise comparison": 83671,
      "approach proposing": 7827,
      "proposing lightweight": 93622,
      "method comparison": 68715,
      "higher correlation": 50171,
      "release crowdsourced": 98444,
      "future evaluations": 46275,
      "baselines complex": 12374,
      "word identification": 133310,
      "languages complex": 60458,
      "identification cwi": 51376,
      "identifying words": 51635,
      "phrases sentence": 87452,
      "difficult understand": 31670,
      "target audience": 117587,
      "cwi shared": 25473,
      "task released": 118628,
      "released data": 98478,
      "data settings": 26439,
      "language seen": 60072,
      "best monolingual": 13382,
      "setting best": 107735,
      "networks multitask": 77666,
      "models submitted": 74111,
      "features simple": 43731,
      "simple learning": 109453,
      "performance result": 86677,
      "baselines future": 12398,
      "development area": 30371,
      "area finally": 8873,
      "inconsistencies annotation": 53434,
      "data explain": 25928,
      "explain results": 41259,
      "word movers": 133367,
      "movers distance": 74889,
      "distance linguistic": 32421,
      "coordination syntactic": 23266,
      "attempt combine": 9736,
      "space particular": 111040,
      "proposed word": 93590,
      "multiple consecutive": 75525,
      "studies proposed": 114266,
      "proposed metric": 93419,
      "proposed measures": 93331,
      "notice significant": 80396,
      "tools text": 122478,
      "data community": 25752,
      "assumption latent": 9662,
      "leveraging information": 63681,
      "word frequencies": 133290,
      "documents study": 33297,
      "study challenge": 114332,
      "recent developments": 96451,
      "area computational": 8866,
      "unsupervised graphbased": 127642,
      "extensive quantitative": 41951,
      "relying semantic": 98766,
      "semantic document": 105035,
      "better capturing": 13530,
      "validate findings": 130721,
      "focused text": 44880,
      "research new": 100569,
      "included data": 53244,
      "evaluation procedures": 39335,
      "realization shared": 96115,
      "information removed": 54919,
      "divide problem": 32907,
      "parts word": 85593,
      "memory based": 68291,
      "subproblem present": 114802,
      "lm based": 64859,
      "apply different": 7170,
      "approach combined": 7426,
      "progress machine": 92160,
      "learning driven": 62526,
      "datasets consistent": 27376,
      "consistent evaluation": 21376,
      "1of100 accuracy": 481,
      "repository contains": 99095,
      "introduce evaluate": 56420,
      "evaluate competitive": 38810,
      "encoder model": 36542,
      "trained entire": 123138,
      "pun generation": 94387,
      "generating pun": 47248,
      "pun sentence": 94389,
      "generating novel": 47243,
      "context strong": 22281,
      "immediate context": 51844,
      "generation ways": 47698,
      "probabilities language": 90848,
      "entity descriptions": 37920,
      "crossdomain knowledge": 24891,
      "graphs wikidata": 48977,
      "queries given": 94922,
      "rapidly increasing": 95733,
      "increasing numbers": 53680,
      "entities knowledge": 37803,
      "graphs fully": 48953,
      "information essential": 54542,
      "model suitable": 72112,
      "mechanism generate": 67988,
      "generate concise": 46916,
      "indepth evaluation": 53799,
      "spoken question": 112016,
      "answering adversarial": 6599,
      "answering sqa": 6701,
      "spoken documents": 111984,
      "documents recent": 33280,
      "catastrophic impact": 15714,
      "asr hypotheses": 9431,
      "adversarial model": 4144,
      "applied domain": 7065,
      "adaptation task": 3272,
      "learn domaininvariant": 62031,
      "domaininvariant features": 33716,
      "qa model": 94507,
      "improve sqa": 52545,
      "translation study": 125299,
      "fundamental problems": 46109,
      "problems lexical": 91335,
      "sense clusters": 105654,
      "resource allows": 100828,
      "test hypotheses": 120459,
      "resources results": 101040,
      "strong empirical": 113669,
      "represents step": 100076,
      "using highlevel": 129740,
      "art sota": 9092,
      "syntax tree": 116566,
      "time paper": 122064,
      "introduce iterative": 56439,
      "large source": 61275,
      "syntax trees": 116567,
      "train semantic": 123010,
      "contextdependent semantic": 22354,
      "22 bleu": 772,
      "scale model": 103736,
      "training extended": 123632,
      "23 bleu": 787,
      "accuracy semantic": 2270,
      "nmt prespecified": 79948,
      "translation leveraging": 124906,
      "constraints decoding": 21594,
      "hurt translation": 51166,
      "various reasons": 131182,
      "augmentation method": 10285,
      "achieves consistent": 2778,
      "improving translation": 53178,
      "popular newspapers": 88112,
      "identification set": 51440,
      "written portuguese": 135143,
      "linguistic analyses": 64420,
      "critical study": 24835,
      "set texts": 107609,
      "published 2008": 94348,
      "2012 2014": 564,
      "2014 2015": 574,
      "tokens new": 122320,
      "coverage vocabulary": 24526,
      "ways incorporating": 132163,
      "average 19": 11175,
      "average 13": 11172,
      "recognizing words": 97081,
      "enhanced representation": 37516,
      "knowledge integration": 58020,
      "integration present": 55814,
      "model enhanced": 71080,
      "enhanced knowledge": 37509,
      "masking strategy": 67315,
      "designed learn": 29606,
      "representation enhanced": 99225,
      "masking strategies": 67314,
      "composed words": 20075,
      "methods achieving": 69290,
      "chinese natural": 16792,
      "including natural": 53332,
      "recognition sentiment": 96994,
      "analysis question": 5792,
      "answering demonstrate": 6623,
      "knowledge inference": 58015,
      "cloze test": 17895,
      "data knowledge": 26061,
      "difficult limited": 31642,
      "generation motivated": 47492,
      "humans tend": 51113,
      "demonstrates strong": 28963,
      "performance provides": 86637,
      "provides good": 94038,
      "good generalization": 48478,
      "design model": 29547,
      "compose coherent": 20058,
      "knowledge just": 58028,
      "domains approach": 33732,
      "achieves reasonable": 2841,
      "reasonable performances": 96211,
      "baseline average": 12192,
      "improvement code": 52692,
      "dnn models": 32932,
      "address various": 3776,
      "types models": 126322,
      "understand various": 126790,
      "greatly improve": 49053,
      "modules building": 74520,
      "test various": 120550,
      "experiments nlp": 41043,
      "nlp datasets": 79597,
      "datasets glue": 27497,
      "compositional data": 20109,
      "aimed providing": 4486,
      "synthetic training": 116649,
      "real training": 96082,
      "useful variety": 128944,
      "rate 87": 95784,
      "diagnostic tasks": 30513,
      "scan dataset": 103780,
      "task applied": 117883,
      "reduces perplexity": 97394,
      "past future": 85642,
      "future neural": 46283,
      "models benefit": 72829,
      "benefit explicitly": 12967,
      "translated past": 124556,
      "untranslated future": 127769,
      "translated untranslated": 124576,
      "untranslated contents": 127768,
      "assignment assignment": 9561,
      "process assign": 91425,
      "translations extensive": 125466,
      "extensive analysis": 41855,
      "demonstrates method": 28947,
      "able recognize": 1681,
      "sentiment lexica": 106755,
      "1000 languages": 175,
      "documentlevel annotations": 33133,
      "common scenario": 18918,
      "languages best": 60427,
      "terms number": 120352,
      "covered languages": 24529,
      "resource ones": 100867,
      "massively parallel": 67347,
      "parallel bible": 84588,
      "information english": 54530,
      "data introduce": 26051,
      "domainspecific embeddings": 33899,
      "twitter domain": 126109,
      "domain evaluate": 33520,
      "sentiment resources": 106788,
      "publication paper": 94280,
      "paper release": 84397,
      "assigning labels": 9557,
      "labels word": 58657,
      "architectures model": 8825,
      "scores words": 104222,
      "minimal taskspecific": 70179,
      "time parser": 122066,
      "f1 wsj": 42701,
      "set achieving": 107349,
      "parsers comparable": 85017,
      "positive sample": 88341,
      "intelligence based": 55827,
      "chemical information": 16707,
      "chemical reactions": 16708,
      "issue negative": 57011,
      "effectively predict": 34838,
      "contextualized embeddings": 22544,
      "despite advances": 29678,
      "advances opendomain": 4009,
      "problem traditional": 91265,
      "valid responses": 130711,
      "words reference": 134161,
      "responses recent": 101293,
      "response given": 101217,
      "referencebased metric": 97547,
      "showed high": 108378,
      "judgments paper": 57448,
      "using contextualized": 129485,
      "relatedness scores": 97925,
      "scores better": 104158,
      "metrics experiments": 69960,
      "experiments evaluation": 40941,
      "attacks study": 9717,
      "great significance": 49024,
      "relevant news": 98566,
      "article extract": 9130,
      "united states": 127228,
      "build bidirectional": 14748,
      "contexts entities": 22391,
      "entities appear": 37744,
      "leveraging large": 63686,
      "large unsupervised": 61320,
      "glove elmo": 48299,
      "words machinetranslated": 134044,
      "role modern": 103204,
      "communication various": 19045,
      "various communities": 131061,
      "communities using": 19065,
      "translation lead": 124897,
      "previous method": 90424,
      "method ignores": 68880,
      "words individual": 134000,
      "individual sentence": 53929,
      "similar experiments": 109092,
      "accuracy respectively": 2263,
      "personal attributes": 87115,
      "conversations opendomain": 23062,
      "conversation work": 22983,
      "web applications": 132222,
      "applications extracting": 6928,
      "problem challenging": 90958,
      "methods inferring": 69554,
      "attribute models": 10181,
      "networks leveraging": 77642,
      "basis output": 12545,
      "texts including": 121528,
      "collection crowdsourced": 18466,
      "dialogues demonstrate": 30826,
      "methods superior": 69784,
      "systems speech": 117151,
      "trained textual": 123313,
      "systems textual": 117195,
      "textual inputs": 121716,
      "literature little": 64758,
      "interesting problems": 56085,
      "asr outputs": 9443,
      "systems test": 117193,
      "encoding approaches": 36682,
      "approaches nmt": 8259,
      "encoderdecoder modeling": 36603,
      "compare translation": 19308,
      "utterances containing": 130629,
      "containing asr": 21783,
      "errors stateoftheart": 38409,
      "encoderdecoder systems": 36627,
      "translation baseline": 124667,
      "outputs better": 83160,
      "better represented": 13698,
      "approaches represent": 8323,
      "linear model": 64355,
      "need identify": 76818,
      "usually limited": 130435,
      "relevant scientific": 98576,
      "scientific articles": 103992,
      "process place": 91546,
      "experiment use": 40517,
      "features domain": 43463,
      "specific information": 111450,
      "using collection": 129449,
      "best detection": 13325,
      "improvement compared": 52694,
      "importance domain": 52056,
      "specific embedding": 111435,
      "detection medical": 29988,
      "teach machines": 119691,
      "introduce set": 56534,
      "challenge tasks": 16108,
      "understanding function": 126848,
      "comprehension specific": 20232,
      "specific types": 111512,
      "types function": 126291,
      "tasks explore": 119126,
      "various pretraining": 131176,
      "pretraining objectives": 90300,
      "encoders language": 36652,
      "results pretraining": 102051,
      "pretraining language": 90268,
      "modeling performs": 72508,
      "tasks supporting": 119540,
      "pretraining objective": 90299,
      "word probing": 133419,
      "tasks highlight": 119159,
      "differences pretraining": 30968,
      "translation processes": 125146,
      "experiments englishfrench": 40937,
      "words human": 133984,
      "characterize different": 16551,
      "knowledge effort": 57886,
      "effort automatically": 35162,
      "automatically classify": 10734,
      "finegrained translation": 44389,
      "englishfrench parallel": 37365,
      "based annotated": 11506,
      "work demonstrates": 134464,
      "demonstrates possible": 28953,
      "possible automatically": 88382,
      "examples experiments": 39833,
      "paraphrase extraction": 84813,
      "extraction bilingual": 42272,
      "systems extract": 116900,
      "extract relations": 42105,
      "relations arguments": 98112,
      "arguments natural": 8987,
      "tasks knowledge": 119234,
      "event schema": 39538,
      "schema induction": 103900,
      "release analyze": 98429,
      "text english": 120911,
      "available date": 10978,
      "contains valuable": 21841,
      "information confidence": 54437,
      "spatial temporal": 111271,
      "information analyze": 54375,
      "corpus comparing": 23701,
      "based wikipedia": 12160,
      "entities present": 37840,
      "open relations": 81927,
      "highly polysemous": 50335,
      "research automated": 100424,
      "automated knowledge": 10447,
      "heterogeneous collections": 49850,
      "matching task": 67435,
      "task binary": 117938,
      "pairs documents": 83519,
      "resources domain": 100967,
      "taskspecific modifications": 119635,
      "clearly outperforms": 17742,
      "outperforms complex": 82871,
      "explicit information": 41325,
      "cognitive state": 18284,
      "posted online": 88467,
      "people need": 85880,
      "automated processing": 10462,
      "considerably improve": 21270,
      "improve effectiveness": 52377,
      "effectiveness efforts": 34889,
      "specific areas": 111405,
      "severely limited": 107884,
      "limited availability": 64209,
      "availability resources": 10925,
      "processing technologies": 91836,
      "systems included": 116949,
      "spanish language": 111162,
      "context awareness": 22018,
      "motivation biomedical": 74877,
      "detection fundamental": 29956,
      "comprehensive biomedical": 20249,
      "biomedical knowledge": 14190,
      "existing event": 40135,
      "models applicable": 72755,
      "knowledge management": 58061,
      "results consider": 101628,
      "detection framework": 29955,
      "framework identify": 45560,
      "identify events": 51499,
      "arguments capture": 8982,
      "context embedding": 22073,
      "attributes derived": 10194,
      "built models": 14928,
      "bionlp shared": 14218,
      "task evaluations": 118161,
      "methods method": 69618,
      "method nearly": 68984,
      "fscore performance": 45914,
      "dataset case": 26779,
      "conducted reveal": 20945,
      "showed best": 108369,
      "subtask b1": 114941,
      "2019 task": 713,
      "2019 approach": 693,
      "participation wmt": 85368,
      "wmt biomedical": 132779,
      "translation shared": 125247,
      "task systems": 118771,
      "participated translation": 85352,
      "directions englishspanish": 31838,
      "pairs create": 83507,
      "create training": 24647,
      "systems achieved": 116716,
      "semantics linguistic": 105431,
      "semantics provides": 105453,
      "empirically induced": 36233,
      "representations successfully": 99913,
      "successfully capture": 115180,
      "capture aspects": 15270,
      "languages shown": 60870,
      "shown large": 108490,
      "work computational": 134423,
      "theoretical linguistics": 121796,
      "critical discussion": 24809,
      "semantic change": 105002,
      "semantics syntax": 105469,
      "computational approaches": 20360,
      "approaches language": 8205,
      "construction method": 21692,
      "technique construct": 119776,
      "construct novel": 21645,
      "utilized field": 130541,
      "construction propose": 21696,
      "propose taskoriented": 93110,
      "algorithm algorithm": 4691,
      "advantages including": 4081,
      "linguistic quality": 64537,
      "high diversity": 50066,
      "method deep": 68759,
      "networks respectively": 77739,
      "results performance": 102033,
      "performance hybrid": 86434,
      "outperforms pure": 82968,
      "resources bhojpuri": 100954,
      "motivation study": 74878,
      "languages chapter": 60439,
      "model nlp": 71585,
      "nlp dependency": 79602,
      "presents brief": 89824,
      "automated detection": 10435,
      "based shared": 12040,
      "shared words": 108119,
      "corpora largely": 23515,
      "process present": 91551,
      "detection approach": 29879,
      "approach matter": 7705,
      "retrieval perspective": 102419,
      "texts relevant": 121592,
      "corpus compilation": 23702,
      "agreement study": 4408,
      "investigate extent": 56756,
      "integration lexical": 55810,
      "retrieving cases": 102470,
      "agreement scores": 4407,
      "considerably improves": 21272,
      "improves retrieval": 53039,
      "retrieval performance": 102418,
      "performance respect": 86674,
      "unified language": 127113,
      "model unilm": 72244,
      "pretrained using": 90208,
      "using types": 130332,
      "types language": 126305,
      "sequencetosequence prediction": 107194,
      "achieved employing": 2618,
      "employing shared": 36320,
      "shared transformer": 108113,
      "transformer network": 124352,
      "selfattention masks": 104881,
      "control context": 22802,
      "context prediction": 22221,
      "bert glue": 13134,
      "glue benchmark": 48309,
      "squad 20": 112073,
      "20 coqa": 496,
      "generation datasets": 47354,
      "generative question": 47755,
      "improvement squad": 52761,
      "question generation": 95160,
      "generation benefit": 47325,
      "benefit learning": 12978,
      "learning structure": 63063,
      "space output": 111036,
      "space structure": 111067,
      "labels learning": 58616,
      "investigate usefulness": 56824,
      "dropout layers": 34177,
      "better capture": 13524,
      "capture structure": 15402,
      "tasks output": 119358,
      "match improve": 67359,
      "recurrent selfattention": 97258,
      "model natural": 71557,
      "language better": 58861,
      "paper tackles": 84468,
      "similar content": 109079,
      "entities wikipedia": 37896,
      "framework timeline": 45715,
      "given collection": 47994,
      "documents related": 33281,
      "timeline summarization": 122157,
      "summarization ts": 115587,
      "form relevant": 45118,
      "topic time": 122578,
      "time previous": 122076,
      "ranking models": 95676,
      "models depends": 73041,
      "handdesigned features": 49361,
      "features complex": 43416,
      "novel dynamic": 80551,
      "generation leveraging": 47460,
      "leveraging distributed": 63674,
      "representations dynamically": 99610,
      "likely sequence": 64143,
      "make sure": 66736,
      "topic global": 122525,
      "demonstrate framework": 28745,
      "extract summaries": 42126,
      "generation outperforms": 47524,
      "outperforms various": 83044,
      "various competitive": 131062,
      "performance unsupervised": 86824,
      "using twodimensional": 130329,
      "twodimensional word": 126137,
      "embedding pretrained": 35470,
      "work super": 134834,
      "promise new": 92259,
      "borrows idea": 14479,
      "generating conversational": 47207,
      "domain dialogues": 33506,
      "dialogues experimental": 30831,
      "results public": 102093,
      "public dataset": 94247,
      "interactive demo": 56015,
      "demo ready": 28640,
      "ready workshop": 96048,
      "nlg critical": 79515,
      "critical research": 24827,
      "nlp field": 79614,
      "utterances natural": 130657,
      "construct corresponding": 21623,
      "corresponding sentences": 24304,
      "given semantics": 48115,
      "new learning": 78990,
      "framework language": 45594,
      "generation dual": 47378,
      "learning providing": 62928,
      "providing way": 94145,
      "tasks exact": 119109,
      "string transduction": 113601,
      "tasks graphemetophoneme": 119152,
      "attention outperform": 9960,
      "outperform popular": 82723,
      "work ask": 134381,
      "ask following": 9295,
      "following question": 44980,
      "helpful inductive": 49800,
      "bias tasks": 13846,
      "tasks develop": 119055,
      "attention sequencetosequence": 10007,
      "latent alignment": 61573,
      "able compute": 1622,
      "based english": 11680,
      "analysis sentiment": 5837,
      "aims determine": 4517,
      "opinions customers": 82119,
      "product service": 92047,
      "great help": 49004,
      "use popular": 128200,
      "popular machine": 88097,
      "waikato environment": 131984,
      "environment knowledge": 38186,
      "knowledge analysis": 57750,
      "analysis weka": 5935,
      "comments experimental": 18813,
      "results precision": 102046,
      "positive comments": 88315,
      "negative comments": 76922,
      "using entity": 129639,
      "entity relations": 38095,
      "novel techniques": 80749,
      "techniques extract": 119886,
      "mining opinions": 70252,
      "reviews customers": 102604,
      "assessment based": 9521,
      "emotional level": 36120,
      "specific product": 111480,
      "exploit features": 41415,
      "features products": 43671,
      "enhanced language": 37510,
      "informative entities": 55137,
      "bert pretrained": 13205,
      "pretrained largescale": 90115,
      "semantic patterns": 105177,
      "consistently improve": 21415,
      "rarely consider": 95754,
      "graphs kgs": 48962,
      "kgs provide": 57666,
      "rich structured": 102793,
      "knowledge facts": 57933,
      "enhance language": 37473,
      "paper utilize": 84489,
      "utilize largescale": 130514,
      "model ernie": 71091,
      "knowledge information": 58016,
      "information simultaneously": 54984,
      "various knowledgedriven": 131111,
      "knowledgedriven tasks": 58267,
      "common nlp": 18900,
      "paper obtained": 84058,
      "adaptation deep": 3214,
      "multilingual transformers": 75394,
      "adaptation multilingual": 3235,
      "masked language": 67287,
      "pretrained bidirectional": 89994,
      "models stateoftheart": 74096,
      "comprehension natural": 20203,
      "inference sentiment": 54216,
      "alternative approaches": 5224,
      "models monolingual": 73581,
      "specific models": 111465,
      "models allow": 72742,
      "allow perform": 5090,
      "perform transfer": 86095,
      "transfer language": 124068,
      "language solve": 60103,
      "solve tasks": 110625,
      "model monolingual": 71530,
      "growth performance": 49193,
      "tasks reading": 119434,
      "comprehension paraphrase": 20212,
      "detection sentiment": 30047,
      "analysis furthermore": 5630,
      "furthermore multilingual": 46194,
      "reduces training": 97402,
      "models russian": 73971,
      "open sourced": 81939,
      "conversation contexts": 22941,
      "content understanding": 21966,
      "microblog posts": 70046,
      "discovery using": 32152,
      "different previous": 31354,
      "effort annotate": 35161,
      "novel sequence": 80722,
      "generation framework": 47412,
      "short sequence": 108234,
      "sparsity issue": 111253,
      "datasets newly": 27596,
      "english twitter": 37324,
      "based classification": 11573,
      "studies demonstrate": 114205,
      "ability effectively": 1512,
      "effectively generate": 34815,
      "rare unseen": 95744,
      "methods predicting": 69676,
      "improve task": 52561,
      "performance biomedical": 86185,
      "require highquality": 100146,
      "highquality annotated": 50368,
      "data specialized": 26489,
      "expert annotations": 41215,
      "introducing noise": 56648,
      "directly modeling": 31890,
      "trained outofdomain": 123228,
      "data taskspecific": 26546,
      "complex biomedical": 19801,
      "using expert": 129653,
      "data instances": 26045,
      "yields small": 135437,
      "scores weight": 104219,
      "instances training": 55638,
      "difficult domain": 31623,
      "effective strategy": 34752,
      "experiments confirm": 40852,
      "expectation specialized": 40384,
      "expert data": 41217,
      "leads improvements": 61940,
      "improvements model": 52875,
      "task commonsense": 117978,
      "inference given": 54148,
      "near humanlevel": 76710,
      "machines perform": 66340,
      "perform human": 86008,
      "inference paper": 54185,
      "models presenting": 73776,
      "new challenge": 78822,
      "dataset questions": 27139,
      "trivial humans": 125829,
      "accuracy stateoftheart": 2287,
      "models struggle": 74108,
      "adversarial filtering": 4123,
      "filtering af": 44081,
      "wrong answers": 135162,
      "proves surprisingly": 93746,
      "surprisingly robust": 116146,
      "key insight": 57583,
      "complexity dataset": 19905,
      "dataset examples": 26903,
      "humans misclassified": 51092,
      "models construction": 72975,
      "deep pretrained": 28391,
      "models broadly": 72865,
      "suggests new": 115380,
      "based speech": 12063,
      "act classification": 3061,
      "political campaign": 87990,
      "target utterance": 117739,
      "new annotation": 78775,
      "annotation schema": 6366,
      "incorporating domainspecific": 53535,
      "novel annotated": 80461,
      "modeled sequential": 72363,
      "classification evaluate": 17199,
      "evaluate techniques": 38936,
      "techniques exploiting": 119885,
      "binary variables": 14140,
      "hand hand": 49319,
      "focus text": 44831,
      "text classifiers": 120790,
      "predictions approach": 89159,
      "problem jointly": 91095,
      "short informative": 108212,
      "text classifier": 120789,
      "input positions": 55394,
      "expected value": 40402,
      "allows directly": 5143,
      "prespecified text": 89952,
      "rate approach": 95785,
      "competitive previous": 19674,
      "extraction explore": 42336,
      "uses attention": 129192,
      "mechanisms neural": 68066,
      "neural interactivepredictive": 77931,
      "multimodal sequence": 75459,
      "present demonstration": 89438,
      "generates text": 47187,
      "predictions different": 89162,
      "different sequence": 31419,
      "revised human": 102645,
      "agent introduces": 4303,
      "form characters": 45071,
      "providing alternative": 94102,
      "provided user": 93990,
      "reduce human": 97331,
      "process implemented": 91505,
      "implemented following": 51962,
      "following clientserver": 44965,
      "developed website": 30320,
      "website communicates": 132292,
      "communicates neural": 19023,
      "opensource code": 82017,
      "developed building": 30257,
      "demonstration hosted": 28986,
      "semantic proximity": 105190,
      "contextualized elmo": 22542,
      "bert word": 13255,
      "embeddings task": 35972,
      "models leverage": 73498,
      "automatically assigned": 10723,
      "assigned word": 9551,
      "relies bidirectional": 98647,
      "representations benchmark": 99516,
      "methods settings": 69744,
      "settings generating": 107805,
      "forms graph": 45246,
      "graph representations": 48864,
      "text entities": 120916,
      "information entities": 54535,
      "uses graph": 129231,
      "network gnn": 77269,
      "architecture incorporate": 8670,
      "relevant entities": 98545,
      "decoder copy": 28000,
      "conceptually simple": 20681,
      "simple mechanism": 109465,
      "generate logical": 46965,
      "stateoftheart tasks": 113001,
      "approaches combined": 8099,
      "wikipedia develop": 132652,
      "develop methods": 30212,
      "text passages": 121172,
      "passages written": 85619,
      "entities provide": 37850,
      "valuable insights": 130764,
      "insights cultural": 55530,
      "cultural differences": 25228,
      "precise alignment": 88781,
      "longer text": 65164,
      "article pair": 9143,
      "pair better": 83420,
      "understand aspects": 126736,
      "benchmark propose": 12850,
      "integrated approach": 55772,
      "alignment using": 5000,
      "generation key": 47448,
      "facilitate semantic": 42792,
      "information regarding": 54910,
      "news social": 79366,
      "knowledge repositories": 58143,
      "events entities": 39577,
      "entitycentric information": 38136,
      "information insufficient": 54694,
      "terms coverage": 120301,
      "events temporal": 39603,
      "temporal knowledge": 120110,
      "makes available": 66760,
      "rdf representation": 95883,
      "representation popular": 99376,
      "relations temporal": 98265,
      "generating concise": 47205,
      "given entity": 48029,
      "article demonstrate": 9120,
      "demonstrate application": 28663,
      "supervision method": 115897,
      "identify relations": 51542,
      "insights characteristics": 55528,
      "language captured": 58866,
      "captured existing": 15429,
      "survey previous": 116184,
      "length present": 63375,
      "walk model": 131997,
      "offers better": 81584,
      "fit data": 44577,
      "kl divergence": 57708,
      "noise inherent": 80054,
      "modeling models": 72478,
      "mdl models": 67589,
      "errors nmt": 38392,
      "nmt quality": 79952,
      "nmt shown": 79960,
      "shown significantly": 108525,
      "sourceside noise": 110941,
      "noise present": 80062,
      "present largescale": 89531,
      "largescale study": 61511,
      "study stateoftheart": 114524,
      "grammar correction": 48627,
      "introduce technique": 56551,
      "distribution caused": 32643,
      "additional insights": 3531,
      "structure documents": 113842,
      "identifying potential": 51611,
      "generating corresponding": 47208,
      "section headings": 104487,
      "structured summarization": 114044,
      "task generate": 118236,
      "context paragraphs": 22214,
      "work formulate": 134541,
      "task hierarchical": 118256,
      "predict sequence": 88936,
      "novel hierarchical": 80590,
      "model named": 71553,
      "attempts capture": 9766,
      "following ways": 44988,
      "ways introduce": 132165,
      "identification second": 51434,
      "second employ": 104405,
      "mechanism ensure": 67979,
      "ensure semantic": 37643,
      "improve consistency": 52357,
      "dataset public": 27132,
      "stateoftheart sequential": 112952,
      "task contextual": 118020,
      "user experience": 128981,
      "ood data": 81862,
      "fostering research": 45340,
      "research release": 100612,
      "release new": 98460,
      "new dialog": 78881,
      "dialog corpora": 30556,
      "corpora augmented": 23414,
      "way method": 132107,
      "models equipped": 73163,
      "mechanism large": 68006,
      "choice subword": 16906,
      "merge operations": 68477,
      "operations neural": 82071,
      "extracted methods": 42164,
      "recipes paper": 96792,
      "paper conduct": 83774,
      "conduct systematic": 20897,
      "systematic exploration": 116674,
      "exploration different": 41500,
      "strategy build": 113506,
      "provide guidance": 93841,
      "lstmbased architectures": 65680,
      "range different": 95559,
      "transformer architectures": 124294,
      "bpe size": 14540,
      "optimal choice": 82161,
      "urge community": 127845,
      "community make": 19080,
      "reduce performance": 97352,
      "contextual data": 22451,
      "vision tasks": 131769,
      "tasks study": 119529,
      "study natural": 114450,
      "translation different": 124752,
      "replace words": 98931,
      "sentence contextual": 105811,
      "words accurately": 133804,
      "word distribution": 132996,
      "weighted combination": 132340,
      "combination multiple": 18571,
      "multiple semantically": 75683,
      "capture richer": 15375,
      "richer information": 102805,
      "scale large": 103732,
      "better attention": 13512,
      "attention powerful": 9974,
      "driving force": 34162,
      "force recent": 45013,
      "models transformerbased": 74242,
      "bert models": 13175,
      "models apply": 72760,
      "parallel attention": 84586,
      "attention head": 9850,
      "focusing different": 44923,
      "input makes": 55360,
      "sophisticated functions": 110669,
      "functions simple": 46079,
      "surprising observation": 116128,
      "multiple heads": 75572,
      "large percentage": 61205,
      "time significantly": 122102,
      "single head": 109739,
      "potential speed": 88587,
      "analyze results": 6003,
      "parts model": 85578,
      "training dynamics": 123595,
      "play role": 87704,
      "gains provided": 46407,
      "legal text": 63316,
      "eu legislation": 38757,
      "multilabel text": 75158,
      "legal domain": 63315,
      "european unions": 38774,
      "dataset substantially": 27223,
      "substantially larger": 114901,
      "datasets suitable": 27740,
      "zeroshot learning": 135509,
      "neural classifiers": 77862,
      "employ labelwise": 36268,
      "labelwise attention": 58660,
      "leads best": 61919,
      "best overall": 13387,
      "harry potter": 49516,
      "prediction challenge": 89039,
      "prediction textual": 89143,
      "descriptions scenes": 29492,
      "text inference": 121050,
      "inference used": 54246,
      "automatically build": 10727,
      "different baselines": 31024,
      "tested models": 120578,
      "models lstmbased": 73537,
      "lstmbased approach": 65679,
      "obtains best": 81454,
      "performance frequent": 86393,
      "memory text": 68388,
      "rnns widely": 102984,
      "used field": 128553,
      "ranging text": 95620,
      "read text": 95941,
      "text beginning": 120681,
      "process long": 91522,
      "reading long": 96025,
      "document categorization": 32958,
      "categorization task": 15770,
      "task topic": 118794,
      "topic categorization": 122499,
      "words irrelevant": 134016,
      "reading texts": 96034,
      "texts step": 121622,
      "word determine": 132988,
      "analysis news": 5720,
      "categorization ontology": 15767,
      "classification benchmark": 17134,
      "sets experimental": 107669,
      "better standard": 13725,
      "performance efficiency": 86330,
      "open research": 81928,
      "action taken": 3082,
      "utterance paper": 130609,
      "user utterance": 129051,
      "utterance identify": 130600,
      "incremental dialog": 53724,
      "dstc2 dataset": 34214,
      "consists assigning": 21469,
      "allows identify": 5163,
      "identify appropriate": 51475,
      "finally implement": 44195,
      "trained new": 123220,
      "new labels": 78973,
      "units paper": 127263,
      "problems related": 91366,
      "phrases work": 87469,
      "units important": 127247,
      "corpus submitted": 24024,
      "detection work": 30098,
      "set polysemous": 107531,
      "example english": 39780,
      "measuring degree": 67908,
      "corpus algorithm": 23652,
      "algorithm involves": 4758,
      "target term": 117729,
      "standard methods": 112261,
      "degree distribution": 28570,
      "ensuring readability": 37648,
      "noun compound": 80423,
      "human machines": 50915,
      "quickly grasp": 95392,
      "grasp informative": 48986,
      "distinctive information": 32560,
      "methods supplement": 69786,
      "supplement information": 115938,
      "information existing": 54547,
      "texts solve": 121614,
      "ensure readability": 37642,
      "data fidelity": 25945,
      "fidelity generated": 43930,
      "generated type": 47145,
      "dataset automatic": 26750,
      "representations unstructured": 99956,
      "text offers": 121153,
      "tool exploring": 122385,
      "application available": 6835,
      "induction word": 54040,
      "wsi task": 135176,
      "word usages": 133622,
      "derived pretrained": 29357,
      "models elmo": 73125,
      "method bert": 68675,
      "prominent methods": 92251,
      "methods propose": 69691,
      "analysis revealing": 5817,
      "sources errors": 110887,
      "novel set": 80725,
      "labeled sentences": 58467,
      "sentences create": 106268,
      "extracted events": 42153,
      "achieves human": 2805,
      "level performance": 63487,
      "performance annotation": 86137,
      "previous event": 90408,
      "systems applied": 116732,
      "geographic contexts": 47849,
      "linking task": 64672,
      "answer open": 6549,
      "question role": 95215,
      "factual accuracy": 42927,
      "accuracy generated": 2172,
      "metric estimate": 69877,
      "text complementary": 120806,
      "evaluation bleu": 39138,
      "bleu bilingual": 14279,
      "bilingual evaluation": 14037,
      "introduce release": 56530,
      "largescale dataset": 61420,
      "wikipedia wikidata": 132688,
      "fact extraction": 42822,
      "shown able": 108439,
      "wikipedia text": 132682,
      "evaluation study": 39410,
      "modeling recurrent": 72523,
      "success various": 115141,
      "various sequential": 131193,
      "sequential tasks": 107257,
      "complex recurrent": 19868,
      "severe computational": 107880,
      "computational inefficiency": 20382,
      "propose lightweight": 92751,
      "uses input": 129235,
      "input forget": 55340,
      "gradient vanishing": 48608,
      "networks apply": 77508,
      "existing recurrent": 40263,
      "sequential models": 107239,
      "models extensive": 73213,
      "yields best": 135397,
      "performance finegrained": 86386,
      "presents computational": 89835,
      "approaches automatically": 8073,
      "user information": 128995,
      "information second": 54959,
      "carefully analyzed": 15515,
      "language tends": 60157,
      "sentences tend": 106519,
      "inspired findings": 55569,
      "developed endtoend": 30271,
      "qualitative results": 94575,
      "characterlevel embedding": 16563,
      "different writing": 31579,
      "languages language": 60670,
      "consist components": 21347,
      "character encoding": 16430,
      "encoding schemes": 36722,
      "representation study": 99423,
      "encoding scheme": 36721,
      "scheme effective": 103918,
      "scheme evaluated": 103920,
      "types result": 126352,
      "adversarial generation": 4125,
      "documents consist": 33207,
      "consist long": 21349,
      "idea learning": 51329,
      "tasks task": 119549,
      "sequence vectors": 107098,
      "lower level": 65439,
      "modeling mlm": 72475,
      "sentence prediction": 105996,
      "tasks bert": 118965,
      "new adversarial": 78765,
      "coherence generated": 18301,
      "representation tree": 99454,
      "pairs word": 83666,
      "task field": 118208,
      "proposed tasks": 93565,
      "pairs single": 83640,
      "focus words": 44842,
      "morphemes work": 74673,
      "description dataset": 29459,
      "process test": 91574,
      "range word": 95613,
      "techniques focus": 119891,
      "systems past": 117054,
      "processing artificial": 91626,
      "generating language": 47227,
      "understandable humans": 126793,
      "teaching machines": 119712,
      "converse humans": 23081,
      "unprecedented growth": 127491,
      "number research": 80955,
      "research articles": 100417,
      "articles published": 9211,
      "specifically problem": 111578,
      "problem activity": 90916,
      "define state": 28493,
      "work provide": 134756,
      "approaches approaches": 8065,
      "comprehensive review": 20268,
      "review building": 102536,
      "building open": 14875,
      "systems important": 116945,
      "approaches building": 8091,
      "seq2seq language": 106894,
      "models architecture": 72769,
      "identify important": 51507,
      "important areas": 52103,
      "research building": 100437,
      "effective dialogue": 34662,
      "provide pointers": 93893,
      "pointers tackle": 87839,
      "tackle open": 117301,
      "generation capabilities": 47328,
      "assessment paper": 9524,
      "multiturn conversations": 75912,
      "coherent conversation": 18315,
      "knowledge selection": 58164,
      "knowledge grounded": 57995,
      "aspects like": 9391,
      "informativeness coherence": 55164,
      "dataset demonstrating": 26862,
      "demonstrating proposed": 28981,
      "generator dual": 47769,
      "information maximization": 54758,
      "parsing aims": 85066,
      "aims transform": 4580,
      "transform natural": 124253,
      "nl utterances": 79510,
      "representations mrs": 99758,
      "generator achieves": 47765,
      "intrinsic connection": 56351,
      "connection tasks": 21135,
      "tasks studied": 119527,
      "paper model": 84046,
      "tasks joint": 119229,
      "tasks concretely": 118999,
      "concretely propose": 20743,
      "regularize learning": 97765,
      "joint distributions": 57268,
      "code generation": 18106,
      "generation summarization": 47648,
      "summarization performance": 115547,
      "consistently improved": 21417,
      "universal conceptual": 127276,
      "conceptual cognitive": 20657,
      "cognitive annotation": 18263,
      "entity event": 37932,
      "compared annotations": 19322,
      "neural ranking": 78642,
      "providing additional": 94100,
      "additional semantics": 3569,
      "words joined": 134017,
      "build dataset": 14760,
      "set approaches": 107363,
      "segmentation accuracy": 104572,
      "method finally": 68839,
      "semantics obtained": 105441,
      "segmentation useful": 104649,
      "increase average": 53589,
      "average recall": 11211,
      "dataset handling": 26959,
      "tabletotext generation": 117273,
      "generation automatically": 47318,
      "automatically constructed": 10742,
      "constructed datasets": 21659,
      "contain reference": 21753,
      "information corresponding": 54458,
      "metrics rely": 70001,
      "rely solely": 98740,
      "bleu rouge": 14321,
      "data computing": 25765,
      "existing text": 40314,
      "generation metrics": 47476,
      "metrics adapt": 69920,
      "adapt evaluate": 3176,
      "wiseman et": 132740,
      "imitation learning": 51838,
      "learning simultaneous": 63029,
      "widely useful": 132596,
      "difficult tasks": 31664,
      "propose simpler": 93070,
      "flexible policies": 44662,
      "better bleu": 13521,
      "vision text": 131770,
      "using domain": 129605,
      "wide use": 132517,
      "use mobile": 128146,
      "method transfer": 69197,
      "paper implemented": 83978,
      "analysis applications": 5495,
      "devices using": 30460,
      "edge devices": 34522,
      "input english": 55331,
      "attention knowledge": 9864,
      "lowresource named": 65537,
      "recognition recent": 96982,
      "achieved tasks": 2714,
      "ner especially": 77041,
      "considerable labeled": 21252,
      "labeled resources": 58464,
      "resources lowresource": 101006,
      "highresource english": 50421,
      "leading poor": 61901,
      "performance ner": 86563,
      "leveraging knowledge": 63682,
      "pretrained highresource": 90043,
      "english models": 37209,
      "languages highresource": 60619,
      "based translation": 12134,
      "translation module": 124969,
      "attention matrices": 9877,
      "align languages": 4901,
      "languages furthermore": 60603,
      "features pretrained": 43665,
      "english model": 37208,
      "features result": 43705,
      "features enrich": 43488,
      "experiments lowresource": 40990,
      "datasets achieves": 27294,
      "consistent significant": 21401,
      "indicates effectiveness": 53859,
      "effectiveness joint": 34899,
      "joint effects": 57269,
      "context user": 22314,
      "online conversation": 81754,
      "online world": 81818,
      "exponential growth": 41676,
      "come play": 18770,
      "play increasingly": 87700,
      "order help": 82336,
      "users better": 129097,
      "better engage": 13572,
      "online study": 81808,
      "study challenging": 114333,
      "user come": 128970,
      "conversations users": 23075,
      "users previous": 129156,
      "neural framework": 77917,
      "layers modeling": 61790,
      "history jointly": 50559,
      "experiment largescale": 40471,
      "datasets collected": 27351,
      "twitter reddit": 126120,
      "methods previous": 69683,
      "multitask semantic": 75889,
      "strategies semantic": 113487,
      "sdp semantic": 104280,
      "acyclic graphs": 3159,
      "algorithm combines": 4704,
      "head words": 49580,
      "words train": 134274,
      "learning taskspecific": 63090,
      "lack various": 58768,
      "various corpora": 131067,
      "propose largescale": 92741,
      "test dataset": 120441,
      "unique language": 127189,
      "blank symbols": 14265,
      "study design": 114352,
      "representation idioms": 99251,
      "affect performance": 4222,
      "models results": 73953,
      "substantially worse": 114917,
      "indicating large": 53872,
      "large space": 61276,
      "research exploiting": 100498,
      "approaches exploit": 8158,
      "context extracted": 22103,
      "encoder layer": 36538,
      "layer improve": 61723,
      "representations individual": 99689,
      "internal layers": 56168,
      "encoder form": 36528,
      "benchmarks model": 12921,
      "model vaswani": 72300,
      "demonstrating necessity": 28979,
      "context nmt": 22201,
      "transformer encoder": 124322,
      "encoder neural": 36546,
      "source representations": 110807,
      "representations wordlevel": 99987,
      "wordlevel subwordlevel": 133756,
      "sequence different": 106931,
      "different subword": 31463,
      "subword vocabulary": 115042,
      "vocabulary sizes": 131905,
      "integrate different": 55749,
      "different segmentations": 31404,
      "model transformer": 72231,
      "transformer propose": 124370,
      "automatic way": 10707,
      "way training": 132137,
      "methods lattice": 69582,
      "used complementary": 128439,
      "performance experiment": 86353,
      "subwordlevel representations": 115052,
      "representations conventional": 99574,
      "conventional transformer": 22911,
      "little knowledge": 64813,
      "knowledge quantitative": 58130,
      "objects events": 81147,
      "events propose": 39597,
      "quantitative information": 94870,
      "information large": 54714,
      "amounts web": 5367,
      "use create": 127974,
      "new large": 78985,
      "resource consisting": 100833,
      "quantities associated": 94895,
      "adjectives verbs": 3864,
      "focused making": 44865,
      "datasets relative": 27667,
      "dataset introduce": 26985,
      "method joint": 68924,
      "style neural": 114585,
      "neural nlg": 78612,
      "popular recent": 88121,
      "generating syntactically": 47270,
      "preserve semantics": 89927,
      "various shortcomings": 131195,
      "systems clear": 116780,
      "outputs simple": 83181,
      "dull repetitive": 34257,
      "repetitive paper": 98919,
      "creating training": 24709,
      "datasets parallel": 27613,
      "representations reference": 99847,
      "texts rich": 121596,
      "user reviews": 129039,
      "systematically exploring": 116696,
      "output present": 83107,
      "representations highly": 99677,
      "texts spanning": 121617,
      "reused generate": 102490,
      "nlg datasets": 79518,
      "datasets domains": 27438,
      "models control": 72984,
      "control important": 22807,
      "aspects including": 9389,
      "output length": 83089,
      "allowing models": 5116,
      "models successfully": 74124,
      "multiple style": 75710,
      "model semisupervised": 71970,
      "sequence matching": 107007,
      "relationship pair": 98305,
      "text sequences": 121288,
      "approaches consider": 8106,
      "consider sequence": 21228,
      "separately proposed": 106876,
      "framework utilizes": 45737,
      "given relationship": 48104,
      "framework facilitate": 45544,
      "semisupervised training": 105630,
      "semantic constraints": 105017,
      "generate semantically": 47009,
      "semantically plausible": 105371,
      "plausible diverse": 87682,
      "diverse sequences": 32842,
      "sequences demonstrate": 107118,
      "qualitative experiments": 94563,
      "results semisupervised": 102152,
      "inference paraphrase": 54187,
      "implicitly encode": 52027,
      "knowledge extent": 57923,
      "limited previous": 64265,
      "strategies optimal": 113480,
      "encode relational": 36447,
      "separate word": 106865,
      "embedding relational": 35481,
      "statistics used": 113210,
      "base available": 11444,
      "available analysis": 10939,
      "information complementary": 54429,
      "topic discussion": 122508,
      "focus mainly": 44786,
      "extend scope": 41795,
      "detect stance": 29820,
      "help predict": 49764,
      "network rcn": 77400,
      "reason information": 96197,
      "comparison empirical": 19541,
      "method discover": 68779,
      "networks widely": 77814,
      "spite great": 111936,
      "success deep": 115067,
      "neglect importance": 76968,
      "importance modeling": 52069,
      "understanding texts": 126999,
      "learn context": 62006,
      "improve text": 52562,
      "classification specifically": 17415,
      "representation classification": 99189,
      "classification compared": 17150,
      "capture context": 15279,
      "extensively evaluate": 41965,
      "method benchmark": 68673,
      "classification experimental": 17202,
      "systems despite": 116842,
      "retrievalbased models": 102445,
      "systems main": 116990,
      "main obstacle": 66447,
      "lowdata regime": 65405,
      "dialogue tasks": 30802,
      "tasks inspired": 119210,
      "success pretraining": 115117,
      "method pretrains": 69070,
      "large generaldomain": 61099,
      "conversational corpora": 22997,
      "finetunes pretrained": 44440,
      "pretrained model": 90128,
      "indomain dataset": 53965,
      "capture nuances": 15352,
      "given dialogue": 48017,
      "evaluation diverse": 39189,
      "domains ranging": 33844,
      "proposed training": 93575,
      "method encode": 68802,
      "systems natural": 117010,
      "represent multiple": 99122,
      "multiple linguistic": 75606,
      "suffer slow": 115245,
      "proposed paradigm": 93517,
      "pairwise similarities": 83683,
      "extend models": 41788,
      "models handle": 73318,
      "lattice structure": 61671,
      "available propose": 11082,
      "positional embeddings": 88300,
      "structures apply": 114057,
      "outperforms examined": 82884,
      "examined baselines": 39762,
      "faster compute": 43167,
      "encoding sentences": 36725,
      "compression machine": 20310,
      "learning attention": 62374,
      "attention given": 9847,
      "methods designed": 69437,
      "designed produce": 29619,
      "rich sentence": 102784,
      "focus solely": 44821,
      "original sequence": 82545,
      "method encoding": 68804,
      "encoding long": 36708,
      "test method": 120470,
      "method sentiment": 69125,
      "contextual spoken": 22500,
      "proven helpful": 93726,
      "typically encoded": 126426,
      "representations previous": 99820,
      "context memory": 22175,
      "objective maximizing": 81092,
      "slu performance": 110121,
      "new dialogue": 78882,
      "multitask framework": 75820,
      "dialogue session": 30740,
      "memory encoder": 68312,
      "retrieval mechanism": 102406,
      "parser domain": 84940,
      "adaptation propose": 3254,
      "based idea": 11763,
      "corpora exploiting": 23484,
      "resources dependency": 100962,
      "simple relying": 109507,
      "current bestperforming": 25267,
      "experiments detailed": 40901,
      "detailed discussion": 29766,
      "datasets biomedical": 27338,
      "create experimental": 24615,
      "math problems": 67449,
      "hierarchical decoder": 49948,
      "threelevel hierarchical": 121940,
      "hierarchical attention": 49930,
      "attention generate": 9846,
      "generate abstractive": 46895,
      "abstractive summaries": 1807,
      "occur sequence": 81500,
      "common occurrence": 18905,
      "chat conversations": 16653,
      "quickly obtain": 95394,
      "overview texts": 83381,
      "texts existing": 121513,
      "major issues": 66569,
      "hierarchical encoderdecoder": 49960,
      "levels information": 63544,
      "words implicitly": 133989,
      "outperforms sota": 82986,
      "refinement algorithm": 97595,
      "introduce use": 56564,
      "poincare embeddings": 87794,
      "taxonomy induction": 119667,
      "results semeval2016": 102147,
      "task 13": 117816,
      "representations supporting": 99918,
      "capture hierarchical": 15309,
      "patterns annotated": 85729,
      "patterns frequency": 85746,
      "close random": 17829,
      "combining supervised": 18752,
      "terms model": 120348,
      "referents pronouns": 97568,
      "pronouns dropped": 92349,
      "chinese sentences": 16818,
      "happens frequently": 49440,
      "essential applications": 38540,
      "pronouns need": 92359,
      "need resolved": 76851,
      "translation chinese": 124693,
      "language work": 60346,
      "based structured": 12080,
      "structured attention": 113992,
      "pronouns utilizing": 92363,
      "utilizing sentencelevel": 130580,
      "sentencelevel wordlevel": 106182,
      "information results": 54928,
      "different conversational": 31065,
      "improvement current": 52700,
      "largescale multilabel": 61470,
      "classification lmtc": 17255,
      "better current": 13557,
      "methods domainspecific": 69458,
      "domainspecific word2vec": 33924,
      "length limit": 63364,
      "finetune bert": 44403,
      "obtaining best": 81434,
      "results zeroshot": 102352,
      "nlp recent": 79676,
      "large networks": 61165,
      "trained abundant": 123063,
      "abundant data": 1833,
      "gains accuracy": 46381,
      "accuracy nlp": 2217,
      "depend availability": 29044,
      "exceptionally large": 39932,
      "large computational": 61058,
      "result models": 101384,
      "models costly": 72994,
      "costly train": 24394,
      "train develop": 122924,
      "cloud compute": 17889,
      "compute time": 20472,
      "environmentally carbon": 38198,
      "carbon footprint": 15500,
      "issue attention": 56991,
      "successful neural": 115166,
      "nlp based": 79580,
      "findings propose": 44307,
      "enhanced deep": 37503,
      "deep transition": 28435,
      "transition architecture": 124481,
      "labeling current": 58490,
      "consecutive hidden": 21163,
      "modeling global": 72437,
      "try address": 125892,
      "state transition": 112536,
      "global representation": 48265,
      "tasks given": 119146,
      "conll03 ner": 21091,
      "chunking task": 16966,
      "outperforms best": 82855,
      "results settings": 102164,
      "settings furthermore": 107803,
      "furthermore leveraging": 46187,
      "bert additional": 13060,
      "f1 ner": 42664,
      "time present": 122073,
      "temporal progression": 120115,
      "methods try": 69817,
      "meaning parts": 67657,
      "change time": 16371,
      "time use": 122133,
      "books corpus": 14423,
      "corpus diachronic": 23757,
      "extracted corpus": 42148,
      "inherent limitations": 55190,
      "limitations using": 64197,
      "using temporal": 130279,
      "information helps": 54654,
      "helps predicting": 49828,
      "time selection": 122097,
      "studies typically": 114293,
      "students learn": 114148,
      "candidate set": 15124,
      "fillintheblank task": 44065,
      "context given": 22117,
      "given paper": 48074,
      "translation leverage": 124905,
      "word achieve": 132903,
      "goal study": 48387,
      "choose appropriate": 16923,
      "appropriate words": 8446,
      "terms sentence": 120380,
      "words view": 134307,
      "armenian language": 9017,
      "evaluation work": 39448,
      "extrinsically evaluate": 42630,
      "embeddings presented": 35872,
      "presented trained": 89799,
      "algorithms adapt": 4826,
      "task intrinsic": 118306,
      "evaluation embeddings": 39199,
      "tasks employed": 119092,
      "network using": 77469,
      "propose corpus": 92607,
      "datasets public": 27653,
      "future models": 46282,
      "things different": 121861,
      "different angle": 30991,
      "information revolution": 54955,
      "significant increase": 108817,
      "data observe": 26184,
      "biases exist": 13871,
      "diverse comprehensive": 32799,
      "comprehensive set": 20270,
      "set perspectives": 107528,
      "responses form": 101268,
      "evidence supporting": 39673,
      "propose address": 92531,
      "given claim": 47991,
      "evidence paragraphs": 39656,
      "evidence making": 39651,
      "use online": 128181,
      "online debate": 81759,
      "create initial": 24618,
      "initial data": 55209,
      "highquality data": 50375,
      "data dataset": 25817,
      "respectively provide": 101159,
      "highlight key": 50264,
      "underlying language": 126682,
      "understanding challenges": 126810,
      "baselines multiple": 12432,
      "multiple subtasks": 75712,
      "community address": 19067,
      "predict novel": 88909,
      "predicting unseen": 89019,
      "models observed": 73653,
      "representations constituents": 99561,
      "evidence model": 39652,
      "model plausibility": 71749,
      "generated novel": 47102,
      "novel combinations": 80511,
      "accurate best": 2326,
      "previously unseen": 90629,
      "independent human": 53771,
      "attention interpretable": 9861,
      "boosted performance": 14440,
      "layers explicitly": 61778,
      "assumption holds": 9661,
      "models analyzing": 72752,
      "impact model": 51879,
      "predict effects": 88883,
      "conclude attention": 20709,
      "components overall": 20042,
      "importance model": 52068,
      "model means": 71508,
      "critical problem": 24822,
      "involves learning": 56897,
      "approach embedding": 7517,
      "form knowledge": 45097,
      "typically make": 126445,
      "use vector": 128355,
      "propose online": 92982,
      "method construct": 68734,
      "construct graph": 21632,
      "information design": 54475,
      "graphical structure": 48933,
      "tasks various": 119595,
      "better representations": 13697,
      "representations example": 99641,
      "respectively using": 101176,
      "nested entity": 77104,
      "ner approaches": 77021,
      "face problem": 42734,
      "mentions paper": 68463,
      "resolve problem": 100802,
      "problem modeling": 91131,
      "headdriven phrase": 49583,
      "phrase structures": 87373,
      "structures entity": 114073,
      "anchor words": 6084,
      "words possible": 134129,
      "word exploiting": 133281,
      "furthermore design": 46161,
      "loss objective": 65288,
      "function train": 46047,
      "detection benchmarks": 29889,
      "lowresource translation": 65571,
      "translation translation": 125396,
      "languages lrls": 60711,
      "terms adequacy": 120266,
      "adequacy fluency": 3825,
      "augmentation utilizing": 10316,
      "utilizing large": 130571,
      "alleviate problems": 5047,
      "translation uses": 125415,
      "targetside monolingual": 117801,
      "related highresource": 97865,
      "language hrl": 59124,
      "specifically experiment": 111547,
      "pivoting method": 87591,
      "method convert": 68743,
      "data lrl": 26103,
      "data distribution": 25860,
      "dictionary second": 30893,
      "framework extensive": 45539,
      "lowresource datasets": 65492,
      "settings data": 107793,
      "techniques improve": 119906,
      "compared supervised": 19472,
      "backtranslation baselines": 11334,
      "stress detection": 113579,
      "detection russian": 30040,
      "trained languages": 123172,
      "applications transfer": 7029,
      "task possible": 118538,
      "model crosslingual": 70942,
      "setting using": 107786,
      "languages improves": 60629,
      "improves quality": 53033,
      "networks social": 77775,
      "entities document": 37770,
      "applications current": 6910,
      "task reach": 118602,
      "reach high": 95891,
      "media domains": 68108,
      "domains present": 33832,
      "systems address": 116722,
      "challenges processing": 16198,
      "processing social": 91795,
      "phonetics phonology": 87305,
      "embeddings partofspeech": 35857,
      "tags features": 117474,
      "model multitask": 71545,
      "multitask endtoend": 75819,
      "network output": 77378,
      "classifiers second": 17633,
      "uses multitask": 129250,
      "learning crf": 62469,
      "systems outperform": 117036,
      "workshop noisy": 135011,
      "noisy usergenerated": 80131,
      "2017 dataset": 632,
      "continual learning": 22598,
      "catastrophic forgetting": 15710,
      "answering study": 6704,
      "study issue": 114418,
      "multimodal approaches": 75415,
      "approaches visual": 8400,
      "answering vqa": 6721,
      "vqa tasks": 131962,
      "types questions": 126342,
      "questions test": 95368,
      "question types": 95233,
      "order matter": 82368,
      "methods mitigate": 69620,
      "learning matching": 62710,
      "systems study": 117164,
      "model response": 71919,
      "designing architecture": 29636,
      "explored existing": 41624,
      "learn robust": 62135,
      "noisy training": 80128,
      "framework simultaneously": 45690,
      "learn matching": 62087,
      "iteration model": 57122,
      "model transfers": 72230,
      "transfers knowledge": 124249,
      "learned training": 62269,
      "training teacher": 123910,
      "teacher student": 119698,
      "student models": 114142,
      "learn improved": 62068,
      "sets indicate": 107677,
      "existing matching": 40171,
      "scientific news": 104012,
      "articles research": 9214,
      "research results": 100620,
      "research paper": 100577,
      "likely occur": 64141,
      "hypothesis construct": 51263,
      "right context": 102830,
      "context features": 22105,
      "features journal": 43576,
      "role identifying": 103175,
      "language hard": 59116,
      "easier model": 34418,
      "model current": 70945,
      "methods prior": 69686,
      "question language": 95173,
      "perform equally": 85990,
      "69 languages": 1198,
      "model obtain": 71599,
      "corpora words": 23623,
      "missing data": 70322,
      "exploiting model": 41472,
      "written language": 135132,
      "fair comparison": 42992,
      "trying answer": 125902,
      "various perspectives": 131167,
      "claim evidence": 17009,
      "combination retrieval": 18581,
      "using recent": 130084,
      "developments natural": 30443,
      "understanding make": 126884,
      "expand coverage": 40354,
      "coverage improve": 24509,
      "decisions time": 27967,
      "employs various": 36337,
      "representations linguistic": 99741,
      "linguistic insight": 64494,
      "multilingual domain": 75248,
      "wikipedia languages": 132663,
      "mapping sentence": 67145,
      "using sentences": 130150,
      "sentences parallel": 106433,
      "information representations": 54921,
      "trained natural": 123211,
      "embeddings context": 35612,
      "context hidden": 22122,
      "words ambiguous": 133814,
      "model deals": 70952,
      "types information": 126300,
      "information represented": 54922,
      "attention simultaneous": 10011,
      "simultaneous machine": 109655,
      "sentence source": 106083,
      "live streaming": 64851,
      "streaming scenarios": 113555,
      "simultaneous systems": 109664,
      "systems carefully": 116770,
      "balance quality": 11398,
      "quality latency": 94703,
      "translation learn": 124900,
      "learn adaptive": 61985,
      "model attends": 70699,
      "attends source": 9786,
      "source tokens": 110855,
      "read far": 95931,
      "sentence soft": 106081,
      "extends monotonic": 41828,
      "schedule allows": 103882,
      "latencyquality tradeoffs": 61571,
      "proposed waitk": 93589,
      "techniques provide": 119963,
      "cambridge university": 15079,
      "university engineering": 127352,
      "elastic weight": 35231,
      "weight consolidation": 132323,
      "consolidation ewc": 21507,
      "forms language": 45247,
      "report substantial": 99052,
      "substantial gains": 114856,
      "gains finetuning": 46393,
      "combination checkpoint": 18552,
      "checkpoint averaging": 16701,
      "sentencelevel transformer": 106179,
      "transformer lm": 124335,
      "based modified": 11856,
      "ngram lm": 79439,
      "translation recent": 125198,
      "especially achieving": 38430,
      "results depending": 101714,
      "input translation": 55460,
      "extend attention": 41781,
      "sequential input": 107230,
      "input propose": 55408,
      "output automatic": 83051,
      "contains multiple": 21832,
      "multiple paths": 75637,
      "information lattice": 54718,
      "mechanism obtain": 68022,
      "obtain latent": 81299,
      "spanishenglish speech": 111179,
      "experiments lattice": 40986,
      "outperforms transformer": 83038,
      "lattice lstm": 61669,
      "approach wmt": 8018,
      "task observe": 118469,
      "observe improvements": 81197,
      "embeddings distinguishing": 35659,
      "approach entirely": 7533,
      "based pretrained": 11944,
      "lexicalsemantic information": 63863,
      "taskspecific information": 119628,
      "trained based": 123076,
      "speed performance": 111874,
      "dependency distance": 29139,
      "short sequences": 108235,
      "graph theoretic": 48874,
      "close sentences": 17832,
      "evidence principle": 39662,
      "dependency treebanks": 29240,
      "maximization introduce": 67507,
      "verify hypothesis": 131589,
      "different families": 31144,
      "analysis qualitative": 5790,
      "qualitative differences": 94560,
      "attempt discover": 9740,
      "problem sentiment": 91215,
      "english provide": 37252,
      "provide challenging": 93772,
      "dataset collect": 26793,
      "study demonstrates": 114348,
      "demonstrates usefulness": 28967,
      "usefulness dataset": 128957,
      "given sentiment": 48119,
      "effect word": 34620,
      "models sentiment": 74004,
      "analysis make": 5690,
      "modeling objective": 72495,
      "problem crosslingual": 90989,
      "crosslingual models": 24977,
      "use bilingual": 127925,
      "embeddings features": 35721,
      "reordering preprocessing": 98901,
      "classification language": 17246,
      "reordering helps": 98898,
      "models cnns": 72914,
      "supervised event": 115752,
      "type pairs": 126217,
      "pairs including": 83563,
      "confusion problem": 21057,
      "training loss": 123691,
      "pairs furthermore": 83552,
      "2017 datasets": 633,
      "chinese event": 16765,
      "diverse decoding": 32806,
      "methods conditional": 69391,
      "conditional language": 20758,
      "models greatly": 73315,
      "highquality natural": 50398,
      "benefit able": 12958,
      "decoding strategies": 28137,
      "candidate list": 15109,
      "cover space": 24490,
      "highquality outputs": 50403,
      "improvements tasks": 52929,
      "candidate outputs": 15114,
      "outputs standard": 83182,
      "high likelihood": 50085,
      "extensive survey": 41959,
      "outputs conditional": 83162,
      "sacrificing quality": 103528,
      "using automatically": 129372,
      "detection common": 29901,
      "common practice": 18909,
      "identify evaluate": 51498,
      "mention boundary": 68416,
      "problem minimum": 91128,
      "solution does": 110552,
      "does scale": 33400,
      "extracting minimum": 42222,
      "consistent manually": 21393,
      "annotated experts": 6196,
      "using minimum": 129886,
      "particular important": 85422,
      "standard coreference": 112215,
      "scores based": 104154,
      "based maximum": 11832,
      "strong supervision": 113721,
      "signal model": 108688,
      "model design": 70972,
      "patterns language": 85751,
      "effectively train": 34852,
      "problem applied": 90935,
      "55 absolute": 1110,
      "results report": 102126,
      "report automatic": 98977,
      "language related": 60031,
      "ancient greek": 6091,
      "model correctly": 70932,
      "questions questionanswering": 95347,
      "generate questions": 46992,
      "sentence paragraph": 105977,
      "setting different": 107742,
      "major aspects": 66537,
      "history using": 50572,
      "turns propose": 125994,
      "modeling coreference": 72409,
      "coreferent mentions": 23382,
      "generated questions": 47110,
      "makes generated": 66784,
      "shifting focus": 108185,
      "parts extensive": 85575,
      "baselines generate": 12403,
      "generate highly": 46951,
      "conversational questions": 23024,
      "questions code": 95283,
      "code implementation": 18108,
      "manipulation strategies": 66932,
      "strategies able": 113447,
      "requires ability": 100240,
      "generate fluent": 46943,
      "fluent text": 44721,
      "factual knowledge": 42935,
      "introduce knowledge": 56442,
      "graph language": 48828,
      "graph relevant": 48862,
      "enable model": 36358,
      "outofvocabulary tokens": 82680,
      "dataset corpus": 26833,
      "text aligned": 120635,
      "achieves significantly": 2863,
      "baseline language": 12235,
      "model additionally": 70632,
      "outperforms large": 82913,
      "large language": 61126,
      "models generating": 73292,
      "probing linguistic": 90892,
      "text recurrent": 121231,
      "generic sequence": 47809,
      "sequence processing": 107038,
      "genuine linguistic": 47841,
      "training present": 123784,
      "present multilingual": 89561,
      "multilingual study": 75377,
      "knowledge encoded": 57892,
      "rnns trained": 102981,
      "word boundaries": 132936,
      "realistic task": 96106,
      "having discover": 49557,
      "useful linguistic": 128901,
      "able solve": 1690,
      "wordlevel knowledge": 133738,
      "rigid word": 102846,
      "conversations hierarchical": 23056,
      "hierarchical latent": 49971,
      "challenge generate": 16037,
      "generate coherent": 46911,
      "conversation takes": 22976,
      "different content": 31057,
      "content little": 21900,
      "hierarchical relationship": 49999,
      "conversational semantic": 23028,
      "contains latent": 21826,
      "captures global": 15444,
      "common topic": 18939,
      "terms fluency": 120328,
      "fluency coherence": 44694,
      "diversity compared": 32869,
      "methods transfer": 69812,
      "learning causal": 62422,
      "detection consider": 29906,
      "sentences express": 106307,
      "causal relations": 15811,
      "instances relation": 55633,
      "bert using": 13248,
      "bidirectional gru": 13927,
      "baseline experiment": 12216,
      "datasets new": 27595,
      "new biomedical": 78817,
      "detection dataset": 29918,
      "learning helps": 62630,
      "datasets larger": 27542,
      "reaches performance": 95915,
      "datasets transfer": 27766,
      "automated reasoning": 10465,
      "based finegrained": 11721,
      "end apply": 36793,
      "partial functions": 85288,
      "kinds context": 57687,
      "meaning occur": 67654,
      "implemented algorithm": 51959,
      "context recognition": 22245,
      "inference machine": 54163,
      "documentlevel sentiment": 33163,
      "paper details": 83863,
      "2019 evaluation": 696,
      "participated task": 85350,
      "deep bilstm": 28209,
      "shows promise": 108610,
      "information negation": 54795,
      "neural sentiment": 78666,
      "results test": 102264,
      "scheduled sampling": 103886,
      "sampling technique": 103615,
      "known problems": 58317,
      "model mix": 71518,
      "step training": 113306,
      "improving model": 53115,
      "rnn transformer": 102938,
      "generation new": 47509,
      "technique propose": 119810,
      "changes allow": 16382,
      "applied transformer": 7133,
      "strategy experiments": 113513,
      "pairs achieve": 83481,
      "technique promising": 119809,
      "stateoftheart vietnamese": 113025,
      "segmentation step": 104636,
      "step tasks": 113303,
      "tasks vietnamese": 119597,
      "paper reviews": 84411,
      "reviews stateoftheart": 102632,
      "building corpora": 14831,
      "corpora developing": 23453,
      "discuss building": 32238,
      "building corpus": 14832,
      "applied solve": 7118,
      "segmentation existing": 104586,
      "existing toolkits": 40317,
      "implementing machine": 51982,
      "accuracy vietnamese": 2317,
      "segmentation according": 104571,
      "existing vietnamese": 40330,
      "segmentation systems": 104638,
      "classification feature": 17209,
      "attribution methods": 10210,
      "users interpret": 129140,
      "interpret predictions": 56214,
      "models approach": 72762,
      "approach integrates": 7644,
      "feature attributions": 43252,
      "function allow": 45996,
      "priors model": 90779,
      "model building": 70791,
      "effectiveness technique": 34965,
      "bias text": 13849,
      "terms improving": 120338,
      "classifier performance": 17570,
      "data setting": 26438,
      "setting forcing": 107750,
      "forcing model": 45023,
      "terms approach": 120269,
      "approach adds": 7337,
      "distance loss": 32422,
      "taskspecific prior": 119638,
      "objective experiments": 81077,
      "model biases": 70776,
      "original task": 82550,
      "task ii": 118271,
      "helps model": 49825,
      "robust machine": 103044,
      "translation domain": 124768,
      "domain sensitive": 33635,
      "report paper": 99021,
      "translation developed": 124747,
      "developed jointly": 30282,
      "baidu research": 11393,
      "oregon state": 82454,
      "state university": 112537,
      "wmt 2019": 132775,
      "translation robustness": 125225,
      "task translation": 118811,
      "style different": 114573,
      "types noises": 126332,
      "corpora extremely": 23486,
      "extremely limited": 42601,
      "use domain": 128014,
      "domains little": 33809,
      "generate parallel": 46985,
      "parallel dataset": 84661,
      "using model": 129888,
      "trained similar": 123276,
      "similar domain": 109086,
      "way achieve": 132053,
      "achieve 10": 2406,
      "translation compared": 124700,
      "model relationship": 71890,
      "comparing similar": 19518,
      "representations initially": 99694,
      "nmt standard": 79964,
      "best given": 13337,
      "sequences does": 107120,
      "empirically method": 36237,
      "learning introduce": 62659,
      "methods semisupervised": 69739,
      "model discriminative": 71007,
      "variables define": 130854,
      "drawing inspiration": 34129,
      "prediction objectives": 89094,
      "discriminative information": 32209,
      "information latent": 54717,
      "including ones": 53343,
      "structure enables": 113850,
      "models consistently": 72966,
      "standard sequential": 112305,
      "datasets improve": 27511,
      "reranking based": 100367,
      "based event": 11692,
      "candidates generated": 15135,
      "generated conversational": 47064,
      "tensor model": 120188,
      "relations limited": 98203,
      "limited event": 64233,
      "knowledge experimental": 57912,
      "interpretation neural": 56271,
      "original goal": 82525,
      "goal jointly": 48359,
      "align translate": 4905,
      "learning interpretable": 62658,
      "learn interpretable": 62072,
      "propose series": 93046,
      "offline online": 81614,
      "parameter update": 84727,
      "method better": 68677,
      "systems performing": 117064,
      "automatic alignment": 10489,
      "alignment tools": 4994,
      "embedding projection": 35474,
      "sentiment model": 106760,
      "model comparisons": 70861,
      "comparisons realworld": 19594,
      "analysis benefits": 5514,
      "benefits large": 13010,
      "large handannotated": 61102,
      "resources order": 101024,
      "test machine": 120467,
      "data hungry": 26010,
      "sentiment tasks": 106803,
      "propose crosslingual": 92609,
      "information bilingual": 54399,
      "representations jointly": 99707,
      "jointly optimizing": 57375,
      "semantics sentiment": 105467,
      "performance sentencelevel": 86699,
      "domains shows": 33857,
      "tasks analysis": 118928,
      "languages demonstrates": 60491,
      "unlabeled monolingual": 127401,
      "data surprisingly": 26524,
      "effect sentiment": 34613,
      "sentiment results": 106789,
      "annotated source": 6235,
      "target leads": 117672,
      "sourcetarget language": 110946,
      "pairs similar": 83637,
      "finally domain": 44175,
      "decreased performance": 28173,
      "performance suggests": 86774,
      "type analysis": 126185,
      "analysis applied": 5496,
      "representing knowledge": 100053,
      "different actions": 30985,
      "human efforts": 50800,
      "process automated": 91426,
      "unique challenges": 127183,
      "information conversational": 54453,
      "conversational text": 23040,
      "text labeled": 121074,
      "lowresource paper": 65551,
      "propose weaklysupervised": 93160,
      "extraction framework": 42342,
      "model longrange": 71472,
      "relations neighbor": 98219,
      "capturing context": 15462,
      "context dependency": 22051,
      "manually annotate": 67023,
      "study morphological": 114442,
      "informationtheoretic measure": 55124,
      "neural transduction": 78715,
      "transduction model": 124025,
      "28 languages": 865,
      "exploratory analyses": 41509,
      "knowledge result": 58157,
      "linguistics literature": 64614,
      "aggregated level": 4352,
      "underlying stems": 126701,
      "simple natural": 109476,
      "technical note": 119749,
      "tools automatic": 122437,
      "machinelearning based": 66322,
      "based using": 12151,
      "impact preprocessing": 51885,
      "networks stateoftheart": 77780,
      "techniques shown": 119984,
      "effects performance": 35002,
      "mt remains": 74988,
      "remains unclear": 98824,
      "paper systematically": 84460,
      "furthermore consider": 46157,
      "range data": 95552,
      "sizes compare": 109958,
      "compare effect": 19241,
      "approaches empirical": 8139,
      "best choice": 13316,
      "largely based": 61334,
      "type model": 126212,
      "size data": 109918,
      "gain significant": 46350,
      "using selection": 130135,
      "representations deep": 99585,
      "relationship sentence": 98307,
      "learned deep": 62210,
      "layers recurrent": 61804,
      "brain regions": 14554,
      "processing sentences": 91791,
      "data utilized": 26614,
      "investigate questions": 56806,
      "questions using": 95374,
      "consider multiple": 21217,
      "multiple neural": 75625,
      "collected human": 18423,
      "deep network": 28345,
      "augment existing": 10256,
      "data best": 25698,
      "work showing": 134799,
      "sentence used": 106119,
      "used distinguish": 128497,
      "earlier words": 34380,
      "exploration use": 41505,
      "generate synthetic": 47025,
      "helps improving": 49819,
      "improving subsequent": 53173,
      "translation sequencetosequence": 125245,
      "alignment target": 4991,
      "target source": 117710,
      "metrics quantify": 69999,
      "quantify quality": 94855,
      "produces best": 91986,
      "transformerbased word": 124440,
      "discovery unsegmented": 32151,
      "phoneme sequences": 87281,
      "segmentation target": 104639,
      "aligned french": 4914,
      "transformer task": 124382,
      "results confirmed": 101622,
      "evaluation alignment": 39117,
      "alignment quality": 4985,
      "use average": 127915,
      "normalized entropy": 80356,
      "best word": 13472,
      "pair collection": 83425,
      "analyzing dialogue": 6038,
      "study modeling": 114440,
      "problems specifically": 91375,
      "specifically address": 111522,
      "problem providing": 91190,
      "help guide": 49730,
      "tasks define": 119032,
      "build recent": 14802,
      "recent successes": 96545,
      "modeling experiments": 72430,
      "tasks report": 119458,
      "careful analysis": 15508,
      "various network": 131148,
      "network design": 77222,
      "finetuning techniques": 44524,
      "techniques lowresource": 119925,
      "languages suffer": 60904,
      "suffer lack": 115228,
      "make hard": 66681,
      "hard apply": 49451,
      "apply stateoftheart": 7218,
      "scarcity problem": 103813,
      "benchmark language": 12833,
      "bert ulmfit": 13244,
      "train robust": 123009,
      "robust classifiers": 103025,
      "classifiers lowresource": 17617,
      "finetuning using": 44526,
      "dialogue learning": 30702,
      "sequential order": 107243,
      "meaningful coherent": 67712,
      "order changes": 82292,
      "utterances lead": 130654,
      "supervised signal": 115838,
      "selfsupervised learning": 104941,
      "capture flow": 15304,
      "utterance pair": 130607,
      "propose samplingbased": 93023,
      "selfsupervised network": 104946,
      "network ssn": 77443,
      "relevant dialogue": 98539,
      "learning adversarial": 62313,
      "dialogue scenarios": 30738,
      "appear training": 6804,
      "challenging learn": 16269,
      "learn accurate": 61984,
      "accurate representations": 2358,
      "observations paper": 81180,
      "formulate learning": 45272,
      "regression problem": 97709,
      "problem address": 90921,
      "observations based": 81175,
      "hierarchical attentionbased": 49938,
      "architecture serve": 8746,
      "function context": 46010,
      "word encoded": 133258,
      "modelagnostic metalearning": 72346,
      "methods constructing": 69401,
      "improves downstream": 52973,
      "tasks embeddings": 119086,
      "modernizing historical": 74434,
      "historical documents": 50524,
      "documents user": 33309,
      "documents given": 33236,
      "historical document": 50520,
      "modernization aims": 74431,
      "written modern": 135135,
      "modern version": 74425,
      "version documents": 131606,
      "difficulty making": 31706,
      "broader audience": 14698,
      "approach profits": 7822,
      "tested approach": 120561,
      "conducted user": 20955,
      "improvement multilingual": 52728,
      "representations main": 99745,
      "sequences recurrent": 107134,
      "cnn architectures": 17987,
      "used encoderdecoder": 128509,
      "language downstream": 58986,
      "intermediate layer": 56133,
      "layer representations": 61744,
      "architectures difficult": 8801,
      "webbased tool": 132277,
      "sentence token": 106106,
      "level present": 63491,
      "present use": 89757,
      "gender issues": 46602,
      "embeddings second": 35923,
      "showing multilingual": 108417,
      "sentences tokens": 106522,
      "layers decoder": 61766,
      "publications using": 94284,
      "deep discourse": 28229,
      "discourse model": 32068,
      "makes automatic": 66759,
      "automatic claim": 10500,
      "information overload": 54819,
      "automated claim": 10432,
      "extraction useful": 42529,
      "scientific knowledge": 104010,
      "dataset 1500": 26709,
      "scientific abstracts": 103990,
      "indicating sentence": 53873,
      "scientific claim": 103993,
      "approach finetuning": 7575,
      "finetuning step": 44517,
      "step allows": 113246,
      "dataset final": 26932,
      "model increases": 71332,
      "percent points": 85919,
      "annotation tool": 6392,
      "applications biomedical": 6901,
      "social cultural": 110333,
      "data article": 25621,
      "issues forefront": 57050,
      "hope provide": 50634,
      "set best": 107378,
      "given diversity": 48022,
      "goal help": 48355,
      "help promote": 49768,
      "believe work": 12706,
      "work data": 134452,
      "exhibit different": 39998,
      "corpus mandarin": 23876,
      "approach characterize": 7417,
      "types tone": 126375,
      "extracted linguistic": 42162,
      "current study": 25374,
      "study context": 114341,
      "phonological information": 87308,
      "theory learning": 121820,
      "logs paper": 65057,
      "techniques enable": 119878,
      "intelligent systems": 55857,
      "consists determining": 21474,
      "dialogue given": 30686,
      "given spoken": 48134,
      "positive impact": 88323,
      "impact accuracy": 51855,
      "approaches models": 8238,
      "datasets ii": 27508,
      "dataset small": 27208,
      "higher accuracies": 50158,
      "achieved use": 2719,
      "content utterances": 21973,
      "model morphological": 71531,
      "work considers": 134434,
      "learning encouraging": 62545,
      "vectors encode": 131431,
      "model end": 71074,
      "end learned": 36814,
      "study short": 114515,
      "processing automatic": 91628,
      "automatic questionanswering": 10637,
      "siri alexa": 109872,
      "applications nowadays": 6980,
      "resources provide": 101031,
      "cost plenty": 24370,
      "new short": 79136,
      "subtask called": 114943,
      "called dialogue": 15056,
      "nugget detection": 80830,
      "aim automatically": 4447,
      "dialogues generated": 30835,
      "paper solve": 84434,
      "network proposed": 77398,
      "structure embedding": 113848,
      "dialogue representation": 30730,
      "level context": 63431,
      "context level": 22167,
      "level furthermore": 63452,
      "furthermore apply": 46144,
      "mechanism utterance": 68049,
      "tried bert": 125775,
      "layer sentence": 61746,
      "representation result": 99397,
      "shows bert": 108554,
      "subtasks outperform": 114975,
      "measures proposed": 67890,
      "measures accuracy": 67849,
      "accuracy precision": 2240,
      "recall f1score": 96337,
      "using traditional": 130299,
      "assessing degree": 9509,
      "event mentioned": 39526,
      "words previous": 134138,
      "novel graphbased": 80586,
      "integrate semantic": 55767,
      "information effectively": 54514,
      "effectively experiments": 34811,
      "systems wmt19": 117239,
      "wmt19 robustness": 132859,
      "robustness task": 103135,
      "submission wmt19": 114741,
      "task mainly": 118370,
      "translating noisy": 124600,
      "posts twitter": 88525,
      "presents different": 89840,
      "typical translation": 126407,
      "combined techniques": 18670,
      "synthetic corpus": 116614,
      "mechanism significantly": 68037,
      "results revealed": 102141,
      "emojis emoticons": 36090,
      "placeholder tokens": 87613,
      "accuracy noisy": 2219,
      "noisy texts": 80127,
      "texts sentiment": 121604,
      "debate transcripts": 27889,
      "information concerning": 54432,
      "political social": 88001,
      "result problem": 101392,
      "different perspectives": 31329,
      "perspectives using": 87179,
      "using varying": 130359,
      "methods relatively": 69720,
      "results systematic": 102249,
      "studies address": 114177,
      "address automatic": 3649,
      "researchers work": 100717,
      "work problems": 134705,
      "problems automatic": 91303,
      "conclude summarizing": 20720,
      "summarizing findings": 115627,
      "possible avenues": 88383,
      "avenues research": 11170,
      "multilingual universal": 75396,
      "semantic retrieval": 105243,
      "retrieval introduce": 102399,
      "retrieval focused": 102393,
      "respectively based": 101125,
      "architectures models": 8827,
      "multitask trained": 75894,
      "using translation": 130319,
      "provide performance": 93892,
      "stateoftheart semantic": 112943,
      "bitext retrieval": 14239,
      "retrieval question": 102424,
      "english transfer": 37313,
      "exceed performance": 39902,
      "monolingual english": 74589,
      "crossdomain generalization": 24890,
      "generalization neural": 46784,
      "neural constituency": 77870,
      "constituency parsers": 21526,
      "benchmark treebanks": 12869,
      "generalize domains": 46808,
      "zeroshot setting": 135526,
      "setting training": 107783,
      "corpus evaluating": 23783,
      "evaluating outofdomain": 39084,
      "outofdomain corpora": 82648,
      "neural nonneural": 78619,
      "pretrained encoder": 90030,
      "encoder representations": 36566,
      "domains does": 33766,
      "treebanks finally": 125660,
      "rich input": 102748,
      "representations learn": 99723,
      "parsers benefit": 85016,
      "benefit structured": 12989,
      "structured output": 114021,
      "prediction output": 89097,
      "analyze generalization": 5975,
      "process obtain": 91541,
      "stateoftheart parsing": 112801,
      "web treebanks": 132269,
      "bias natural": 13821,
      "inference popular": 54195,
      "popular natural": 88106,
      "nli datasets": 79540,
      "datasets shown": 27708,
      "hypothesisonly biases": 51286,
      "spurious correlations": 112063,
      "correlations data": 24252,
      "evaluate adversarial": 38784,
      "statistical empirical": 113094,
      "models chinese": 72900,
      "role alleviating": 103155,
      "chinese words": 16842,
      "especially scenario": 38504,
      "fundamental problem": 46105,
      "sequence previous": 107034,
      "real application": 96051,
      "application characteristics": 6837,
      "ngram smoothing": 79453,
      "approach tackle": 7952,
      "models probabilistic": 73803,
      "probabilistic neural": 90824,
      "better real": 13689,
      "approach individual": 7636,
      "approach bring": 7399,
      "automated word": 10481,
      "russian study": 103503,
      "study address": 114305,
      "problem automated": 90945,
      "russian using": 103509,
      "simple bidirectional": 109378,
      "achieve accuracy": 2413,
      "accuracy 90": 2081,
      "experiment training": 40513,
      "based multilingual": 11864,
      "embeddings automatically": 35562,
      "including dialects": 53284,
      "consider possible": 21221,
      "aligned english": 4912,
      "corpus parallel": 23925,
      "mined data": 70150,
      "languages pairs": 60774,
      "achieving strong": 2988,
      "scores language": 104182,
      "languages need": 60760,
      "2019 largescale": 702,
      "largescale documentlevel": 61428,
      "documentlevel neural": 33153,
      "describes microsoft": 29415,
      "submissions wmt19": 114759,
      "wmt19 news": 132854,
      "deep transformer": 28428,
      "models start": 74091,
      "start strong": 112401,
      "backtranslation backtranslation": 11332,
      "explore finetuning": 41547,
      "deeper models": 28448,
      "ensembling strategies": 37632,
      "synthetic parallel": 116636,
      "train transformer": 123045,
      "transformer translation": 124388,
      "language monolingual": 59691,
      "sentencelevel documentlevel": 106156,
      "documentlevel systems": 33166,
      "based preliminary": 11941,
      "preliminary human": 89273,
      "systems score": 117126,
      "score higher": 104079,
      "higher human": 50182,
      "naver labs": 76678,
      "labs europes": 58673,
      "europes systems": 38776,
      "robustness noise": 103119,
      "noise social": 80066,
      "informal language": 54338,
      "spelling mistakes": 111908,
      "orthographic variations": 82599,
      "extracted social": 42182,
      "japaneseenglish translation": 57213,
      "sets source": 107713,
      "according automatic": 1966,
      "bleu human": 14285,
      "proposed single": 93551,
      "single ensemble": 109732,
      "choices present": 16920,
      "adaptation language": 3226,
      "comparison network": 19558,
      "network topology": 77456,
      "modeling relations": 72528,
      "relations languages": 98197,
      "language characteristics": 58873,
      "languages automated": 60415,
      "large textual": 61294,
      "novel statistical": 80735,
      "statistical studies": 113169,
      "improving crosslingual": 53080,
      "crosslingual natural": 24982,
      "techniques work": 120016,
      "weighted network": 132355,
      "structure used": 113979,
      "used crosslingual": 128466,
      "different network": 31285,
      "metrics empirically": 69954,
      "hundreds thousands": 51153,
      "languages seen": 60858,
      "seen novel": 104536,
      "linguistic studies": 64559,
      "corpus speech": 24012,
      "radio broadcasts": 95447,
      "march 2019": 67176,
      "corpus intended": 23843,
      "intended use": 55872,
      "researchers fields": 100692,
      "fields natural": 44002,
      "corpus encompasses": 23769,
      "automatically transcribed": 10829,
      "geographical location": 47855,
      "speaker turn": 111300,
      "descriptive statistics": 29505,
      "representation mapping": 99323,
      "relation detection": 97967,
      "detection knowledge": 29978,
      "answering relation": 6690,
      "answering previous": 6662,
      "problem current": 90990,
      "questions relations": 95354,
      "data unseen": 26595,
      "unseen relations": 127537,
      "performance drop": 86324,
      "main reason": 66455,
      "representations unseen": 99955,
      "mapping method": 67138,
      "adapter learn": 3296,
      "relations based": 98113,
      "based previously": 11951,
      "previously learned": 90605,
      "learned relation": 62248,
      "employ adversarial": 36250,
      "reconstruction objective": 97122,
      "simplequestion dataset": 109550,
      "relations experiments": 98165,
      "method greatly": 68866,
      "performance seen": 86694,
      "stateoftheart code": 112610,
      "lstms learn": 65712,
      "question question": 95208,
      "present linguistic": 89536,
      "neural parser": 78626,
      "parser adding": 84920,
      "agreement information": 4394,
      "main verbs": 66476,
      "motivated theoretical": 74863,
      "particular work": 85466,
      "capture similar": 15395,
      "typologically different": 126493,
      "languages parser": 60788,
      "parser learns": 84970,
      "learns different": 63203,
      "bilstms used": 14113,
      "used architecture": 128395,
      "used explanations": 128536,
      "explanations case": 41291,
      "learned network": 62234,
      "dependency representations": 29225,
      "benefits using": 13026,
      "best way": 13469,
      "way integrate": 132097,
      "predicting human": 88988,
      "future paper": 46288,
      "explore task": 41593,
      "range everyday": 95566,
      "everyday activities": 39626,
      "embedding framework": 35410,
      "perform automatic": 85949,
      "given user": 48167,
      "text previous": 121201,
      "additionally explore": 3607,
      "inferred user": 54264,
      "framework natural": 45623,
      "suggest models": 115319,
      "current benchmarks": 25264,
      "phenomena make": 87230,
      "make case": 66629,
      "evaluation natural": 39304,
      "supplementary training": 115945,
      "data multitask": 26147,
      "using evaluation": 129644,
      "used pretraining": 128695,
      "pretraining task": 90332,
      "task does": 118114,
      "universal representations": 127328,
      "learning incremental": 62650,
      "transformer deliberation": 124320,
      "conversations task": 23073,
      "responses chatting": 101257,
      "given document": 48023,
      "document knowledge": 33028,
      "role document": 103171,
      "novel transformerbased": 80756,
      "transformerbased architecture": 124401,
      "architecture multiturn": 8700,
      "particular devise": 85404,
      "knowledge related": 58138,
      "related documents": 97855,
      "documents motivated": 33255,
      "motivated human": 74852,
      "human cognitive": 50777,
      "cognitive process": 18275,
      "process design": 91457,
      "improve context": 52358,
      "generated model": 47093,
      "baselines context": 12376,
      "shorttext conversation": 108353,
      "important linguistic": 52181,
      "users purpose": 129160,
      "specific sentence": 111490,
      "collect new": 18387,
      "function new": 46036,
      "functions including": 46076,
      "including information": 53302,
      "retrievalbased neural": 102446,
      "present contribution": 89420,
      "task quality": 118592,
      "sentence documentlevel": 105828,
      "pairs englishgerman": 83531,
      "englishgerman englishrussian": 37379,
      "linear neural": 64358,
      "using bert": 129388,
      "bert xlm": 13258,
      "xlm pretrained": 135197,
      "new ensemble": 78901,
      "simple technique": 109528,
      "considerable margin": 21253,
      "word polysemy": 133410,
      "framework distributional": 45500,
      "contexts large": 22406,
      "traditionally considered": 122887,
      "highly flexible": 50323,
      "usage patterns": 127868,
      "embeddings recent": 35900,
      "recent generation": 96464,
      "methods offer": 69650,
      "motivated word": 74864,
      "human generated": 50862,
      "shown correlate": 108457,
      "correlate poorly": 24197,
      "poorly human": 88068,
      "collect human": 18380,
      "experiments use": 41190,
      "new technologies": 79212,
      "drastically change": 34100,
      "techniques research": 119974,
      "interactive systems": 56033,
      "systems help": 116930,
      "studies aim": 114179,
      "studies limited": 114248,
      "respect number": 101093,
      "process fact": 91484,
      "project help": 92220,
      "collected corpus": 18411,
      "answering set": 6698,
      "set questions": 107552,
      "questions propose": 95344,
      "aims predicting": 4558,
      "containing salient": 21805,
      "information modeled": 54771,
      "f1scores previous": 42723,
      "audio video": 10242,
      "multimodal fusion": 75430,
      "captured attention": 15425,
      "mechanisms suggest": 68070,
      "model potentially": 71753,
      "potentially used": 88620,
      "finding key": 44277,
      "languages usually": 60958,
      "usually small": 130453,
      "large distantly": 61080,
      "finetune pretrained": 44411,
      "languages lowresources": 60709,
      "lowresources languages": 65577,
      "languages improve": 60627,
      "scenarios empirical": 103843,
      "finetuning pretrained": 44492,
      "small gold": 110154,
      "gold corpus": 48432,
      "compare supervised": 19300,
      "transfer useful": 124206,
      "transfer using": 124207,
      "transfer monolingual": 124155,
      "monolingual lm": 74595,
      "dataset exploiting": 26916,
      "exploiting characterlevel": 41459,
      "characterlevel input": 16567,
      "input bidirectional": 55303,
      "source multilingual": 110796,
      "representations provides": 99835,
      "gender case": 46593,
      "languages support": 60908,
      "unsupervised neural": 127680,
      "text readability": 121228,
      "set novel": 107512,
      "approaches determining": 8124,
      "unsupervised setting": 127721,
      "leverage neural": 63608,
      "supervised setting": 115836,
      "classification architectures": 17123,
      "tested proposed": 120583,
      "neural unsupervised": 78727,
      "robust transferable": 103079,
      "languages allows": 60399,
      "architectures number": 8835,
      "benchmark new": 12845,
      "study offers": 114458,
      "approaches readability": 8304,
      "rely extensive": 98693,
      "tend degenerate": 120146,
      "distributed narrow": 32615,
      "narrow cone": 76188,
      "largely limits": 61342,
      "representation power": 99379,
      "power word": 88658,
      "embeddings analyze": 35553,
      "analyze conditions": 5957,
      "causes problem": 15840,
      "novel regularization": 80703,
      "problem achieve": 90913,
      "baseline algorithms": 12185,
      "network input": 77285,
      "layer achieves": 61698,
      "achieves higher": 2800,
      "accuracy dialogue": 2138,
      "conversational artificial": 22994,
      "goal dialogue": 48349,
      "conversational partner": 23020,
      "recent dialogue": 96457,
      "dialogue managers": 30708,
      "bagofwords input": 11375,
      "experiments dialogue": 40904,
      "conversational dataset": 23000,
      "improves models": 52996,
      "continual pretraining": 22600,
      "pretraining framework": 90261,
      "recently pretrained": 96730,
      "tasks indicates": 119198,
      "pretraining largescale": 90274,
      "processing current": 91652,
      "current pretraining": 25325,
      "usually focus": 130421,
      "focus training": 44832,
      "simple tasks": 109527,
      "cooccurrence words": 23240,
      "relations order": 98227,
      "pretraining tasks": 90335,
      "20 outperforms": 517,
      "outperforms bert": 82854,
      "bert xlnet": 13261,
      "english tasks": 37304,
      "tasks glue": 119147,
      "glue benchmarks": 48313,
      "codes pretrained": 18210,
      "cuni systems": 25239,
      "systems unsupervised": 117217,
      "2019 paper": 705,
      "used unsupervised": 128834,
      "news shared": 79362,
      "translation wmt19": 125428,
      "follow strategy": 44943,
      "al 2018b": 4639,
      "crosslingual embedding": 24947,
      "embedding mappings": 35436,
      "trained monolingual": 123204,
      "data followed": 25953,
      "data synthetic": 26531,
      "monolingual corpus": 74561,
      "iterative backtranslation": 57126,
      "entities vocabulary": 37895,
      "embedding mapping": 35435,
      "wmt19 shared": 132861,
      "reasoning word": 96325,
      "armed conflict": 9014,
      "texts extend": 121517,
      "armed conflicts": 9015,
      "attempting predict": 9760,
      "predict new": 88906,
      "new relations": 79102,
      "past events": 85641,
      "source semantic": 110809,
      "diachronic word": 30492,
      "texts simple": 121611,
      "technique improve": 119790,
      "task demonstrated": 118064,
      "demonstrated using": 28935,
      "decrease number": 28170,
      "evaluation historical": 39231,
      "evaluation bert": 39136,
      "metric using": 69910,
      "bert bidirectional": 13078,
      "bidirectional encoder": 13907,
      "representations transformers": 99944,
      "dataset metric": 27025,
      "documentlevel transformer": 33168,
      "transformer nmt": 124366,
      "implemented tensor2tensor": 51977,
      "aimed improving": 4485,
      "context source": 22273,
      "target instead": 117626,
      "instead translating": 55689,
      "sentence independently": 105905,
      "possibly overlapping": 88449,
      "isolated sentences": 56974,
      "potential effect": 88549,
      "lexical coherence": 63740,
      "analysis revealed": 5816,
      "draw conclusions": 34112,
      "aspectbased opinion": 9353,
      "proposed summarize": 93559,
      "content online": 21915,
      "online customer": 81757,
      "current approach": 25258,
      "approach opinion": 7758,
      "sentiment scores": 106793,
      "mining framework": 70233,
      "customer review": 25441,
      "topic word": 122585,
      "advantages existing": 4079,
      "domain allow": 33468,
      "words set": 134208,
      "reviews restaurants": 102628,
      "empirical findings": 36169,
      "analysis customer": 5554,
      "networks conversational": 77551,
      "conversational machine": 23012,
      "comprehension conversational": 20169,
      "comprehension mc": 20196,
      "proven significantly": 93730,
      "significantly challenging": 108897,
      "challenging compared": 16233,
      "better utilization": 13764,
      "history existing": 50554,
      "approaches effectively": 8137,
      "effectively capture": 34793,
      "handling questions": 49419,
      "coreference ellipsis": 23345,
      "passage text": 85611,
      "simply treat": 109627,
      "effective graph": 34685,
      "question conversation": 95141,
      "aware context": 11259,
      "context graph": 22119,
      "conversation turn": 22981,
      "novel recurrent": 80702,
      "flow mechanism": 44684,
      "mechanism model": 68012,
      "model temporal": 72151,
      "temporal dependencies": 120100,
      "sequence context": 106926,
      "coqa quac": 23304,
      "benchmarks addition": 12883,
      "offer good": 81563,
      "good interpretability": 48484,
      "reasoning process": 96295,
      "networks effective": 77577,
      "tool used": 122405,
      "text deep": 120862,
      "gives high": 48195,
      "semantic encoding": 105042,
      "encoding allows": 36681,
      "visualization tool": 131830,
      "interactions using": 56006,
      "using speech": 130206,
      "annotation frameworks": 6316,
      "automated recognition": 10466,
      "acoustic lexical": 3011,
      "speech patterns": 111721,
      "factors gender": 42889,
      "behavior classification": 12648,
      "results indicates": 101861,
      "techniques incorporating": 119910,
      "incorporating contextual": 53523,
      "similar language": 109103,
      "task statistical": 118742,
      "statistical vs": 113175,
      "problem similar": 91231,
      "translation area": 124648,
      "research years": 100668,
      "performance popular": 86607,
      "popular approaches": 88076,
      "yield similar": 135355,
      "performance varies": 86836,
      "pair statistical": 83456,
      "difference bleu": 30940,
      "pair proposed": 83449,
      "case language": 15590,
      "additionally report": 3632,
      "results combination": 101585,
      "1st place": 485,
      "2nd place": 889,
      "task semisupervised": 118683,
      "unit used": 127221,
      "used extracting": 128545,
      "extracting valuable": 42244,
      "valuable information": 130763,
      "information longer": 54739,
      "explicit sentence": 41348,
      "embedding local": 35434,
      "local representation": 64934,
      "representation capture": 99185,
      "focus keywords": 44776,
      "representation obtained": 99360,
      "scarcity labeled": 103806,
      "investigate adapt": 56712,
      "allowing utilize": 5123,
      "utilize unlabeled": 130534,
      "segmentation experiments": 104588,
      "74 105": 1235,
      "respectively applied": 101123,
      "achieving relative": 2972,
      "studies revealed": 114275,
      "structure logical": 113902,
      "logical structure": 65021,
      "amr abstract": 5381,
      "formalism representing": 45186,
      "correct prediction": 24115,
      "attractive core": 10170,
      "resulting framework": 101444,
      "framework similar": 45688,
      "especially textual": 38518,
      "documents called": 33197,
      "technological advances": 120019,
      "better use": 13760,
      "information recent": 54900,
      "recent techniques": 96548,
      "techniques natural": 119935,
      "nlp deep": 79599,
      "efficiently process": 35154,
      "large volume": 61325,
      "obtain relevant": 81313,
      "classify text": 17662,
      "text applications": 120652,
      "applications context": 6908,
      "represents challenge": 100071,
      "understanding search": 126950,
      "search suitable": 104354,
      "representations specific": 99903,
      "models requires": 73938,
      "representative corpora": 100007,
      "access material": 1918,
      "review main": 102556,
      "techniques deep": 119862,
      "protolanguage reconstruction": 93680,
      "reconstruction historical": 97119,
      "historical linguists": 50530,
      "sound change": 110701,
      "comparative method": 19208,
      "method utilizes": 69217,
      "languages process": 60815,
      "provide novel": 93882,
      "task far": 118203,
      "embeddings reveals": 35920,
      "reveals models": 102520,
      "learning crossdomain": 62470,
      "nlpcc 2019": 79797,
      "semisupervised domain": 105600,
      "representation vectors": 99462,
      "network source": 77441,
      "domains product": 33833,
      "results target": 102251,
      "evaluation documentlevel": 39191,
      "documentlevel nmt": 33158,
      "increasingly difficult": 53696,
      "output translation": 83142,
      "aimed assessing": 4479,
      "phenomena mt": 87232,
      "task manually": 118375,
      "documentlevel translation": 33170,
      "morphosyntactic analysis": 74814,
      "analysis introduce": 5661,
      "treebanks evaluate": 125659,
      "morphosyntactic tasks": 74819,
      "tasks pos": 119384,
      "task shared": 118706,
      "shared multitask": 107993,
      "pos morphological": 88210,
      "finally investigate": 44202,
      "results analyses": 101506,
      "syntactic abstraction": 116361,
      "abstraction english": 1789,
      "ud treebanks": 126535,
      "annotations using": 6479,
      "capture patterns": 15360,
      "elements representing": 35282,
      "eliminating need": 35301,
      "recognition present": 96967,
      "present corpus": 89422,
      "corpus finnish": 23805,
      "articles manually": 9197,
      "extracted archives": 42139,
      "news source": 79368,
      "purposes present": 94449,
      "present baseline": 89390,
      "experiments corpus": 40858,
      "role nlp": 103210,
      "especially recurrent": 38501,
      "recently increasing": 96694,
      "representations offered": 99789,
      "modules used": 74531,
      "used explain": 128535,
      "reasoning models": 96273,
      "models prediction": 73766,
      "recent paper": 96486,
      "paper claims": 83761,
      "2019 challenge": 694,
      "needs account": 76885,
      "experimental design": 40530,
      "multiple random": 75653,
      "random seed": 95509,
      "endtoend adversarial": 36868,
      "meaningful interpretation": 67721,
      "models reliable": 73917,
      "work does": 134482,
      "establishing strong": 38622,
      "baselines new": 12434,
      "bert paper": 13197,
      "contextualized embedding": 22543,
      "known bert": 58287,
      "bert task": 13238,
      "approach enhance": 7530,
      "enhance model": 37475,
      "model efficiency": 71048,
      "using token": 130294,
      "12 datasets": 251,
      "experiments bert": 40806,
      "bestperforming models": 13477,
      "models 25": 72636,
      "bert embeddings": 13111,
      "helps understanding": 49833,
      "rich representation": 102776,
      "codes available": 18198,
      "researchers improve": 100694,
      "comprehension machine": 20190,
      "attend words": 9781,
      "using syntax": 130262,
      "guide text": 49247,
      "text modeling": 121127,
      "incorporating explicit": 53536,
      "explicit syntactic": 41353,
      "syntactic constraints": 116383,
      "mechanism better": 67953,
      "representations selfattention": 99870,
      "transformerbased encoder": 124408,
      "encoder introduce": 36535,
      "original transformer": 82559,
      "architecture better": 8629,
      "popular benchmarks": 88082,
      "helps achieve": 49807,
      "substantial performance": 114872,
      "improvement strong": 52770,
      "baselines evaluation": 12391,
      "trained backtranslation": 123075,
      "augmentation technique": 10310,
      "technique leverages": 119799,
      "improvements test": 52931,
      "test examples": 120454,
      "examples source": 39880,
      "backtranslated training": 11326,
      "occurring text": 81519,
      "evidence support": 39672,
      "preferred humans": 89250,
      "capture human": 15311,
      "human preferences": 50934,
      "score measure": 104091,
      "lexical grammatical": 63767,
      "subordinate clause": 114795,
      "collect largescale": 18385,
      "develop model": 30213,
      "induce semantic": 54003,
      "semantic types": 105340,
      "subordinate clauses": 114796,
      "relationship types": 98312,
      "detection existing": 29946,
      "heavily rely": 49656,
      "rely humanannotated": 98710,
      "humanannotated data": 50996,
      "data expensive": 25918,
      "expensive obtain": 40426,
      "tackle training": 117312,
      "data bottleneck": 25709,
      "investigate methods": 56780,
      "manual labeling": 67006,
      "construct largescale": 21636,
      "deleting words": 28595,
      "propose selfsupervised": 93030,
      "selfsupervised pretraining": 104949,
      "tasks tagging": 119545,
      "noisy words": 80133,
      "ii sentence": 51715,
      "classification distinguish": 17188,
      "tasks jointly": 119231,
      "train network": 122976,
      "network pretrained": 77394,
      "finetuned using": 44438,
      "data experimental": 25921,
      "results commonly": 101588,
      "using 1000": 129306,
      "1000 sentences": 176,
      "dataset significantly": 27203,
      "methods reducing": 69713,
      "reducing error": 97414,
      "effective noisy": 34727,
      "channel modeling": 16407,
      "neural noisy": 78616,
      "models incrementally": 73393,
      "based partial": 11923,
      "source available": 110713,
      "entire source": 37716,
      "source models": 110792,
      "perform remarkably": 86056,
      "trained billions": 123083,
      "billions words": 14085,
      "direct model": 31793,
      "model 32": 70501,
      "bleu wmt17": 14346,
      "wmt17 germanenglish": 132840,
      "righttoleft reranking": 102844,
      "models ensembles": 73160,
      "direct models": 31794,
      "commonsense reasoning": 19005,
      "abductive reasoning": 1473,
      "reasoning inference": 96256,
      "example jenny": 39786,
      "little research": 64827,
      "research support": 100633,
      "abductive natural": 1470,
      "inference generation": 54146,
      "study investigates": 114415,
      "introduce challenge": 56389,
      "explanations based": 41290,
      "based dataset": 11622,
      "multiplechoice question": 75755,
      "task choosing": 117962,
      "conditional generation": 20755,
      "task explaining": 118180,
      "explaining given": 41274,
      "best language": 13359,
      "analysis leads": 5673,
      "leads new": 61947,
      "types reasoning": 126343,
      "reasoning deep": 96244,
      "models continue": 72981,
      "making automated": 66823,
      "issue model": 57007,
      "training text": 123919,
      "performance dataset": 86272,
      "performs different": 87006,
      "performance independently": 86460,
      "independently different": 53787,
      "different demographic": 31089,
      "measuring bias": 67907,
      "bias comparing": 13797,
      "using metrics": 129881,
      "bias dataset": 13799,
      "language named": 59699,
      "like english": 64039,
      "using latest": 129801,
      "stateoftheart architecture": 112586,
      "50 compared": 1072,
      "model 10": 70494,
      "10 improvement": 134,
      "data key": 26059,
      "systems realworld": 117098,
      "expected work": 40404,
      "technique achieve": 119761,
      "setup using": 107863,
      "dataset evaluate": 26895,
      "stanford multidomain": 112368,
      "scheduling weather": 103890,
      "weather information": 132216,
      "fewshot approach": 43897,
      "consistently outperforming": 21428,
      "model terms": 72153,
      "bleu entity": 14280,
      "dataefficient requiring": 26685,
      "incorporating word": 53572,
      "using language": 129780,
      "model rescoring": 71915,
      "translation track": 125384,
      "translation pbsmt": 125105,
      "pbsmt model": 85823,
      "wordlevel neural": 133746,
      "subwordlevel nmt": 115051,
      "propose solve": 93076,
      "morphological richness": 74729,
      "aligned using": 4928,
      "conneau et": 21113,
      "fluency consistency": 44696,
      "translation candidates": 124683,
      "generated beam": 47057,
      "preprocessing postprocessing": 89332,
      "quality final": 94668,
      "final translations": 44136,
      "translations language": 125472,
      "language graph": 59110,
      "language challenging": 58870,
      "challenging lack": 16264,
      "pairs previous": 83609,
      "solve lowresource": 110600,
      "transfer multilingual": 124157,
      "propose concept": 92592,
      "graph design": 48792,
      "novel graph": 80585,
      "boosts accuracy": 14448,
      "accuracy lowresource": 2204,
      "forward backward": 45323,
      "experiments ted": 41172,
      "talks multilingual": 117564,
      "multilingual dataset": 75235,
      "specifically improve": 111558,
      "message passing": 68497,
      "networks document": 77574,
      "document understanding": 33101,
      "emerged effective": 36050,
      "effective framework": 34680,
      "framework processing": 45657,
      "framework nlp": 45629,
      "network document": 77229,
      "conducted 10": 20910,
      "architectures competitive": 8794,
      "studies reveal": 114274,
      "performance code": 86211,
      "fast endtoend": 43136,
      "performance runtime": 86690,
      "method estimating": 68812,
      "opening door": 81995,
      "validity proposed": 130756,
      "approach finegrained": 7573,
      "attention general": 9845,
      "general task": 46718,
      "textual sentiment": 121735,
      "stateoftheart relation": 112886,
      "model surprisingly": 72127,
      "reasonable performance": 96210,
      "improved model": 52616,
      "performance robust": 86687,
      "defined data": 28498,
      "rigorous analysis": 102849,
      "metrics results": 70004,
      "attention does": 9823,
      "faithful explanations": 43002,
      "trained attention": 123071,
      "attention task": 10021,
      "large pretrained": 61215,
      "approach discriminative": 7498,
      "adapt pretrained": 3190,
      "model arbitrary": 70681,
      "similar techniques": 109157,
      "setting negative": 107761,
      "adapting pretrained": 3322,
      "pretrained transformer": 90194,
      "models sensitive": 73995,
      "parameter changes": 84711,
      "attention approach": 9793,
      "attention experiments": 9834,
      "shown complementary": 108454,
      "exploit rich": 41439,
      "features suffer": 43746,
      "benefit global": 12971,
      "networks continuous": 77549,
      "representations basic": 99515,
      "embeddings allow": 35552,
      "global sentence": 48270,
      "representations benefit": 99519,
      "equivalent terms": 38252,
      "error profile": 38314,
      "representations help": 99673,
      "search errors": 104313,
      "exploit inherent": 41420,
      "13 languages": 284,
      "evaluating contextualized": 39044,
      "54 languages": 1105,
      "lemmatization dependency": 63337,
      "dependencies 23": 29062,
      "flair elmo": 44632,
      "embedding inputs": 35418,
      "baseline udpipe": 12328,
      "udpipe 20": 126538,
      "20 bestperforming": 489,
      "bestperforming systems": 13478,
      "systems conll": 116802,
      "task overall": 118495,
      "overall winner": 83269,
      "2018 present": 669,
      "comparison contextualized": 19534,
      "methods comparison": 69384,
      "embeddings endtoend": 35690,
      "characterlevel word": 16588,
      "embeddings report": 35908,
      "report stateoftheart": 99047,
      "compared results": 19436,
      "ud 22": 126528,
      "growing social": 49184,
      "processing computational": 91644,
      "computational argumentation": 20361,
      "argumentation natural": 8967,
      "prediction addition": 89026,
      "addition analyze": 3399,
      "concepts labeled": 20624,
      "pretrained representation": 90177,
      "data demonstrated": 25823,
      "demonstrated helpful": 28916,
      "helpful improving": 49797,
      "nmt current": 79850,
      "knowledge generating": 57950,
      "data extracting": 25939,
      "sentencelevel contextual": 106150,
      "knowledge complex": 57834,
      "complex diverse": 19808,
      "playing important": 87717,
      "leverage monolingual": 63604,
      "contextual representations": 22492,
      "representations design": 99590,
      "design framework": 29541,
      "sentencelevel representations": 106170,
      "representations nmt": 99783,
      "quality experimental": 94660,
      "chineseenglish germanenglish": 16853,
      "germanenglish machine": 47934,
      "transformer baselines": 124307,
      "baselines experiments": 12394,
      "approach lowresource": 7696,
      "largest dataset": 61531,
      "analysis recognizing": 5798,
      "structured machinereadable": 114012,
      "applications deep": 6912,
      "datasets currently": 27389,
      "currently publicly": 25414,
      "learning base": 62382,
      "articles publicly": 9209,
      "available pubmed": 11087,
      "typical document": 126397,
      "annotated experiments": 6194,
      "accurately recognize": 2390,
      "different document": 31102,
      "document domain": 32995,
      "advanced models": 3949,
      "models document": 73101,
      "sequencetosequence pretraining": 107195,
      "pretraining text": 90340,
      "new sequencetosequence": 79131,
      "seq2seq pretraining": 106907,
      "pretraining method": 90283,
      "tasks unlike": 119584,
      "openai gpt": 81950,
      "pretraining approaches": 90234,
      "subsequent finetuning": 114812,
      "finetuning stage": 44516,
      "stage design": 112144,
      "pointergenerator networks": 87833,
      "backbone architecture": 11286,
      "experiments text": 41177,
      "tasks abstractive": 118912,
      "using taskspecific": 130269,
      "taskspecific techniques": 119641,
      "techniques significantly": 119985,
      "significantly speed": 109052,
      "translation sequence": 125241,
      "like neural": 64076,
      "crossentropy loss": 24913,
      "loss leads": 65281,
      "leads models": 61946,
      "local optima": 64932,
      "scenarios experiments": 103848,
      "2014 englishgerman": 576,
      "englishgerman englishfrench": 37376,
      "loss general": 65277,
      "improvement comparison": 52696,
      "comparison strong": 19579,
      "explicit dialogue": 41315,
      "act labels": 3065,
      "represent utterances": 99151,
      "utterances explicitly": 130639,
      "models strongly": 74104,
      "strongly relies": 113750,
      "quality input": 94691,
      "input dialogue": 55319,
      "approach dialogue": 7491,
      "act prediction": 3066,
      "prediction auxiliary": 89033,
      "utterance representations": 130613,
      "alleviates need": 5060,
      "accuracy points": 2235,
      "strong competitors": 113660,
      "corpus performs": 23929,
      "performs par": 87013,
      "forms word": 45266,
      "word label": 133329,
      "propose perform": 92988,
      "forms novel": 45255,
      "inflections word": 54299,
      "languages approach": 60407,
      "approach promising": 7824,
      "labeling neural": 58517,
      "does lead": 33363,
      "transition model": 124492,
      "model crf": 70940,
      "neural encoding": 77909,
      "better representing": 13699,
      "label sequences": 58412,
      "explicitly leverages": 41376,
      "embeddings captures": 35586,
      "label dependency": 58390,
      "incrementally refined": 53748,
      "label distributions": 58392,
      "attention results": 9997,
      "tagging ner": 117428,
      "improves overall": 53002,
      "accuracy similar": 2276,
      "models conversational": 72987,
      "tasks general": 119141,
      "provide responses": 93909,
      "responses relevant": 101294,
      "work conduct": 134426,
      "conduct indepth": 20880,
      "indepth survey": 53802,
      "survey recent": 116185,
      "recent literature": 96469,
      "general conversation": 46644,
      "approaches different": 8127,
      "affects quality": 4249,
      "information conversation": 54452,
      "propose ideas": 92715,
      "domain currently": 33497,
      "currently stateoftheart": 25417,
      "corpus secondly": 23983,
      "secondly augment": 104474,
      "architectures specifically": 8848,
      "feed additional": 43815,
      "model like": 71460,
      "finally conduct": 44158,
      "working natural": 134903,
      "languages field": 60584,
      "field linguists": 43963,
      "recognition especially": 96872,
      "text parsing": 121169,
      "languages little": 60702,
      "little annotation": 64790,
      "entity recognizers": 38087,
      "ner rely": 77073,
      "rely availability": 98678,
      "amounts labeled": 5342,
      "data making": 26115,
      "making challenging": 66829,
      "extend new": 41790,
      "lowerresourced languages": 65461,
      "approaches involving": 8200,
      "involving crosslingual": 56912,
      "learns highly": 63213,
      "resourced languages": 100893,
      "languages active": 60386,
      "question given": 95165,
      "experimentation using": 40743,
      "real human": 96067,
      "approach best": 7390,
      "transferred model": 124230,
      "model performing": 71734,
      "effort results": 35184,
      "little data": 64800,
      "annotation strategy": 6377,
      "modeling predict": 72512,
      "correctly answer": 24171,
      "questions according": 95268,
      "work lowresource": 134621,
      "gives opportunity": 48201,
      "problem inspired": 91085,
      "modeling method": 72473,
      "datasets multitask": 27585,
      "improving prediction": 53142,
      "baselines lowresource": 12425,
      "dataselection algorithms": 26701,
      "finetuning neural": 44482,
      "specifically trained": 111602,
      "trained particular": 123237,
      "particular characteristics": 85390,
      "finetuning technique": 44523,
      "adapting nmt": 3321,
      "use technique": 128320,
      "technique adapt": 119763,
      "set particular": 107524,
      "using transductive": 130306,
      "transductive data": 124027,
      "selection algorithms": 104764,
      "retrieve sentences": 102454,
      "time test": 122122,
      "set provided": 107549,
      "adapted small": 3292,
      "text attributes": 120668,
      "product information": 92038,
      "performance sentiment": 86701,
      "facto standard": 42848,
      "standard method": 112259,
      "incorporate additional": 53449,
      "mechanism performance": 68025,
      "extending model": 41817,
      "classifier instead": 17551,
      "instead focus": 55664,
      "incorporated model": 53497,
      "weight matrices": 132326,
      "attributes experiments": 10196,
      "approach attention": 7368,
      "model finally": 71182,
      "finally representations": 44231,
      "representations transfer": 99942,
      "model implementation": 71299,
      "claim generation": 17011,
      "augmented inventing": 10326,
      "inventing inventors": 56686,
      "techniques envision": 119881,
      "autocomplete function": 10395,
      "inventors conceive": 56692,
      "conceive better": 20549,
      "better inventions": 13612,
      "intelligence order": 55834,
      "order generate": 82328,
      "generate patent": 46986,
      "fundamental question": 46111,
      "question measure": 95182,
      "claim language": 17012,
      "language rarely": 60021,
      "explored nlp": 41629,
      "contains rich": 21836,
      "explicit implicit": 41324,
      "annotations work": 6483,
      "propose spanbased": 93078,
      "spanbased approach": 111141,
      "order study": 82412,
      "study effectiveness": 114363,
      "spans generated": 111202,
      "generated patent": 47108,
      "concept natural": 20580,
      "classifier implemented": 17549,
      "pretrained google": 90041,
      "gpt2 model": 48564,
      "model way": 72311,
      "stateoftheart pretrained": 112868,
      "quantitative metric": 94873,
      "generation particularly": 47538,
      "roberta liu": 102999,
      "sentences fed": 106315,
      "similar pair": 109116,
      "10000 sentences": 180,
      "sentences requires": 106479,
      "50 million": 1076,
      "bert makes": 13160,
      "similarity search": 109305,
      "unsupervised tasks": 127733,
      "pretrained bert": 89983,
      "bert network": 13194,
      "semantically meaningful": 105370,
      "bert roberta": 13220,
      "maintaining accuracy": 66507,
      "accuracy bert": 2107,
      "tasks transfer": 119572,
      "tasks outperforms": 119357,
      "lstmcrf model": 65692,
      "recognition based": 96825,
      "based lstmcrf": 11819,
      "architecture combines": 8639,
      "combines word": 18703,
      "embeddings bilstm": 35579,
      "learning raw": 62939,
      "raw word": 95867,
      "input using": 55466,
      "information contributes": 54451,
      "information named": 54783,
      "recognition needs": 96921,
      "simultaneously using": 109698,
      "different granularity": 31170,
      "sizable improvements": 109908,
      "general representations": 46706,
      "text fundamental": 120969,
      "researchers proposed": 100707,
      "proposed use": 93582,
      "pretraining multitask": 90293,
      "robust representations": 103066,
      "representations methods": 99754,
      "optimizationbased metalearning": 82220,
      "algorithms paper": 4872,
      "metalearning algorithm": 68544,
      "algorithm maml": 4770,
      "tasks validate": 119593,
      "tasks efficiently": 119083,
      "crosslingual features": 24957,
      "zeroshot crosslingual": 135493,
      "availability corpora": 10902,
      "corpora train": 23603,
      "advances field": 3989,
      "explore zeroshot": 41618,
      "parsing train": 85266,
      "german dutch": 47888,
      "semantic graphbank": 105066,
      "used conjunction": 128447,
      "structure directly": 113836,
      "directly encoding": 31874,
      "encoding input": 36701,
      "context gates": 22113,
      "transformer machine": 124337,
      "translation context": 124708,
      "effective control": 34653,
      "target contexts": 117593,
      "nmt challenging": 79846,
      "rnn paper": 102928,
      "identify source": 51556,
      "contexts introduce": 22404,
      "introduce gate": 56426,
      "mechanism control": 67961,
      "transformer addition": 124290,
      "bias problem": 13836,
      "mechanism paper": 68024,
      "method guide": 68868,
      "supervision automatically": 115868,
      "using pointwise": 130024,
      "information extensive": 54554,
      "score strong": 104128,
      "baseline attention": 12190,
      "core module": 23328,
      "module ner": 74499,
      "use bilstm": 127926,
      "resources gazetteers": 100983,
      "improve ner": 52433,
      "ner paper": 77062,
      "paper instead": 83984,
      "focuses analyzing": 44886,
      "bring improvements": 14629,
      "remedy problem": 98861,
      "wnut 2017": 132880,
      "clear consistent": 17721,
      "foundation future": 45344,
      "shallow syntax": 107932,
      "computationally cheap": 20448,
      "syntaxaware representations": 116580,
      "contextual embeddings": 22458,
      "method involves": 68922,
      "features obtained": 43636,
      "obtained automatically": 81346,
      "baselines analysis": 12353,
      "embeddings transfer": 35988,
      "transfer linguistic": 124146,
      "linguistic tasks": 64564,
      "findings evidence": 44293,
      "representations make": 99746,
      "make additional": 66620,
      "interactive annotation": 56011,
      "feedback generation": 43829,
      "reasoning proposed": 96300,
      "labeling training": 58553,
      "process given": 91496,
      "data leverage": 26085,
      "nlp model": 79637,
      "interactive approach": 56012,
      "data creation": 25809,
      "creation annotation": 24712,
      "annotation results": 6364,
      "results user": 102300,
      "user studies": 129043,
      "dialogue settings": 30741,
      "mental states": 68414,
      "dialogue participants": 30719,
      "dialogue task": 30798,
      "task producing": 118569,
      "neural chatbot": 77857,
      "framing problem": 45751,
      "add constraint": 3349,
      "task level": 118350,
      "data induce": 26036,
      "learning latent": 62682,
      "human response": 50955,
      "response patterns": 101225,
      "dnns work": 32938,
      "qualitative analyses": 94550,
      "large positive": 61213,
      "positive correlations": 88317,
      "parameters training": 84791,
      "filtering using": 44090,
      "data outperforms": 26199,
      "finally highlight": 44191,
      "highlight cases": 50256,
      "datasets human": 27505,
      "human conceptual": 50784,
      "conceptual knowledge": 20667,
      "research semantic": 100622,
      "datasets limited": 27549,
      "words contrast": 133884,
      "modelling techniques": 72628,
      "effective vector": 34776,
      "large lexicons": 61132,
      "knowledge distributional": 57875,
      "word2vec architecture": 133668,
      "task modelling": 118401,
      "space makes": 111019,
      "makes easy": 66776,
      "ranking candidate": 95666,
      "arbitrary words": 8593,
      "previous approach": 90379,
      "tasks finally": 119130,
      "discuss method": 32266,
      "develop efficient": 30193,
      "efficient sampling": 35108,
      "sampling techniques": 103616,
      "reliable way": 98624,
      "way online": 132112,
      "turning point": 125987,
      "white supremacist": 132475,
      "theory suggests": 121827,
      "empirically tested": 36245,
      "differences language": 30960,
      "youtube channels": 135462,
      "possible changes": 88389,
      "models measure": 73560,
      "video transcripts": 131662,
      "larger extent": 61365,
      "changes language": 16385,
      "shows increase": 108589,
      "implications understanding": 51996,
      "achieved new": 2657,
      "performance translating": 86814,
      "disambiguation paper": 31981,
      "predict translation": 88948,
      "correct given": 24102,
      "representation ambiguous": 99164,
      "ambiguous noun": 5294,
      "encoder hidden": 36530,
      "encode relevant": 36449,
      "information disambiguation": 54495,
      "weights attention": 132373,
      "ambiguous nouns": 5295,
      "distribute attention": 32609,
      "attention context": 9808,
      "decoder hidden": 28018,
      "models recurrent": 73892,
      "upcoming words": 127788,
      "syntactically complex": 116506,
      "unexpectedly high": 127058,
      "high probabilities": 50108,
      "increasing size": 53685,
      "corpus trained": 24043,
      "gains increasing": 46395,
      "corpus need": 23896,
      "large models": 61154,
      "match human": 67358,
      "performance comparison": 86243,
      "gpt bert": 48553,
      "transformerbased models": 124425,
      "results make": 101904,
      "case data": 15579,
      "zeroshot text": 135529,
      "entailment approach": 37653,
      "approach zeroshot": 8032,
      "problem little": 91109,
      "text domain": 120891,
      "domain aspect": 33471,
      "emotion event": 36102,
      "literature make": 64760,
      "unified datasets": 127105,
      "contributions include": 22789,
      "provide facilitate": 93826,
      "diverse aspects": 32795,
      "aspects topic": 9413,
      "joy anger": 57416,
      "given dataset": 48013,
      "labels test": 58647,
      "challenging realistic": 16301,
      "chang et": 16354,
      "al 2008": 4600,
      "classifying text": 17680,
      "specific training": 111507,
      "data iii": 26015,
      "formulation study": 45296,
      "study way": 114551,
      "detection location": 29985,
      "approach addresses": 7335,
      "pun detection": 94386,
      "jointly sequence": 57387,
      "new tagging": 79192,
      "scheme model": 103931,
      "capable performing": 15211,
      "performing joint": 86954,
      "task useful": 118827,
      "information properly": 54879,
      "properly captured": 92430,
      "captured proposed": 15436,
      "explicit crosslingual": 41312,
      "pretraining unsupervised": 90349,
      "translation pretraining": 125137,
      "pretraining proven": 90309,
      "effective unsupervised": 34767,
      "method unsupervised": 69204,
      "crosslingual training": 25026,
      "training signals": 123861,
      "ngram embeddings": 79431,
      "new pretraining": 79080,
      "pretraining model": 90286,
      "crosslingual masked": 24973,
      "model cmlm": 70836,
      "text stream": 121329,
      "candidates time": 15148,
      "step experiments": 113264,
      "models encoder": 73147,
      "ordinal regression": 82449,
      "analysis campaign": 5522,
      "scale propose": 103751,
      "semisupervised settings": 105625,
      "results demonstrating": 101711,
      "baseline approaches": 12189,
      "approaches analyze": 8059,
      "analyze utility": 6015,
      "prediction provide": 89113,
      "provide qualitative": 93900,
      "analysis terms": 5892,
      "neural contextualized": 77872,
      "representation chinese": 99188,
      "understanding pretrained": 126923,
      "great successes": 49033,
      "capacity capture": 15229,
      "capture deep": 15285,
      "contextualized information": 22551,
      "text pretraining": 121199,
      "tasks current": 119023,
      "current version": 25388,
      "based bert": 11546,
      "relative positional": 98365,
      "word masking": 133357,
      "inference xnli": 54249,
      "chain conditional": 15971,
      "aims ground": 4535,
      "ground entity": 49095,
      "caption image": 15253,
      "image corresponding": 51778,
      "dependencies different": 29076,
      "methods aim": 69303,
      "capture dependencies": 15287,
      "dependencies need": 29099,
      "losses paper": 65309,
      "task treat": 118813,
      "labels use": 58651,
      "crfs model": 24771,
      "model dependencies": 70968,
      "contrast standard": 22714,
      "multiple correct": 75529,
      "gold labels": 48435,
      "algorithm enables": 4724,
      "endtoend training": 36986,
      "entities dataset": 37763,
      "dataset analysis": 26732,
      "entity dependencies": 37918,
      "dependencies captured": 29068,
      "training regime": 123807,
      "representations representational": 99854,
      "representational similarity": 99478,
      "similarity analysis": 109195,
      "analysis rsa": 5824,
      "technique developed": 119782,
      "interpretation language": 56264,
      "language encoders": 59003,
      "encoders based": 36633,
      "eyetracking data": 42639,
      "showcasing potential": 108366,
      "shall know": 107909,
      "company keeps": 19122,
      "nlp information": 79623,
      "information individuals": 54683,
      "texts short": 121605,
      "approaches modelling": 8237,
      "modelling social": 72622,
      "attention social": 10014,
      "ignores fact": 51680,
      "situations present": 109900,
      "based graph": 11746,
      "user representation": 129037,
      "target task": 117720,
      "task combines": 117972,
      "combines linguistic": 18691,
      "alternative models": 5238,
      "models analyse": 72747,
      "evaluating crosslingual": 39046,
      "crosslingual effectiveness": 24946,
      "shown capable": 108447,
      "capable translating": 15221,
      "english single": 37277,
      "model improved": 71312,
      "performance low": 86503,
      "potential crosslingual": 88543,
      "effectiveness representations": 34943,
      "representations encoder": 99627,
      "multilingual nmt": 75315,
      "model downstream": 71028,
      "downstream classification": 33994,
      "classification sequence": 17397,
      "tasks covering": 119020,
      "compare strong": 19298,
      "baseline multilingual": 12267,
      "multilingual bert": 75201,
      "bert mbert": 13164,
      "learning scenarios": 62982,
      "zeroshot transfer": 135532,
      "tasks discriminative": 119067,
      "discriminative neural": 32212,
      "alignment introduce": 4964,
      "novel discriminative": 80546,
      "transformerbased machine": 124419,
      "experiments based": 40789,
      "based small": 12052,
      "number labeled": 80903,
      "major improvements": 66567,
      "unsupervised baselines": 127606,
      "chinese ner": 16795,
      "lead higher": 61853,
      "perform ablation": 85939,
      "ablation analysis": 1582,
      "analysis annotation": 5493,
      "annotation experiment": 6307,
      "manual alignment": 66972,
      "king queen": 57696,
      "word relationships": 133433,
      "orthogonal transformations": 82588,
      "transformations embedding": 124278,
      "property word": 92507,
      "linear substructures": 64373,
      "turn allows": 125969,
      "analogies solved": 5412,
      "suggests models": 115377,
      "trained word": 123337,
      "embeddings easily": 35673,
      "easily learn": 34459,
      "evidence models": 39653,
      "represent relationships": 99131,
      "alternative way": 5250,
      "example given": 39784,
      "given translation": 48165,
      "orthogonal matrix": 82585,
      "reasoning using": 96324,
      "accurate using": 2373,
      "vector arithmetic": 131296,
      "findings suggest": 44314,
      "translation vectors": 125423,
      "contextaware neural": 22342,
      "years studies": 135304,
      "studies neural": 114256,
      "use documentlevel": 128012,
      "documentlevel context": 33134,
      "previous sentences": 90459,
      "sentences incorporate": 106353,
      "incorporate context": 53457,
      "context useful": 22313,
      "context considered": 22035,
      "architecture used": 8767,
      "contexts study": 22429,
      "context improve": 22130,
      "weight sharing": 132328,
      "decoder states": 28055,
      "calculates attention": 15025,
      "decoder state": 28054,
      "subword language": 115023,
      "model query": 71836,
      "models slow": 74060,
      "models fast": 73239,
      "generation query": 47578,
      "deal issues": 27860,
      "result model": 101383,
      "maintaining similar": 66525,
      "similar quality": 109127,
      "generated results": 47116,
      "characters model": 16616,
      "analysis ablation": 5479,
      "ablation study": 1585,
      "investigating relationship": 56857,
      "refers phenomenon": 97584,
      "speakers tend": 111329,
      "features describing": 43447,
      "gender diversity": 46596,
      "predicting perceived": 88998,
      "twostage approach": 126158,
      "entities medical": 37816,
      "medical documents": 68202,
      "entity contains": 37915,
      "problem subtasks": 91254,
      "entities based": 37749,
      "based representation": 11992,
      "neural components": 77866,
      "components designed": 20024,
      "learned jointly": 62221,
      "using shared": 130167,
      "shared encoder": 107978,
      "encoder text": 36580,
      "external features": 42002,
      "features previous": 43667,
      "embedding distribution": 35397,
      "work models": 134638,
      "space present": 111045,
      "using generative": 129701,
      "learn transformation": 62170,
      "fast named": 43145,
      "integrated various": 55781,
      "embeddings available": 35563,
      "knowledge free": 57942,
      "models knowledge": 73439,
      "progress pretraining": 92176,
      "led surge": 63291,
      "knowledge models": 58066,
      "present training": 89744,
      "able answer": 1607,
      "answer queries": 6557,
      "queries structured": 94940,
      "open class": 81884,
      "class relations": 17048,
      "easy extend": 34478,
      "data require": 26343,
      "human supervision": 50972,
      "train present": 122997,
      "present indepth": 89514,
      "models finetuning": 73254,
      "finetuning bert": 44449,
      "ii bert": 51697,
      "bert does": 13103,
      "answering supervised": 6706,
      "supervised baseline": 115731,
      "types factual": 126289,
      "surprisingly strong": 116148,
      "demonstrates potential": 28954,
      "systems code": 116786,
      "romanian language": 103251,
      "contains 26000": 21815,
      "classes sentences": 17076,
      "styles corpus": 114617,
      "language space": 60107,
      "space specifically": 111065,
      "specifically targeted": 111594,
      "recognition available": 96820,
      "free use": 45770,
      "use extend": 128040,
      "approach taskoriented": 7964,
      "approach bypasses": 7407,
      "use explicit": 128038,
      "explicit semantics": 41347,
      "form taskspecific": 45133,
      "trained hundreds": 123158,
      "examples extracted": 39834,
      "appropriate different": 8417,
      "text visual": 121408,
      "responses according": 101252,
      "available different": 10982,
      "languages contextaware": 60469,
      "translation modern": 124968,
      "sentencelevel nmt": 106165,
      "sentences context": 106262,
      "model correct": 70931,
      "sentencelevel translations": 106181,
      "translations sentences": 125492,
      "context training": 22304,
      "requires monolingual": 100304,
      "documentlevel data": 33135,
      "monolingual sequencetosequence": 74615,
      "model maps": 71503,
      "sentences consistent": 106257,
      "ones consistent": 81681,
      "consistent groups": 21380,
      "obtained sampling": 81407,
      "sentence approach": 105774,
      "using contrastive": 129491,
      "contrastive evaluation": 22730,
      "evaluation large": 39251,
      "contextual phenomena": 22485,
      "englishrussian translation": 37426,
      "score conduct": 104064,
      "conduct human": 20878,
      "evaluation strong": 39408,
      "strong preference": 113702,
      "analyze discourse": 5965,
      "hard capture": 49453,
      "practices lowresource": 88740,
      "development sets": 30425,
      "using available": 129376,
      "aim answer": 4446,
      "set early": 107426,
      "training epochs": 123612,
      "development languages": 30395,
      "lead overestimation": 61864,
      "multiple experiments": 75563,
      "experiments recent": 41110,
      "lowresource nlp": 65547,
      "sets average": 107649,
      "average languages": 11198,
      "absolute accuracy": 1722,
      "14 languages": 315,
      "languages tasks": 60913,
      "highlight importance": 50261,
      "faster learning": 43181,
      "adaptive policies": 3336,
      "challenging previous": 16294,
      "work falls": 134527,
      "training combine": 123386,
      "approaches propose": 8294,
      "framework learn": 45596,
      "generated parallel": 47106,
      "available source": 11106,
      "context provides": 22239,
      "information read": 54895,
      "read source": 95939,
      "word experiments": 133280,
      "experiments germanenglish": 40953,
      "germanenglish method": 47937,
      "learn flexible": 62050,
      "better modeling": 13630,
      "modeling hierarchical": 72440,
      "ordered neurons": 82433,
      "hybrid selfattention": 51194,
      "outperforms individual": 82911,
      "rnns propose": 102977,
      "propose enhance": 92656,
      "enhance strength": 37494,
      "models advanced": 72727,
      "neurons lstm": 78747,
      "benchmark machine": 12837,
      "targeted linguistic": 117768,
      "linguistic evaluation": 64471,
      "logical inference": 65011,
      "processing based": 91629,
      "lack generalization": 58710,
      "modelling language": 72607,
      "current input": 25287,
      "input previous": 55400,
      "interactions inputs": 55987,
      "model viewed": 72307,
      "transition function": 124489,
      "art datasets": 9046,
      "close large": 17823,
      "large gap": 61098,
      "lstm transformer": 65671,
      "2019 knowledge": 700,
      "triple extraction": 125809,
      "approach extract": 7562,
      "document offers": 33045,
      "offers simple": 81593,
      "solution challenge": 110549,
      "text provides": 121223,
      "information triple": 55059,
      "degree betweenness": 28567,
      "types jointly": 126302,
      "learning align": 62328,
      "models state": 74094,
      "typically provide": 126450,
      "models remain": 73921,
      "solution paper": 110561,
      "accurate translations": 2370,
      "discrete alignments": 32161,
      "alignments attention": 5009,
      "attention probabilities": 9978,
      "training leverage": 123686,
      "optimize translation": 82228,
      "produces competitive": 91990,
      "sacrificing translation": 103529,
      "previous attempts": 90387,
      "finally incorporating": 44197,
      "model alignments": 70655,
      "accuracies compared": 2058,
      "settings study": 107839,
      "dialogue turns": 30809,
      "turns using": 125996,
      "systems specialized": 117150,
      "results endtoend": 101752,
      "dialogue propose": 30729,
      "average absolute": 11179,
      "insights improving": 55536,
      "media newspaper": 68131,
      "influence social": 54317,
      "nature influence": 76656,
      "positively correlated": 88351,
      "knowledge language": 58031,
      "tasks requiring": 119465,
      "requiring significant": 100356,
      "question best": 95133,
      "knowledge explore": 57921,
      "experimental methods": 40543,
      "polarity item": 87912,
      "npi licensing": 80799,
      "like negation": 64075,
      "bert significant": 13229,
      "knowledge features": 57935,
      "widely different": 132542,
      "models grammatical": 73312,
      "given domain": 48025,
      "writing assistance": 135088,
      "environments language": 38201,
      "human writers": 50990,
      "writers benefit": 135083,
      "framework achieving": 45413,
      "integrate latest": 55759,
      "latest developments": 61652,
      "result users": 101418,
      "text effective": 120900,
      "transformer networks": 124359,
      "tracking entities": 122756,
      "entities entities": 37774,
      "encoders like": 36657,
      "handle nuances": 49397,
      "untested paper": 127765,
      "tracking tasks": 122765,
      "text test": 121366,
      "test standard": 120539,
      "prediction pretrained": 89105,
      "pretrained transformers": 90206,
      "simple baselines": 109375,
      "results attained": 101525,
      "second assess": 104388,
      "networks capture": 77530,
      "entity references": 38089,
      "references different": 97558,
      "shallow context": 107915,
      "context clues": 22028,
      "form complex": 45074,
      "supervised knowledgebased": 115760,
      "answering kbqa": 6641,
      "question logical": 95175,
      "supervision provided": 115909,
      "forms model": 45250,
      "complex question": 19865,
      "typically involves": 126438,
      "involves huge": 56893,
      "huge search": 50726,
      "main problems": 66453,
      "reduce success": 97361,
      "degrade quality": 28559,
      "quality training": 94813,
      "problems lead": 91333,
      "search method": 104328,
      "method weakly": 69224,
      "questions search": 95364,
      "space constrained": 110985,
      "search paths": 104335,
      "larger proportion": 61379,
      "forms generated": 45244,
      "improving precision": 53141,
      "articles event": 9186,
      "extraction news": 42413,
      "article summarization": 9163,
      "summarization article": 115480,
      "aggregation lack": 4361,
      "available methods": 11047,
      "news datasets": 79328,
      "did provide": 30920,
      "provide indepth": 93850,
      "syntactic domainspecific": 116400,
      "rules automatically": 103381,
      "relevant phrases": 98571,
      "phrases english": 87420,
      "provide answers": 93758,
      "expert evaluation": 41220,
      "overall precision": 83250,
      "summarize main": 115607,
      "analysis dont": 5584,
      "analysis morphological": 5707,
      "translate rare": 124534,
      "zipfian distribution": 135546,
      "common work": 18950,
      "investigate stateoftheart": 56815,
      "stateoftheart bilingual": 112603,
      "morphological forms": 74704,
      "drops considerably": 34187,
      "morphological inflections": 74711,
      "adding simple": 3392,
      "simple morphological": 109475,
      "constraint training": 21586,
      "better encoding": 13570,
      "comparison methods": 19555,
      "lowresource dependency": 65493,
      "languages require": 60845,
      "require lots": 100171,
      "lots training": 65337,
      "improving lowresource": 53112,
      "diverse lowresource": 32822,
      "available crosslingual": 10968,
      "training helpful": 123647,
      "inference introduce": 54154,
      "assessments demonstrate": 9533,
      "collecting annotations": 18444,
      "given premise": 48085,
      "nli data": 79538,
      "approach human": 7611,
      "performance demonstrating": 86285,
      "demonstrating models": 28978,
      "capable subtle": 15220,
      "current nli": 25316,
      "nli tasks": 79554,
      "questions multiturn": 95331,
      "comparisons dialogue": 19590,
      "remains important": 98798,
      "make real": 66715,
      "judgments human": 57447,
      "measure used": 67829,
      "novel procedure": 80692,
      "human judge": 50875,
      "asked pay": 9302,
      "attention speaker": 10016,
      "pairwise judgment": 83675,
      "optimized maximize": 82234,
      "resulting better": 101432,
      "tests tests": 120613,
      "faster cheaper": 43165,
      "efficient sentence": 35112,
      "using discrete": 129584,
      "structure complex": 113822,
      "complex sequential": 19877,
      "sequential convolutional": 107216,
      "networks potentially": 77700,
      "superior classification": 115681,
      "efficient alternative": 35056,
      "alternative propose": 5240,
      "use discrete": 128005,
      "patterns sentences": 85780,
      "sentences results": 106482,
      "benefit syntactic": 12990,
      "features results": 43708,
      "embeddings preserve": 35873,
      "information compared": 54428,
      "complexity model": 19925,
      "selfattention neural": 104897,
      "mechanism work": 68051,
      "approaches incorporate": 8193,
      "incorporate syntactic": 53489,
      "knowledge transformer": 58214,
      "especially long": 38466,
      "sentences lowresource": 106387,
      "englishjapanese translation": 37416,
      "bert perform": 13199,
      "successive text": 115199,
      "text compression": 120810,
      "compression text": 20327,
      "diverse applications": 32794,
      "applications summarization": 7022,
      "comprehension text": 20239,
      "text editing": 120896,
      "autoencoder models": 10405,
      "source input": 110768,
      "adapting different": 3306,
      "output usually": 83146,
      "retraining model": 102374,
      "model scratch": 71953,
      "propose fully": 92688,
      "bert score": 13223,
      "score candidate": 104058,
      "based average": 11540,
      "step apply": 113248,
      "task extractive": 118196,
      "compression model": 20313,
      "indomain examples": 53966,
      "examples similar": 39877,
      "compression ratio": 20320,
      "evaluations verify": 39490,
      "verify model": 131590,
      "produces highquality": 91998,
      "corpus american": 23655,
      "studies based": 114187,
      "manual coding": 66983,
      "football games": 45006,
      "mentions linked": 68457,
      "identify major": 51521,
      "confounding factors": 21044,
      "racial bias": 95427,
      "perform computational": 85974,
      "science studies": 103985,
      "regularization based": 97746,
      "reasoning ability": 96224,
      "reasoning sentence": 96308,
      "essential natural": 38558,
      "improvement tasks": 52775,
      "prediction nsp": 89092,
      "bert learns": 13157,
      "sentencepair input": 106185,
      "despite effectiveness": 29686,
      "remedy propose": 98862,
      "task includes": 118284,
      "determine sentence": 30133,
      "improves ability": 52944,
      "remarkable improvement": 98832,
      "improvement vanilla": 52783,
      "vanilla bert": 130812,
      "incorporate documentlevel": 53464,
      "hans dataset": 49434,
      "task promising": 118570,
      "parsing ner": 85174,
      "context recently": 22243,
      "embeddings bert": 35572,
      "ner tasks": 77090,
      "czech named": 25494,
      "corpus 11": 23630,
      "11 20": 214,
      "20 report": 526,
      "results mentioned": 101914,
      "tasks corpora": 119018,
      "entity relation": 38090,
      "contextualized span": 22570,
      "representations examine": 99640,
      "unified multitask": 127122,
      "recognition relation": 96984,
      "extraction event": 42328,
      "capture local": 15332,
      "datasets variety": 27786,
      "domains perform": 33829,
      "comparing different": 19502,
      "representations contextualized": 99569,
      "like bert": 64021,
      "relationships entities": 98320,
      "updates model": 127806,
      "representations predicted": 99810,
      "coreference links": 23349,
      "model disambiguate": 71001,
      "easily adapted": 34431,
      "task recent": 118608,
      "transfer highresource": 124058,
      "languages unclear": 60934,
      "unclear knowledge": 126634,
      "knowledge transferred": 58213,
      "ner experiments": 77044,
      "learning works": 63174,
      "embeddings investigate": 35767,
      "investigate model": 56781,
      "knowledge wikipedia": 58232,
      "direction improve": 31821,
      "performances results": 86894,
      "results shed": 102169,
      "light future": 63989,
      "content documents": 21877,
      "quantify extent": 94848,
      "extent given": 41976,
      "given short": 48125,
      "specific person": 111475,
      "dataset tweets": 27249,
      "useful generating": 128890,
      "twitter profiles": 126119,
      "profiles understanding": 92102,
      "author text": 10369,
      "text report": 121244,
      "report test": 99055,
      "approaches problem": 8289,
      "use models": 128148,
      "authors texts": 10387,
      "strong correlations": 113662,
      "texts demonstrate": 121493,
      "demonstrate statistically": 28871,
      "significant correlation": 108749,
      "study pretrained": 114478,
      "sequential sentence": 107252,
      "classification step": 17420,
      "sequence sentences": 107048,
      "corresponding categories": 24276,
      "categories task": 15755,
      "context document": 22064,
      "recent successful": 96546,
      "task used": 118826,
      "incorporate dependencies": 53461,
      "work pretrained": 134700,
      "2018 particular": 668,
      "task capture": 117951,
      "capture contextual": 15281,
      "need hierarchical": 76816,
      "construct joint": 21634,
      "representation allows": 99163,
      "bert transformer": 13242,
      "transformer layers": 124334,
      "sentences approach": 106214,
      "structured scientific": 114038,
      "tend suffer": 120163,
      "suffer paucity": 115232,
      "success transfer": 115135,
      "involves training": 56906,
      "modeling training": 72572,
      "trained network": 123216,
      "far richer": 43098,
      "linear word": 64388,
      "lms predominantly": 64888,
      "tasks help": 119155,
      "predictions based": 89160,
      "document test": 33095,
      "test performance": 120479,
      "performance types": 86818,
      "types pretrained": 126339,
      "lms investigate": 64884,
      "investigate possible": 56796,
      "experiments suggest": 41158,
      "criterion used": 24800,
      "training pretrained": 123785,
      "beneficial neural": 12951,
      "modeling knowledge": 72454,
      "text contain": 120815,
      "contain explicit": 21739,
      "entities propose": 37847,
      "method embed": 68795,
      "multiple knowledge": 75585,
      "models enhance": 73156,
      "enhance representations": 37488,
      "representations structured": 99908,
      "humancurated knowledge": 51016,
      "use integrated": 128098,
      "entity linker": 37952,
      "retrieve relevant": 102452,
      "approaches entity": 8146,
      "selfsupervised language": 104939,
      "objective jointly": 81089,
      "endtoend multitask": 36933,
      "multitask setting": 75890,
      "bert knowledge": 13148,
      "enhanced bert": 37502,
      "improved perplexity": 52630,
      "recall facts": 96338,
      "task downstream": 118120,
      "downstream performance": 34020,
      "relationship extraction": 98297,
      "extraction entity": 42325,
      "finegrained knowledge": 44360,
      "labeling domain": 58497,
      "labeling previous": 58524,
      "entire target": 37717,
      "domain considering": 33486,
      "considering diversity": 21323,
      "domain samples": 33631,
      "lead negative": 61861,
      "negative transfer": 76959,
      "transfer results": 124181,
      "important characteristic": 52120,
      "different elements": 31119,
      "elements given": 35279,
      "given sample": 48109,
      "domain relevance": 33625,
      "requires consideration": 100255,
      "propose finegrained": 92677,
      "fusion model": 46243,
      "modeling scheme": 72537,
      "balance learning": 11397,
      "learning target": 63083,
      "experiments sequence": 41127,
      "tasks finegrained": 119133,
      "networks like": 77643,
      "read recent": 95937,
      "research understanding": 100658,
      "probes models": 90882,
      "known priori": 58316,
      "novel abstractive": 80454,
      "imposing prior": 52308,
      "prior finetuning": 90703,
      "classifier able": 17516,
      "order predict": 82384,
      "method diverse": 68782,
      "diverse tasks": 32854,
      "models contrast": 72982,
      "models reproduce": 73931,
      "evidence extraction": 39645,
      "extraction evidence": 42330,
      "providing justification": 94121,
      "seek build": 104517,
      "models scientific": 73977,
      "extracting text": 42243,
      "primary research": 90655,
      "evidence presented": 39661,
      "learning pretrained": 62906,
      "pretrained biomedical": 90000,
      "discourse structures": 32094,
      "text results": 121264,
      "biomedical experimental": 14185,
      "articles demonstrate": 9182,
      "tagging datasets": 117381,
      "transferability new": 124214,
      "potential using": 88596,
      "quality scientific": 94783,
      "scientific claims": 103994,
      "concept representations": 20586,
      "tasks lack": 119237,
      "lack semantic": 58744,
      "representation extracted": 99236,
      "set relations": 107561,
      "term results": 120241,
      "performance distributional": 86306,
      "dynamic fusion": 34307,
      "nmt used": 80004,
      "fluent output": 44714,
      "models investigated": 73429,
      "models predictions": 73767,
      "model handcrafted": 71271,
      "adopt language": 3894,
      "regard translation": 97651,
      "line approach": 64324,
      "model prediction": 71760,
      "adequacy translation": 3828,
      "accordingly work": 2018,
      "work employs": 134489,
      "model attentive": 70705,
      "architecture language": 8681,
      "model auxiliary": 70715,
      "model compared": 70853,
      "englishjapanese machine": 37414,
      "proposed dynamic": 93260,
      "fusion mechanism": 46240,
      "scores additionally": 104151,
      "analyses attention": 5440,
      "mechanism allows": 67944,
      "predictive language": 89203,
      "appropriate grammatical": 8422,
      "based ensemble": 11681,
      "application document": 6846,
      "like dialog": 64035,
      "personalization knowledge": 87138,
      "important components": 52127,
      "components better": 20018,
      "section finally": 104486,
      "engaging experience": 37021,
      "visual analysis": 131784,
      "toolkit text": 122421,
      "captioning video": 15266,
      "usually relies": 130448,
      "taskspecific metrics": 119632,
      "aligned human": 4915,
      "human assessment": 50753,
      "patterns paper": 85767,
      "text references": 121234,
      "jupyter notebook": 57459,
      "covers common": 24554,
      "based metrics": 11843,
      "metrics bertscore": 69928,
      "tag set": 117341,
      "set wikipedia": 107636,
      "wikipedia great": 132657,
      "general world": 46732,
      "step goal": 113269,
      "categorization dataset": 15763,
      "large multilingual": 61162,
      "japanese english": 57187,
      "evaluate dataset": 38815,
      "using best": 129393,
      "tag sets": 117342,
      "systems critical": 116824,
      "systems naturally": 117015,
      "filled pauses": 44026,
      "process especially": 91477,
      "especially important": 38461,
      "make effective": 66663,
      "use current": 127977,
      "need research": 76850,
      "including incremental": 53301,
      "inform research": 54336,
      "translation semantic": 125236,
      "estimation recent": 38666,
      "work demonstrated": 134462,
      "directly improve": 31880,
      "substantially improve": 114892,
      "limitations doesnt": 64172,
      "semantically correct": 105367,
      "english training": 37311,
      "translations evaluated": 125462,
      "optimization procedure": 82206,
      "converges faster": 22931,
      "suggests proposed": 115381,
      "querying knowledge": 95005,
      "long challenge": 65061,
      "dialogue previous": 30724,
      "previous sequencetosequence": 90462,
      "work treats": 134862,
      "query attention": 94949,
      "attention entire": 9831,
      "entities consistent": 37759,
      "steps improve": 113321,
      "consistency generated": 21359,
      "entities step": 37875,
      "inspired observation": 55573,
      "single kb": 109745,
      "retrieval component": 102387,
      "component explicitly": 19978,
      "filter irrelevant": 44071,
      "entities second": 37864,
      "perform attention": 85948,
      "mechanism address": 67943,
      "training feasible": 123633,
      "retrieval data": 102390,
      "technique experiments": 119786,
      "datasets effectiveness": 27440,
      "news coverage": 79325,
      "scenarios data": 103842,
      "events taking": 39602,
      "order account": 82265,
      "training evaluate": 123614,
      "humancurated dataset": 51015,
      "news topics": 79383,
      "baselines based": 12359,
      "involves tasks": 56904,
      "tasks intent": 119219,
      "intent detection": 55902,
      "detection slot": 30054,
      "works existing": 134945,
      "relations slots": 98259,
      "slots intents": 110089,
      "performance address": 86127,
      "stacking multiple": 112135,
      "information exchange": 54546,
      "atis snips": 9682,
      "domain gap": 33538,
      "gap crosslingual": 46447,
      "data prohibits": 26285,
      "languages recent": 60833,
      "crosslingual understanding": 25042,
      "understanding xlu": 127020,
      "progress area": 92139,
      "bridge language": 14596,
      "language universal": 60297,
      "consider setting": 21229,
      "semisupervised crosslingual": 105597,
      "available target": 11121,
      "language combine": 58889,
      "stateoftheart crosslingual": 112626,
      "unsupervised data": 127619,
      "close language": 17821,
      "baselines achieve": 12345,
      "domain transfer": 33681,
      "oriented dialogue": 82494,
      "tracking dst": 122755,
      "predicting user": 89020,
      "stateoftheart dst": 112647,
      "dst models": 34207,
      "manual annotations": 66979,
      "turn level": 125973,
      "costly obtain": 24388,
      "domains address": 33728,
      "limitations propose": 64190,
      "based reinforcement": 11987,
      "improve models": 52419,
      "work language": 134599,
      "training large": 123676,
      "quite difficult": 95400,
      "present techniques": 89737,
      "techniques training": 120001,
      "models implement": 73356,
      "model parallel": 71702,
      "approach enables": 7522,
      "training transformer": 123935,
      "pipeline model": 87546,
      "fully implemented": 45962,
      "transformer based": 124303,
      "billion parameters": 14079,
      "strong single": 113713,
      "single gpu": 109738,
      "advance state": 3931,
      "transformer language": 124330,
      "bertlike models": 13283,
      "models critical": 73003,
      "increased performance": 53631,
      "achieve sota": 2550,
      "sota results": 110695,
      "compared sota": 19445,
      "datasets bert": 27332,
      "achieves sota": 2871,
      "results race": 102107,
      "race dataset": 95422,
      "graphical neural": 48931,
      "networks text": 77794,
      "syntax structure": 116562,
      "models graph": 73313,
      "natural structure": 76622,
      "networks gnn": 77613,
      "provides powerful": 94061,
      "tool model": 122394,
      "networks gcn": 77608,
      "text organized": 121156,
      "form graph": 45092,
      "graph proposed": 48859,
      "model lstm": 71482,
      "used dynamically": 128505,
      "dynamically decide": 34345,
      "neighbor information": 76984,
      "information transmitted": 55058,
      "upper layers": 127821,
      "problem furthermore": 91063,
      "experiments single": 41145,
      "label text": 58418,
      "significantly datasets": 108903,
      "problem improving": 91080,
      "approach incorporate": 7632,
      "incorporate syntax": 53491,
      "syntax natural": 116542,
      "uses contextual": 129210,
      "broadly applicable": 14704,
      "bert mtdnn": 13183,
      "nli benchmarks": 79537,
      "takes character": 117529,
      "representation train": 99444,
      "train language": 122942,
      "unit word": 127222,
      "unsupervised segmentation": 127709,
      "including syntactic": 53390,
      "implicit discourse": 52005,
      "relation recognition": 98065,
      "recognition textual": 97031,
      "meaningful improvement": 67719,
      "complementary alternative": 19737,
      "techniques model": 119932,
      "estimation quality": 38665,
      "description natural": 29469,
      "assess feasibility": 9478,
      "quality requirements": 94774,
      "vaes text": 130704,
      "text especially": 120919,
      "generation strong": 47639,
      "strong autoregressive": 113624,
      "translation explicitly": 124815,
      "autoencoder cvae": 10400,
      "propose modified": 92790,
      "results transformer": 102284,
      "decoder result": 28052,
      "result proposed": 101396,
      "quality demonstrating": 94629,
      "efficiency robustness": 35045,
      "parsing unlabeled": 85274,
      "auxiliary languages": 10877,
      "learning important": 62638,
      "fundamental techniques": 46125,
      "techniques transfer": 120002,
      "transfer languages": 124069,
      "embeddings contextual": 35613,
      "propose leverage": 92750,
      "leverage unannotated": 63629,
      "unannotated sentences": 126587,
      "sentences auxiliary": 106224,
      "learning languageagnostic": 62677,
      "representations specifically": 99904,
      "specifically explore": 111548,
      "learning contextual": 62459,
      "contextual encoders": 22462,
      "invariant representations": 56682,
      "languages facilitate": 60580,
      "facilitate crosslingual": 42761,
      "experiments crosslingual": 40861,
      "parser source": 84995,
      "range target": 95608,
      "demonstrate adversarial": 28659,
      "analysis evaluate": 5603,
      "training working": 123956,
      "explores task": 41652,
      "information shown": 54980,
      "typological information": 126487,
      "learning leverage": 62688,
      "better transfer": 13748,
      "transfer performance": 124176,
      "typological similarity": 126490,
      "crosslingual transferability": 25040,
      "parsing generating": 85116,
      "common task": 18932,
      "results classic": 101577,
      "algorithms learning": 4864,
      "text simple": 121297,
      "usually tradeoff": 130460,
      "present ways": 89764,
      "ways automatically": 132147,
      "overcoming issue": 83307,
      "extensions improve": 41847,
      "algorithm propose": 4788,
      "propose dynamically": 92637,
      "algorithms implemented": 4859,
      "users preferred": 129154,
      "models focused": 73259,
      "valuable new": 130767,
      "template learning": 120079,
      "systems change": 116776,
      "users twitter": 129177,
      "literature proposed": 64766,
      "attribute change": 10178,
      "april 2019": 8476,
      "experiment design": 40460,
      "output overcome": 83102,
      "use human": 128090,
      "demonstrated human": 28920,
      "design evaluation": 29538,
      "judgments conduct": 57441,
      "understand impact": 126750,
      "output addition": 83047,
      "continuous scale": 22629,
      "experiment novel": 40480,
      "bestworst scaling": 13491,
      "study 40": 114303,
      "crowdsourced workers": 25114,
      "continuous scales": 22630,
      "task prior": 118564,
      "augmentation introduce": 10282,
      "introduce hybrid": 56434,
      "extractions based": 42543,
      "distances entities": 32439,
      "simplified architecture": 109600,
      "risk mining": 102878,
      "expert evaluations": 41221,
      "sentences preferred": 106449,
      "information increases": 54681,
      "engine paper": 37033,
      "approach automated": 7371,
      "generation aims": 47300,
      "models help": 73323,
      "help encoding": 49716,
      "text named": 121136,
      "task biomedical": 117941,
      "domain named": 33588,
      "mentions ambiguous": 68444,
      "spans annotated": 111196,
      "multiple entity": 75555,
      "order based": 82284,
      "remains significant": 98821,
      "significant source": 108867,
      "source error": 110755,
      "error address": 38278,
      "issue generating": 56998,
      "concept mentions": 20579,
      "mentions multiple": 68459,
      "provide context": 93792,
      "context mention": 22176,
      "approached task": 8035,
      "task classification": 117963,
      "problem developed": 91010,
      "developed deep": 30265,
      "identify likely": 51516,
      "specifically develop": 111541,
      "connected layer": 21126,
      "layer final": 61719,
      "classification priority": 17342,
      "substantial increase": 114869,
      "generation pretraining": 47557,
      "pretraining work": 90352,
      "supervision signals": 115917,
      "nlg tasks": 79529,
      "propose pretrain": 92996,
      "crosslingual settings": 25014,
      "objective encourages": 81073,
      "model represent": 71906,
      "languages shared": 60868,
      "conduct zeroshot": 20908,
      "pretraining procedure": 90306,
      "procedure use": 91404,
      "use monolingual": 128151,
      "data finetune": 25949,
      "language directly": 58968,
      "evaluated language": 38982,
      "results question": 102103,
      "generation abstractive": 47288,
      "summarization model": 115535,
      "methods zeroshot": 69860,
      "crosslingual generation": 24959,
      "data implementation": 26018,
      "word concept": 132963,
      "benchmarks various": 12941,
      "tasks basis": 118956,
      "embeddings vectors": 36018,
      "vectors low": 131453,
      "space learned": 111013,
      "corpora various": 23616,
      "various algorithms": 131036,
      "relations embeddings": 98143,
      "evaluated tasks": 39012,
      "tasks predicting": 119390,
      "concept pair": 20581,
      "using knowledge": 129776,
      "ii evaluating": 51707,
      "extent pretrained": 41981,
      "relations evaluate": 98153,
      "captured current": 15427,
      "efficient adaptation": 35053,
      "text task": 121361,
      "universally used": 127342,
      "use downstream": 128018,
      "cover small": 24489,
      "set entity": 107430,
      "deep structured": 28425,
      "datasets jointly": 27532,
      "identify entity": 51497,
      "leveraging multiple": 63691,
      "robust input": 103040,
      "combining models": 18736,
      "predictions test": 89193,
      "datasets testing": 27756,
      "testing datasets": 120591,
      "present speech": 89714,
      "text transcription": 121374,
      "resource egyptian": 100841,
      "high resource": 50128,
      "unlabeled audio": 127371,
      "audio data": 10226,
      "collected youtube": 18442,
      "network acoustic": 77125,
      "cnn followed": 18004,
      "layers time": 61811,
      "network tdnn": 77452,
      "sequence discriminative": 106933,
      "training acoustic": 123346,
      "model ngram": 71583,
      "list rescoring": 64699,
      "lowest word": 65465,
      "speech dataset": 111669,
      "importance subword": 52083,
      "morphological tasks": 74751,
      "tasks truly": 119578,
      "learning subwords": 63069,
      "models value": 74315,
      "lowdata regimes": 65406,
      "subwords word": 115054,
      "scenarios representative": 103865,
      "type data": 126193,
      "stem lack": 113232,
      "lack taskspecific": 58757,
      "data lack": 26067,
      "required train": 100225,
      "language type": 60195,
      "subwordinformed word": 115046,
      "method main": 68949,
      "main results": 66460,
      "types large": 126308,
      "embeddings suggest": 35964,
      "suggest effective": 115309,
      "largely depends": 61336,
      "depends language": 29287,
      "data critical": 25811,
      "improving pretrained": 53144,
      "models vocabulary": 74332,
      "remarkable success": 98847,
      "pretrain deep": 89972,
      "deep language": 28240,
      "model largescale": 71417,
      "language instead": 59211,
      "pretraining monolingual": 90289,
      "monolingual language": 74591,
      "multilingual deep": 75237,
      "hundreds languages": 51150,
      "size language": 109928,
      "hinders performance": 50457,
      "tokenlevel sentencelevel": 122293,
      "inspired previous": 55575,
      "settings investigate": 107813,
      "approaches joint": 8201,
      "mapping based": 67127,
      "addressing outofvocabulary": 3814,
      "including partofspeech": 53346,
      "using mixture": 129887,
      "promising best": 92267,
      "work attempts": 134386,
      "oov issue": 81864,
      "settings improving": 107809,
      "architectures recently": 8842,
      "improvements traditional": 52932,
      "neural semantic": 78661,
      "appear relatively": 6801,
      "high recall": 50126,
      "large beam": 61037,
      "beam sizes": 12610,
      "improvement respect": 52758,
      "generator produces": 47778,
      "produces list": 92002,
      "list potential": 64697,
      "similarity candidate": 109207,
      "improves parsing": 53005,
      "performance extensive": 86366,
      "experiment model": 40472,
      "parsing datasets": 85095,
      "geo atis": 47847,
      "atis overnight": 9681,
      "sourcetarget domain": 110945,
      "world different": 135026,
      "strikingly different": 113595,
      "different cultures": 31071,
      "different things": 31495,
      "world work": 135056,
      "context machine": 22171,
      "particularly low": 85486,
      "regions world": 97682,
      "local events": 64918,
      "formalize concept": 45196,
      "metric quantify": 69896,
      "organic text": 82459,
      "text produced": 121216,
      "differences conclude": 30951,
      "phonetic sequences": 87295,
      "understanding speech": 126969,
      "speech production": 111732,
      "syllabification methods": 116300,
      "shown perform": 108506,
      "ner speech": 77080,
      "network constructed": 77211,
      "crf output": 24755,
      "rarely evaluated": 95758,
      "evaluated multiple": 38994,
      "competitive state": 19689,
      "art systems": 9095,
      "english dutch": 37122,
      "italian french": 57092,
      "dialogs agents": 30623,
      "reach agreement": 95887,
      "challenges modeling": 16175,
      "using finite": 129680,
      "state transducers": 112535,
      "transducers fsts": 124022,
      "history states": 50570,
      "dialog policy": 30579,
      "planning generation": 87639,
      "humans better": 51069,
      "generation ability": 47287,
      "better deal": 13561,
      "best strategy": 13456,
      "largely depend": 61335,
      "depend context": 29046,
      "context current": 22046,
      "identify set": 51552,
      "context set": 22266,
      "humanhuman dialogs": 51030,
      "smooth inverse": 110264,
      "inverse frequency": 56700,
      "critically review": 24847,
      "wikipedia contain": 132645,
      "contain large": 21743,
      "text need": 121141,
      "dynamically changing": 34341,
      "articles challenging": 9176,
      "constrained generation": 21569,
      "task output": 118494,
      "solution identify": 110556,
      "model expand": 71122,
      "text consistent": 120814,
      "consistent given": 21379,
      "model copy": 70925,
      "achieving highest": 2954,
      "furthermore demonstrate": 46160,
      "toolkit present": 122418,
      "present universal": 89753,
      "semantics uds": 105475,
      "sets single": 107712,
      "semantic graph": 105065,
      "node edge": 80019,
      "using sophisticated": 130197,
      "graphs using": 48975,
      "power understanding": 88655,
      "developing better": 30336,
      "vulnerabilities natural": 131969,
      "characteristics syntax": 16537,
      "text alterations": 120639,
      "result text": 101414,
      "ones demonstrate": 81682,
      "changes lexical": 16387,
      "datasets representing": 27674,
      "representing different": 100047,
      "levels lexical": 63546,
      "nmt approach": 79832,
      "quality criteria": 94623,
      "generate translations": 47038,
      "input natural": 55371,
      "sampling strategy": 103614,
      "weak feedback": 132178,
      "experiments sentiment": 41125,
      "data german": 25980,
      "performance classification": 86203,
      "transformers introduce": 124449,
      "architecture selfattention": 8741,
      "sequence dialogue": 106929,
      "context argue": 22012,
      "encoding sequence": 36726,
      "sequence single": 107071,
      "single conversation": 109719,
      "multiple overlapping": 75632,
      "discourse segments": 32087,
      "multiple topics": 75732,
      "performance transformer": 86813,
      "designed overcome": 29616,
      "systems develop": 116845,
      "multilayer representation": 75186,
      "meaning dialogue": 67631,
      "variety experiments": 130980,
      "obtains promising": 81472,
      "annotated dialogue": 6182,
      "nlu dataset": 79808,
      "providing overall": 94127,
      "performance higher": 86428,
      "stateoftheart tools": 113008,
      "example average": 39774,
      "entity tagging": 38108,
      "compositional semantics": 20132,
      "study semantic": 114509,
      "logical inferences": 65012,
      "complexity syntactic": 19939,
      "inference patterns": 54189,
      "patterns study": 85782,
      "english semantic": 37272,
      "effectively handles": 34817,
      "modeling confidence": 72401,
      "recently significant": 96768,
      "achieved various": 2721,
      "models aiming": 72737,
      "models assess": 72776,
      "quality output": 94738,
      "use similarity": 128276,
      "similarity training": 109324,
      "models confidence": 72959,
      "methods solely": 69764,
      "measures investigate": 67876,
      "internal alignment": 56158,
      "external alignment": 41993,
      "methods downstream": 69460,
      "confidence estimation": 20986,
      "tokens addition": 122297,
      "addition methods": 3450,
      "task able": 117826,
      "20 data": 497,
      "sports news": 112039,
      "closely follow": 17855,
      "follow underlying": 44944,
      "challenge automated": 15997,
      "news generation": 79338,
      "generation realworld": 47581,
      "demonstrate generation": 28748,
      "dataset source": 27211,
      "limits learning": 64318,
      "sentences meaning": 106395,
      "requires expensive": 100267,
      "expensive annotation": 40413,
      "complex structures": 19883,
      "learning policy": 62895,
      "samples labeled": 103586,
      "maximize performance": 67512,
      "performance test": 86795,
      "selection process": 104819,
      "stochastic nature": 113360,
      "improving data": 53082,
      "efficiency learning": 35030,
      "languages highly": 60618,
      "highly ambiguous": 50292,
      "forms morphologically": 45251,
      "standard orthography": 112289,
      "features identify": 43543,
      "better context": 13549,
      "lexical choices": 63738,
      "different modeling": 31264,
      "make joint": 66691,
      "modeling difficult": 72418,
      "different features": 31147,
      "model surface": 72120,
      "use arabic": 127897,
      "arabic test": 8558,
      "arabic dialectal": 8507,
      "text level": 121087,
      "level graph": 63455,
      "classification recently": 17370,
      "preserving global": 89934,
      "corpus level": 23861,
      "support online": 115983,
      "gnn based": 48323,
      "model builds": 70792,
      "builds graphs": 14905,
      "graphs input": 48959,
      "graph corpus": 48791,
      "method removes": 69102,
      "corpus support": 24029,
      "preserve global": 89922,
      "information build": 54403,
      "build graphs": 14770,
      "memory attention": 68287,
      "ner used": 77094,
      "paper formally": 83959,
      "improvements especially": 52848,
      "mentions given": 68454,
      "ucca semantic": 126526,
      "masked sequence": 67302,
      "sequence tagger": 107074,
      "builds graph": 14904,
      "graph nodes": 48850,
      "using masking": 129862,
      "reflects decisions": 97625,
      "lingual transfer": 64414,
      "transfer problem": 124178,
      "model french": 71207,
      "framenet semantic": 45397,
      "using adversarial": 129324,
      "adapted process": 3290,
      "building previous": 14877,
      "effectiveness adversarial": 34864,
      "domain generalization": 33540,
      "context semantic": 22256,
      "written documents": 135124,
      "propose extend": 92670,
      "underlying question": 126693,
      "question study": 95221,
      "study adversarial": 114307,
      "used combine": 128432,
      "level abstraction": 63412,
      "abstraction order": 1791,
      "increase robustness": 53613,
      "robustness lexical": 103105,
      "lexical stylistic": 63823,
      "stylistic variations": 114634,
      "errors proposed": 38404,
      "strategy evaluated": 113509,
      "evaluated french": 38976,
      "annotated framenet": 6198,
      "framenet paradigm": 45395,
      "learning increases": 62646,
      "increases models": 53646,
      "models generalization": 73272,
      "generalization capabilities": 46769,
      "translating clean": 124587,
      "clean texts": 17705,
      "noise input": 80055,
      "improving nmt": 53132,
      "models robustness": 73969,
      "corpus provides": 23950,
      "pairs data": 83509,
      "approaches heavily": 8178,
      "heavily dependent": 49651,
      "firstly propose": 44566,
      "limited noisy": 64255,
      "explore effect": 41542,
      "effect utilizing": 34619,
      "form speech": 45126,
      "linear relations": 64368,
      "spaces present": 111101,
      "present introductory": 89520,
      "datasets natural": 27590,
      "inference using": 54247,
      "method look": 68945,
      "vector differences": 131302,
      "differences embeddings": 30955,
      "sindhi language": 109702,
      "aims extract": 4527,
      "existing information": 40143,
      "following categories": 44963,
      "location date": 64962,
      "short forms": 108210,
      "important aspect": 52104,
      "surface level": 116063,
      "level understanding": 63515,
      "entities involved": 37799,
      "involved document": 56881,
      "different arabic": 31001,
      "like arabic": 64018,
      "paper explains": 83911,
      "ner framework": 77045,
      "approach ner": 7735,
      "semisupervised neural": 105618,
      "e2e systems": 34364,
      "trained deep": 123111,
      "strong deep": 113666,
      "need large": 76827,
      "carefully annotated": 15516,
      "data reach": 26316,
      "reach satisfactory": 95899,
      "performance acquiring": 86121,
      "timeconsuming task": 122148,
      "semisupervised deep": 105599,
      "nonannotated data": 80152,
      "open way": 81947,
      "datasets current": 27388,
      "development new": 30406,
      "transformers stateoftheart": 124463,
      "processing recent": 91780,
      "progress natural": 92162,
      "driven advances": 34148,
      "pretraining transformer": 90347,
      "models pretraining": 73791,
      "opensource library": 82020,
      "art transformer": 9098,
      "unified api": 127102,
      "clinical sentiment": 17787,
      "risk prediction": 102879,
      "prediction predicting": 89103,
      "30 days": 901,
      "information clinical": 54418,
      "clinical decisionmaking": 17765,
      "task previously": 118561,
      "previously explored": 90598,
      "paper assess": 83743,
      "patients efficient": 85706,
      "automatically unlabeled": 10838,
      "increase diversity": 53594,
      "methods attempt": 69322,
      "data good": 25984,
      "good coverage": 48472,
      "features utterance": 43780,
      "selection experiments": 104782,
      "performance addition": 86125,
      "data cases": 25720,
      "right answer": 102827,
      "reason recent": 96204,
      "learn exploit": 62044,
      "exploit annotation": 41406,
      "biases current": 13866,
      "current datasets": 25271,
      "measuring progress": 67923,
      "task evaluating": 118157,
      "derivations explanations": 29333,
      "answers present": 6740,
      "rc datasets": 95870,
      "create publicly": 24636,
      "questions annotated": 95269,
      "annotated reference": 6222,
      "derivations experiments": 29332,
      "multiple reference": 75658,
      "benchmark zeroshot": 12874,
      "zeroshot dependency": 135495,
      "offtheshelf deep": 81624,
      "development unsupervised": 30435,
      "experiments outperform": 41050,
      "systems shared": 117138,
      "changing training": 16401,
      "languages ii": 60624,
      "languages zeroshot": 60986,
      "finegrained classification": 44341,
      "contemporary datasets": 21847,
      "public health": 94261,
      "health mentions": 49622,
      "create dataset": 24609,
      "based presence": 11943,
      "based semantics": 12013,
      "created annotated": 24654,
      "annotated based": 6141,
      "hierarchical methods": 49975,
      "methods dataset": 69418,
      "dataset designing": 26866,
      "designing experiments": 29638,
      "understanding sentiment": 126955,
      "makes dataset": 66768,
      "monolingual alignment": 74548,
      "problem adapting": 90917,
      "word vectorbased": 133632,
      "reference data": 97511,
      "data adapt": 25568,
      "slightly different": 110021,
      "language distribution": 58975,
      "frame adaptation": 45381,
      "vector alignment": 131294,
      "simply average": 109621,
      "models alignment": 72740,
      "align vectors": 4906,
      "results simple": 102190,
      "algorithm allows": 4692,
      "adapting generalpurpose": 3309,
      "generalpurpose models": 46888,
      "models changing": 72897,
      "distributions evaluation": 32761,
      "communication people": 19038,
      "malicious purposes": 66886,
      "fake review": 43022,
      "using texts": 130288,
      "humanwritten texts": 51134,
      "meanings proposed": 67747,
      "text based": 120680,
      "evaluation detecting": 39179,
      "method detects": 68771,
      "accuracy higher": 2176,
      "accuracy achieve": 2091,
      "results fscore": 101801,
      "experiments related": 41113,
      "related japanese": 97869,
      "richresource languages": 102819,
      "20 challenging": 495,
      "fewshot relation": 43910,
      "classification present": 17337,
      "task investigate": 118310,
      "models adapt": 72714,
      "adapt new": 3187,
      "relations construct": 98128,
      "fewrel dataset": 43895,
      "choice new": 16893,
      "experimental analysis": 40524,
      "analysis stateoftheart": 5873,
      "stateoftheart fewshot": 112662,
      "challenges research": 16204,
      "dataset baselines": 26760,
      "baselines released": 12454,
      "wngt 2019": 132877,
      "efficiency task": 35047,
      "network vaswani": 77470,
      "able eliminate": 1634,
      "11 bleu": 215,
      "challenge datasets": 16018,
      "datasets taskoriented": 27747,
      "conversational scenarios": 23027,
      "promising performances": 92288,
      "address interpretability": 3683,
      "creating challenge": 24694,
      "using dialogue": 129564,
      "allows generating": 5159,
      "generating large": 47228,
      "large synthetic": 61286,
      "data taking": 26536,
      "control generation": 22805,
      "process neural": 91536,
      "patterns propose": 85773,
      "built single": 14940,
      "recent advancements": 96390,
      "use semisupervised": 128263,
      "cuttingedge methods": 25467,
      "architectures bert": 8783,
      "learn nuances": 62105,
      "bayes decision": 12567,
      "coupled traditional": 24459,
      "lstm able": 65588,
      "transformer variants": 124393,
      "leverage knowledge": 63591,
      "knowledge gained": 57944,
      "performance target": 86783,
      "providing succinct": 94139,
      "understanding recent": 126934,
      "learning potential": 62897,
      "knowledge gap": 57945,
      "ontologybased semantic": 81855,
      "ontology knowledge": 81844,
      "evaluated impact": 38980,
      "pretraining bert": 90236,
      "performance finetuning": 86387,
      "causal reasoning": 15809,
      "knowledge obtained": 58089,
      "obtaining new": 81444,
      "tasks linguistic": 119270,
      "evaluation germanenglish": 39223,
      "suite present": 115418,
      "results application": 101509,
      "107 phenomena": 195,
      "phenomena organized": 87239,
      "organized 14": 82475,
      "14 categories": 312,
      "test items": 120464,
      "particular systems": 85454,
      "context feature": 22104,
      "entities belonging": 37750,
      "class based": 17030,
      "task knowledge": 118328,
      "numerous downstream": 81017,
      "answering web": 6723,
      "set previous": 107542,
      "approaches make": 8228,
      "core challenge": 23313,
      "freetext corpora": 45795,
      "corpora lead": 23516,
      "tackles problem": 117320,
      "method selects": 69117,
      "unsupervised ensemble": 127633,
      "set based": 107372,
      "basic tasks": 12536,
      "tasks computational": 118997,
      "identify word": 51576,
      "unsupervised monolingual": 127673,
      "settings small": 107837,
      "small available": 110135,
      "possible remedy": 88425,
      "wellresourced language": 132429,
      "language exist": 59023,
      "exist data": 40019,
      "explore compare": 41531,
      "compare ways": 19315,
      "ways exploit": 132157,
      "function jointly": 46030,
      "segmentation experiment": 104587,
      "techniques effectively": 119876,
      "effectively control": 34802,
      "control output": 22815,
      "output segmentation": 83116,
      "variance adversarial": 130863,
      "adversarial datasets": 4110,
      "datasets recently": 27663,
      "recently designed": 96678,
      "perform targeted": 86085,
      "inference types": 54243,
      "work liu": 134617,
      "al 2019a": 4647,
      "proposed datasets": 93252,
      "focus improve": 44773,
      "improve training": 52564,
      "different distribution": 31099,
      "distribution having": 32656,
      "complexity level": 19922,
      "level work": 63524,
      "conclusions models": 20728,
      "learn generalize": 62055,
      "inference phenomena": 54194,
      "dative alternation": 27838,
      "enables building": 36378,
      "better challenge": 13531,
      "datasets creating": 27387,
      "creating robust": 24707,
      "unigram distribution": 127155,
      "word2vec popular": 133683,
      "popular model": 88103,
      "widely investigated": 132547,
      "tasks proposing": 119419,
      "task resulting": 118655,
      "current understanding": 25384,
      "detection tackle": 30072,
      "problem relation": 91202,
      "relation graph": 98039,
      "existing structured": 40300,
      "task targets": 118777,
      "events use": 39608,
      "detect event": 29804,
      "overlapping nested": 83337,
      "search process": 104339,
      "global properties": 48263,
      "f1score performance": 42717,
      "unsupervised multilingual": 127676,
      "multilingual contextualized": 75223,
      "contextualized representations": 22565,
      "work crosslingual": 134450,
      "tasks requires": 119464,
      "large linguistic": 61133,
      "semantic alignment": 104973,
      "languageindependent representations": 60369,
      "labeling methods": 58510,
      "advantage deep": 4039,
      "ner pos": 77067,
      "comparison quality": 19567,
      "quality indicators": 94688,
      "predicting quality": 89003,
      "task presented": 118551,
      "problem measuring": 91125,
      "performance document": 86309,
      "features including": 43560,
      "content textual": 21962,
      "features directly": 43460,
      "available text": 11126,
      "features inform": 43566,
      "usergenerated data": 129072,
      "data tweets": 26584,
      "academic articles": 1854,
      "showing features": 108404,
      "mrc framework": 74926,
      "nested ner": 77109,
      "flat ner": 44636,
      "developed tasks": 30307,
      "ner able": 77017,
      "label particular": 58405,
      "assigned labels": 9549,
      "capable handling": 15202,
      "tasks instead": 119215,
      "instead treating": 55690,
      "treating task": 125565,
      "task ner": 118450,
      "ner sequence": 77076,
      "formulate machine": 45273,
      "comprehension mrc": 20199,
      "mrc task": 74930,
      "task example": 118164,
      "extracting answer": 42200,
      "answer spans": 6582,
      "question person": 95198,
      "person mentioned": 87098,
      "entities different": 37767,
      "additionally query": 3627,
      "query encodes": 94956,
      "encodes informative": 36670,
      "informative prior": 55146,
      "knowledge strategy": 58187,
      "strategy facilitates": 113514,
      "facilitates process": 42803,
      "process entity": 91473,
      "extraction leading": 42370,
      "leading better": 61885,
      "performance boost": 86188,
      "current sota": 25337,
      "sota models": 110692,
      "ace04 ace05": 2399,
      "respectively english": 101134,
      "conll 2003": 21073,
      "2003 english": 544,
      "english ontonotes": 37229,
      "msra chinese": 74952,
      "present automated": 89386,
      "combines various": 18702,
      "classifier human": 17547,
      "metrics measuring": 69986,
      "datasets evaluating": 27456,
      "developed contextual": 30262,
      "examine different": 39741,
      "methods aggregating": 69302,
      "classification attentionbased": 17128,
      "finetuning training": 44525,
      "labeled text": 58472,
      "substantial time": 114878,
      "time converge": 122001,
      "aim develop": 4452,
      "develop effective": 30192,
      "model goal": 71257,
      "goal provide": 48385,
      "feature extractors": 43283,
      "propose attentionbased": 92558,
      "finetuning algorithm": 44444,
      "selects relevant": 104863,
      "contextualized features": 22550,
      "uses features": 129229,
      "methods widelyused": 69848,
      "benchmarking datasets": 12878,
      "datasets achieve": 27292,
      "performance introduce": 86469,
      "approach endtoend": 7527,
      "endtoend algorithm": 36869,
      "given pretrained": 48086,
      "doing multitask": 33419,
      "total training": 122701,
      "pronouns english": 92350,
      "english texts": 37308,
      "english challenging": 37082,
      "task potential": 118539,
      "written english": 135125,
      "english does": 37120,
      "distinguish cases": 32567,
      "languages spanish": 60885,
      "english phrases": 37240,
      "make distinction": 66660,
      "use obtain": 128180,
      "domains following": 33783,
      "lot room": 65328,
      "improvement especially": 52704,
      "extremely challenging": 42596,
      "video game": 131657,
      "models existing": 73191,
      "datatotext datasets": 27812,
      "limited diversity": 64230,
      "models advantage": 72728,
      "corpus 7k": 23636,
      "act types": 3072,
      "making suitable": 66866,
      "explores domain": 41646,
      "video games": 131658,
      "supporting rich": 116031,
      "models poor": 73752,
      "heavytailed distribution": 49669,
      "distribution propose": 32675,
      "yield high": 135339,
      "distance vectors": 32435,
      "conducted evaluate": 20921,
      "largescale realworld": 61504,
      "dataset 10000": 26705,
      "written description": 135120,
      "employ deep": 36257,
      "networks predict": 77701,
      "opensource data": 82018,
      "community present": 19085,
      "language variations": 60322,
      "interactions interlocutors": 55988,
      "precisely capture": 88785,
      "specifically employ": 111544,
      "contexts different": 22387,
      "memory enhance": 68313,
      "finally construct": 44162,
      "construct corpus": 21622,
      "easyfirst dependency": 34488,
      "embeddings characterlevel": 35591,
      "vietnamese vietnamese": 131700,
      "uses deep": 129218,
      "model reported": 71905,
      "embeddings method": 35805,
      "framework building": 45447,
      "chat dialogue": 16655,
      "framework developing": 45494,
      "chat dialogues": 16656,
      "dialogues effective": 30829,
      "systems useful": 117220,
      "developed evaluated": 30273,
      "reasonably good": 96218,
      "model structures": 72094,
      "tasks training": 119571,
      "training generative": 123641,
      "challenges building": 16138,
      "methods tend": 69795,
      "tend use": 120164,
      "use metalearning": 128141,
      "metalearning framework": 68550,
      "task fewshot": 118207,
      "fewshot setting": 43914,
      "shared module": 107990,
      "module gating": 74491,
      "module private": 74504,
      "private module": 90786,
      "modules shared": 74529,
      "structures better": 114061,
      "capture characteristics": 15273,
      "corresponding task": 24309,
      "consistency response": 21366,
      "adaptation robustness": 3259,
      "robustness share": 103131,
      "task encourage": 118144,
      "encourage research": 36762,
      "robustness domain": 103094,
      "adaptation realworld": 3257,
      "greatly beneficial": 49047,
      "discuss challenges": 32240,
      "good baseline": 48463,
      "latest techniques": 61659,
      "robustness perform": 103123,
      "human shows": 50963,
      "online systems": 81810,
      "systems finally": 116906,
      "generalization memorization": 46779,
      "model nearest": 71561,
      "drawn text": 34137,
      "including original": 53345,
      "stateoftheart perplexity": 112866,
      "model particularly": 71715,
      "particularly helpful": 85483,
      "learning similarity": 63027,
      "sequences text": 107141,
      "text easier": 120895,
      "modeling long": 72466,
      "generation produce": 47569,
      "produce sequence": 91930,
      "general model": 46679,
      "arbitrary order": 8588,
      "order model": 82372,
      "decoding order": 28117,
      "order result": 82400,
      "result training": 101416,
      "models number": 73652,
      "captioning machine": 15261,
      "entailment machine": 37666,
      "generated translation": 47143,
      "translation provided": 125158,
      "learning determine": 62497,
      "candidate reference": 15118,
      "translation generating": 124838,
      "relying simple": 98767,
      "simple ngram": 109486,
      "ngram overlap": 79451,
      "implementation transformer": 51951,
      "mnli corpus": 70462,
      "apply evaluation": 7174,
      "better correlation": 13553,
      "score compared": 104062,
      "traditional metrics": 122841,
      "pretraining chinese": 90239,
      "text encoder": 120903,
      "representations pretraining": 99819,
      "text encoders": 120904,
      "word pieces": 133408,
      "characters chinese": 16601,
      "information carried": 54408,
      "text granularity": 121030,
      "easily adapt": 34429,
      "certain combinations": 15935,
      "loss important": 65278,
      "information especially": 54541,
      "especially problematic": 38492,
      "considered training": 21307,
      "training result": 123820,
      "potential word": 88600,
      "encoder bert": 36499,
      "comprehensive information": 20264,
      "information character": 54412,
      "character sequence": 16463,
      "phrases contains": 87416,
      "chinese nlp": 16797,
      "corpus important": 23830,
      "pretraining techniques": 90339,
      "generalization tasks": 46797,
      "models semantics": 73992,
      "using naturalistic": 129942,
      "naturalistic corpora": 76629,
      "adversarial methods": 4143,
      "methods models": 69624,
      "evaluated new": 38996,
      "new examples": 78914,
      "known semantic": 58322,
      "given data": 48011,
      "taskspecific models": 119634,
      "translation community": 124698,
      "community focus": 19074,
      "fact good": 42825,
      "space existing": 111000,
      "existing translation": 40325,
      "directly translating": 31913,
      "new categories": 78821,
      "pronunciation sentence": 92368,
      "translating text": 124607,
      "text sentence": 121281,
      "compare traditional": 19305,
      "text translation": 121377,
      "translations comparable": 125457,
      "coreference phenomena": 23355,
      "access explicit": 1907,
      "genres news": 47837,
      "outputs evaluate": 83165,
      "coreference chain": 23342,
      "typology aims": 126504,
      "incorrect word": 53581,
      "features coreference": 43429,
      "automatic translations": 10705,
      "translations compared": 125459,
      "compared source": 19446,
      "shows stronger": 108634,
      "outputs human": 83170,
      "systems user": 117221,
      "critical metric": 24817,
      "systems prior": 117076,
      "using heuristically": 129732,
      "heuristically constructed": 49879,
      "features number": 43633,
      "number turns": 80988,
      "high agreement": 50038,
      "scores predicted": 104192,
      "learned data": 62209,
      "scores improve": 104179,
      "systems shown": 117142,
      "structure pretrained": 113936,
      "problem multilingual": 91133,
      "text multiple": 121134,
      "study factors": 114381,
      "factors influence": 42893,
      "effective crosslingual": 34654,
      "contrary previously": 22684,
      "previously hypothesized": 90601,
      "text comes": 120801,
      "comes different": 18777,
      "parameters layers": 84762,
      "independently trained": 53790,
      "noncontextual word": 80181,
      "automatically discovered": 10761,
      "process improving": 91508,
      "bidirectional decoding": 13905,
      "decoding dynamic": 28086,
      "translation generally": 124835,
      "fail exploit": 42958,
      "semantics information": 105426,
      "information usually": 55081,
      "decoder bidirectional": 27997,
      "decoding alleviate": 28075,
      "regularization term": 97759,
      "translations experimental": 125464,
      "model fewer": 71177,
      "understanding knowledge": 126872,
      "nonautoregressive machine": 80155,
      "translation nonautoregressive": 125066,
      "translation nat": 124999,
      "nat systems": 76208,
      "tokens parallel": 122325,
      "speed compared": 111863,
      "compared autoregressive": 19325,
      "autoregressive models": 10869,
      "existing nat": 40219,
      "nat models": 76207,
      "data pretrained": 26270,
      "autoregressive model": 10868,
      "performance knowledge": 86477,
      "empirically useful": 36246,
      "models reason": 73869,
      "paper design": 83859,
      "systematic experiments": 116673,
      "experiments investigate": 40966,
      "training knowledge": 123669,
      "model variations": 72296,
      "output data": 83064,
      "quality based": 94604,
      "propose approaches": 92551,
      "wmt14 ende": 132808,
      "benchmark improving": 12831,
      "improving grammatical": 53100,
      "data synthesis": 26528,
      "synthesis method": 116598,
      "correction based": 24144,
      "poor good": 88057,
      "language learner": 59239,
      "tends generate": 120172,
      "model generally": 71232,
      "fluent grammatically": 44707,
      "model respectively": 71918,
      "pair translations": 83463,
      "pseudo parallel": 94182,
      "approach capable": 7409,
      "rule set": 103354,
      "set seed": 107572,
      "task iwslt": 118322,
      "iwslt 2019": 57163,
      "present submission": 89725,
      "aims study": 4576,
      "input embeddings": 55329,
      "transformerbased encoderdecoder": 124409,
      "able use": 1703,
      "data constrained": 25785,
      "bert external": 13124,
      "model configuration": 70883,
      "bert architecture": 13065,
      "architecture training": 8762,
      "regarding training": 97661,
      "text set": 121290,
      "news commentary": 79318,
      "observed using": 81239,
      "bleu test": 14341,
      "translation performances": 125118,
      "sparse text": 111236,
      "desirable property": 29661,
      "property representations": 92504,
      "dense representations": 29017,
      "progresses recent": 92188,
      "based dense": 11634,
      "transformation paper": 124270,
      "transformation method": 124266,
      "facilitate nlp": 42783,
      "space jointly": 111010,
      "idea proposed": 51336,
      "representations sparse": 99902,
      "space performed": 111041,
      "perform downstream": 85984,
      "diverges training": 32791,
      "domain key": 33554,
      "statistical smt": 113164,
      "nmt paper": 79939,
      "sets unknown": 107721,
      "suffer different": 115216,
      "adequate fluent": 3832,
      "unrelated source": 127504,
      "problem empirically": 91022,
      "methods reported": 69725,
      "improve adequacy": 52334,
      "indomain robustness": 53980,
      "robustness terms": 103136,
      "improving domain": 53092,
      "slightly increase": 110024,
      "adequacy translations": 3829,
      "smt general": 110284,
      "despite success": 29734,
      "simultaneous neural": 109659,
      "translation snmt": 125276,
      "framework contains": 45472,
      "stopping criterion": 113375,
      "determines stop": 30143,
      "translation corpora": 124714,
      "nmt perform": 79943,
      "open natural": 81916,
      "recently largescale": 96704,
      "trained parser": 123236,
      "annotation new": 6342,
      "quality particularly": 94749,
      "research evaluation": 100489,
      "selection training": 104842,
      "data consolidation": 25782,
      "new gold": 78938,
      "annotations good": 6437,
      "sampling selftraining": 103606,
      "selftraining neural": 104959,
      "tasks order": 119353,
      "models truly": 74252,
      "truly useful": 125865,
      "correctly generating": 24175,
      "mrs test": 74938,
      "time practice": 122072,
      "forms semantic": 45259,
      "fail generate": 42963,
      "faithful input": 43003,
      "method sample": 69113,
      "utterance pairs": 130608,
      "training augmented": 123366,
      "augmented data": 10324,
      "stateoftheart outputs": 112798,
      "quality decoding": 94628,
      "strategies affect": 113449,
      "text recent": 121229,
      "progress pretrained": 92173,
      "models led": 73495,
      "works investigated": 134955,
      "correctness models": 24186,
      "models unclear": 74262,
      "extent generated": 41975,
      "generated stateoftheart": 47125,
      "models generated": 73290,
      "generated sentence": 47119,
      "strongly depends": 113744,
      "depends decoding": 29284,
      "ability generating": 1525,
      "generating wikipedia": 47284,
      "topk nucleus": 122670,
      "nucleus sampling": 80828,
      "effective decoding": 34658,
      "comparison previously": 19565,
      "used decoding": 128476,
      "generation unseen": 47688,
      "tasks focus": 119136,
      "focus generating": 44769,
      "input tables": 55448,
      "aims test": 4578,
      "test generalization": 120458,
      "types appear": 126250,
      "training construct": 123394,
      "table schemas": 117261,
      "table representations": 117259,
      "demonstrates model": 28948,
      "methods large": 69576,
      "margin addition": 67189,
      "comparison standard": 19574,
      "proposed task": 93564,
      "reasoning sentences": 96309,
      "relationships entailment": 98319,
      "alignment mechanism": 4967,
      "mechanism effectively": 67971,
      "helps nli": 49827,
      "parts similar": 85583,
      "segments sentence": 104674,
      "entailment contradiction": 37656,
      "multiple perspectives": 75638,
      "holistic view": 50602,
      "propose multiperspective": 92802,
      "novel nli": 80675,
      "perspectives different": 87174,
      "parts sentences": 85582,
      "makes final": 66782,
      "introduce auxiliary": 56385,
      "perspectives experiments": 87177,
      "snli multinli": 110316,
      "bert ebert": 13105,
      "bert present": 13204,
      "novel way": 80774,
      "knowledge entities": 57902,
      "entity vectors": 38132,
      "aligned entity": 4913,
      "vectors wordpiece": 131491,
      "wordpiece vectors": 133793,
      "vectors resulting": 131473,
      "version bert": 131604,
      "expensive pretraining": 40429,
      "bert encoder": 13113,
      "unsupervised question": 127698,
      "classification rc": 17365,
      "linking el": 64656,
      "original bert": 82505,
      "form entity": 45085,
      "entity names": 37976,
      "mitigates problem": 70380,
      "generalpurpose pretrained": 46890,
      "ai applications": 4419,
      "task effectively": 118129,
      "model convert": 70919,
      "selection tasks": 104839,
      "finally pretrained": 44220,
      "pretrained representations": 90178,
      "representations proposed": 99831,
      "proposed encoder": 93264,
      "yielding strong": 135388,
      "results diverse": 101735,
      "dual encoders": 34236,
      "reduced size": 97374,
      "performance believe": 86169,
      "believe model": 12699,
      "canonical correlation": 15152,
      "datasets gains": 27492,
      "different applications": 30997,
      "present thorough": 89743,
      "thorough error": 121882,
      "analysis dependency": 5567,
      "language vietnamese": 60332,
      "present effective": 89459,
      "evaluation vietnamese": 39441,
      "achieved improvement": 2649,
      "documentlevel model": 33151,
      "argument spans": 8959,
      "creation new": 24721,
      "types demonstrate": 126272,
      "demonstrate strong": 28874,
      "model rams": 71840,
      "covers wide": 24558,
      "single unified": 109808,
      "unified format": 127107,
      "labeling spans": 58542,
      "spans relations": 111208,
      "parsing syntax": 85254,
      "semantics relation": 105455,
      "aspect based": 9316,
      "achieving performance": 2967,
      "stateoftheart specialized": 112965,
      "demonstrate benefits": 28678,
      "benefits multitask": 13014,
      "analyze differences": 5963,
      "differences similarities": 30973,
      "testbed evaluating": 120556,
      "models generalized": 73280,
      "language multilingual": 59696,
      "nmt machine": 79888,
      "pose question": 88255,
      "question perform": 95197,
      "perform zeroshot": 86100,
      "translation original": 125077,
      "text original": 121157,
      "data original": 26195,
      "sentencelevel classifiers": 106148,
      "data nmt": 26173,
      "using technique": 130273,
      "technique bias": 119773,
      "bias model": 13819,
      "produce natural": 91911,
      "gains human": 46394,
      "demonstrate possible": 28818,
      "quality analyze": 94591,
      "analyze models": 5987,
      "translationese output": 125447,
      "models ubiquitous": 74259,
      "ubiquitous natural": 126510,
      "data concatenation": 25766,
      "english limited": 37195,
      "feasibility training": 43238,
      "training monolingual": 123717,
      "transformerbased language": 124412,
      "parsing named": 85166,
      "crawled data": 24597,
      "use wikipedia": 128358,
      "results good": 101815,
      "improves state": 53046,
      "outputs generating": 83169,
      "input specification": 55442,
      "implement different": 51930,
      "control mechanisms": 22812,
      "improve semantic": 52536,
      "maintaining fluency": 66514,
      "fluency common": 44695,
      "highquality parallel": 50405,
      "sentences web": 106544,
      "bitext mining": 14238,
      "sentence space": 106086,
      "space applied": 110970,
      "billions sentences": 14084,
      "crawl corpus": 24594,
      "using unified": 130335,
      "unified approach": 127103,
      "languages able": 60381,
      "20 language": 508,
      "million parallel": 70101,
      "quality mined": 94719,
      "human translated": 50977,
      "wmt19 test": 132863,
      "translation english": 124791,
      "evaluation uses": 39434,
      "achieve excellent": 2455,
      "results distant": 101733,
      "outperforming best": 82802,
      "best submission": 13457,
      "2019 workshop": 717,
      "translation wat": 125425,
      "defining words": 28514,
      "words textual": 134266,
      "useful task": 128938,
      "task practical": 118540,
      "distributional hypothesis": 32701,
      "argue natural": 8923,
      "sequencetosequence task": 107209,
      "implement approach": 51928,
      "transformerbased sequencetosequence": 124438,
      "model proposal": 71809,
      "generation endtoend": 47386,
      "results contextual": 101634,
      "contextual noncontextual": 22484,
      "models handling": 73319,
      "widely utilized": 132597,
      "utilized various": 130549,
      "local contextual": 64913,
      "units enhance": 127240,
      "enhance ability": 37463,
      "model sentencelevel": 71976,
      "documentlevel modeling": 33152,
      "classification reading": 17367,
      "various stateoftheart": 131204,
      "showing promising": 108425,
      "tasks enhanced": 119098,
      "transferring knowledge": 124238,
      "ner existing": 77043,
      "methods directly": 69449,
      "directly transfer": 31911,
      "propose finetune": 92680,
      "finetune learned": 44408,
      "examples given": 39841,
      "examples end": 39829,
      "model parameter": 71703,
      "fast adapt": 43128,
      "construct multiple": 21640,
      "sentence similarities": 106072,
      "ability different": 1507,
      "languages introduce": 60649,
      "masking scheme": 67313,
      "maximum term": 67552,
      "generation literature": 47462,
      "review recently": 102568,
      "gaining increasing": 46376,
      "effectively exploit": 34812,
      "unlabelled corpus": 127416,
      "power unsupervised": 88656,
      "learned pretraining": 62244,
      "pretraining learning": 90276,
      "nontrivial problem": 80314,
      "problem review": 91208,
      "summarizes recent": 115621,
      "recent efforts": 96458,
      "integration pretrained": 55815,
      "lines work": 64408,
      "using contextdependent": 129480,
      "embeddings lexical": 35788,
      "process correcting": 91446,
      "wordlevel models": 133745,
      "models seldom": 73984,
      "seldom used": 104686,
      "knowledge research": 58150,
      "investigated effectiveness": 56835,
      "results characterlevel": 101569,
      "deeplearning based": 28457,
      "embeddings capable": 35583,
      "capable outperforming": 15209,
      "models certain": 72891,
      "scenarios finally": 103850,
      "dataset community": 26803,
      "recent challenges": 96437,
      "focuses applying": 44887,
      "applying endtoend": 7240,
      "dialog technologies": 30608,
      "visual sceneaware": 131808,
      "sceneaware dialog": 103879,
      "describes task": 29438,
      "definition provided": 28526,
      "setup track": 107859,
      "track summarize": 122736,
      "summarize results": 115613,
      "results submitted": 102220,
      "systems highlight": 116931,
      "highlight overall": 50267,
      "trends stateoftheart": 125755,
      "stateoftheart technologies": 113004,
      "technologies tasks": 120028,
      "hierarchical network": 49986,
      "distractors reading": 32604,
      "comprehension reading": 20224,
      "requires deep": 100258,
      "deep understanding": 28437,
      "question traditional": 95228,
      "creative text": 24731,
      "distractor generation": 32600,
      "ignore important": 51667,
      "question making": 95180,
      "generated distractors": 47075,
      "relevant question": 98573,
      "semantically relevant": 105374,
      "relevant article": 98526,
      "set meaningful": 107490,
      "hierarchical architecture": 49928,
      "guide decoder": 49231,
      "problem add": 90919,
      "add additional": 3348,
      "similarity loss": 109252,
      "baselines automatic": 12356,
      "metrics achieving": 69918,
      "compared distractors": 19359,
      "distractors generated": 32602,
      "generated baselines": 47056,
      "baselines controlling": 12377,
      "controlling neural": 22856,
      "supervision work": 115931,
      "aims produce": 4560,
      "produce translations": 91943,
      "language content": 58918,
      "problem neural": 91144,
      "training triplets": 123939,
      "labeled target": 58471,
      "examples limited": 39854,
      "different styles": 31460,
      "scheme multitask": 103932,
      "models automatically": 72795,
      "automatically generates": 10780,
      "generates synthetic": 47181,
      "missing element": 70324,
      "comprehensive automatic": 20247,
      "models producing": 73812,
      "translations better": 125455,
      "better match": 13625,
      "match desired": 67356,
      "preserving source": 89941,
      "source meaning": 110790,
      "generation challenges": 47330,
      "explainable ai": 41265,
      "uncertainty data": 126615,
      "highlight specific": 50274,
      "robust extraction": 103033,
      "idiomatic expressions": 51651,
      "expressions like": 41755,
      "range difficulties": 95560,
      "difficulties natural": 31683,
      "work annotation": 134371,
      "annotation extraction": 6312,
      "coverage different": 24500,
      "research progress": 100590,
      "required addition": 100211,
      "potential source": 88585,
      "valuable linguistic": 130766,
      "variability propose": 130832,
      "corpora investigating": 23505,
      "feasibility using": 43239,
      "reliable source": 98622,
      "high reliability": 50127,
      "fleiss kappa": 44644,
      "accurate performance": 2352,
      "combining complementary": 18712,
      "languages given": 60610,
      "given sufficient": 48142,
      "tools available": 122439,
      "available empirical": 10989,
      "infectious diseases": 54095,
      "necessary step": 76759,
      "step automatic": 113251,
      "articles available": 9172,
      "form news": 45111,
      "articles usually": 9226,
      "loosely structured": 65233,
      "sentence achieve": 105762,
      "text achieves": 120628,
      "highest recall": 50232,
      "information help": 54652,
      "accuracy analysis": 2099,
      "analysis observation": 5725,
      "useful range": 128919,
      "range domains": 95561,
      "psychological research": 94214,
      "machine learningbased": 65837,
      "learningbased approaches": 63179,
      "task estimating": 118154,
      "controlling information": 22855,
      "studied especially": 114158,
      "especially spoken": 38513,
      "determines appropriate": 30142,
      "human expert": 50855,
      "reallife conversations": 96128,
      "annotated large": 6206,
      "large diverse": 61081,
      "set behavior": 107376,
      "observed behaviors": 81217,
      "negative positive": 76944,
      "language findings": 59047,
      "consistent different": 21375,
      "word recognition": 133429,
      "systems increasingly": 116953,
      "increasingly important": 53700,
      "important given": 52167,
      "recognition framework": 96883,
      "framework captures": 45453,
      "framework employs": 45516,
      "employs sequencetosequence": 36333,
      "word output": 133388,
      "output given": 83079,
      "wordlevel bidirectional": 133730,
      "network conduct": 77207,
      "experiments verify": 41200,
      "effectiveness framework": 34894,
      "framework outperforms": 45639,
      "role word": 103224,
      "conversation neural": 22964,
      "bland generic": 14263,
      "researchers propose": 100705,
      "use conditional": 127955,
      "model expected": 71123,
      "expected generate": 40390,
      "potential improvement": 88564,
      "unsatisfactory paper": 127512,
      "introduce discrete": 56412,
      "meaning improve": 67640,
      "major advantage": 66531,
      "model exploit": 71141,
      "exploit semantic": 41440,
      "sampling approach": 103600,
      "approach enable": 7521,
      "enable efficient": 36352,
      "selection large": 104793,
      "indicate model": 53841,
      "important sentences": 52251,
      "field research": 43977,
      "identify summarize": 51564,
      "reports paper": 99085,
      "investigate important": 56767,
      "important properties": 52218,
      "crossvalidation using": 25088,
      "machine algorithm": 65739,
      "algorithm shows": 4800,
      "shows using": 108642,
      "using raw": 130081,
      "suggest using": 115337,
      "new contexts": 78839,
      "contexts humans": 22397,
      "humans rapidly": 51105,
      "new linguistic": 78996,
      "linguistic conventions": 64455,
      "humans introduce": 51084,
      "introduce interactive": 56438,
      "repeated reference": 98910,
      "reference task": 97536,
      "benchmark models": 12843,
      "communication propose": 19040,
      "accurately efficiently": 2381,
      "time evaluate": 122024,
      "reference game": 97518,
      "game experiments": 46419,
      "human partners": 50925,
      "detection data": 29916,
      "augmentation based": 10274,
      "intelligent assistant": 55844,
      "assistant systems": 9574,
      "systems essential": 116875,
      "utterances automatically": 130625,
      "handle noisy": 49396,
      "typical approach": 126394,
      "approach introducing": 7648,
      "examples combined": 39814,
      "text samples": 121271,
      "utterances usually": 130675,
      "performance largely": 86490,
      "based sampling": 12002,
      "sample size": 103565,
      "detection accuracy": 29866,
      "experiments given": 40954,
      "performance improves": 86454,
      "structure modeling": 113907,
      "using masked": 129861,
      "hierarchical transformer": 50017,
      "useful understanding": 128943,
      "understanding nature": 126901,
      "providing features": 94115,
      "conversations work": 23078,
      "modeling identifying": 72443,
      "make accurate": 66618,
      "model aggregate": 70646,
      "reddit dataset": 97300,
      "addition report": 3471,
      "experiments new": 41039,
      "including bert": 53266,
      "model datasets": 70950,
      "dataset human": 26967,
      "human dialogue": 50792,
      "different sentences": 31414,
      "sentences used": 106535,
      "humancomputer interactions": 51007,
      "meaning utterance": 67703,
      "speakers utterance": 111333,
      "utterance explicitly": 130597,
      "dialogue snippets": 30742,
      "context utterance": 22317,
      "utterances collected": 130628,
      "listening comprehension": 64709,
      "tests like": 120609,
      "toefl test": 122242,
      "test english": 120451,
      "dialogues movie": 30838,
      "scripts available": 104267,
      "enabling neural": 36414,
      "target prediction": 117691,
      "translations noisy": 125479,
      "solve challenges": 110588,
      "words according": 133801,
      "according word": 2009,
      "sentences additional": 106201,
      "parameters primary": 84778,
      "implement models": 51935,
      "demonstrate combination": 28688,
      "texts research": 121593,
      "difficulties posed": 31687,
      "nonstandard writing": 80297,
      "simple extension": 109430,
      "introduce concept": 56394,
      "words added": 133807,
      "standard words": 112340,
      "models noisy": 73648,
      "range evaluation": 95565,
      "tasks intrinsic": 119220,
      "effectiveness sequencetosequence": 34950,
      "generating generic": 47224,
      "responses responses": 101295,
      "generation enhance": 47388,
      "enhance generation": 37470,
      "performs additional": 86982,
      "tokenlevel representation": 122292,
      "query representation": 94983,
      "generation generic": 47423,
      "responses experimental": 101265,
      "superiority proposed": 115708,
      "metrics diversity": 69953,
      "explicit phrase": 41340,
      "phrase alignment": 87330,
      "unable capture": 126568,
      "hard interpret": 49464,
      "models key": 73437,
      "idea build": 51317,
      "nmt phrase": 79945,
      "design new": 29552,
      "new decoding": 78873,
      "algorithm easily": 4720,
      "lexical structural": 63822,
      "constraints experiments": 21598,
      "approach makes": 7701,
      "process nmt": 91540,
      "interpretable sacrificing": 56252,
      "quality addition": 94587,
      "using connectionist": 129474,
      "connectionist temporal": 21139,
      "temporal classification": 120093,
      "translation starts": 125290,
      "process end": 91470,
      "accuracy latency": 2194,
      "performance work": 86863,
      "adaptive manner": 3333,
      "manner proposed": 66958,
      "special token": 111371,
      "model chooses": 70824,
      "input token": 55455,
      "generating output": 47244,
      "output token": 83137,
      "function handle": 46022,
      "optimized using": 82236,
      "using algorithm": 129328,
      "classification ctc": 17166,
      "ctc use": 25196,
      "possible output": 88415,
      "sequences including": 107126,
      "japanese investigate": 57191,
      "collection bilingual": 18461,
      "dataset opensource": 27071,
      "python package": 94484,
      "sentencelevel parallel": 106166,
      "lexicon extraction": 63882,
      "words highly": 133983,
      "quality language": 94699,
      "library supports": 63964,
      "supported language": 116023,
      "person names": 87101,
      "languages source": 60881,
      "language introduce": 59224,
      "ranking results": 95684,
      "used noisy": 128664,
      "origin language": 82500,
      "mining social": 70259,
      "media sites": 68164,
      "results context": 101633,
      "english hebrew": 37166,
      "online web": 81817,
      "annotation existing": 6304,
      "written norwegian": 135142,
      "norwegian bokmaal": 80374,
      "bokmaal nynorsk": 14403,
      "types including": 126298,
      "geopolitical entities": 47872,
      "entities products": 37842,
      "products events": 92071,
      "events addition": 39561,
      "present details": 89450,
      "annotation effort": 6301,
      "longterm dependency": 65202,
      "simplified english": 109602,
      "modeling popular": 72510,
      "base task": 11484,
      "new architectures": 78803,
      "faster better": 43164,
      "better training": 13747,
      "languages larger": 60677,
      "larger dataset": 61360,
      "higher average": 50161,
      "times word": 122191,
      "word appears": 132924,
      "dataset datasets": 26845,
      "sizes different": 109959,
      "datasets perform": 27619,
      "search dataset": 104305,
      "smaller ones": 110241,
      "helps sentiment": 49830,
      "multitask attention": 75811,
      "network performs": 77388,
      "performs sentiment": 87021,
      "analysis emotion": 5591,
      "analysis proposed": 5780,
      "source external": 110758,
      "knowledge improve": 58006,
      "improve sentiment": 52538,
      "emotion prediction": 36109,
      "build meaningful": 14784,
      "task compare": 117984,
      "systems stance": 117154,
      "stance sentiment": 112175,
      "emotion corpus": 36099,
      "performance emotion": 86333,
      "encode semantic": 36450,
      "network nn": 77374,
      "largely rely": 61347,
      "dense word": 29022,
      "rich morphological": 102769,
      "large population": 61210,
      "plays essential": 87728,
      "essential role": 38564,
      "developing language": 30343,
      "embeddings corpus": 35620,
      "corpus challenging": 23690,
      "preprocessing pipeline": 89331,
      "skipgram sg": 109993,
      "words cbow": 133844,
      "representations intrinsic": 99703,
      "demonstrate high": 28753,
      "sg cbow": 107900,
      "studies text": 114291,
      "classification focused": 17212,
      "challenging multilingual": 16278,
      "informal noisy": 54340,
      "text learning": 121084,
      "aims learn": 4546,
      "lexical normalization": 63788,
      "transliteration model": 125525,
      "relies entirely": 98650,
      "text external": 120942,
      "utilized learning": 130545,
      "learning purpose": 62929,
      "text dataset": 120855,
      "dataset developed": 26870,
      "roman urdu": 103246,
      "urdu english": 127840,
      "dataset outperforms": 27079,
      "previous model": 90431,
      "languages internet": 60648,
      "german paper": 47905,
      "german text": 47919,
      "half million": 49297,
      "sentences generated": 106329,
      "using customized": 129528,
      "pages used": 83412,
      "fundamental importance": 46097,
      "processing experimental": 91670,
      "improvements task": 52928,
      "content approach": 21857,
      "understanding intent": 126863,
      "dialog managers": 30571,
      "conversational skills": 23030,
      "common user": 18945,
      "intent users": 55912,
      "forms utterances": 45264,
      "lack data": 58698,
      "comprehension performance": 20213,
      "demonstrate necessity": 28805,
      "necessity extracting": 76770,
      "accurate parsing": 2351,
      "building parallel": 14876,
      "following guidelines": 44970,
      "korean corpus": 58345,
      "type propose": 126220,
      "class imbalance": 17038,
      "applications corpus": 6909,
      "corpus generation": 23817,
      "method multilingual": 68975,
      "benchmark linguistic": 12835,
      "minimal pairs": 70171,
      "english introduce": 37174,
      "introduce benchmark": 56386,
      "set evaluating": 107433,
      "grammatical phenomena": 48715,
      "containing 1000": 21780,
      "morphology semantics": 74803,
      "semantics data": 105405,
      "generated according": 47046,
      "lms stateoftheart": 64890,
      "distribution quantifiers": 32676,
      "annotation manuals": 6335,
      "framework intended": 45582,
      "types simply": 126358,
      "10 dataset": 127,
      "dataset complex": 26810,
      "complex sentence": 19874,
      "dataset intended": 26983,
      "use languages": 128113,
      "languages dataset": 60483,
      "10 words": 155,
      "dataset able": 26715,
      "test semantic": 120489,
      "properties sentence": 92481,
      "preliminary analysis": 89263,
      "systems text": 117194,
      "generation amr": 47307,
      "reflect meaning": 97608,
      "amr annotations": 5385,
      "used decode": 128475,
      "directly encode": 31873,
      "local graph": 64925,
      "graph context": 48775,
      "addition english": 3426,
      "acquiring knowledge": 3043,
      "bert large": 13153,
      "scale unlabeled": 103760,
      "data finetuning": 25950,
      "model labeled": 71400,
      "far different": 43076,
      "different monolingual": 31268,
      "monolingual pretrained": 74607,
      "gap leads": 46461,
      "using finetuning": 129679,
      "finetuning nmt": 44487,
      "utilize prior": 130524,
      "includes modules": 53258,
      "adapted general": 3283,
      "nmt network": 79926,
      "network knowledge": 77293,
      "process proposed": 91555,
      "approach integrate": 7642,
      "nmt experimental": 79862,
      "wmt english": 132785,
      "german german": 47895,
      "review survey": 102582,
      "mt automatic": 74957,
      "translation written": 125436,
      "paradigm shift": 84548,
      "shift recent": 108180,
      "mainly relies": 66493,
      "relies various": 98664,
      "countbased models": 24419,
      "translation single": 125265,
      "single neural": 109773,
      "nmt architectures": 79835,
      "architectures word": 8859,
      "encoderdecoder network": 36607,
      "trends field": 125746,
      "effective data": 34656,
      "augmentation approaches": 10273,
      "approaches endtoend": 8144,
      "training taskoriented": 123905,
      "work augments": 134388,
      "expensive crowdsourcing": 40418,
      "different automatic": 31020,
      "augmentation word": 10317,
      "approaches able": 8039,
      "baseline terms": 12323,
      "terms success": 120388,
      "approaches achieves": 8043,
      "analyses confirm": 5445,
      "curated word": 25247,
      "kind texts": 57685,
      "texts online": 121564,
      "online multilingual": 81783,
      "multilingual resources": 75354,
      "resources wikipedia": 101066,
      "languages evaluation": 60559,
      "highresourced languages": 50435,
      "multilingual representations": 75351,
      "focus african": 44734,
      "african languages": 4275,
      "way word": 132143,
      "curated corpora": 25243,
      "collect high": 18378,
      "representations surface": 99919,
      "pairs dataset": 83510,
      "provide corpora": 93795,
      "corpora embeddings": 23467,
      "embeddings test": 35977,
      "finegrained emotion": 44347,
      "emotion classification": 36098,
      "chinese microblogs": 16786,
      "feelings daily": 43864,
      "analysis sa": 5825,
      "syntax information": 116537,
      "structures chinese": 114063,
      "happiness sadness": 49443,
      "anger disgust": 6106,
      "disgust fear": 32362,
      "fear surprise": 43233,
      "model reaches": 71853,
      "utilize information": 130509,
      "information dependency": 54471,
      "annotate new": 6131,
      "classification open": 17312,
      "sequencelevel knowledge": 107107,
      "compression technique": 20325,
      "teacher models": 119697,
      "models train": 74190,
      "train smaller": 123024,
      "smaller models": 110239,
      "data points": 26241,
      "case models": 15600,
      "simplified datasets": 109601,
      "augmentation regularization": 10298,
      "augmentation strategies": 10306,
      "achieve bleu": 2436,
      "evaluation perspective": 39322,
      "systems close": 116781,
      "alleviating problem": 5063,
      "systems evaluated": 116878,
      "evaluation dialogue": 39182,
      "requires significantly": 100321,
      "neural module": 78206,
      "module networks": 74500,
      "reasoning text": 96317,
      "text answering": 120649,
      "answering compositional": 6611,
      "compositional questions": 20126,
      "questions require": 95356,
      "require multiple": 100182,
      "multiple steps": 75704,
      "steps reasoning": 113332,
      "learn parse": 62107,
      "executable programs": 39967,
      "visual qa": 131801,
      "domains challenging": 33738,
      "learn models": 62092,
      "questions opendomain": 95338,
      "opendomain text": 81981,
      "model needs": 71564,
      "paragraph text": 84572,
      "symbolic reasoning": 116319,
      "sorting counting": 110685,
      "numbers dates": 81000,
      "help extract": 49723,
      "arguments associated": 8980,
      "question program": 95205,
      "supervision provides": 115910,
      "provides sufficient": 94084,
      "drop dataset": 34168,
      "reasoning challenges": 96233,
      "interpretability word": 56234,
      "embeddings substantially": 35958,
      "substantially successful": 114913,
      "successful capturing": 115153,
      "modeling provides": 72519,
      "provides intuitive": 94045,
      "intuitive way": 56671,
      "way evaluate": 132075,
      "evaluate embeddings": 38832,
      "definitions corresponding": 28535,
      "corresponding words": 24316,
      "words task": 134258,
      "task great": 118249,
      "practical application": 88700,
      "modeling generate": 72435,
      "generate reasonable": 46997,
      "embeddings generate": 35730,
      "semantics better": 105398,
      "setting investigate": 107757,
      "investigate joint": 56774,
      "modeling experimental": 72428,
      "modeling multitask": 72483,
      "growing need": 49176,
      "texts task": 121632,
      "task typically": 118816,
      "typically modeled": 126447,
      "bilstm models": 14099,
      "provide stateoftheart": 93925,
      "rnn sequence": 102930,
      "modeling terms": 72561,
      "terms performance": 120357,
      "performance computational": 86247,
      "directions previous": 31852,
      "previous future": 90413,
      "previous context": 90400,
      "performance making": 86522,
      "inference time": 54238,
      "capturing global": 15468,
      "global dependencies": 48234,
      "regardless input": 97667,
      "model adopts": 70641,
      "adopts graph": 3912,
      "sentences instead": 106358,
      "systems benchmark": 116755,
      "datasets demonstrating": 27418,
      "model conduct": 70880,
      "conduct thorough": 20900,
      "entities help": 37791,
      "help performance": 49763,
      "used named": 128644,
      "aspects main": 9393,
      "use informal": 128093,
      "entities social": 37870,
      "compared regular": 19434,
      "analysis named": 5710,
      "entities publiclyavailable": 37853,
      "dataset turkish": 27248,
      "provide finergrained": 93830,
      "names different": 76141,
      "representations yielded": 100001,
      "yielded significant": 135367,
      "improvements nlp": 52884,
      "tasks mainly": 119284,
      "question improved": 95167,
      "improved relation": 52633,
      "types representations": 126351,
      "representations generated": 99668,
      "models replacing": 73924,
      "embeddings flair": 35727,
      "replacing static": 98949,
      "embeddings contextualized": 35614,
      "representations lead": 99721,
      "requires timeconsuming": 100325,
      "combination representations": 18580,
      "representations survey": 99921,
      "methods evaluation": 69484,
      "reduces reliance": 97396,
      "networks translation": 77803,
      "translations obtained": 125481,
      "years ago": 135242,
      "sentences independently": 106354,
      "documentlevel machine": 33149,
      "neural revolution": 78657,
      "state future": 112495,
      "literature based": 64747,
      "architectures training": 8854,
      "evaluation strategies": 39406,
      "document mt": 33040,
      "conclude presenting": 20715,
      "exploration research": 41504,
      "recognition detect": 96858,
      "sentences extract": 106312,
      "involves subtasks": 56903,
      "component extraction": 19979,
      "extraction recent": 42461,
      "standard multitask": 112270,
      "learning effective": 62528,
      "effective chinese": 34642,
      "framework neural": 45624,
      "iteratively performs": 57143,
      "designed baselines": 29583,
      "baselines gains": 12401,
      "recent breakthroughs": 96436,
      "effectiveness selfsupervised": 34946,
      "tasks addition": 118920,
      "standard syntactic": 112315,
      "largescale language": 61454,
      "method capture": 68688,
      "bert capture": 13086,
      "using zeroshot": 130393,
      "supervised pretraining": 115823,
      "incorporate knowledge": 53477,
      "knowledge realworld": 58133,
      "applied downstream": 7067,
      "answering datasets": 6620,
      "triviaqa searchqa": 125835,
      "27 f1": 851,
      "f1 improvements": 42659,
      "standard finegrained": 112240,
      "typing dataset": 126472,
      "accuracy gains": 2169,
      "generation despite": 47365,
      "nearly existing": 76733,
      "response users": 101248,
      "users utterance": 129182,
      "specifically adopt": 111524,
      "models furthermore": 73266,
      "networks process": 77707,
      "structure extensive": 113861,
      "quality dialogue": 94630,
      "generation proposed": 47574,
      "publicly released": 94332,
      "released codes": 98477,
      "dialog policies": 30578,
      "interactions building": 55980,
      "building machine": 14861,
      "driven spoken": 34156,
      "careful design": 15511,
      "algorithms models": 4868,
      "robust handle": 103038,
      "interaction work": 55976,
      "better baseline": 13516,
      "prediction zeroshot": 89155,
      "generative language": 47729,
      "zeroshot model": 135513,
      "text metadata": 121114,
      "task provide": 118583,
      "tasks input": 119209,
      "input train": 55457,
      "answer natural": 6545,
      "objective allows": 81066,
      "tasks need": 119335,
      "need multiple": 76837,
      "zeroshot performance": 135518,
      "performance generative": 86418,
      "supervision benchmark": 115872,
      "improvement classification": 52690,
      "accuracy random": 2253,
      "random majority": 95501,
      "class baselines": 17032,
      "language serve": 60084,
      "serve simple": 107299,
      "points way": 87884,
      "way new": 132110,
      "strategies text": 113490,
      "architectures pretraining": 8837,
      "pretraining methods": 90285,
      "contextualized multilingual": 22559,
      "challenge field": 16034,
      "focus zeroshot": 44844,
      "scarcity lowresource": 103811,
      "comprehensive comparison": 20252,
      "compare multiple": 19261,
      "multiple baselines": 75503,
      "baselines performs": 12442,
      "performs stateoftheart": 87028,
      "settings finally": 107802,
      "knowledge sharing": 58172,
      "sharing languages": 108133,
      "languages joint": 60655,
      "training setting": 123857,
      "lead bias": 61845,
      "zeroshot abstractive": 135489,
      "news summarization": 79373,
      "generating summary": 47269,
      "teaching model": 119713,
      "model discriminate": 71006,
      "summarization models": 115538,
      "largescale unlabeled": 61517,
      "predicting leading": 88990,
      "article collect": 9115,
      "corpus conduct": 23712,
      "data cleaning": 25731,
      "cleaning filtering": 17716,
      "pretraining dataset": 90248,
      "dataset existing": 26904,
      "bart t5": 11440,
      "approach dramatically": 7509,
      "dramatically improve": 34091,
      "improve summarization": 52554,
      "summarization quality": 115554,
      "dataset rouge1": 27175,
      "rouge1 score": 103303,
      "model microsoft": 71514,
      "provide public": 93899,
      "demo website": 28643,
      "government agencies": 48543,
      "media organizations": 68136,
      "analysis rich": 5822,
      "used adapting": 128378,
      "marketing strategies": 67246,
      "decision makings": 27938,
      "analysis extensively": 5616,
      "studied various": 114174,
      "success methods": 115098,
      "limits applicability": 64313,
      "data containing": 25788,
      "languages enables": 60524,
      "knowledge skills": 58176,
      "different deep": 31082,
      "various pretrained": 131174,
      "dataset addition": 26724,
      "proposed ensemble": 93269,
      "result f1": 101373,
      "problem better": 90952,
      "better solutions": 13723,
      "semantic expansion": 105047,
      "expansion paper": 40374,
      "task called": 117950,
      "case given": 15588,
      "given seed": 48110,
      "seed term": 104507,
      "suggest terms": 115334,
      "terms fit": 120327,
      "fit context": 44576,
      "interesting applications": 56067,
      "computerassisted writing": 20506,
      "previous explorations": 90410,
      "study demonstrate": 114346,
      "annotations task": 6471,
      "manner dataset": 66941,
      "sentences derived": 106274,
      "propose network": 92814,
      "architecture encodes": 8654,
      "encodes context": 36667,
      "separately suggesting": 106877,
      "context encoder": 22075,
      "encoder architecture": 36493,
      "demonstrate competitive": 28691,
      "achieved appropriate": 2593,
      "encoder attention": 36494,
      "scoring function": 104226,
      "text layout": 121081,
      "document image": 33022,
      "understanding pretraining": 126926,
      "years despite": 135253,
      "despite widespread": 29742,
      "pretraining models": 90288,
      "exclusively focus": 39960,
      "interactions text": 56004,
      "layout information": 61820,
      "scanned documents": 103783,
      "image features": 51784,
      "incorporate words": 53495,
      "words visual": 134308,
      "time text": 122123,
      "single framework": 109736,
      "pretraining achieves": 90232,
      "results downstream": 101740,
      "simple concept": 109392,
      "work exists": 134502,
      "exists natural": 40346,
      "network ann": 77135,
      "written words": 135158,
      "words pronunciation": 134143,
      "tested model": 120577,
      "interestingly model": 56094,
      "essential building": 38541,
      "processing word": 91858,
      "problem effectively": 91021,
      "special subset": 111368,
      "subset words": 114841,
      "underexplored existing": 126662,
      "scarce paper": 103793,
      "handle outofvocabulary": 49398,
      "embeddings represented": 35912,
      "approaches skipgram": 8343,
      "evaluated methods": 38989,
      "showed effectiveness": 108374,
      "similarity embedding": 109225,
      "transformer auxiliary": 124299,
      "improved text": 52648,
      "texts semantically": 121602,
      "novel supervised": 80741,
      "yields stateoftheart": 135441,
      "performance collection": 86215,
      "collection benchmark": 18460,
      "embeddings demonstrate": 35636,
      "zeroshot language": 135507,
      "languages unseen": 60941,
      "training adapting": 123347,
      "adapting deep": 3304,
      "popular tasks": 88135,
      "classification work": 17503,
      "text explore": 120939,
      "explore feasibility": 41546,
      "language informal": 59204,
      "text aim": 120634,
      "aim achieve": 4441,
      "models compared": 72936,
      "existing multilingual": 40215,
      "characterbased embeddings": 16488,
      "equivalent performance": 38250,
      "embeddings sentiment": 35930,
      "sentiments words": 106836,
      "supervised information": 115759,
      "general semantic": 46709,
      "words occurring": 134091,
      "help capture": 49701,
      "capture domainspecific": 15295,
      "information supervised": 55020,
      "words indicative": 133999,
      "supervised features": 115755,
      "success rates": 115124,
      "combinations contextual": 18601,
      "generate original": 46981,
      "vectors combine": 131413,
      "twitter datasets": 126108,
      "datasets turkish": 27768,
      "generate document": 46935,
      "vectors employ": 131430,
      "employ support": 36284,
      "method utilising": 69215,
      "vectors approaches": 131408,
      "margin evaluated": 67192,
      "english corpora": 37097,
      "conventional word": 22914,
      "words fixed": 133960,
      "patterns words": 85794,
      "power representations": 88654,
      "separately different": 106871,
      "relations address": 98106,
      "model easily": 71037,
      "according various": 2008,
      "various relations": 131184,
      "words result": 134175,
      "center embedding": 15890,
      "embedding represent": 35482,
      "overall semantics": 83257,
      "represent relational": 99128,
      "dependencies compared": 29069,
      "effectively distinguish": 34804,
      "words respect": 134174,
      "respect different": 101076,
      "different relations": 31379,
      "relations introducing": 98191,
      "relations use": 98275,
      "effectiveness experiments": 34891,
      "acquisition word": 3052,
      "embeddings need": 35827,
      "original embedding": 82514,
      "negation using": 76918,
      "using transformerbased": 130315,
      "transformerbased architectures": 124402,
      "component systems": 20002,
      "especially biomedical": 38435,
      "domain previous": 33616,
      "detection scope": 30042,
      "scope resolution": 104041,
      "rulebased systems": 103374,
      "systems deep": 116834,
      "apply popular": 7202,
      "xlnet roberta": 135211,
      "task publicly": 118588,
      "bioscope corpus": 14221,
      "sfu review": 107897,
      "review corpus": 102539,
      "improvements previously": 52901,
      "model multiple": 71541,
      "datasets confirm": 27372,
      "negation detection": 76911,
      "bioscope papers": 14222,
      "bioscope abstracts": 14220,
      "03 f1": 17,
      "model dutch": 71032,
      "prediction correct": 89046,
      "native nonnative": 76230,
      "research conducted": 100450,
      "correction models": 24159,
      "resolution specifically": 100784,
      "separate datasets": 106848,
      "datasets built": 27340,
      "texts firstly": 121520,
      "predicts correct": 89214,
      "simultaneously predicts": 109691,
      "partofspeech tag": 85541,
      "tag model": 117337,
      "model containing": 70906,
      "combination sentence": 18582,
      "architecture results": 8737,
      "accuracy partofspeech": 2232,
      "layer integrated": 61726,
      "knowledge positively": 58110,
      "positively affects": 88350,
      "results serve": 102162,
      "point future": 87800,
      "online community": 81751,
      "scale difficult": 103713,
      "difficult expensive": 31626,
      "expensive paper": 40427,
      "verify feasibility": 131587,
      "data argue": 25619,
      "various purposes": 131178,
      "exploiting common": 41460,
      "context effective": 22072,
      "human readable": 50945,
      "task learn": 118346,
      "context independent": 22138,
      "representations entities": 99630,
      "entities mentioned": 37818,
      "models allows": 72746,
      "allows learn": 5167,
      "learn high": 62062,
      "quality entity": 94645,
      "demonstrate successful": 28876,
      "successful results": 115171,
      "novel fewshot": 80570,
      "task existing": 118165,
      "existing entity": 40128,
      "match stateoftheart": 67368,
      "features obtain": 43635,
      "tackbp 2010": 117280,
      "using alias": 129330,
      "base domain": 11452,
      "domain training": 33679,
      "global entity": 48237,
      "finegrained type": 44390,
      "pretraining paper": 90301,
      "introduces novel": 56621,
      "novel selfsupervised": 80712,
      "selfsupervised objective": 104947,
      "objective named": 81098,
      "onestepahead prediction": 81723,
      "predicts tokens": 89224,
      "simultaneously based": 109671,
      "based previous": 11950,
      "context tokens": 22299,
      "tokens time": 122341,
      "explicitly encourages": 41369,
      "future tokens": 46318,
      "prevent overfitting": 90366,
      "scale dataset": 103711,
      "experiments cnndailymail": 40833,
      "squad 11": 112072,
      "summarization question": 115555,
      "scale pretraining": 103749,
      "pretraining corpus": 90242,
      "transliteration using": 125527,
      "gained attention": 46357,
      "focus translating": 44833,
      "techniques neural": 119941,
      "providing single": 94133,
      "large neural": 61166,
      "network attention": 77151,
      "mechanism sequencetosequence": 68036,
      "despite significant": 29727,
      "progress domain": 92146,
      "foreign words": 45045,
      "challenge current": 16014,
      "richness language": 102815,
      "word different": 132991,
      "translation low": 124915,
      "language like": 59252,
      "train work": 123053,
      "maximization em": 67505,
      "unsupervised language": 127647,
      "words parallel": 134106,
      "need train": 76865,
      "phrasebased hierarchical": 87381,
      "translation include": 124864,
      "include lstm": 53229,
      "hierarchical dirichlet": 49954,
      "dirichlet process": 31925,
      "process approach": 91424,
      "approach linguistic": 7689,
      "linguistic relationships": 64542,
      "relationships paper": 98327,
      "languages display": 60510,
      "demonstrated qualitative": 28931,
      "conclusions stemming": 20730,
      "lack explicit": 58706,
      "distribution data": 32646,
      "language additionally": 58815,
      "lexical level": 63779,
      "mechanism introduces": 68001,
      "different conditioning": 31053,
      "remains underexplored": 98827,
      "overcome problems": 83298,
      "methodology allows": 69242,
      "sound changes": 110702,
      "outline future": 82630,
      "comparing rulebased": 19516,
      "information electronic": 54516,
      "medical research": 68223,
      "deidentification methods": 28583,
      "automatically remove": 10824,
      "information medical": 54761,
      "unavailability labeled": 126595,
      "constrained english": 21568,
      "medical text": 68228,
      "methods languages": 69575,
      "study construct": 114339,
      "test generalizability": 120457,
      "existing rulebased": 40274,
      "rulebased method": 103368,
      "dutch language": 34281,
      "architecture performs": 8726,
      "performs strongly": 87029,
      "strongly languages": 113747,
      "domains limited": 33805,
      "data compared": 25755,
      "neural method": 78140,
      "make code": 66632,
      "deidentification models": 28584,
      "datasets enable": 27445,
      "future benchmarks": 46259,
      "benchmarks lexical": 12914,
      "capturing local": 15477,
      "local semantic": 64936,
      "semantic correspondence": 105024,
      "construction update": 21699,
      "explore approach": 41520,
      "model baseline": 70751,
      "famous sememe": 43060,
      "learn local": 62084,
      "codes paper": 18207,
      "aggregation transformer": 4370,
      "matching transformer": 67438,
      "pair sequences": 83454,
      "unnecessary noise": 127479,
      "sequence pair": 107024,
      "pair matching": 83439,
      "levels experiments": 63535,
      "embeddings syntactic": 35969,
      "portuguese word": 88203,
      "vectors represent": 131470,
      "space vectors": 111081,
      "vectors able": 131407,
      "capture useful": 15414,
      "useful syntactic": 128937,
      "approaches like": 8219,
      "like word2vec": 64112,
      "meanings paper": 67746,
      "techniques generate": 119893,
      "outperform traditional": 82766,
      "cloze questions": 17893,
      "tasks solved": 119506,
      "supervised counterpart": 115744,
      "input examples": 55335,
      "help language": 49744,
      "understand given": 126748,
      "phrases used": 87461,
      "used assign": 128397,
      "examples finally": 39836,
      "training performed": 123778,
      "outperforms supervised": 83033,
      "approaches lowresource": 8222,
      "settings large": 107816,
      "critical decisions": 24808,
      "public policy": 94268,
      "opinions topic": 82135,
      "solely text": 110536,
      "opinions propose": 82129,
      "express similar": 41707,
      "polarity specific": 87929,
      "specifically unsupervised": 111606,
      "approaches similarly": 8342,
      "considerably better": 21266,
      "20 increase": 507,
      "extant approaches": 41778,
      "similarity sentiment": 109309,
      "range downstream": 95562,
      "construction query": 21697,
      "existing corpusbased": 40096,
      "incorporating lexical": 53545,
      "negative sets": 76956,
      "sets provided": 107701,
      "suffer semantic": 115242,
      "semantic drift": 105038,
      "seed set": 104506,
      "guidance propose": 49227,
      "related target": 97900,
      "set users": 107630,
      "performs multiple": 87011,
      "generating auxiliary": 47201,
      "expansion process": 40375,
      "phrases using": 87463,
      "sequence binary": 106920,
      "binary decision": 14129,
      "requires identification": 100276,
      "internal structures": 56178,
      "presents unsupervised": 89911,
      "tagged text": 117350,
      "extraction main": 42381,
      "data process": 26280,
      "messages social": 68515,
      "multimodal sentiment": 75457,
      "years nlp": 135282,
      "models power": 73761,
      "systems current": 116827,
      "stateoftheart algorithms": 112577,
      "paper implement": 83977,
      "labeling word": 58556,
      "paper hypothesize": 83974,
      "hypothesize crosslingual": 51289,
      "languages verify": 60968,
      "making models": 66853,
      "improve adaptation": 52333,
      "adaptation performance": 3246,
      "languages reduce": 60837,
      "sequence encoders": 106938,
      "observe performance": 81206,
      "addition based": 3403,
      "hypothesis propose": 51271,
      "method finetuning": 68842,
      "finetuning multilingual": 44480,
      "bert downstream": 13104,
      "results dialogue": 101724,
      "dialogue natural": 30714,
      "tasks reducing": 119446,
      "fitted model": 44588,
      "better zeroshot": 13773,
      "performance furthermore": 86398,
      "furthermore proposed": 46208,
      "strong crosslingual": 113663,
      "parameter space": 84725,
      "propose bayesian": 92572,
      "model space": 72071,
      "task infer": 118293,
      "data seen": 26380,
      "inference enables": 54137,
      "prediction time": 89144,
      "perform accurate": 85941,
      "experiment typologically": 40514,
      "yields comparable": 135404,
      "stateoftheart zeroshot": 113033,
      "accuracy proposed": 2249,
      "framework offers": 45634,
      "code located": 18109,
      "machine human": 65758,
      "difficult extend": 31628,
      "opendomain scenarios": 81979,
      "scenarios recently": 103864,
      "systems hand": 116927,
      "responses lack": 101280,
      "conversation able": 22935,
      "advantage interactions": 4047,
      "tackle challenges": 117287,
      "contributions twofold": 22795,
      "challenges presented": 16196,
      "develop interactive": 30207,
      "ask questions": 9299,
      "questions training": 95369,
      "online fashion": 81772,
      "text sentiment": 121283,
      "vision learning": 131762,
      "architecture nlp": 8713,
      "improve extraction": 52383,
      "called hybrid": 15059,
      "cnn performs": 18017,
      "approach large": 7666,
      "comparative linguistics": 19207,
      "project aimed": 92210,
      "overcome challenge": 83275,
      "language relationships": 60033,
      "automated comparison": 10434,
      "techniques techniques": 119996,
      "phonetic representation": 87294,
      "key words": 57613,
      "words concept": 133873,
      "word sets": 133550,
      "distance calculated": 32407,
      "calculated based": 15021,
      "levenshtein distance": 63571,
      "minimum number": 70211,
      "insertions deletions": 55515,
      "words preserved": 134136,
      "evaluation hierarchical": 39230,
      "mean standard": 67612,
      "language sets": 60085,
      "capsule network": 15248,
      "character representation": 16460,
      "representation ability": 99159,
      "network method": 77323,
      "bidirectional gated": 13921,
      "ner proposed": 77070,
      "encoder representation": 36559,
      "representation transformersbert": 99453,
      "capture sequence": 15391,
      "features selfattention": 43714,
      "proposed different": 93259,
      "captured hidden": 15430,
      "recognition evaluated": 96874,
      "study best": 114329,
      "specific document": 111431,
      "document related": 33061,
      "methods calculating": 69358,
      "calculating distance": 15028,
      "available apache": 10943,
      "framework starting": 45696,
      "new frontier": 78929,
      "solution capable": 110548,
      "dialogue agent": 30637,
      "achieve specific": 2552,
      "openended conversations": 81986,
      "mainly rely": 66494,
      "learning simple": 63028,
      "considering semantic": 21332,
      "results poor": 102040,
      "rate work": 95811,
      "work adopt": 134360,
      "responses introducing": 101278,
      "learning knowledge": 62664,
      "candidate keywords": 15108,
      "drives conversation": 34160,
      "considers semantic": 21341,
      "prediction discourse": 89051,
      "discourse help": 32059,
      "response retrieval": 101236,
      "retrieval module": 102411,
      "better retrieval": 13710,
      "performance meaningful": 86523,
      "meaningful conversations": 67714,
      "novel dual": 80550,
      "higher success": 50209,
      "push research": 94456,
      "research boundary": 100434,
      "better introduce": 13611,
      "sina weibo": 109700,
      "method produce": 69073,
      "conversations significantly": 23067,
      "improving stateoftheart": 53168,
      "mining annotation": 70224,
      "design models": 29548,
      "data gathered": 25969,
      "statistical data": 113089,
      "empirical observations": 36175,
      "consistent structure": 21403,
      "structure strong": 113962,
      "interdisciplinary research": 56061,
      "ways building": 132148,
      "narrative information": 76173,
      "structure narratives": 113910,
      "identification concepts": 51373,
      "differences respect": 30970,
      "respect gender": 101078,
      "accounts significant": 2050,
      "concepts used": 20649,
      "enables researchers": 36401,
      "insights global": 55534,
      "early twentieth": 34395,
      "twentieth century": 126077,
      "networks distributed": 77571,
      "information nearby": 54790,
      "transfer research": 124180,
      "regardless distances": 97665,
      "investigate introduce": 56772,
      "networks mtn": 77664,
      "automatically derive": 10750,
      "corpora study": 23593,
      "online communication": 81747,
      "online medical": 81782,
      "graph hierarchical": 48820,
      "adopts hybrid": 3914,
      "graph text": 48873,
      "framework build": 45446,
      "answering complex": 6610,
      "complex medical": 19831,
      "medical questions": 68220,
      "similarity deep": 109218,
      "similar question": 109129,
      "question large": 95174,
      "qa dataset": 94502,
      "dataset compare": 26805,
      "representation transformers": 99451,
      "transformers bert": 124444,
      "duplicate questions": 34270,
      "questions dataset": 95297,
      "dataset medical": 27016,
      "nonautoregressive neural": 80160,
      "maximum mutual": 67543,
      "information mmi": 54767,
      "mmi models": 70457,
      "bidirectional dependency": 13906,
      "backward probability": 11351,
      "used objective": 128669,
      "model direct": 70998,
      "list generated": 64691,
      "generated given": 47080,
      "inevitably results": 54083,
      "decoding experimental": 28091,
      "produces diverse": 91992,
      "diverse coherent": 32798,
      "responses yielding": 101304,
      "yielding substantive": 135390,
      "substantive gains": 114920,
      "gains bleu": 46386,
      "aims use": 4583,
      "using lowlevel": 129840,
      "end develop": 36804,
      "language level": 59248,
      "edit operations": 34535,
      "user evaluation": 128980,
      "utilizing proposed": 130577,
      "identify object": 51526,
      "segmentation key": 104594,
      "demonstrates advantages": 28940,
      "processing benchmarks": 91631,
      "benchmarks glue": 12906,
      "benchmarking platform": 12879,
      "time cost": 122004,
      "benchmarking results": 12880,
      "models differ": 73066,
      "lot different": 65323,
      "comparing multitask": 19508,
      "platform available": 87652,
      "translation joint": 124880,
      "representation early": 99216,
      "target units": 117738,
      "efficiency paper": 35036,
      "inefficiency issue": 54073,
      "efficient attention": 35060,
      "sequence modelling": 107017,
      "framework outperform": 45638,
      "outperform transformer": 82768,
      "tasks bleu": 118969,
      "systematic model": 116681,
      "approach allowing": 7347,
      "parameters code": 84739,
      "detection subjective": 30067,
      "bias detection": 13800,
      "detection critical": 29914,
      "propaganda detection": 92381,
      "content recommendation": 21930,
      "recommendation sentiment": 97087,
      "experiments detecting": 40902,
      "using bertbased": 129391,
      "bertbased models": 13276,
      "labeled instances": 58460,
      "bias propose": 13837,
      "propose bertbased": 92573,
      "ensembles outperform": 37627,
      "languages transferring": 60929,
      "transferring pretrained": 124244,
      "models demonstrated": 73035,
      "effectiveness downstream": 34885,
      "tasks availability": 118950,
      "multilingual pretrained": 75336,
      "models enables": 73145,
      "focuses heavily": 44900,
      "model english": 71077,
      "languages limited": 60698,
      "computational budget": 20363,
      "bert base": 13069,
      "bert zeroshot": 13263,
      "zeroshot tasks": 135528,
      "modeling information": 72449,
      "extraction introduce": 42358,
      "simple new": 109485,
      "semantics end": 105413,
      "modify transformer": 74458,
      "architecture way": 8772,
      "layout features": 61819,
      "way use": 132138,
      "use raw": 128222,
      "finetuned downstream": 44423,
      "endtoend information": 36907,
      "baseline roberta": 12304,
      "public leaderboard": 94263,
      "key information": 57580,
      "dataset improving": 26976,
      "representations human": 99678,
      "cognitive theories": 18285,
      "theories human": 121810,
      "work analyze": 134370,
      "accuracy significantly": 2275,
      "human concept": 50782,
      "models human": 73341,
      "modelling latent": 72608,
      "present generative": 89505,
      "set latent": 107477,
      "tasks explicitly": 119125,
      "explicitly modelling": 41387,
      "task embedding": 118133,
      "positive transfer": 88344,
      "tasks efficient": 119082,
      "tasks instantiate": 119214,
      "covering broad": 24532,
      "comparing performance": 19511,
      "latent task": 61617,
      "sequencetosequence baselines": 107158,
      "average tasks": 11216,
      "adaptation based": 3211,
      "based inference": 11769,
      "inference latent": 54159,
      "standard finetuning": 112241,
      "parameter adaptation": 84710,
      "adaptation performs": 3248,
      "performance finally": 86383,
      "learnt model": 63249,
      "natural way": 76624,
      "dialogue interaction": 30692,
      "behavior using": 12670,
      "model producing": 71801,
      "accurate responses": 2359,
      "human beings": 50767,
      "past works": 85656,
      "fluent response": 44716,
      "response according": 101188,
      "query context": 94952,
      "context models": 22181,
      "response utterance": 101249,
      "messages readability": 68512,
      "readability instead": 95947,
      "instead long": 55668,
      "step intelligent": 113274,
      "model help": 71277,
      "decide wait": 27927,
      "response directly": 101195,
      "speaking style": 111339,
      "respectively generate": 101143,
      "possible utterances": 88445,
      "combining dialogue": 18721,
      "user directly": 128976,
      "compared approach": 19323,
      "approach popular": 7801,
      "ending prediction": 36855,
      "models incorporating": 73387,
      "information adaptive": 54361,
      "attention text": 10022,
      "training classifiers": 123384,
      "handcraft features": 49334,
      "knowledge contrast": 57841,
      "statistical features": 113096,
      "frequency distribution": 45840,
      "shows simple": 108631,
      "information enhance": 54531,
      "significantly compared": 108898,
      "mechanism named": 68016,
      "model global": 71254,
      "mechanism incorporates": 67995,
      "features latent": 43587,
      "features attention": 43370,
      "captures dependency": 15443,
      "cnnbased rnnbased": 18031,
      "modern english": 74388,
      "linguistic search": 64552,
      "investigate question": 56804,
      "vastly increase": 131285,
      "tools nlp": 122462,
      "queries automatically": 94913,
      "tagger parser": 117359,
      "similar text": 109159,
      "metrics accuracy": 69917,
      "query searches": 94986,
      "parsed data": 84897,
      "detection ed": 29932,
      "key subtask": 57600,
      "subtask information": 114945,
      "extraction aims": 42258,
      "aims recognize": 4567,
      "effectiveness integrating": 34898,
      "usually ignore": 130423,
      "dependency label": 29149,
      "rich useful": 102799,
      "exploits syntactic": 41494,
      "module designed": 74483,
      "representations aggregating": 99496,
      "dependency types": 29250,
      "types furthermore": 126292,
      "fully explore": 45959,
      "clues hidden": 17919,
      "module introduced": 74495,
      "relation representations": 98067,
      "experiments widely": 41206,
      "ace2005 dataset": 2403,
      "competitive baseline": 19637,
      "lexical relation": 63794,
      "classification lexical": 17252,
      "form relation": 45117,
      "relation triples": 98076,
      "concepts challenging": 20607,
      "framework address": 45414,
      "task lexical": 118352,
      "learn concept": 62004,
      "rich lexical": 102763,
      "knowledge injected": 58017,
      "probabilistic distribution": 90804,
      "task distribution": 118107,
      "distribution supervised": 32681,
      "classifier experiments": 17543,
      "methods detecting": 69439,
      "impact linguistic": 51876,
      "structural knowledge": 113775,
      "knowledge social": 58177,
      "providing access": 94098,
      "sensitive information": 105739,
      "provide means": 93872,
      "apply linguistic": 7180,
      "resources lexical": 100997,
      "lexical conceptual": 63743,
      "detection leverage": 29984,
      "leverage structural": 63625,
      "confidence results": 20995,
      "framing detection": 45750,
      "detection identification": 29966,
      "users social": 129170,
      "transformers pretrained": 124459,
      "variants achieved": 130873,
      "varieties nlp": 130958,
      "millions parameters": 70121,
      "reallife applications": 96127,
      "compress large": 20295,
      "transformer vaswani": 124394,
      "2017 based": 628,
      "small model": 110172,
      "model student": 72095,
      "selfattention module": 104892,
      "module plays": 74502,
      "plays vital": 87743,
      "networks large": 77639,
      "model teacher": 72149,
      "module transformer": 74514,
      "transformer layer": 124333,
      "effective flexible": 34679,
      "scaled dotproduct": 103765,
      "module new": 74501,
      "works introducing": 134954,
      "baselines different": 12386,
      "parameter size": 84724,
      "models particular": 73717,
      "99 accuracy": 1451,
      "obtain competitive": 81267,
      "detecting potential": 29847,
      "potential topics": 88590,
      "better user": 13762,
      "names locations": 76146,
      "use algorithms": 127891,
      "given english": 48026,
      "convey important": 23113,
      "wikipedia titles": 132684,
      "f1 especially": 42655,
      "thesaurus embedding": 121839,
      "cohyponymy detection": 18339,
      "distributionally similar": 32751,
      "community paper": 19084,
      "network embedding": 77234,
      "embedding distributional": 35398,
      "relations extensive": 98171,
      "models binary": 72856,
      "modeling alleviate": 72373,
      "alleviate shortage": 5052,
      "utterances domains": 130636,
      "user goals": 128992,
      "provide user": 93950,
      "user simulator": 129042,
      "compare evaluate": 19245,
      "models corpus": 72991,
      "make suitable": 66734,
      "tasks crossdomain": 119021,
      "user simulation": 129041,
      "shallow parser": 107927,
      "languages effectively": 60520,
      "languages arising": 60410,
      "ordering words": 82441,
      "sentence separately": 106067,
      "generation taskoriented": 47663,
      "crucial component": 25135,
      "component taskoriented": 20003,
      "nlg module": 79523,
      "response natural": 101223,
      "typically relies": 126451,
      "controllable generation": 22826,
      "evaluations modeling": 39475,
      "translation existing": 124803,
      "networks generate": 77611,
      "trained translation": 123318,
      "model tends": 72152,
      "tends focus": 120171,
      "generated target": 47135,
      "translation target": 125326,
      "word contextual": 132970,
      "boost training": 14438,
      "training nmt": 123749,
      "future context": 46263,
      "representation current": 99200,
      "word decoding": 132981,
      "results widelyused": 102335,
      "including wmt14": 53405,
      "strong transformerbased": 113727,
      "baseline automatic": 12191,
      "resource exploring": 100845,
      "propose statistical": 93082,
      "collect corpus": 18375,
      "substantial agreement": 114850,
      "agreement fleiss": 4392,
      "formulated automatic": 45286,
      "automatic segmentation": 10647,
      "models unified": 74270,
      "autoregressive language": 10864,
      "text masked": 121101,
      "masked tokens": 67307,
      "corrupted tokens": 24328,
      "masked spans": 67303,
      "welldesigned position": 132402,
      "position embeddings": 88285,
      "embeddings selfattention": 35924,
      "redundant computation": 97478,
      "information position": 54851,
      "model bidirectional": 70777,
      "decoder respectively": 28051,
      "respectively experiments": 101139,
      "results wide": 102326,
      "tasks widely": 119602,
      "used benchmarks": 128410,
      "transformerbased model": 124424,
      "language morphologically": 59693,
      "english given": 37160,
      "arabic natural": 8533,
      "models languagespecific": 73459,
      "languagespecific bert": 60989,
      "bert based": 13070,
      "provided pretrained": 93979,
      "corpus models": 23886,
      "bert specifically": 13234,
      "specifically arabic": 111528,
      "language performance": 59808,
      "compared multilingual": 19394,
      "performance tested": 86797,
      "research applications": 100405,
      "efficient graph": 35080,
      "views words": 131726,
      "words nodes": 134082,
      "despite successes": 29736,
      "suffers drawbacks": 115255,
      "difficult words": 31673,
      "conduct different": 20836,
      "addition integrate": 3440,
      "information serves": 54975,
      "serves input": 107315,
      "datasets various": 27787,
      "model brings": 70786,
      "brings significant": 14650,
      "highperformance models": 50360,
      "transformer achieving": 124289,
      "generation current": 47349,
      "systems mainly": 116991,
      "human conversations": 50787,
      "complexity training": 19944,
      "varies greatly": 130950,
      "queryresponse pairs": 95009,
      "learning converse": 62463,
      "complex ones": 19854,
      "dynamically adjust": 34339,
      "learning progress": 62919,
      "progress paper": 92170,
      "conducted stateoftheart": 20950,
      "demonstrate learning": 28769,
      "disfluent speech": 32360,
      "nlp perspective": 79661,
      "addition previous": 3461,
      "works used": 134988,
      "depending input": 29272,
      "speech making": 111707,
      "making difficult": 66834,
      "different contributions": 31063,
      "clinical nlp": 17779,
      "performance textbased": 86800,
      "word span": 133580,
      "information finally": 54616,
      "audio features": 10232,
      "features inspired": 43571,
      "features experimentally": 43498,
      "learning pretraining": 62908,
      "understanding experiments": 126842,
      "chinese release": 16813,
      "new chinese": 78829,
      "used chinese": 128421,
      "chinese bert": 16739,
      "cost memory": 24364,
      "memory works": 68397,
      "works good": 134951,
      "corpus achieves": 23642,
      "accelerating training": 1878,
      "compared bertbase": 19338,
      "based selfattention": 12007,
      "separately trained": 106878,
      "trained embeddings": 123125,
      "paper devise": 83870,
      "devise new": 30463,
      "mechanism achieves": 67941,
      "transformer terms": 124383,
      "recognized benchmarks": 97060,
      "users easily": 129111,
      "audio corpora": 10225,
      "following specific": 44987,
      "rules implemented": 103411,
      "parsers finally": 85025,
      "problem dialogue": 91012,
      "modeling systems": 72557,
      "systems works": 117242,
      "massive training": 67339,
      "data realworld": 26322,
      "train dialogue": 122926,
      "pretraining large": 90272,
      "adapting target": 3325,
      "framework paper": 45647,
      "fast domain": 43132,
      "task dstc8": 118124,
      "based gpt2": 11743,
      "gpt2 finetuned": 48561,
      "dataset robust": 27174,
      "diverse response": 32836,
      "evaluation improvement": 39239,
      "attaining competitive": 9727,
      "performance adaptation": 86122,
      "multiwoz dataset": 75956,
      "hyperdimensional computing": 51211,
      "ngram statistics": 79454,
      "learning led": 62686,
      "led significant": 63282,
      "performance increase": 86457,
      "models computationally": 72947,
      "efficient algorithms": 35055,
      "representations formed": 99660,
      "formed using": 45228,
      "representations serve": 99885,
      "input standard": 55444,
      "classifiers investigate": 17613,
      "investigate applicability": 56715,
      "datasets classification": 27345,
      "using classifiers": 129441,
      "decreasing time": 28177,
      "classifiers small": 17634,
      "dataset memory": 27017,
      "classifiers large": 17615,
      "dimensionality representation": 31762,
      "7th dialog": 1280,
      "presents systems": 89908,
      "long documents": 65072,
      "stateoftheart solutions": 112959,
      "range contexts": 95551,
      "like sentencelevel": 64089,
      "context documentlevel": 22067,
      "short documents": 108209,
      "realworld documents": 96166,
      "document page": 33047,
      "presence multiple": 89350,
      "encourage progress": 36761,
      "datasets utilizing": 27783,
      "utilizing textual": 130583,
      "features nlp": 43631,
      "formal documents": 45157,
      "propose pipeline": 92990,
      "pipeline method": 87544,
      "baseline different": 12210,
      "analyze behavior": 5952,
      "tools neural": 122461,
      "greatly advanced": 49043,
      "methods depends": 69433,
      "systems case": 116772,
      "gold data": 48433,
      "segmentation words": 104655,
      "words subword": 134245,
      "units used": 127269,
      "years recently": 135293,
      "networks benefit": 77521,
      "simpler languages": 109560,
      "subword model": 115026,
      "english finnish": 37137,
      "north sami": 80370,
      "sets approach": 107648,
      "model original": 71621,
      "training algorithm": 123352,
      "optimization leads": 82194,
      "leads higher": 61935,
      "implementations new": 51955,
      "writing pattern": 135093,
      "make spelling": 66729,
      "based manual": 11826,
      "manual generation": 67001,
      "research present": 100583,
      "bengali word": 13038,
      "word generation": 133297,
      "used bengali": 128411,
      "bengali words": 13039,
      "phonetically similar": 87303,
      "generating errors": 47217,
      "experiment new": 40479,
      "relative word": 98384,
      "vector offsets": 131333,
      "making different": 66833,
      "unlabeled tokens": 127412,
      "separate target": 106862,
      "represent relationship": 99130,
      "fasttext glove": 43204,
      "various word": 131238,
      "syntax dependency": 116533,
      "dependency grammars": 29144,
      "word position": 133411,
      "pseudo labeling": 94179,
      "learning largescale": 62681,
      "domains provided": 33840,
      "target labels": 117628,
      "pseudo labels": 94180,
      "errors incorrect": 38377,
      "incorrectly predicted": 53583,
      "intelligent conversational": 55846,
      "hypothesis reranking": 51274,
      "evaluation multilingual": 39302,
      "crosslingual lexical": 24970,
      "largescale lexical": 61459,
      "benchmark covering": 12748,
      "languages mandarin": 60719,
      "chinese spanish": 16824,
      "pairs providing": 83620,
      "adjectives adverbs": 3861,
      "datasets extensive": 27479,
      "language coverage": 58926,
      "analysis monolingual": 5706,
      "crosslingual benchmarks": 24930,
      "benchmarks evaluate": 12902,
      "analyze wide": 6017,
      "wide array": 132489,
      "crosslingual representation": 24998,
      "embeddings fasttext": 35719,
      "mbert xlm": 67571,
      "representations fully": 99662,
      "public release": 94270,
      "results indepth": 101852,
      "learning available": 62381,
      "website encourage": 132294,
      "community effort": 19072,
      "languages largescale": 60678,
      "largescale semantic": 61507,
      "semantic resource": 105241,
      "analysis order": 5732,
      "paper time": 84474,
      "provides neural": 94055,
      "using widely": 130374,
      "word2vec fasttext": 133672,
      "generated neural": 47098,
      "using intrinsic": 129766,
      "lack publicly": 58737,
      "available benchmark": 10951,
      "datasets provides": 27651,
      "neutral classes": 78754,
      "classes provide": 17073,
      "benchmark baseline": 12734,
      "performance presented": 86613,
      "adapt diverse": 3175,
      "diverse machine": 32823,
      "learning support": 63075,
      "machine logistic": 65840,
      "regression naive": 97705,
      "learning convolutional": 62464,
      "network hybrid": 77275,
      "approaches effectiveness": 8138,
      "using distinct": 129595,
      "representation approaches": 99172,
      "respectively finally": 101141,
      "methodology outperforms": 69265,
      "adapted machine": 3286,
      "terms f1score": 120326,
      "pretrain word": 89980,
      "provide useful": 93947,
      "task rely": 118633,
      "cross attention": 24857,
      "mechanism detect": 67968,
      "parts sentence": 85581,
      "pairs relevant": 83628,
      "assessing quality": 9517,
      "contribution concerns": 22769,
      "finegrained scores": 44376,
      "quality empirical": 94641,
      "dataset neural": 27052,
      "featurebased methods": 43335,
      "robust methods": 103045,
      "web corpora": 132227,
      "corpora order": 23550,
      "provide robust": 93913,
      "tasks proper": 119404,
      "led success": 63290,
      "classification web": 17499,
      "considerable differences": 21246,
      "lexicon obtained": 63899,
      "obtained corpus": 81355,
      "corpus second": 23982,
      "kinds texts": 57694,
      "tools results": 122473,
      "work identify": 134558,
      "generated content": 47063,
      "factchecking process": 42846,
      "process work": 91590,
      "order able": 82263,
      "articles verified": 9227,
      "stanza python": 112387,
      "python natural": 94481,
      "processing toolkit": 91842,
      "existing widely": 40331,
      "fully neural": 45967,
      "analysis including": 5654,
      "recognition trained": 97032,
      "dependencies treebanks": 29120,
      "multilingual corpora": 75224,
      "python interface": 94479,
      "corenlp software": 23386,
      "tasks coreference": 119016,
      "code documentation": 18098,
      "66 languages": 1184,
      "evaluating multilingual": 39077,
      "better humanmachine": 13602,
      "english greatly": 37164,
      "greatly limits": 49058,
      "experiment multilingual": 40473,
      "baselines evaluate": 12389,
      "advantage having": 4046,
      "languages hand": 60616,
      "hand stateoftheart": 49327,
      "inferior performance": 54260,
      "models showing": 74024,
      "modeling challenging": 72387,
      "accelerate research": 1870,
      "data selftraining": 26389,
      "pidgin english": 87492,
      "language significantly": 60092,
      "west africa": 132454,
      "75 million": 1249,
      "million speakers": 70108,
      "proper machine": 92416,
      "techniques targeted": 119994,
      "english english": 37125,
      "descriptions structured": 29495,
      "techniques unsupervised": 120005,
      "alignment human": 4961,
      "selftraining technique": 104963,
      "technique improves": 119791,
      "text fluency": 120955,
      "fluency relevance": 44699,
      "relevance analysis": 98502,
      "words word2vec": 134320,
      "stateoftheart implementation": 112673,
      "understood work": 127027,
      "work derive": 134466,
      "model establish": 71095,
      "close relationship": 17830,
      "learning addition": 62310,
      "addition provide": 3468,
      "global optimal": 48255,
      "optimal solution": 82171,
      "nsurl2019 task": 80814,
      "future researches": 46312,
      "teams teams": 119737,
      "systems best": 116758,
      "score based": 104053,
      "evaluation seven": 39393,
      "person organization": 87102,
      "time money": 122052,
      "decoder encoder": 28006,
      "encoder layers": 36539,
      "wide attention": 132490,
      "focuses using": 44918,
      "source linguistic": 110789,
      "features encoder": 43480,
      "information decoder": 54466,
      "decoder layers": 28028,
      "embeddings surprisingly": 35966,
      "approach project": 7823,
      "trained frozen": 123146,
      "level transformer": 63513,
      "transformer decoder": 124318,
      "measure word": 67830,
      "findings motivate": 44305,
      "number decoder": 80864,
      "experiments case": 40817,
      "deep encoder": 28230,
      "joint approach": 57262,
      "detection applications": 29877,
      "require efficient": 100131,
      "noun compounds": 80424,
      "smaller components": 110227,
      "idiomatic nature": 51652,
      "learningbased approach": 63178,
      "detection german": 29958,
      "train using": 123049,
      "using newly": 129964,
      "endtoend entity": 36898,
      "embeddings entity": 35698,
      "process identifying": 91504,
      "corpus contrast": 23727,
      "class labels": 17042,
      "labels existing": 58599,
      "models operate": 73666,
      "mention level": 68422,
      "key factors": 57568,
      "factors make": 42898,
      "context windows": 22326,
      "window size": 132712,
      "drawbacks propose": 34124,
      "using bigru": 129408,
      "contextualised embeddings": 22531,
      "resources propose": 101029,
      "method estimate": 68811,
      "contextual embedding": 22456,
      "space approach": 110971,
      "approach fully": 7580,
      "purely datadriven": 94413,
      "strong statistical": 113716,
      "different rankings": 31372,
      "rankings derived": 95698,
      "containing different": 21786,
      "word finally": 133285,
      "unsupervised nature": 127679,
      "language code": 58883,
      "accepted long": 1897,
      "long paper": 65079,
      "eacl 2021": 34368,
      "crossdomain data": 24886,
      "important cases": 52110,
      "data short": 26442,
      "short supply": 108244,
      "aim learn": 4464,
      "learn source": 62152,
      "data abundant": 25555,
      "domain pairs": 33603,
      "crossdomain performance": 24897,
      "performance advantage": 86129,
      "gains downstream": 46391,
      "speech natural": 111713,
      "report recent": 99032,
      "various personal": 131166,
      "dataset majority": 27010,
      "millions tokens": 70123,
      "estimation text": 38669,
      "structure information": 113881,
      "generalpurpose text": 46893,
      "matching deep": 67400,
      "procedure unsupervised": 91403,
      "maximizing mutual": 67525,
      "information learn": 54722,
      "learn text": 62166,
      "matching approach": 67397,
      "approach tasks": 7966,
      "identification answer": 51362,
      "approaches experiments": 8155,
      "integrated mutual": 55779,
      "learns better": 63196,
      "better text": 13743,
      "reducing repetitive": 97433,
      "utterance generation": 130599,
      "models prone": 73816,
      "prone generating": 92334,
      "utterances work": 130677,
      "repetition problem": 98915,
      "end consider": 36799,
      "task problem": 118565,
      "prominent tasks": 92254,
      "study influence": 114407,
      "influence model": 54313,
      "manage achieve": 66895,
      "achieve lower": 2496,
      "does generalize": 33354,
      "models transformers": 74243,
      "tokens difficult": 122306,
      "learn learning": 62079,
      "likely happen": 64139,
      "individual token": 53935,
      "tokens lower": 122316,
      "tokens different": 122305,
      "weighting functions": 132362,
      "deterministic nature": 30162,
      "nature hinders": 76654,
      "network rnnbased": 77422,
      "receptive field": 96786,
      "explore types": 41602,
      "global latent": 48248,
      "datasets automatic": 27322,
      "metric human": 69882,
      "improve standard": 52546,
      "standard transformers": 112331,
      "terms diversity": 120312,
      "bert various": 13252,
      "tasks benchmark": 118958,
      "bert build": 13085,
      "industry use": 54064,
      "noise data": 80048,
      "data bert": 25697,
      "bert performed": 13201,
      "bert performs": 13202,
      "explore sensitivity": 41585,
      "work commonly": 134417,
      "mistakes typos": 70350,
      "degradation performance": 28552,
      "performance bert": 86176,
      "berts performance": 13287,
      "analysis textual": 5898,
      "drops significantly": 34189,
      "identify shortcomings": 51553,
      "shortcomings existing": 108283,
      "existing bert": 40083,
      "drop performance": 34169,
      "performance findings": 86384,
      "datasets finetuning": 27486,
      "approaches computational": 8103,
      "documentation endangered": 33120,
      "interpretability recordings": 56231,
      "investigate choice": 56732,
      "choice translation": 16908,
      "documentation work": 33123,
      "produced bilingual": 91951,
      "mass multilingual": 67321,
      "multilingual speech": 75374,
      "al 2020": 4649,
      "apply task": 7221,
      "lowresource unsupervised": 65572,
      "choice language": 16889,
      "aligned translations": 4926,
      "clues extracted": 17918,
      "segmentation neural": 104605,
      "godard et": 48422,
      "increases translation": 53655,
      "challenging language": 16265,
      "pairs corpus": 83503,
      "oriented dialogues": 82496,
      "known suffer": 58323,
      "suffer inconsistent": 115223,
      "general difficulty": 46650,
      "relative lack": 98360,
      "lack labeled": 58721,
      "data personality": 26234,
      "new labeled": 78971,
      "dataset domain": 26882,
      "dialogue based": 30642,
      "additional layer": 3539,
      "annotation potentially": 6353,
      "useful training": 128942,
      "baseline endtoend": 12213,
      "players european": 87715,
      "tools services": 122474,
      "sets resources": 107705,
      "fully operational": 45968,
      "technologies data": 120023,
      "text general": 120972,
      "entity attributes": 37907,
      "leveraging pretrained": 63696,
      "joint prediction": 57312,
      "prediction entity": 89054,
      "transitions model": 124516,
      "previous current": 90402,
      "states predict": 113051,
      "predict transition": 88947,
      "span text": 111136,
      "predefined set": 88835,
      "comprehension datasets": 20176,
      "dataset language": 26994,
      "language engineering": 59004,
      "focused morphosyntactic": 44867,
      "sentence aligned": 105766,
      "encoded xml": 36489,
      "encoding initiative": 36699,
      "initiative guidelines": 55253,
      "czech english": 25489,
      "english estonian": 37128,
      "romanian russian": 103252,
      "history development": 50550,
      "encoding components": 36684,
      "pairs low": 83578,
      "dataset automatically": 26751,
      "commercially available": 18844,
      "support translation": 115994,
      "english assess": 37072,
      "assess current": 9472,
      "score results": 104118,
      "translation promising": 125149,
      "promising low": 92279,
      "dataset useful": 27258,
      "useful research": 128923,
      "benchmark evaluate": 12819,
      "case basque": 15578,
      "tasks unfortunately": 119582,
      "research groups": 100528,
      "languages models": 60732,
      "lower quality": 65447,
      "available best": 10954,
      "languages included": 60632,
      "particularly true": 85501,
      "smaller languages": 110235,
      "number monolingual": 80915,
      "fasttext word": 43207,
      "bert language": 13149,
      "corpora produce": 23560,
      "results publicly": 102100,
      "including topic": 53397,
      "benchmarks models": 12922,
      "used work": 128855,
      "work publicly": 134758,
      "assessing human": 9511,
      "translations french": 125467,
      "study present": 114474,
      "methods assessing": 69320,
      "aligned texts": 4924,
      "texts learning": 121540,
      "models underresourced": 74266,
      "written spoken": 135149,
      "certain kinds": 15944,
      "instructions human": 55712,
      "quality work": 94832,
      "work comparative": 134418,
      "document similarity": 33080,
      "information technology": 55033,
      "resources news": 101021,
      "unigram bigram": 127153,
      "adopted text": 3902,
      "implemented python": 51972,
      "python programming": 94485,
      "language toolkits": 60173,
      "values lower": 130796,
      "documents better": 33195,
      "similarity documents": 109224,
      "increases distance": 53641,
      "text gives": 121028,
      "gives accurate": 48178,
      "result better": 101362,
      "clustering ranking": 17954,
      "languages corpus": 60471,
      "global language": 48247,
      "datadriven language": 26662,
      "language mapping": 59274,
      "english arabic": 37069,
      "arabic russian": 8546,
      "identification model": 51398,
      "local languages": 64928,
      "data ii": 26014,
      "dataset total": 27238,
      "crawl web": 24595,
      "paper addition": 83699,
      "addition describing": 3421,
      "analysis relationship": 5801,
      "sources digital": 110885,
      "digital data": 31720,
      "webcrawled social": 132280,
      "media sources": 68166,
      "population demographics": 88165,
      "world datasets": 135025,
      "datasets representative": 27673,
      "datasets represent": 27671,
      "data makes": 26114,
      "makes better": 66761,
      "green ideas": 49080,
      "influence context": 54304,
      "acceptability ratings": 1892,
      "sentences judged": 106363,
      "context irrelevant": 22153,
      "irrelevant context": 56946,
      "context results": 22253,
      "relevant contexts": 98533,
      "bidirectional models": 13965,
      "model achieving": 70619,
      "processing central": 91633,
      "modelling text": 72629,
      "text discourse": 120878,
      "causal inference": 15806,
      "knowledge does": 57877,
      "everyday scenario": 39630,
      "knowledge induced": 58014,
      "measures correlation": 67859,
      "events corpus": 39571,
      "causal effect": 15805,
      "output method": 83093,
      "generation case": 47329,
      "studying deep": 114555,
      "effectiveness machine": 34902,
      "expressed english": 41712,
      "generation languages": 47455,
      "encoded neural": 36479,
      "corpora typically": 23607,
      "typically small": 126461,
      "benefit greatly": 12972,
      "pretraining based": 90235,
      "czech morphologically": 25493,
      "language pretraining": 59824,
      "judged automatic": 57427,
      "low data": 65355,
      "data scenarios": 26376,
      "unseen slot": 127543,
      "slot values": 110072,
      "online users": 81815,
      "attention key": 9863,
      "entities concepts": 37757,
      "categories existing": 15737,
      "volume entities": 131940,
      "entities categories": 37755,
      "topics language": 122634,
      "containing large": 21798,
      "various granularities": 131102,
      "methods compared": 69382,
      "compared variety": 19486,
      "variety baselines": 130965,
      "produced attention": 91949,
      "deployed various": 29313,
      "news recommendation": 79359,
      "limited authentic": 64208,
      "authentic parallel": 10356,
      "studies proved": 114267,
      "translation bt": 124680,
      "noisy generated": 80103,
      "generated imperfect": 47086,
      "data greatly": 25988,
      "greatly diminished": 49050,
      "scale monolingual": 103737,
      "noise synthetic": 80069,
      "achieve larger": 2493,
      "larger improvements": 61369,
      "improvements experimental": 52855,
      "effective improve": 34690,
      "deen translation": 28193,
      "crucial building": 25134,
      "building engaging": 14840,
      "engaging dialogue": 37019,
      "attracted research": 10159,
      "content quality": 21929,
      "relevance fluency": 98509,
      "enable better": 36349,
      "quality achieved": 94586,
      "impact usability": 51899,
      "framework uses": 45732,
      "uses information": 129233,
      "response prototype": 101231,
      "prototype retrieved": 93687,
      "generator takes": 47781,
      "input obtain": 55377,
      "obtain highquality": 81291,
      "indomain crossdomain": 53958,
      "tokens text": 122340,
      "semantics text": 105473,
      "read article": 95929,
      "article study": 9161,
      "endofsequence eos": 36859,
      "generation specifically": 47628,
      "specifically pretrained": 111577,
      "model gpt2": 71262,
      "story generation": 113415,
      "proportion words": 92514,
      "ranking given": 95669,
      "infrequent ones": 55174,
      "method termed": 69182,
      "domain identify": 33544,
      "identify suitable": 51563,
      "compared average": 19327,
      "method explainable": 68826,
      "explainable authorship": 41266,
      "algorithms utilize": 4889,
      "identify types": 51570,
      "media authors": 68082,
      "algorithms employ": 4845,
      "obtained social": 81412,
      "generative dialogue": 47722,
      "attention past": 9966,
      "challenge problem": 16086,
      "problem far": 91055,
      "methods evaluate": 69481,
      "lack systematic": 58756,
      "effective paper": 34728,
      "demonstrated effective": 28912,
      "sampling mechanism": 103601,
      "extremely imbalanced": 42599,
      "metric significantly": 69901,
      "improve correlation": 52361,
      "judgments using": 57452,
      "using augmented": 129361,
      "method stateoftheart": 69163,
      "world binary": 135021,
      "learning rank": 62933,
      "data dialogue": 25838,
      "view train": 131711,
      "train binary": 122906,
      "response candidate": 101191,
      "relevant irrelevant": 98561,
      "challenging work": 16346,
      "constructed human": 21663,
      "retrieval models": 102410,
      "propose multilevel": 92798,
      "method simple": 69152,
      "datasets stateoftheart": 27732,
      "stateoftheart matching": 112697,
      "learning summarize": 63073,
      "wikipedia revision": 132672,
      "revision histories": 102649,
      "histories paper": 50543,
      "automatically constructing": 10744,
      "particular method": 85431,
      "main body": 66399,
      "sentences added": 106199,
      "pairs quality": 83621,
      "training validation": 123945,
      "validation set": 130743,
      "summarization systems": 115573,
      "systems proposed": 117084,
      "proposed dataset": 93251,
      "dataset dataset": 26843,
      "contrast sets": 22713,
      "sets supervised": 107714,
      "systematic gaps": 116675,
      "decision rules": 27942,
      "perform test": 86090,
      "capabilities propose": 15175,
      "particular dataset": 85400,
      "dataset authors": 26749,
      "meaningful ways": 67734,
      "sets contrast": 107659,
      "sets provide": 107700,
      "local view": 64941,
      "decision boundary": 27935,
      "used accurately": 128375,
      "models true": 74251,
      "linguistic capabilities": 64433,
      "10 diverse": 130,
      "diverse nlp": 32828,
      "drop reading": 34170,
      "imdb sentiment": 51834,
      "analysis contrast": 5549,
      "cases release": 15664,
      "encourage future": 36755,
      "similar annotation": 109073,
      "annotation processes": 6357,
      "present newspaper": 89595,
      "guidelines introduce": 49263,
      "model serve": 71990,
      "work step": 134818,
      "text adventure": 120632,
      "learn optimal": 62106,
      "given state": 48135,
      "models direct": 73078,
      "direct access": 31781,
      "train state": 123026,
      "particularly pronounced": 85492,
      "deep qlearning": 28399,
      "siamese neural": 108654,
      "architecture novel": 8714,
      "adventure game": 4095,
      "convergence performance": 22921,
      "15 higher": 347,
      "higher baselines": 50164,
      "baselines able": 12344,
      "models representation": 73929,
      "systematic analysis": 116658,
      "work describes": 134468,
      "experiments probe": 41065,
      "bertstyle models": 13292,
      "discrete linguistic": 32169,
      "features feature": 43512,
      "representations attention": 99509,
      "specific attention": 111409,
      "subjectverb agreement": 114711,
      "largescale opensource": 61490,
      "media dialog": 68106,
      "existing conversational": 40090,
      "datasets consist": 27375,
      "natural speech": 76621,
      "collected news": 18429,
      "interview transcripts": 56329,
      "transcripts compared": 124011,
      "outofdomain performance": 82658,
      "demonstrating usefulness": 28983,
      "realworld conversations": 96155,
      "role annotations": 103156,
      "facilitating development": 42809,
      "fact experiments": 42821,
      "labels improves": 58610,
      "generate specific": 47018,
      "aims identify": 4537,
      "human labels": 50894,
      "healthcare applications": 49631,
      "billing codes": 14077,
      "modalities text": 70482,
      "considerable amounts": 21241,
      "zeroshot sequence": 135524,
      "labeling method": 58509,
      "labeling convolutional": 58488,
      "convolutional decomposition": 23154,
      "case available": 15577,
      "available documentlevel": 10986,
      "finegrained features": 44358,
      "health record": 49626,
      "healthcare workers": 49634,
      "learning selfsupervised": 62989,
      "data comparable": 25753,
      "model makes": 71493,
      "representation types": 99456,
      "readability index": 95946,
      "learning wikipedia": 63163,
      "data suitable": 26517,
      "high school": 50135,
      "school students": 103966,
      "undergraduate students": 126669,
      "corpus finegrained": 23803,
      "finegrained named": 44367,
      "events important": 39582,
      "management extracting": 66903,
      "pertaining specific": 87188,
      "challenge work": 16118,
      "documents annotated": 33180,
      "annotated finegrained": 6197,
      "annotated set": 6234,
      "allows training": 5191,
      "training evaluating": 123615,
      "algorithms aim": 4827,
      "corpus study": 24022,
      "study annotation": 114320,
      "types relations": 126348,
      "texts important": 121527,
      "guidelines annotation": 49260,
      "relation mentions": 98054,
      "present preliminary": 89646,
      "preliminary corpus": 89264,
      "web social": 132259,
      "kornli korsts": 58349,
      "key tasks": 57608,
      "languages publicly": 60827,
      "motivated construct": 74845,
      "construct release": 21649,
      "following previous": 44977,
      "use propose": 128217,
      "challenge task": 16107,
      "given written": 48176,
      "situation real": 109891,
      "aspect human": 9331,
      "understanding ability": 126795,
      "ability use": 1575,
      "models finetuned": 73253,
      "examples best": 39813,
      "does worse": 33410,
      "worse low": 135065,
      "reveals language": 102517,
      "errors hard": 38373,
      "generative setting": 47758,
      "showing room": 108430,
      "room progress": 103269,
      "streaming simultaneous": 113556,
      "source content": 110735,
      "content available": 21861,
      "available study": 11117,
      "related problem": 97888,
      "live captioning": 64846,
      "streaming approaches": 113553,
      "source token": 110854,
      "translation scratch": 125232,
      "allow revisions": 5091,
      "systems experiment": 116889,
      "experiment showing": 40509,
      "showing large": 108410,
      "dynamic data": 34302,
      "selection weighting": 104846,
      "utilize monolingual": 130516,
      "backtranslation improve": 11337,
      "require resulting": 100193,
      "data high": 26001,
      "domain achieve": 33425,
      "achieve goals": 2467,
      "strategies proposed": 113485,
      "domain dissimilar": 33509,
      "generaldomain text": 46737,
      "dynamic curriculum": 34300,
      "backtranslation models": 11339,
      "strategies based": 113452,
      "iteration evaluate": 57121,
      "models domain": 73104,
      "adaptation lowresource": 3228,
      "lowresource highresource": 65499,
      "mt settings": 74992,
      "success pretrained": 115113,
      "particular models": 85433,
      "models employ": 73141,
      "tokenization methods": 122279,
      "methods notably": 69644,
      "unigram language": 127158,
      "kudo 2018": 58357,
      "text best": 120685,
      "literature does": 64752,
      "contain direct": 21737,
      "compare finetuned": 19248,
      "finetuned task": 44437,
      "tokenization method": 122278,
      "method matches": 68956,
      "matches outperforms": 67384,
      "pretrained lms": 90121,
      "lm method": 64866,
      "problem multiple": 91134,
      "employ bert": 36252,
      "comparison bert": 19532,
      "result new": 101385,
      "surpasses previous": 116108,
      "work substantial": 134827,
      "substantial margin": 114871,
      "margin exploring": 67193,
      "parameterefficient transfer": 84729,
      "learning finetuning": 62585,
      "downstream language": 34000,
      "downstream generation": 33999,
      "diverse language": 32819,
      "tasks just": 119232,
      "parameters task": 84788,
      "model maintain": 71488,
      "maintain improve": 66503,
      "investigate relationship": 56810,
      "13 total": 289,
      "matrix factorization": 67475,
      "better data": 13558,
      "utterance according": 130589,
      "offers advantages": 81581,
      "domain extracting": 33533,
      "labeling sequence": 58540,
      "approaches presented": 8286,
      "detection error": 29940,
      "correction approaches": 24143,
      "present sequence": 89689,
      "approach error": 7535,
      "correction approach": 24142,
      "approaches trained": 8378,
      "trained used": 123323,
      "endtoend avoid": 36879,
      "detection correction": 29913,
      "got accuracy": 48531,
      "realworld test": 96183,
      "data designing": 25832,
      "robust dialogue": 103029,
      "achieve moderate": 2498,
      "moderate correlation": 74373,
      "pretrained masked": 90122,
      "different tag": 31473,
      "ner fundamental": 77046,
      "component modern": 19988,
      "modern language": 74394,
      "understanding pipeline": 126919,
      "public ner": 94266,
      "ner resources": 77074,
      "resources annotated": 100942,
      "given particular": 48080,
      "types users": 126383,
      "leverage multiple": 63607,
      "multiple resources": 75672,
      "sets paper": 107692,
      "training unified": 123940,
      "ner model": 77053,
      "contrast recent": 22711,
      "models original": 73671,
      "flexibility makes": 44649,
      "makes easier": 66775,
      "sensitive domains": 105738,
      "different ner": 31283,
      "ner architectures": 77022,
      "par strong": 84527,
      "crf baseline": 24744,
      "selfattentive model": 104910,
      "selfattentive neural": 104912,
      "neural syntactic": 78694,
      "parsers using": 85052,
      "produce stateoftheart": 91935,
      "joint parsing": 57305,
      "additional unlabeled": 3594,
      "selftraining semisupervised": 104960,
      "semisupervised technique": 105627,
      "technique incorporating": 119793,
      "incorporating unlabeled": 53571,
      "stateoftheart selfattentive": 112942,
      "selfattentive parser": 104913,
      "detection demonstrating": 29920,
      "provides benefits": 94002,
      "pretrained contextualized": 90010,
      "provides gains": 94036,
      "existing tasks": 40309,
      "tasks second": 119479,
      "stateoftheart bert": 112601,
      "bert representations": 13218,
      "task conduct": 118000,
      "competitive task": 19698,
      "model integrated": 71351,
      "erroneous sentence": 38272,
      "contained input": 21773,
      "approach integrated": 7643,
      "syntactic word": 116499,
      "improve classification": 52348,
      "sentences accurately": 106197,
      "advertisement text": 4195,
      "search advertising": 104293,
      "need search": 76857,
      "focus general": 44768,
      "ignoring diverse": 51687,
      "generate candidate": 46909,
      "needs based": 76887,
      "based queries": 11963,
      "field based": 43932,
      "based external": 11708,
      "added input": 3364,
      "serve bridges": 107293,
      "transfer ability": 124032,
      "association model": 9624,
      "use various": 128354,
      "pretrained natural": 90169,
      "patterns general": 85747,
      "general text": 46722,
      "makes existing": 66780,
      "useful practice": 128915,
      "practice work": 88738,
      "modeling better": 72382,
      "dialogue behavior": 30643,
      "incorporate user": 53492,
      "task pretrained": 118553,
      "dialogue applications": 30640,
      "intention recognition": 55915,
      "recognition dialogue": 96860,
      "prediction response": 89119,
      "ability mitigate": 1542,
      "mitigate data": 70360,
      "problem taskoriented": 91261,
      "logic framework": 64990,
      "topdown bottomup": 122489,
      "architecture designed": 8644,
      "designed integrate": 29603,
      "analysis nonstandard": 5722,
      "utterance used": 130618,
      "context architecture": 22011,
      "concrete use": 20740,
      "answers written": 6749,
      "german learners": 47903,
      "comprehension questions": 20222,
      "questions reading": 95350,
      "reading context": 96016,
      "context possible": 22220,
      "possible target": 88436,
      "target answers": 117582,
      "answers given": 6730,
      "integrates existing": 55784,
      "existing nlp": 40237,
      "components produce": 20045,
      "levels linguistic": 63548,
      "linguistic modeling": 64510,
      "framework maximum": 45610,
      "architecture builds": 8631,
      "representations amrs": 99499,
      "amrs represent": 5401,
      "provided target": 93984,
      "apply architecture": 7160,
      "architecture tasks": 8758,
      "generation selfsupervised": 47609,
      "mass bart": 67320,
      "generation existing": 47394,
      "existing pretraining": 40259,
      "train transformerbased": 123046,
      "answering conversational": 6613,
      "new text": 79219,
      "context work": 22332,
      "novel scheme": 80710,
      "text conditioned": 120812,
      "conditioned context": 20798,
      "new scheme": 79118,
      "benchmarks covering": 12895,
      "rank official": 95626,
      "generation squad": 47633,
      "movie dialogues": 74896,
      "bert learn": 13156,
      "models targeted": 74158,
      "information interdocument": 54698,
      "scientific documents": 104001,
      "documents classification": 33200,
      "embeddings power": 35868,
      "end tasks": 36833,
      "citation graph": 16990,
      "taskspecific finetuning": 119627,
      "research documentlevel": 100478,
      "documentlevel tasks": 33167,
      "tasks ranging": 119432,
      "baselines benchmark": 12360,
      "parser transfer": 85002,
      "focus predicting": 44804,
      "selection framework": 104783,
      "delexicalized parser": 28603,
      "empirical observation": 36174,
      "instance level": 55604,
      "level end": 63440,
      "selection baselines": 104770,
      "similarity manually": 109253,
      "encoding syntactic": 36729,
      "properties languages": 92457,
      "parsers 20": 85010,
      "20 diverse": 501,
      "test languages": 120466,
      "languages respectively": 60849,
      "instancelevel model": 55615,
      "present characterbased": 89406,
      "words varying": 134303,
      "varying amounts": 131253,
      "tools model": 122456,
      "structures database": 114068,
      "morphology model": 74792,
      "words parts": 134111,
      "constituent structure": 21548,
      "possible generate": 88407,
      "generate optimal": 46980,
      "task split": 118734,
      "corpus manually": 23877,
      "split word": 111945,
      "enriching transformer": 37568,
      "factors lowresource": 42897,
      "translation introducing": 124876,
      "information referring": 54907,
      "known improve": 58305,
      "certain settings": 15954,
      "recurrent architectures": 97160,
      "study proposes": 114491,
      "translation architecture": 124646,
      "architecture transformer": 8765,
      "allows introduce": 5165,
      "particular proposed": 85445,
      "features study": 43745,
      "different architectural": 31002,
      "performance combining": 86216,
      "features embedding": 43474,
      "level encoder": 63439,
      "level experiment": 63442,
      "combination strategies": 18585,
      "08 bleu": 74,
      "baseline transformer": 12327,
      "benchmark includes": 12832,
      "languages obtain": 60768,
      "obtain improvement": 81294,
      "extraction dataset": 42294,
      "arguments appear": 8979,
      "sentences argue": 106215,
      "role proposed": 103214,
      "traditional tasks": 122879,
      "tasks considering": 119005,
      "metric evaluate": 69878,
      "performance representative": 86672,
      "representative methods": 100012,
      "bestperforming model": 13475,
      "leads gains": 61933,
      "gains standard": 46408,
      "evaluation settings": 39388,
      "datadriven dependency": 26658,
      "attributed lack": 10190,
      "lack availability": 58684,
      "datadriven machine": 26665,
      "models originally": 73672,
      "training focus": 123637,
      "specific features": 111443,
      "explicitly models": 41388,
      "work analyse": 134369,
      "performance parsers": 86602,
      "using indomain": 129751,
      "impact word": 51902,
      "sentences provided": 106461,
      "provided input": 93971,
      "input systems": 55447,
      "agents ability": 4318,
      "work introduced": 134582,
      "introduced tasks": 56599,
      "datasets aim": 27304,
      "aim help": 4459,
      "agents learn": 4323,
      "investigate ways": 56830,
      "ways combine": 132149,
      "combine models": 18632,
      "ranging simple": 95618,
      "model aggregation": 70647,
      "aggregation schemes": 4367,
      "schemes require": 103953,
      "minimal additional": 70154,
      "training various": 123946,
      "training schemes": 123830,
      "particular capabilities": 85386,
      "capabilities results": 15176,
      "twostage approaches": 126159,
      "constructed avoid": 21653,
      "avoid unwanted": 11243,
      "using static": 130219,
      "like statistical": 64098,
      "methods extract": 69502,
      "translations nmt": 125478,
      "nmt approaches": 79833,
      "propose word": 93161,
      "idea leverage": 51330,
      "leverage multilingual": 63606,
      "embeddings static": 35953,
      "data relying": 26336,
      "relying parallel": 98764,
      "superior comparable": 115682,
      "comparable language": 19146,
      "pairs compared": 83497,
      "achieve word": 2587,
      "alignment f1": 4959,
      "knowledge graphbased": 57976,
      "graphbased dialogue": 48896,
      "generation improved": 47434,
      "metalearning knowledge": 68551,
      "systems capable": 116767,
      "generating informative": 47226,
      "reasoning mechanisms": 96270,
      "incompleteness knowledge": 53427,
      "kg paper": 57660,
      "proposes dynamic": 93598,
      "knowledge triples": 58217,
      "problem adversarial": 90923,
      "adversarial attack": 4101,
      "knowledgeaware dialogue": 58244,
      "generation train": 47682,
      "using minimal": 129884,
      "minimal training": 70180,
      "model initialize": 71346,
      "adapt previous": 3192,
      "unseen knowledge": 127529,
      "lexicon development": 63879,
      "support social": 115991,
      "demonstrate improvements": 28763,
      "approach resource": 7863,
      "performance state": 86739,
      "linguistic diversity": 64468,
      "diversity inclusion": 32879,
      "world language": 135035,
      "diversity world": 32900,
      "world small": 135048,
      "types languages": 126307,
      "languages resources": 60847,
      "followed time": 44955,
      "terms resources": 120375,
      "calls question": 15075,
      "acl community": 2996,
      "pretraining recently": 90312,
      "models follow": 73260,
      "paradigm achieved": 84532,
      "great performance": 49009,
      "various downstream": 131085,
      "pretraining stage": 90326,
      "usually suffers": 130459,
      "threestage framework": 121943,
      "modeling indomain": 72448,
      "learn domainspecific": 62033,
      "domainspecific patterns": 33911,
      "novel selective": 80711,
      "strategy learn": 113526,
      "specifically design": 111538,
      "design method": 29546,
      "measure importance": 67802,
      "importance token": 52087,
      "important tokens": 52280,
      "achieve comparable": 2439,
      "performance 50": 86105,
      "computation cost": 20347,
      "indicates method": 53861,
      "effective efficient": 34669,
      "present easy": 89458,
      "allows create": 5137,
      "create multilingual": 24629,
      "use original": 128184,
      "original monolingual": 82533,
      "methods training": 69811,
      "models relatively": 73913,
      "samples new": 103587,
      "hardware requirements": 49499,
      "requirements training": 100238,
      "available learning": 11030,
      "learning relation": 62954,
      "distant supervised": 32457,
      "extraction relation": 42466,
      "relations critical": 98132,
      "failing capture": 42977,
      "topology structure": 122686,
      "specifically build": 111531,
      "build graph": 14769,
      "global cooccurrence": 48233,
      "graph learn": 48829,
      "learn correlation": 62013,
      "relations finally": 98176,
      "capable modeling": 15207,
      "global relation": 48264,
      "baselines furthermore": 12397,
      "graph used": 48880,
      "existing relation": 40266,
      "systems improve": 116946,
      "transformers recently": 124461,
      "achieved considerable": 2613,
      "considerable success": 21260,
      "studied investigate": 114165,
      "various linguistic": 131122,
      "focusing analyzing": 44921,
      "output attention": 83050,
      "input vectors": 55471,
      "analyses bert": 5442,
      "transformerbased neural": 124430,
      "include following": 53225,
      "ii reasonable": 51714,
      "findings provide": 44308,
      "insights inner": 55537,
      "workings transformers": 134911,
      "multilingual unsupervised": 75397,
      "translation unsupervised": 125407,
      "translation unmt": 125404,
      "unmt recently": 127469,
      "results language": 101873,
      "pairs time": 83647,
      "single encoder": 109731,
      "single decoder": 109723,
      "use multilingual": 128154,
      "distillation methods": 32511,
      "performance experiments": 86360,
      "surpassing strong": 116120,
      "strong unsupervised": 113729,
      "baselines achieving": 12350,
      "pairs supervised": 83643,
      "supervised graphemetophoneme": 115757,
      "graphemetophoneme g2p": 48924,
      "g2p conversion": 46331,
      "schwa deletion": 103970,
      "using prosodic": 130062,
      "phonetic analysis": 87289,
      "present statistical": 89718,
      "lexicon extracted": 63881,
      "language punjabi": 60007,
      "diversity quality": 32891,
      "tasks storytelling": 119523,
      "right decoding": 102832,
      "presently exists": 89813,
      "decoding procedure": 28126,
      "best criteria": 13323,
      "holtzman et": 50605,
      "counterintuitive observation": 24430,
      "evaluate algorithm": 38785,
      "paper frame": 83962,
      "task access": 117828,
      "access labeled": 1913,
      "pairs training": 83651,
      "predict concept": 88877,
      "concept relations": 20585,
      "relations graph": 98179,
      "method notably": 68992,
      "graphbased model": 48902,
      "representations task": 99928,
      "expand existing": 40356,
      "existing corpus": 40095,
      "topics learning": 122635,
      "learning classify": 62438,
      "classify intents": 17651,
      "intents slot": 55924,
      "slot labels": 110065,
      "labels given": 58608,
      "examples intent": 39848,
      "classification ic": 17223,
      "filling sf": 44052,
      "core components": 23316,
      "small propose": 110189,
      "classes seen": 17075,
      "datasets atis": 27320,
      "meta learning": 68532,
      "prototypical networks": 93691,
      "finetuning baseline": 44448,
      "baseline benchmark": 12195,
      "networks achieves": 77504,
      "atis datasets": 9680,
      "datasets addition": 27298,
      "addition demonstrate": 3420,
      "disentangled representation": 32348,
      "generation learning": 47459,
      "focused supervised": 44879,
      "training taskspecific": 123909,
      "measures reflecting": 67893,
      "similarity data": 109215,
      "attributes apply": 10192,
      "disentangle semantics": 32343,
      "approaches general": 8169,
      "framework applicable": 45425,
      "tasks dense": 119046,
      "embeddings order": 35843,
      "create suitable": 24643,
      "knowledge sense": 58169,
      "information interactions": 54696,
      "multihead selfattention": 75129,
      "token dependencies": 122248,
      "input prior": 55401,
      "measures fail": 67865,
      "propose selfattention": 93029,
      "method interpret": 68919,
      "transformer bert": 124308,
      "example conduct": 39776,
      "studies firstly": 114229,
      "apply selfattention": 7214,
      "heads pruned": 49610,
      "marginal performance": 67208,
      "results used": 102298,
      "error corrections": 38301,
      "involves tradeoff": 56905,
      "cost training": 24376,
      "data example": 25912,
      "example machine": 39788,
      "learning weak": 63161,
      "talks english": 117562,
      "requiring significantly": 100357,
      "significantly human": 108923,
      "used successfully": 128790,
      "readers access": 95957,
      "employ recurrent": 36275,
      "rnn combined": 102910,
      "ctc loss": 25193,
      "data avoid": 25686,
      "different loss": 31239,
      "function improve": 46026,
      "source parallel": 110802,
      "text available": 120678,
      "possibility generating": 88369,
      "model capability": 70801,
      "utilizes context": 130552,
      "error best": 38290,
      "best configuration": 13322,
      "measure contribution": 67787,
      "context learning": 22164,
      "data error": 25901,
      "multiple segmentations": 75680,
      "white spaces": 132474,
      "complexity building": 19901,
      "englishthai parallel": 37435,
      "encoding different": 36687,
      "obtained experiments": 81369,
      "datasets performance": 27621,
      "autoregressive generation": 10863,
      "models gpt": 73309,
      "uniform prior": 127141,
      "model main": 71487,
      "main advantage": 66393,
      "surprisingly good": 116137,
      "potentially enable": 88610,
      "applications traditional": 7028,
      "unidirectional generation": 127093,
      "generation pretrained": 47553,
      "set downstream": 107425,
      "downstream nlu": 34017,
      "stochastic gates": 113354,
      "information subset": 55013,
      "encoder states": 36577,
      "investigate effects": 56749,
      "experiments depending": 40899,
      "depending task": 29274,
      "potential improving": 88566,
      "improving decoding": 53084,
      "decoding efficiency": 28087,
      "standard transformer": 112328,
      "certain word": 15963,
      "observations explore": 81176,
      "outputs based": 83159,
      "optimization recently": 82209,
      "vae based": 130699,
      "impressive progress": 52321,
      "improving diversity": 53089,
      "diversity generated": 32875,
      "improvements paper": 52892,
      "novel multimodal": 80658,
      "relevance diversity": 98507,
      "simple gaussian": 109438,
      "capture complex": 15278,
      "data posterior": 26251,
      "response relevance": 101234,
      "prior posterior": 90723,
      "optimization strategy": 82212,
      "training criteria": 123412,
      "hard experimental": 49459,
      "results widely": 102331,
      "learning unlabeled": 63140,
      "better tackle": 13738,
      "tackle named": 117295,
      "ner problem": 77069,
      "learned source": 62259,
      "rich labeled": 102754,
      "data previous": 26274,
      "ner based": 77023,
      "methods applicable": 69310,
      "train student": 123029,
      "model unlabeled": 72246,
      "language proposed": 60000,
      "multisource crosslingual": 75783,
      "similarity measuring": 109264,
      "better weight": 13768,
      "supervision different": 115879,
      "languages benchmark": 60424,
      "largescale pretrained": 61493,
      "models gpt2": 73310,
      "lefttoright language": 63307,
      "text largely": 121079,
      "unexplored work": 127063,
      "work takes": 134843,
      "step understanding": 113311,
      "outputs terms": 83184,
      "terms discourse": 120307,
      "comprehensive study": 20272,
      "explicit discourse": 41316,
      "generation finetuned": 47409,
      "scenarios results": 103867,
      "gpt2 does": 48560,
      "scenario propose": 103829,
      "modeling discourse": 72419,
      "access external": 1908,
      "different modalities": 31260,
      "storing knowledge": 113403,
      "knowledge propose": 58126,
      "transformer neural": 124363,
      "read operation": 95936,
      "knowledge apply": 57755,
      "modules generative": 74524,
      "generative dialog": 47720,
      "performance measured": 86525,
      "measured automatic": 67833,
      "neural modeling": 78157,
      "datasets offer": 27602,
      "gap building": 46444,
      "highquality dataset": 50376,
      "datasets german": 27495,
      "spanish portuguese": 111164,
      "available project": 11080,
      "various heuristics": 131103,
      "analysis extracted": 5617,
      "showing better": 108397,
      "zeroshot finetuning": 135502,
      "settings training": 107840,
      "opensubtitles dataset": 82037,
      "little additional": 64786,
      "web demo": 132230,
      "matching retrieving": 67430,
      "translation memories": 124930,
      "systems matching": 116998,
      "based edit": 11667,
      "introduce sentence": 56533,
      "encoders improve": 36649,
      "improve matching": 52414,
      "efficient solution": 35116,
      "distance based": 32406,
      "based algorithms": 11502,
      "current context": 25268,
      "methods obtaining": 69649,
      "rely complicated": 98680,
      "simple fixed": 109434,
      "policies design": 87942,
      "algorithm achieve": 4688,
      "lower latency": 65434,
      "change forms": 16360,
      "level meaning": 63475,
      "meaning challenging": 67619,
      "verbs paper": 131559,
      "computationally simple": 20458,
      "share pypi": 107960,
      "information knowledge": 54709,
      "knowledge common": 57828,
      "information pair": 54820,
      "introduce word": 56569,
      "translation generate": 124836,
      "train embedding": 122929,
      "wordlevel sentencelevel": 133755,
      "singleturn multiturn": 109854,
      "supervised pos": 115821,
      "taggers perform": 117369,
      "languages partofspeech": 60795,
      "supervision crosslingual": 115876,
      "supervision combination": 115874,
      "perform supervised": 86081,
      "supervised ones": 115818,
      "commonly evaluated": 18960,
      "evaluated languages": 38983,
      "languages taggers": 60911,
      "use sources": 128287,
      "languages train": 60924,
      "highlight need": 50266,
      "need new": 76838,
      "approaches pos": 8283,
      "representation position": 99377,
      "position encoding": 88287,
      "tasks generating": 119144,
      "languages modeling": 60731,
      "positional relationships": 88304,
      "augment sans": 10265,
      "sentence specifically": 106089,
      "specifically utilize": 111609,
      "information encourage": 54528,
      "significantly consistently": 108899,
      "quality strong": 94803,
      "baselines extensive": 12395,
      "extensive analyses": 41852,
      "confirm performance": 21016,
      "gains come": 46388,
      "words improve": 133991,
      "efficiency main": 35031,
      "inaccurate results": 53192,
      "results suboptimal": 102224,
      "performance modeling": 86538,
      "modeling sentences": 72542,
      "proposed explicitly": 93276,
      "information mi": 54763,
      "reconstruction loss": 97120,
      "loss based": 65256,
      "domains results": 33853,
      "vanilla transformer": 130824,
      "preserving high": 89935,
      "accuracy efficiency": 2143,
      "bert lowresource": 13159,
      "huge success": 50728,
      "supervised zeroshot": 115858,
      "learning success": 63070,
      "104 languages": 191,
      "languages wikipedia": 60969,
      "trained paper": 123232,
      "approach extend": 7557,
      "benefit new": 12983,
      "languages mbert": 60722,
      "experiments named": 41030,
      "average increase": 11196,
      "23 f1": 789,
      "logical operations": 65013,
      "correctness textual": 24187,
      "words symbolic": 134251,
      "capable leveraging": 15206,
      "largescale benchmark": 61400,
      "dataset built": 26774,
      "semistructured tables": 105589,
      "network built": 77178,
      "automatically derives": 10752,
      "heterogeneous graph": 49856,
      "related contexts": 97848,
      "used obtain": 128670,
      "ablation experiments": 1583,
      "important obtain": 52200,
      "results document": 101736,
      "classification graph": 17219,
      "classification fundamental": 17215,
      "fundamental natural": 46100,
      "inductive learning": 54049,
      "document use": 33103,
      "finegrained word": 44394,
      "produce embeddings": 91886,
      "embeddings unseen": 35999,
      "document finally": 33009,
      "word nodes": 133371,
      "embedding extensive": 35405,
      "stateoftheart text": 113006,
      "nonlinear dynamic": 80223,
      "dynamic systems": 34328,
      "run paper": 103462,
      "paper perform": 84066,
      "perform systematic": 86084,
      "systematic empirical": 116669,
      "suggest future": 115312,
      "work direction": 134478,
      "represented languages": 100030,
      "data research": 26352,
      "research resources": 100618,
      "collection curation": 18468,
      "task creating": 118035,
      "creating datasets": 24697,
      "datasets focused": 27488,
      "news headlines": 79343,
      "news topic": 79381,
      "datasets study": 27734,
      "document work": 33111,
      "propose baselines": 92571,
      "classification investigate": 17238,
      "investigate approach": 56717,
      "performance classifiers": 86206,
      "analyzing political": 6047,
      "device used": 30455,
      "critical purposes": 24824,
      "widespread phenomenon": 132617,
      "media popular": 68151,
      "computational study": 20435,
      "new publicly": 79091,
      "set tweets": 107619,
      "tweets real": 126053,
      "different genders": 31160,
      "finally identify": 44193,
      "important improving": 52170,
      "translational equivalence": 125440,
      "meaning languages": 67645,
      "computational lexical": 20387,
      "field suffers": 43986,
      "framework define": 45484,
      "define properties": 28490,
      "proposes unifying": 93619,
      "treatment relations": 125567,
      "validated experiments": 130733,
      "expansion language": 40372,
      "model probing": 71790,
      "task benefits": 117930,
      "benefits downstream": 13007,
      "answering query": 6677,
      "construction existing": 21687,
      "existing set": 40285,
      "expansion methods": 40373,
      "selecting context": 104747,
      "features extracting": 43511,
      "extracting new": 42224,
      "entities key": 37802,
      "challenge entity": 16031,
      "later iterations": 61646,
      "novel iterative": 80605,
      "address semantic": 3763,
      "probing pretrained": 90903,
      "candidate entity": 15103,
      "entity based": 37908,
      "based selected": 12006,
      "framework generates": 45553,
      "bert shown": 13226,
      "variants proposed": 130880,
      "chinese pretrained": 16806,
      "models examine": 73181,
      "model series": 71989,
      "community propose": 19086,
      "carried extensive": 15538,
      "tasks revisit": 119476,
      "performances nlp": 86888,
      "findings help": 44294,
      "help future": 49726,
      "dialogue corpora": 30652,
      "largescale dialogue": 61426,
      "recently available": 96665,
      "quality utterance": 94824,
      "proposed scoring": 93542,
      "scoring method": 104231,
      "method designed": 68767,
      "designed based": 29582,
      "widely shared": 132553,
      "good correlation": 48470,
      "furthermore method": 46189,
      "long context": 65065,
      "state generation": 112497,
      "predicts dialogue": 89216,
      "tagging technique": 117455,
      "task taskoriented": 118779,
      "enabling model": 36413,
      "context approaches": 22010,
      "approaches attempt": 8068,
      "baseline significantly": 12310,
      "significantly drops": 108907,
      "context sequence": 22265,
      "establishing new": 38616,
      "stateoftheart joint": 112678,
      "joint goal": 57278,
      "goal accuracy": 48326,
      "multiwoz 20": 75952,
      "20 dataset": 498,
      "dataset syntaxaware": 27229,
      "effective performance": 34730,
      "generating additional": 47198,
      "additional bilingual": 3494,
      "strategy neural": 113531,
      "methods simply": 69758,
      "choose words": 16928,
      "roles sentence": 103235,
      "tree input": 125597,
      "effective clue": 34645,
      "dataset result": 27164,
      "proposed syntaxaware": 93561,
      "significant translation": 108873,
      "plausible explanations": 87683,
      "explanations models": 41296,
      "faithful explanation": 43001,
      "predictions work": 89198,
      "explanation models": 41285,
      "carry meaning": 15556,
      "does affect": 33327,
      "affect models": 4220,
      "experiments wide": 41204,
      "explanation predictions": 41287,
      "modified lstm": 74448,
      "different time": 31498,
      "resulting attention": 101430,
      "provide precise": 93896,
      "importance ranking": 52075,
      "indicative words": 53877,
      "distributions learned": 32764,
      "predictions code": 89161,
      "italian language": 57095,
      "years pretrained": 135288,
      "architectures provided": 8841,
      "mainly english": 66484,
      "english develop": 37110,
      "architecture provide": 8735,
      "means automatic": 67756,
      "automatic humanbased": 10562,
      "humanbased evaluation": 51001,
      "automatic assessment": 10497,
      "sentences human": 106343,
      "simpler language": 109559,
      "systems response": 117120,
      "systems correlate": 116821,
      "evaluation focus": 39216,
      "construct test": 21650,
      "evaluating systems": 39106,
      "evaluation compared": 39150,
      "compared widely": 19489,
      "english current": 37101,
      "domain models": 33585,
      "especially usergenerated": 38525,
      "drops performance": 34188,
      "stateoftheart tagging": 112996,
      "reddit forum": 97301,
      "forum discussions": 45304,
      "specifically use": 111607,
      "large benchmark": 61039,
      "results training": 102279,
      "data tested": 26554,
      "amounts indomain": 5339,
      "data outperform": 26197,
      "web domains": 132233,
      "domains make": 33811,
      "make progress": 66712,
      "performance tagging": 86782,
      "closer reality": 17876,
      "novel multilingual": 80657,
      "adaptation approach": 3208,
      "adapter modules": 3297,
      "enables learn": 36390,
      "language embeddings": 58996,
      "allows easy": 5146,
      "existing linguistic": 40162,
      "network resulting": 77411,
      "strong monolingual": 113689,
      "multilingual baselines": 75199,
      "highresource lowresource": 50431,
      "languages showing": 60869,
      "proposed adaptation": 93168,
      "typological features": 126485,
      "features key": 43577,
      "key success": 57601,
      "supervised word": 115857,
      "span target": 111135,
      "squad v20": 112081,
      "style question": 114589,
      "task solve": 118724,
      "bert finetuned": 13129,
      "created gold": 24669,
      "accuracy adding": 2095,
      "context token": 22298,
      "question experiments": 95156,
      "alignment datasets": 4952,
      "datasets chinese": 27344,
      "english proposed": 37251,
      "significantly outperformed": 108996,
      "outperformed previous": 82786,
      "previous supervised": 90506,
      "achieved f1": 2621,
      "133 points": 300,
      "higher previous": 50194,
      "methods bilingual": 69349,
      "extract bilingual": 42071,
      "texts automatically": 121461,
      "extract span": 42123,
      "span use": 111138,
      "utilize pretrained": 130523,
      "using integer": 129761,
      "sentence extraction": 105869,
      "extraction experiment": 42332,
      "accuracy baseline": 2105,
      "newspaper articles": 79406,
      "articles proposed": 9208,
      "bert achieves": 13058,
      "arabic information": 8517,
      "extraction multilingual": 42396,
      "transformers mbert": 124456,
      "crosslingual zeroshot": 25050,
      "performance arabic": 86149,
      "studied paper": 114167,
      "bilingual bert": 14022,
      "learning study": 63066,
      "zeroshort transfer": 135487,
      "argument role": 8956,
      "mbert xlmroberta": 67574,
      "exploiting sentence": 41478,
      "document alignment": 32949,
      "method incorporates": 68897,
      "information candidate": 54404,
      "rescoring method": 100391,
      "61 relative": 1160,
      "reduction error": 97447,
      "best previously": 13416,
      "downstream mt": 34003,
      "mt performance": 74983,
      "corpora method": 23526,
      "uses multilingual": 129248,
      "embeddings demonstrating": 35638,
      "evaluation languages": 39250,
      "frame task": 45389,
      "copy input": 23287,
      "best case": 13315,
      "matches human": 67380,
      "simple intuitive": 109449,
      "39 languages": 987,
      "languages outperforms": 60772,
      "data explore": 25933,
      "submission wmt": 114738,
      "interesting facts": 56072,
      "facts logical": 42919,
      "records provided": 97137,
      "hard existing": 49458,
      "nlg generation": 79519,
      "involving common": 56911,
      "great challenges": 48996,
      "challenges models": 16176,
      "understand semantics": 126774,
      "paired examples": 83474,
      "models analyze": 72751,
      "dataset encourage": 26889,
      "building advanced": 14821,
      "generation dataset": 47353,
      "knowledge injection": 58018,
      "generation language": 47453,
      "generation successfully": 47645,
      "successfully learned": 115190,
      "produce general": 91892,
      "knowledge develop": 57858,
      "different prior": 31357,
      "assumption model": 9663,
      "sufficient knowledge": 115281,
      "people topics": 85897,
      "needed knowledge": 76875,
      "knowledge address": 57747,
      "approach inject": 7638,
      "inject knowledge": 55256,
      "train largescale": 122949,
      "textual knowledge": 121717,
      "model access": 70525,
      "limited knowledge": 64243,
      "work producing": 134707,
      "producing coherent": 92016,
      "coherent informative": 18318,
      "come long": 18768,
      "long way": 65147,
      "developing novel": 30356,
      "advancements neural": 3971,
      "summarization dialog": 115493,
      "evaluated synthetic": 39010,
      "synthetic tasks": 116647,
      "representative performance": 100016,
      "applications investigate": 6948,
      "tasks downstream": 119076,
      "downstream use": 34063,
      "experiments benchmarking": 40804,
      "models synthetic": 74148,
      "synthetic sentence": 116642,
      "tasks contrast": 119014,
      "contrast performance": 22701,
      "correlation model": 24240,
      "alignment core": 4951,
      "core unsupervised": 23336,
      "injection statistical": 55269,
      "approaches superior": 8364,
      "remains key": 98799,
      "output training": 83140,
      "autoencoders text": 10418,
      "simplification ts": 109596,
      "sentences simplified": 106496,
      "models heavily": 73321,
      "quality parallel": 94747,
      "leverage large": 63594,
      "modeling automatic": 72381,
      "generation parallel": 47532,
      "data iterative": 26055,
      "generate appropriate": 46904,
      "simple complex": 109388,
      "quite similar": 95404,
      "sentences tackle": 106515,
      "denoising methods": 29000,
      "sentences separate": 106491,
      "complexity modeling": 19926,
      "autoencoders introduce": 10417,
      "introduce different": 56410,
      "manner automatic": 66939,
      "simplification systems": 109592,
      "dual learning": 34237,
      "learning modular": 62799,
      "corresponding natural": 24296,
      "representations dual": 99609,
      "performance dual": 86328,
      "work learned": 134607,
      "general learning": 46672,
      "performance nlu": 86576,
      "generation exploring": 47405,
      "graph previous": 48856,
      "generation input": 47441,
      "rdf triples": 95884,
      "keyvalue pairs": 57632,
      "wikibio webnlg": 132628,
      "text practice": 121187,
      "input knowledge": 55352,
      "introduce largescale": 56446,
      "largescale challenging": 61402,
      "dataset facilitate": 26924,
      "facilitate study": 42793,
      "practical scenario": 88711,
      "knowledge various": 58230,
      "types main": 126316,
      "large knowledge": 61120,
      "graph kg": 48824,
      "graphtosequence models": 48984,
      "models severely": 74017,
      "severely suffer": 107886,
      "suffer problems": 115240,
      "problems information": 91326,
      "generating descriptions": 47211,
      "structure able": 113803,
      "able represent": 1683,
      "represent original": 99124,
      "original graph": 82526,
      "information comprehensively": 54430,
      "furthermore incorporate": 46181,
      "aggregation methods": 4363,
      "learn extract": 62045,
      "existing approach": 40047,
      "solves problem": 110636,
      "multidomain learning": 75094,
      "training domains": 123594,
      "propose improve": 92718,
      "taskspecific model": 119633,
      "improve knowledge": 52399,
      "experiments domains": 40923,
      "results addition": 101500,
      "transferability outperforming": 124215,
      "model 124": 70495,
      "adapting new": 3319,
      "domain little": 33576,
      "global warming": 48280,
      "nlp despite": 79603,
      "despite considerable": 29681,
      "detection fake": 29952,
      "mutlitask learning": 75969,
      "dependencies method": 29095,
      "method developing": 68775,
      "broadcoverage semantic": 14695,
      "annotated resource": 6223,
      "transfer supervised": 124188,
      "use supervised": 128306,
      "different multitask": 31274,
      "singletask baseline": 109844,
      "baseline setting": 12309,
      "setting english": 107745,
      "baseline 18": 12175,
      "set observe": 107515,
      "factor improving": 42859,
      "inference large": 54157,
      "datasets nlp": 27598,
      "progress languages": 92156,
      "language prohibitively": 59993,
      "commercial machine": 18835,
      "systems robust": 117125,
      "automatically paper": 10807,
      "paper offer": 84059,
      "large english": 61087,
      "english nli": 37224,
      "labels using": 58653,
      "using datasets": 129536,
      "datasets address": 27300,
      "address core": 3670,
      "core issues": 23326,
      "large finally": 61091,
      "finally models": 44207,
      "machinetranslated datasets": 66349,
      "share code": 107948,
      "code models": 18118,
      "scalable framework": 103691,
      "generation introduce": 47446,
      "augment text": 10266,
      "learning contextaware": 62457,
      "demonstrate value": 28906,
      "attributes high": 10198,
      "series automatic": 107270,
      "nlg model": 79521,
      "dialogue virtual": 30815,
      "virtual assistants": 131742,
      "assistants google": 9578,
      "google assistant": 48516,
      "alexa apple": 4669,
      "apple siri": 6817,
      "enable users": 36368,
      "number services": 80964,
      "services apis": 107328,
      "web using": 132271,
      "propose schemaguided": 93026,
      "linearly number": 64402,
      "number slots": 80968,
      "combination simple": 18583,
      "model subsequently": 72100,
      "coherent natural": 18323,
      "text automatic": 120676,
      "baselines robust": 12456,
      "robust outofdomain": 103058,
      "inputs shows": 55496,
      "shows improved": 108586,
      "sample efficiency": 103560,
      "semantic recognition": 105192,
      "segmentation segment": 104630,
      "treated separately": 125558,
      "task effective": 118127,
      "expression identification": 41739,
      "identification classification": 51370,
      "streusle corpus": 113587,
      "crf sequence": 24758,
      "generalizes previous": 46841,
      "model generalizes": 71230,
      "surpasses existing": 116104,
      "establishes baseline": 38608,
      "modeling lexical": 72460,
      "does data": 33339,
      "augmentation improve": 10279,
      "exploit superficial": 41444,
      "central challenge": 15897,
      "proposed using": 93586,
      "augmentation generating": 10278,
      "generating training": 47280,
      "encouraging models": 36778,
      "models prefer": 73768,
      "features design": 43448,
      "design series": 29563,
      "hypothesis data": 51264,
      "learn stronger": 62156,
      "strong feature": 113676,
      "representations syntax": 99925,
      "constituency dependency": 21522,
      "dependency structure": 29228,
      "networks significant": 77768,
      "sensitivity syntactic": 105754,
      "structure perform": 113932,
      "networks treebased": 77804,
      "constituency parse": 21523,
      "performance subjectverb": 86765,
      "agreement prediction": 4398,
      "network generalizes": 77265,
      "does yield": 33412,
      "yield improvement": 135343,
      "models substantially": 74114,
      "substantially improved": 114897,
      "improved finetuning": 52606,
      "finetuning small": 44512,
      "constructed data": 21657,
      "syntactic biases": 116374,
      "models lacking": 73447,
      "corpora word": 23622,
      "corpus largescale": 23857,
      "words 10": 133798,
      "10 indian": 135,
      "trained corpora": 123100,
      "corpora create": 23445,
      "available pretrained": 11075,
      "embedding multiple": 35460,
      "multiple evaluation": 75557,
      "tasks hope": 119160,
      "selecting informative": 104751,
      "finetuning language": 44468,
      "essential modern": 38557,
      "finetuning limited": 44475,
      "finetuning method": 44476,
      "metric training": 69908,
      "trained approximate": 123070,
      "selects informative": 104862,
      "informative examples": 55138,
      "ones method": 81696,
      "finetuning present": 44491,
      "generality method": 46748,
      "encourage researchers": 36765,
      "release pretrained": 98468,
      "effective finetuning": 34678,
      "finetuning improving": 44464,
      "performance reducing": 86663,
      "crosslinguistic syntactic": 25055,
      "syntactic evaluation": 116402,
      "models range": 73850,
      "sentences high": 106340,
      "evidence english": 39643,
      "english investigate": 37175,
      "learn syntax": 62161,
      "sets english": 107667,
      "hebrew russian": 49673,
      "grammars develop": 48673,
      "lstms achieved": 65698,
      "poor accuracy": 88053,
      "object relative": 81054,
      "generally higher": 46858,
      "high syntactic": 50141,
      "accuracy english": 2149,
      "answering using": 6720,
      "evaluation generative": 39222,
      "answering genqa": 6633,
      "difficult assess": 31610,
      "assess correctness": 9471,
      "correctness generated": 24185,
      "generated answers": 47047,
      "freeform answer": 45776,
      "ngram similarity": 79452,
      "metrics fail": 69962,
      "incorrect answers": 53575,
      "assigns different": 9565,
      "keyphrase prediction": 57621,
      "reference answer": 97505,
      "answer evaluate": 6527,
      "evaluate metric": 38859,
      "highquality human": 50385,
      "covid19 information": 24567,
      "translation day": 124737,
      "health professionals": 49623,
      "general public": 46697,
      "effective cases": 34641,
      "cases language": 15651,
      "help people": 49762,
      "information published": 54889,
      "networks task": 77791,
      "problem introduce": 91089,
      "new technique": 79210,
      "metrics different": 69951,
      "technique better": 119772,
      "medical entity": 68204,
      "type prediction": 126219,
      "datasets medical": 27562,
      "text existing": 120927,
      "methods adopt": 69298,
      "list candidate": 64688,
      "candidate concepts": 15101,
      "concept paper": 20583,
      "paper probe": 84212,
      "based predicted": 11940,
      "type entity": 126198,
      "improves entity": 52975,
      "linking performance": 64666,
      "dearth annotated": 27883,
      "data medical": 26124,
      "largescale medical": 61463,
      "linking datasets": 64653,
      "demonstrate pretraining": 28825,
      "performance make": 86519,
      "make source": 66727,
      "code datasets": 18094,
      "available medical": 11045,
      "current endtoend": 25280,
      "models inherently": 73403,
      "lack flexibility": 58707,
      "process resulting": 91564,
      "uninteresting responses": 127175,
      "come expense": 18763,
      "relevance informativeness": 98511,
      "framework controllable": 45475,
      "lexical control": 63750,
      "mechanism trained": 68041,
      "comprehensive benchmark": 20248,
      "fast pace": 43149,
      "nlu benchmarks": 79806,
      "comparison proposed": 19566,
      "methods benchmarks": 69342,
      "benchmarks available": 12887,
      "handful languages": 49366,
      "introduce comprehensive": 56393,
      "multitask benchmark": 75812,
      "consists diverse": 21476,
      "recognition questionanswering": 96977,
      "ecommerce domain": 34505,
      "generalize different": 46807,
      "domains applications": 33731,
      "additionally release": 3631,
      "standard baselines": 112199,
      "proposed multilingual": 93501,
      "multilingual transformerbased": 75392,
      "computational social": 20432,
      "causal conclusions": 15804,
      "observational data": 81173,
      "data contains": 25789,
      "potential causes": 88542,
      "potential effects": 88550,
      "observed text": 81235,
      "history social": 50568,
      "scattered different": 103817,
      "evaluation practices": 39328,
      "examples provide": 39868,
      "provide guide": 93842,
      "despite increased": 29696,
      "increased attention": 53626,
      "regex synthesis": 97672,
      "limited complexity": 64219,
      "complexity compared": 19904,
      "tasks users": 119588,
      "users post": 129152,
      "datasets simple": 27715,
      "probabilistic grammar": 90808,
      "observed realworld": 81230,
      "second obtain": 104434,
      "linguistically diverse": 64584,
      "diverse natural": 32825,
      "language augment": 58850,
      "similar real": 109135,
      "prior datasets": 90697,
      "techniques highlight": 119901,
      "challenge presented": 16084,
      "dataset including": 26979,
      "including nonlocal": 53341,
      "multimodal inputs": 75435,
      "models plms": 73749,
      "complex words": 19897,
      "study investigating": 114416,
      "optimized bert": 82231,
      "generation dg": 47367,
      "vocabulary units": 131915,
      "attracts attention": 10175,
      "decoding technique": 28140,
      "words step": 134240,
      "information time": 55044,
      "timely fashion": 122162,
      "high translation": 50142,
      "experiments technique": 41171,
      "technique achieves": 119762,
      "englishtochinese translation": 37438,
      "outofdomain dependency": 82652,
      "monolingual dependency": 74585,
      "vector allows": 131295,
      "time build": 121992,
      "accuracy test": 2301,
      "languages match": 60721,
      "multilingual annotations": 75195,
      "annotations given": 6436,
      "does work": 33409,
      "work practice": 134672,
      "naive training": 76024,
      "training ner": 123729,
      "underperforms models": 126718,
      "having access": 49552,
      "point paper": 87809,
      "paper simple": 84431,
      "polyglot models": 88013,
      "data consistently": 25778,
      "consistently significantly": 21440,
      "outperform monolingual": 82718,
      "monolingual counterparts": 74562,
      "compared monolingual": 19391,
      "share parameters": 107958,
      "languages finetuning": 60591,
      "given complexity": 47995,
      "setting work": 107787,
      "attempt explore": 9742,
      "perform experimental": 85994,
      "testing model": 120596,
      "nlp experiment": 79613,
      "experimental settings": 40729,
      "predictions unseen": 89195,
      "unseen languages": 127530,
      "baselines human": 12409,
      "predictions experimental": 89168,
      "accelerate progress": 1869,
      "linguistic generalization": 64485,
      "position paper": 88289,
      "identically distributed": 51356,
      "paradigm consists": 84534,
      "consists stages": 21495,
      "stages pretraining": 112162,
      "arbitrary size": 8590,
      "set paradigm": 107522,
      "particular data": 85398,
      "generalize examples": 46809,
      "magnitude data": 66379,
      "data systems": 26533,
      "architectures generalize": 8807,
      "analysis contextual": 5546,
      "models lens": 73496,
      "models come": 72927,
      "vastly different": 131283,
      "architectures use": 8855,
      "novel similarity": 80728,
      "information deep": 54468,
      "models facilitate": 73229,
      "facilitate investigation": 42777,
      "affect model": 4219,
      "requiring external": 100338,
      "models family": 73237,
      "surprisingly different": 116134,
      "different individual": 31175,
      "observed differences": 81218,
      "lower higher": 65430,
      "finetuning downstream": 44454,
      "limitations crosslingual": 64171,
      "evaluation evaluation": 39205,
      "evaluation crosslingual": 39166,
      "usually performed": 130443,
      "performed zeroshot": 86938,
      "tasks unsupervised": 119585,
      "unsupervised crosslingual": 127617,
      "evaluation directly": 39187,
      "directly compare": 31866,
      "holds promise": 50596,
      "range metrics": 95580,
      "based stateoftheart": 12070,
      "pretrained mbert": 90126,
      "semantic encoders": 105041,
      "referencefree mt": 97552,
      "evaluation identify": 39236,
      "identify key": 51512,
      "limitations semantic": 64191,
      "propose partial": 92986,
      "evaluation best": 39137,
      "unconditional text": 126641,
      "model real": 71858,
      "real sentences": 96077,
      "sentences generate": 106328,
      "generate novel": 46978,
      "novel sentences": 80721,
      "diversity training": 32897,
      "used comparing": 128438,
      "comparing methods": 19506,
      "contradictory conclusions": 22671,
      "problem novel": 91148,
      "unavailable propose": 126601,
      "metrics bilingual": 69931,
      "understudy bleu": 127032,
      "embedding distance": 35396,
      "generative pretrained": 47750,
      "syntactic transformations": 116494,
      "tasks utility": 119592,
      "accurately reflect": 2391,
      "reflect syntactic": 97611,
      "relations make": 98206,
      "make semantic": 66723,
      "relations explicit": 98166,
      "representations lack": 99712,
      "work standalone": 134814,
      "pipeline evaluated": 87538,
      "using documentlevel": 129604,
      "documents previous": 33270,
      "focus exclusively": 44762,
      "focuses data": 44893,
      "documentlevel evaluation": 33140,
      "metric used": 69909,
      "minimum risk": 70212,
      "risk training": 102880,
      "maximumlikelihood training": 67555,
      "training robust": 123824,
      "15 million": 350,
      "related questions": 97892,
      "questions written": 95378,
      "lack fundamental": 58709,
      "appear natural": 6799,
      "new named": 79044,
      "ner corpus": 77033,
      "types trained": 126376,
      "increase 10": 53585,
      "10 f1": 132,
      "achieves overall": 2829,
      "score code": 104059,
      "code token": 18157,
      "model code": 70837,
      "measuring linguistic": 67921,
      "popular nlp": 88114,
      "probes supervised": 90884,
      "structure models": 113909,
      "probe structural": 90878,
      "structural probe": 113786,
      "extent syntactic": 41987,
      "contextualised word": 22534,
      "novel design": 80543,
      "parsing literature": 85145,
      "traditional parser": 122858,
      "controllable text": 22830,
      "generation important": 47432,
      "work controllable": 134441,
      "new schema": 79117,
      "process classifying": 91436,
      "process requires": 91561,
      "overview different": 83364,
      "modules provide": 74528,
      "analysis advantages": 5489,
      "paper understanding": 84484,
      "linking nel": 64665,
      "specifically task": 111595,
      "task entails": 118150,
      "statistical phrase": 113153,
      "phrase detection": 87343,
      "effective understanding": 34766,
      "base population": 11467,
      "unlike english": 127432,
      "english knowledge": 37180,
      "bases wikidata": 12508,
      "rely automatically": 98676,
      "growing size": 49183,
      "working paper": 134905,
      "linking module": 64664,
      "information relation": 54913,
      "modules designed": 74522,
      "fusion techniques": 46253,
      "minimal intervention": 70165,
      "presented knowledge": 89787,
      "population present": 88167,
      "present gold": 89508,
      "gold dataset": 48434,
      "dataset benchmarking": 26764,
      "language consisting": 58911,
      "efficiency proposed": 35041,
      "established baselines": 38598,
      "datasets make": 27557,
      "datasets benchmark": 27331,
      "filipino second": 44024,
      "bert distilbert": 13100,
      "setting introduce": 107756,
      "analyze pretrained": 5993,
      "comparing models": 19507,
      "models aimed": 72736,
      "release models": 98459,
      "datasets research": 27677,
      "success text": 115132,
      "quality quantity": 94767,
      "complexsimple sentence": 19947,
      "corpus extracted": 23798,
      "aligning sentences": 4938,
      "evaluate improve": 38843,
      "create manually": 24624,
      "simplification corpora": 109579,
      "wikipedia propose": 132670,
      "parallel documents": 84665,
      "utilizes neural": 130555,
      "pair model": 83443,
      "similarity experiments": 109232,
      "alignment task": 4992,
      "simplification datasets": 109580,
      "trained datasets": 123109,
      "evaluation efficient": 39198,
      "efficient strategies": 35117,
      "steps predict": 113330,
      "studies focused": 114231,
      "deal hierarchical": 27857,
      "efficient ways": 35134,
      "propose auxiliary": 92569,
      "auxiliary synthetic": 10885,
      "synthetic task": 116646,
      "external dictionaries": 42000,
      "embeddings additional": 35544,
      "prediction layer": 89072,
      "adapted beam": 3281,
      "provide large": 93864,
      "combination auxiliary": 18550,
      "accuracy efficient": 2144,
      "efficient approaches": 35058,
      "approaches outperform": 8270,
      "studies using": 114296,
      "drastically reduced": 34105,
      "wellknown english": 132420,
      "tasks 2018": 118911,
      "2018 2019": 651,
      "little difference": 64802,
      "absolute performance": 1744,
      "data introduced": 26052,
      "2018 dataset": 661,
      "data substantial": 26516,
      "systems advantage": 116725,
      "speech directed": 111676,
      "directed children": 31815,
      "variation human": 130892,
      "helps language": 49821,
      "information speech": 55001,
      "initial stages": 55224,
      "learning eventually": 62558,
      "acoustic properties": 3024,
      "synthetic speech": 116645,
      "patterns syntax": 85784,
      "pair parallel": 83448,
      "corpus building": 23687,
      "detailed picture": 29782,
      "detailed corpus": 29763,
      "danish language": 25535,
      "corpora scale": 23580,
      "result focused": 101375,
      "effort provide": 35181,
      "provide diverse": 93809,
      "billion word": 14081,
      "corpus covers": 23733,
      "time periods": 122068,
      "socioeconomic status": 110476,
      "type recent": 126222,
      "elicited speech": 35293,
      "speech samples": 111797,
      "healthy controls": 49638,
      "lms trained": 64891,
      "transcripts speech": 124014,
      "single feature": 109734,
      "performance little": 86499,
      "previously developed": 90594,
      "lexical frequency": 63763,
      "associated lexical": 9600,
      "model resulting": 71922,
      "improves current": 52966,
      "standard training": 112327,
      "proposed mitigate": 93421,
      "wellknown problem": 132422,
      "nmt tendency": 79987,
      "datasets multiple": 27584,
      "multiple test": 75726,
      "test domains": 120447,
      "performance deterioration": 86291,
      "deterioration increasing": 30111,
      "increasing beam": 53665,
      "reduce exposure": 97327,
      "performance indomain": 86463,
      "indomain test": 53982,
      "increase model": 53604,
      "systematic assessment": 116660,
      "syntactic generalization": 116410,
      "remains unknown": 98830,
      "humanlike syntactic": 51057,
      "knowledge furthermore": 57943,
      "produce proper": 91917,
      "proper syntactic": 92423,
      "syntactic generalizations": 116414,
      "present systematic": 89732,
      "models testing": 74181,
      "models underperforming": 74265,
      "dataset size": 27206,
      "variability syntactic": 130834,
      "performance substantially": 86769,
      "learning detect": 62495,
      "machine translations": 66310,
      "tremendously recent": 125738,
      "translation crosslingual": 124723,
      "scale turn": 103759,
      "help learn": 49745,
      "compositional sentiment": 20133,
      "variant bert": 130866,
      "bert effectively": 13106,
      "effectively captures": 34799,
      "binary constituency": 14128,
      "phraselevel sentiment": 87403,
      "phraselevel annotations": 87402,
      "transferred sentiment": 124235,
      "tasks emotion": 119087,
      "visualization methods": 131826,
      "methods understand": 69820,
      "relation model": 98055,
      "crisis situation": 24787,
      "applying intuition": 7250,
      "forwards backwards": 45332,
      "jointly optimize": 57372,
      "user demands": 128974,
      "unlabeled instances": 127398,
      "instances annotation": 55617,
      "spend time": 111924,
      "reading paper": 96028,
      "particular user": 85463,
      "users context": 129104,
      "making possible": 66857,
      "use directly": 128003,
      "knowledge fixed": 57936,
      "model clearly": 70833,
      "limitations previous": 64189,
      "approaches successfully": 8359,
      "information outside": 54818,
      "weights using": 132385,
      "supervised architectures": 115727,
      "combine information": 18627,
      "component paper": 19998,
      "integrate information": 55752,
      "retrieval pretrained": 102420,
      "model purely": 71832,
      "purely unsupervised": 94420,
      "models way": 74334,
      "performance resulting": 86678,
      "competitive supervised": 19697,
      "context different": 22059,
      "different segment": 31402,
      "classifier determine": 17538,
      "determine context": 30124,
      "predictions robust": 89190,
      "robust noisy": 103056,
      "language application": 58832,
      "facts expressed": 42909,
      "language computational": 58906,
      "distinct existing": 32535,
      "learned distributional": 62212,
      "reading models": 96026,
      "performance questions": 86646,
      "domain contrast": 33488,
      "experiments support": 41161,
      "support discussion": 115965,
      "challenges facing": 16160,
      "interesting realworld": 56088,
      "realworld task": 96181,
      "able utilize": 1704,
      "words ones": 134094,
      "subjectverbobject svo": 114716,
      "structures model": 114087,
      "model induces": 71339,
      "model retains": 71925,
      "word group": 133307,
      "space effectively": 110995,
      "effectively applied": 34787,
      "applied number": 7104,
      "tasks reasoning": 119438,
      "work reducing": 134766,
      "categories deep": 15734,
      "data abstract": 25553,
      "contents paper": 21990,
      "paper short": 84427,
      "background objective": 11310,
      "fields like": 43999,
      "search recommendation": 104344,
      "makes challenging": 66763,
      "learning solutions": 63036,
      "automatic discourselevel": 10519,
      "learning particular": 62880,
      "particular define": 85401,
      "observe accuracy": 81187,
      "promising solution": 92306,
      "solution automatic": 110546,
      "models known": 73443,
      "significantly enhanced": 108909,
      "document previous": 33054,
      "integrates information": 55785,
      "context generating": 22115,
      "responses mind": 101283,
      "using teacherstudent": 130272,
      "teacherstudent framework": 119703,
      "category consists": 15783,
      "documents including": 33240,
      "annotations nlp": 6447,
      "information format": 54625,
      "format converters": 45207,
      "make tools": 66739,
      "tools able": 122425,
      "able read": 1679,
      "produce different": 91882,
      "different output": 31314,
      "idea document": 51323,
      "theory nlp": 121822,
      "conceptual framework": 20661,
      "theory successfully": 121826,
      "successfully apply": 115179,
      "semeval2020 task": 105550,
      "building multimodal": 14866,
      "humor paper": 51140,
      "regression baseline": 97694,
      "baseline bilstm": 12198,
      "bert subtasks": 13236,
      "features employ": 43476,
      "possible solutions": 88433,
      "potential provide": 88581,
      "designed promote": 29620,
      "positive psychology": 88338,
      "describes architecture": 29387,
      "architecture supports": 8755,
      "supports range": 116038,
      "covid19 pandemic": 24571,
      "rapid deployment": 95712,
      "semantic structural": 105308,
      "detection identifying": 29968,
      "posts social": 88522,
      "mining public": 70257,
      "public sentiment": 94273,
      "relationship modeling": 98301,
      "properly handle": 92433,
      "structure content": 113826,
      "posts comments": 88515,
      "task determining": 118080,
      "text entails": 120915,
      "entails hypothesis": 37687,
      "using notion": 129977,
      "use nli": 128174,
      "contain complex": 21734,
      "learningbased models": 63182,
      "performing natural": 86961,
      "techniques despite": 119868,
      "attractive features": 10171,
      "argue existing": 8918,
      "drawbacks approach": 34122,
      "approach question": 7832,
      "selection lexical": 104796,
      "approach nlg": 7741,
      "purely statistical": 94418,
      "learning probabilistic": 62909,
      "effectiveness capturing": 34874,
      "multivariate gaussian": 75932,
      "gaussian distribution": 46542,
      "paraphrases demonstrate": 84839,
      "performance overall": 86591,
      "examples using": 39897,
      "using controlled": 129492,
      "generalization training": 46798,
      "classification limited": 17254,
      "especially applications": 38431,
      "applications output": 6982,
      "classifiers generalize": 17611,
      "limited evidence": 64234,
      "generalization task": 46796,
      "primarily relies": 90643,
      "relies pretrained": 98655,
      "embeddings map": 35802,
      "map words": 67117,
      "training similar": 123864,
      "parsed corpora": 84895,
      "based available": 11539,
      "select features": 104698,
      "features offers": 43638,
      "feedforward network": 43845,
      "network convolutional": 77214,
      "remain competitive": 98769,
      "presence occupational": 89352,
      "structure online": 113925,
      "online texts": 81811,
      "models research": 73941,
      "research social": 100626,
      "social stratification": 110458,
      "closely linked": 17857,
      "associated different": 9595,
      "research focuses": 100520,
      "results reproduced": 102131,
      "paper gives": 83968,
      "questions results": 95363,
      "obtained text": 81422,
      "literature social": 64774,
      "corpus chinese": 23692,
      "form large": 45100,
      "18th century": 431,
      "20 million": 513,
      "million characters": 70095,
      "analysis historical": 5642,
      "available opensource": 11067,
      "classical chinese": 17086,
      "new opensource": 79066,
      "creative commons": 24726,
      "commons license": 18980,
      "study analyzing": 114319,
      "analysis focus": 5625,
      "meaningful semantic": 67727,
      "used future": 128562,
      "span extraction": 111123,
      "lightweight model": 64013,
      "task formulation": 118225,
      "formulation allows": 45294,
      "allows simple": 5187,
      "simple integration": 109446,
      "henderson et": 49838,
      "especially useful": 38524,
      "representations scratch": 99868,
      "slotfilling task": 110081,
      "new challenging": 78826,
      "challenging data": 16239,
      "research demonstrated": 100462,
      "demonstrated distributional": 28911,
      "2012 paper": 567,
      "statistical sources": 113166,
      "sources required": 110920,
      "demonstrate direct": 28700,
      "language necessary": 59708,
      "contextual elements": 22455,
      "models simulate": 74055,
      "preceding turn": 88778,
      "dialogue sds": 30739,
      "listening tests": 64710,
      "models aspect": 72772,
      "aspect recent": 9344,
      "years common": 135245,
      "explanation paper": 41286,
      "used recurrent": 128719,
      "propose remedy": 93018,
      "attention provide": 9982,
      "human sentence": 50961,
      "processing recurrence": 91783,
      "rnns long": 102973,
      "architecture outperforms": 8719,
      "rnns natural": 102974,
      "known ability": 58284,
      "transformer rnnbased": 124375,
      "measures human": 67868,
      "outperform rnns": 82740,
      "selfpaced reading": 104928,
      "reading times": 96036,
      "neural activity": 77827,
      "activity reading": 3135,
      "reading english": 96020,
      "challenging widely": 16345,
      "processing provides": 91779,
      "creating multimodal": 24703,
      "datasets sequential": 27701,
      "text video": 121407,
      "task aligning": 117870,
      "align different": 4898,
      "sets challenging": 107654,
      "tend contain": 120144,
      "contain far": 21740,
      "challenges use": 16219,
      "unsupervised alignment": 127601,
      "algorithm learns": 4766,
      "instructions different": 55710,
      "research multimodal": 100562,
      "commonsense information": 18989,
      "information largescale": 54716,
      "largescale multidocument": 61466,
      "dataset wikipedia": 27279,
      "summarization mds": 115529,
      "content large": 21898,
      "short summaries": 108242,
      "summaries important": 115443,
      "generation lack": 47452,
      "address use": 3775,
      "total number": 122698,
      "number document": 80876,
      "humanwritten summaries": 51133,
      "source articles": 110712,
      "articles automatically": 9171,
      "dataset empirical": 26885,
      "examining stateoftheart": 39770,
      "summarization previous": 115549,
      "summarization tls": 115584,
      "task generally": 118234,
      "individual subtasks": 53932,
      "date selection": 27830,
      "effective combination": 34646,
      "benchmarks robust": 12936,
      "evaluation present": 39330,
      "larger spans": 61383,
      "previous datasets": 90405,
      "applying transformer": 7279,
      "transformer characterlevel": 124316,
      "transformer shown": 124379,
      "outperform recurrent": 82736,
      "models various": 74321,
      "inflection generation": 54287,
      "historical text": 50535,
      "using transformer": 130314,
      "role performance": 103211,
      "large batch": 61035,
      "normalization transformer": 80352,
      "transformer outperforms": 124367,
      "baseline characterlevel": 12199,
      "alternative language": 5232,
      "bert gpt2": 13137,
      "improvements outofdomain": 52890,
      "research transfer": 100651,
      "cultural heritage": 25230,
      "specific topics": 111506,
      "topics results": 122654,
      "scenario users": 103836,
      "hot topic": 50666,
      "topic research": 122569,
      "nlp including": 79622,
      "fields machine": 44000,
      "availability bilingual": 10901,
      "limited bilingual": 64216,
      "eu project": 38758,
      "small collection": 110141,
      "multilingual bilingual": 75215,
      "pairs germanenglish": 83554,
      "germanenglish chineseenglish": 47931,
      "extracted bilingual": 42142,
      "analysis better": 5518,
      "general evaluation": 46656,
      "follow standard": 44942,
      "researchers use": 100715,
      "features unsupervised": 43770,
      "output test": 83134,
      "expert annotated": 41214,
      "data computation": 25762,
      "access additional": 1901,
      "dataset enables": 26888,
      "latin used": 61666,
      "corpus comprising": 23708,
      "comprehensive understanding": 20281,
      "form interpretation": 45096,
      "underlying lexical": 126684,
      "neural morphological": 78209,
      "morphological taggers": 74741,
      "achieved state": 2699,
      "results morphological": 101961,
      "evaluate efficacy": 38831,
      "language label": 59234,
      "systems explicitly": 116894,
      "labels seen": 58638,
      "label bias": 58385,
      "openended text": 81989,
      "produce repetitive": 91921,
      "repetitive generic": 98918,
      "samplingbased decoding": 103619,
      "topk sampling": 122672,
      "standard seq2seq": 112302,
      "locally normalized": 64952,
      "normalized probability": 80358,
      "bias major": 13815,
      "major reason": 66584,
      "training label": 123671,
      "quantitatively measure": 94892,
      "bias test": 13848,
      "ability discriminate": 1509,
      "meaningful texts": 67730,
      "approach terms": 7968,
      "terms automatic": 120270,
      "metrics analysis": 69922,
      "improving segmentation": 53160,
      "technical support": 119753,
      "problems long": 91338,
      "long complex": 65064,
      "contain various": 21770,
      "crucial information": 25145,
      "information problem": 54871,
      "resolution correctly": 100756,
      "questions formulate": 95308,
      "formulate problem": 45278,
      "problem sequence": 91217,
      "task study": 118747,
      "classification baseline": 17133,
      "art supervised": 9094,
      "combining contextual": 18713,
      "pretrained different": 90023,
      "marked improvement": 67225,
      "embeddings single": 35941,
      "single pretrained": 109784,
      "answer retrieval": 6572,
      "unlabeled learning": 127400,
      "learning named": 62821,
      "understudied problem": 127030,
      "challenging linguistic": 16271,
      "low coverage": 65354,
      "coverage existing": 24506,
      "resources address": 100939,
      "positiveunlabeled learning": 88355,
      "domainspecific linguistic": 33908,
      "quickly efficiently": 95390,
      "seed dictionary": 104501,
      "exhibits better": 40012,
      "language vice": 60330,
      "share information": 107952,
      "information social": 54988,
      "activities events": 3131,
      "health status": 49629,
      "share online": 107956,
      "directly indirectly": 31881,
      "valuable data": 130761,
      "risk factors": 102877,
      "adverse drug": 4188,
      "drug reactions": 34198,
      "medical experts": 68206,
      "medical concept": 68196,
      "presents baseline": 89823,
      "baseline data": 12206,
      "german introduce": 47898,
      "set medical": 107491,
      "textual networks": 121722,
      "works implemented": 134952,
      "models little": 73514,
      "little focus": 64808,
      "node attributes": 80018,
      "information leads": 54721,
      "classification result": 17383,
      "result test": 101412,
      "set comparison": 107395,
      "networks jointly": 77632,
      "objectives result": 81141,
      "allows creation": 5138,
      "citation networks": 16991,
      "model textual": 72167,
      "best empirical": 13327,
      "successfully employed": 115184,
      "factors including": 42892,
      "data quantity": 26309,
      "general language": 46667,
      "model taskspecific": 72148,
      "adaptation affect": 3207,
      "thoroughly studied": 121906,
      "based experimental": 11699,
      "graph parsing": 48854,
      "effective encoding": 34670,
      "dependencies shared": 29115,
      "task iwpt": 118320,
      "iwpt 2020": 57159,
      "models enhanced": 73157,
      "parser adapted": 84919,
      "train single": 123021,
      "using gold": 129709,
      "sentence splitting": 106091,
      "bert encoding": 13115,
      "severe drop": 107882,
      "4th place": 1065,
      "official ranking": 81602,
      "average elas": 11185,
      "bertbased framework": 13271,
      "framework spoken": 45695,
      "given passage": 48081,
      "question multiple": 95184,
      "speech machine": 111706,
      "correct choice": 24096,
      "choice answer": 16885,
      "contain useful": 21767,
      "text achieve": 120627,
      "certain level": 15945,
      "caused automatic": 15828,
      "line research": 64327,
      "framework inherits": 45576,
      "inherits advantages": 55198,
      "contextualized language": 22553,
      "learned bert": 62204,
      "information distilled": 54499,
      "experiments demonstrates": 40897,
      "models begun": 72823,
      "allowing build": 5108,
      "build powerful": 14798,
      "powerful language": 88666,
      "bert increasingly": 13143,
      "popular stateoftheart": 88132,
      "monolingual bert": 74551,
      "massive dataset": 67327,
      "dataset different": 26873,
      "obtains higher": 81466,
      "higher scores": 50207,
      "scores datasets": 104165,
      "existing ones": 40245,
      "outperforming multilingual": 82810,
      "works sentiment": 134975,
      "classification named": 17296,
      "common questions": 18913,
      "bert baseline": 13073,
      "accuracy trained": 2305,
      "trained 20": 123062,
      "task bert": 117931,
      "triplet loss": 125824,
      "baseline achieved": 12179,
      "direct use": 31809,
      "evaluation generating": 39221,
      "semantically valid": 105384,
      "valid adversarial": 130709,
      "adversarial questions": 4159,
      "help evaluate": 49718,
      "evaluate extent": 38836,
      "language questions": 60019,
      "questions difficult": 95301,
      "single character": 109711,
      "perception paper": 85928,
      "meaning original": 67655,
      "entity delexicalization": 37917,
      "use gumbelsoftmax": 128079,
      "adversarial loss": 4141,
      "attack models": 9702,
      "attack success": 9704,
      "performance robustness": 86688,
      "central component": 15898,
      "arguments explicitly": 8985,
      "sentence gives": 105893,
      "renders difficult": 98891,
      "machines interpret": 66336,
      "resources represent": 101035,
      "nlp make": 79634,
      "form paper": 45114,
      "build robust": 14805,
      "important issues": 52175,
      "issues regarding": 57072,
      "ud annotations": 126529,
      "original revised": 82540,
      "using biaffine": 129395,
      "biaffine attention": 13786,
      "corpus shows": 23996,
      "trained previous": 123245,
      "learn relations": 62123,
      "critical errors": 24810,
      "growing nlp": 49177,
      "questions raised": 95349,
      "tasks deemed": 119028,
      "respect paper": 101094,
      "paper automatic": 83750,
      "automatic legal": 10576,
      "emnlp 2019": 36084,
      "recognition french": 96884,
      "french treebank": 45828,
      "main source": 66463,
      "does include": 33359,
      "include explicit": 53224,
      "related named": 97882,
      "applications largescale": 6954,
      "entity annotations": 37905,
      "annotations contain": 6417,
      "information automatic": 54385,
      "automatic preannotation": 10627,
      "guidelines provide": 49265,
      "online press": 81790,
      "coronavirus pandemic": 23395,
      "ii does": 51705,
      "does come": 33334,
      "impact language": 51875,
      "language primary": 59827,
      "primary communication": 90648,
      "communication tool": 19044,
      "frequency lists": 45847,
      "application enables": 6849,
      "knowledge corpus": 57844,
      "statistical analyses": 113076,
      "test capabilities": 120424,
      "powerful neural": 88673,
      "nlp problem": 79670,
      "parsing addition": 85065,
      "existing transitionbased": 40324,
      "datasets previous": 27630,
      "stateoftheart graphbased": 112671,
      "task quite": 118597,
      "design simple": 29564,
      "simple meaning": 109464,
      "representation captures": 99186,
      "brief overview": 14613,
      "overview existing": 83365,
      "annotation methods": 6338,
      "selftraining unsupervised": 104964,
      "unsupervised parsing": 127691,
      "task like": 118353,
      "like language": 64055,
      "modeling work": 72589,
      "model supervision": 72117,
      "predictions training": 89194,
      "extend recent": 41794,
      "shen et": 108167,
      "al 2018a": 4638,
      "semisupervised fashion": 105602,
      "predicted model": 88963,
      "unlabeled training": 127413,
      "16 f1": 379,
      "semisupervised parsing": 105620,
      "classification opendomain": 17313,
      "step subsequent": 113302,
      "particular complex": 85394,
      "content features": 21881,
      "information enrich": 54532,
      "enrich utterance": 37560,
      "representation rich": 99399,
      "rich domains": 102742,
      "available entities": 10992,
      "complement model": 19730,
      "data augment": 25627,
      "limited amounts": 64202,
      "bases generate": 12500,
      "generate additional": 46899,
      "labeled utterances": 58479,
      "openly available": 81999,
      "humanhuman conversational": 51027,
      "approach previous": 7816,
      "humanmachine conversations": 51061,
      "collected amazon": 18396,
      "improves topic": 53054,
      "quantitative results": 94883,
      "results detailed": 101718,
      "modeling generation": 72436,
      "modeling introduce": 72452,
      "statistical dependencies": 113091,
      "accuracy unfortunately": 2311,
      "learning informative": 62655,
      "prohibitively large": 92205,
      "latent codes": 61579,
      "ignored translation": 51678,
      "impose strong": 52304,
      "distribution latent": 32665,
      "latent code": 61578,
      "propose apply": 92546,
      "framework stateoftheart": 45698,
      "transformer introduce": 124328,
      "approximate posterior": 8459,
      "posterior based": 88480,
      "models fewshot": 73246,
      "gains nlp": 46400,
      "tasks benchmarks": 118961,
      "text followed": 120958,
      "followed finetuning": 44947,
      "thousands examples": 121922,
      "humans generally": 51081,
      "perform new": 86037,
      "examples simple": 39878,
      "fewshot performance": 43908,
      "performance reaching": 86653,
      "finetuning approaches": 44446,
      "approaches specifically": 8347,
      "specifically train": 111601,
      "gradient updates": 48607,
      "interaction model": 55953,
      "methodological issues": 69237,
      "gpt3 generate": 48573,
      "articles human": 9192,
      "written humans": 135131,
      "discuss broader": 32237,
      "improve dialogue": 52373,
      "resources specifically": 101046,
      "based background": 11541,
      "movie plot": 74897,
      "reddit comments": 97296,
      "capturing structural": 15495,
      "task received": 118606,
      "attention community": 9804,
      "architecture uses": 8768,
      "ability bert": 1488,
      "explicit structure": 41350,
      "information specifically": 55000,
      "capture structural": 15399,
      "information ii": 54664,
      "lstms capture": 65706,
      "capture sequential": 15392,
      "information iii": 54665,
      "bert deep": 13093,
      "analyze proposed": 5995,
      "effectively combine": 34800,
      "information series": 54974,
      "tasks structural": 119525,
      "poorly task": 88071,
      "task leads": 118345,
      "ways combining": 132150,
      "outperforms deep": 82878,
      "representations elmo": 99616,
      "benefits additional": 13003,
      "information explicitly": 54552,
      "surprising given": 116127,
      "recent claims": 96438,
      "bert captures": 13087,
      "captures structural": 15455,
      "gives improvement": 48200,
      "number publications": 80945,
      "superficial cues": 115671,
      "evaluated data": 38966,
      "evolving research": 39705,
      "weaknesses models": 132210,
      "datasets methods": 27568,
      "discuss findings": 32251,
      "possible future": 88404,
      "resource researchers": 100873,
      "tools assess": 122434,
      "quality data": 94626,
      "extremely powerful": 42608,
      "augmentation paper": 10296,
      "achieved sota": 2698,
      "improved information": 52609,
      "models providing": 73838,
      "providing insights": 94119,
      "subjective information": 114694,
      "significant efforts": 108754,
      "bases work": 12510,
      "opinion phrases": 82095,
      "node represents": 80022,
      "represents set": 100074,
      "extracted reviews": 42176,
      "groups similar": 49155,
      "combining clusters": 18710,
      "quality labeled": 94695,
      "understand documents": 126746,
      "datasets visual": 27793,
      "visual textual": 131813,
      "insufficient paper": 55725,
      "finegrained tokenlevel": 44387,
      "tokenlevel annotations": 122287,
      "latex documents": 61662,
      "documents available": 33189,
      "modalities compared": 70476,
      "build strong": 14810,
      "information variety": 55084,
      "languages report": 60843,
      "languages example": 60562,
      "abilities neural": 1479,
      "language exploring": 59034,
      "bert named": 13191,
      "input consists": 55310,
      "consists sentence": 21492,
      "capture longdistance": 15335,
      "longdistance relationships": 65151,
      "crosssentence information": 25072,
      "exploring use": 41668,
      "information ner": 54796,
      "context form": 22108,
      "bert input": 13144,
      "input allows": 55295,
      "allows study": 5189,
      "majority voting": 66611,
      "require changes": 100117,
      "ner benchmarks": 77026,
      "demonstrates proposed": 28955,
      "results german": 101813,
      "bertbased approaches": 13268,
      "work open": 134662,
      "open licenses": 81915,
      "web document": 132231,
      "heavy use": 49667,
      "use world": 128370,
      "high volumes": 50148,
      "growing amounts": 49159,
      "probabilistic classification": 90798,
      "problem mainly": 91118,
      "high classification": 50040,
      "handle high": 49382,
      "dimensional sparse": 31753,
      "representation common": 99190,
      "common lack": 18887,
      "data largescale": 26077,
      "order solve": 82405,
      "method web": 69226,
      "class improve": 17041,
      "precision using": 88807,
      "documents experimental": 33227,
      "comprehension behavior": 20166,
      "reading behavior": 95972,
      "takes human": 117533,
      "probability context": 90857,
      "deployed real": 29308,
      "naturalistic text": 76631,
      "dataset sizes": 27207,
      "features models": 43613,
      "general better": 46635,
      "models ngram": 73642,
      "lstm structurally": 65668,
      "structurally supervised": 113801,
      "supervised neural": 115814,
      "eye movement": 42632,
      "movement data": 74886,
      "finally compare": 44155,
      "depth syntactic": 29323,
      "methods controlled": 69405,
      "controlled psycholinguistic": 22845,
      "psycholinguistic experiments": 94204,
      "significant relationship": 108857,
      "suggest different": 115307,
      "approaches required": 8326,
      "required best": 100212,
      "language comprehension": 58905,
      "linguistic materials": 64506,
      "recent computational": 96439,
      "research enabled": 100486,
      "larger scale": 61380,
      "novel computational": 80512,
      "quantifies semantic": 94842,
      "semantics concept": 105401,
      "concept approach": 20561,
      "space enables": 110998,
      "framework empirical": 45515,
      "concept domain": 20566,
      "interaction semantic": 55968,
      "factors language": 42895,
      "sign languages": 108680,
      "quality standards": 94798,
      "explore differences": 41538,
      "parameters transfer": 84794,
      "modelling automatic": 72597,
      "recognition spoken": 97011,
      "great advances": 48993,
      "power large": 88633,
      "common languages": 18889,
      "work transfer": 134858,
      "methods model": 69622,
      "techniques finetuning": 119890,
      "standard stacked": 112308,
      "stacked lstm": 112128,
      "corpus standard": 24017,
      "treebank corpus": 125631,
      "use probabilistic": 128214,
      "functional distributional": 46056,
      "semantics functional": 105418,
      "computationally tractable": 20459,
      "learning truthconditional": 63127,
      "truthconditional semantics": 125888,
      "semantics corpus": 105404,
      "corpus previous": 23939,
      "probabilistic version": 90836,
      "bayesian inference": 12581,
      "previous formulation": 90412,
      "problem treating": 91272,
      "acts framework": 3142,
      "donkey sentences": 33943,
      "sentences finally": 106318,
      "syntactic search": 116462,
      "graphs contrast": 48944,
      "introduce lightweight": 56450,
      "query language": 94970,
      "underlying syntactic": 126705,
      "representations instead": 99701,
      "coupled simple": 24458,
      "speed efficient": 111865,
      "engine allows": 37026,
      "wikipedia available": 132643,
      "richly annotated": 102812,
      "serve training": 107307,
      "contains different": 21820,
      "annotated manually": 6208,
      "come different": 18761,
      "treebank contains": 125630,
      "sentences morphological": 106402,
      "diversity texts": 32896,
      "annotations serve": 6464,
      "linguistic research": 64545,
      "involve reasoning": 56877,
      "textual spans": 121749,
      "recognition coreference": 96847,
      "extensive research": 41952,
      "spans text": 111211,
      "comprehensive empirical": 20256,
      "task vary": 118838,
      "vary different": 131243,
      "different facets": 31142,
      "individual tasks": 53934,
      "encoder pretraining": 36555,
      "especially true": 38520,
      "years driven": 135254,
      "art variety": 9102,
      "standard nlp": 112284,
      "tasks classification": 118977,
      "texttotext tasks": 121670,
      "situation different": 109887,
      "come close": 18759,
      "terms corpus": 120299,
      "based popular": 11936,
      "popular bert": 88083,
      "larger model": 61370,
      "consisting billion": 21448,
      "methodology collecting": 69252,
      "approaches modeling": 8236,
      "structure documentlevel": 113841,
      "community machine": 19079,
      "input document": 55323,
      "document obtain": 33044,
      "obtain discourse": 81277,
      "introduce transformerbased": 56560,
      "encoder embed": 36522,
      "embed discourse": 35348,
      "finally combine": 44152,
      "embedding fed": 35409,
      "fed encoder": 43802,
      "encoder experimental": 36524,
      "domainspecific mt": 33910,
      "mt shown": 74993,
      "number errors": 80881,
      "compared mt": 19393,
      "mt ht": 74967,
      "flag errors": 44629,
      "postediting effort": 88473,
      "number segments": 80961,
      "evaluation protocols": 39343,
      "trending topic": 125743,
      "reliable evaluation": 98613,
      "state affairs": 112421,
      "various evaluation": 131090,
      "evaluation protocol": 39342,
      "set place": 107530,
      "place paper": 87605,
      "presents comprehensive": 89833,
      "automated human": 10444,
      "methods dialogue": 69445,
      "identifying shortcomings": 51625,
      "accumulating evidence": 2055,
      "evaluation dimensions": 39186,
      "analyze types": 6012,
      "types evaluation": 126280,
      "dimensions used": 31774,
      "compared expert": 19372,
      "midresource languages": 70074,
      "extracted common": 42145,
      "train monolingual": 122967,
      "trained wikipedia": 123336,
      "parsing languages": 85140,
      "improve multilingual": 52425,
      "diverse corpus": 32802,
      "benefit multilingual": 12980,
      "multilingual embedding": 75249,
      "text uses": 121395,
      "space efficient": 110996,
      "consider approaches": 21189,
      "text case": 120695,
      "case exact": 15582,
      "features lowresource": 43599,
      "document sentence": 33074,
      "classification best": 17137,
      "results continuous": 101635,
      "compressed representations": 20298,
      "performs particularly": 87015,
      "generation challenging": 47331,
      "evaluation models": 39300,
      "models merely": 73562,
      "compare generated": 19250,
      "approach resolve": 7861,
      "idea learn": 51328,
      "unlabeled conversation": 127375,
      "corpus approach": 23666,
      "approach considers": 7445,
      "use twitter": 128345,
      "speakers conversations": 111309,
      "conversations test": 23074,
      "terms high": 120334,
      "scores model": 104186,
      "trained twitter": 123319,
      "dialogues additional": 30821,
      "provide code": 93776,
      "learned parameters": 62240,
      "parameters used": 84795,
      "future challenges": 46260,
      "problem nlp": 91145,
      "number native": 80916,
      "review paper": 102560,
      "resolving problem": 100817,
      "potential improvements": 88565,
      "text people": 121175,
      "predictions future": 89173,
      "future events": 46276,
      "consistently better": 21411,
      "question makes": 95179,
      "better paper": 13645,
      "language people": 59806,
      "open online": 81919,
      "financial analysts": 44253,
      "present number": 89627,
      "linguistic metrics": 64509,
      "metrics computed": 69940,
      "computed text": 20474,
      "text associated": 120664,
      "readability emotion": 95945,
      "studying linguistic": 114558,
      "factors associated": 42882,
      "language potentially": 59816,
      "political texts": 88004,
      "procedure based": 91390,
      "features prior": 43668,
      "increase ability": 53586,
      "predict unknown": 88950,
      "words containing": 133878,
      "matching algorithm": 67394,
      "processing step": 91802,
      "postprocessing technique": 88510,
      "vietnamese datasets": 131673,
      "method obtained": 68995,
      "requires reasoning": 100314,
      "context explicitly": 22099,
      "explicitly provided": 41390,
      "autoencoder encoder": 10402,
      "discrete variables": 32182,
      "discrete representations": 32174,
      "representations enable": 99623,
      "automatically selecting": 10827,
      "relevant evidence": 98546,
      "generation provides": 47576,
      "provides natural": 94053,
      "extracting nary": 42223,
      "crosssentence relations": 25080,
      "using constrained": 129477,
      "work relation": 134768,
      "extraction deals": 42296,
      "relations occurring": 98225,
      "sentence having": 105896,
      "arguments propose": 8992,
      "new formulation": 78923,
      "task relations": 118625,
      "intrasentence relations": 56342,
      "sentences arguments": 106216,
      "sequence representation": 107042,
      "various classifiers": 131059,
      "variant generalized": 130867,
      "approach datasets": 7471,
      "domains biomedical": 33736,
      "biomedical general": 14187,
      "transformer recurrent": 124373,
      "annotation using": 6403,
      "output stateoftheart": 83126,
      "news test": 79377,
      "best transformer": 13464,
      "31 reduction": 922,
      "errors 10": 38356,
      "relevant translation": 98597,
      "embeddings lowresource": 35794,
      "structural design": 113764,
      "design choices": 29526,
      "sentences real": 106466,
      "representations increase": 99688,
      "results probing": 102054,
      "designing probing": 29645,
      "largescale annotated": 61398,
      "data highquality": 26003,
      "task design": 118070,
      "design english": 29536,
      "embeddings cases": 35589,
      "size annotated": 109911,
      "probing dataset": 90889,
      "dataset type": 27251,
      "classifier used": 17590,
      "multilingual tasks": 75381,
      "tasks primarily": 119399,
      "distinct aspects": 32532,
      "figurative language": 44014,
      "analyses proposed": 5462,
      "pragmatic features": 88749,
      "choosing transfer": 16933,
      "analysis role": 5823,
      "structural position": 113785,
      "build work": 14817,
      "argue addition": 8908,
      "context lexical": 22168,
      "plays significant": 87741,
      "significant role": 108863,
      "role determining": 103168,
      "order investigate": 82350,
      "role lexical": 103198,
      "order conduct": 82299,
      "effect size": 34614,
      "semantics results": 105458,
      "semantic factors": 105048,
      "influence word": 54319,
      "does language": 33362,
      "language reports": 60036,
      "seek address": 104514,
      "joint supervision": 57324,
      "annotations models": 6446,
      "used tools": 128816,
      "tools understanding": 122479,
      "building lowresource": 14860,
      "nlp key": 79627,
      "key problems": 57594,
      "problems lack": 91330,
      "pretraining corpora": 90241,
      "performance suffers": 86772,
      "annotations provided": 6457,
      "annotators prior": 6496,
      "30 participants": 904,
      "par better": 84519,
      "better crosslingual": 13556,
      "methods built": 69357,
      "representations potential": 99807,
      "detection current": 29915,
      "new event": 78910,
      "types work": 126389,
      "set provide": 107548,
      "dataset fewshot": 26930,
      "entities work": 37899,
      "opendomain dialogues": 81970,
      "particular address": 85377,
      "annotating modeling": 6269,
      "corpus dataset": 23744,
      "dialogues annotated": 30822,
      "corpus indicates": 23839,
      "previously introduced": 90602,
      "introduced dialogue": 56576,
      "model entities": 71086,
      "models combining": 72926,
      "information yield": 55111,
      "yield best": 135330,
      "best performances": 13397,
      "online catalog": 81744,
      "taxonomy enrichment": 119666,
      "applications various": 7037,
      "search despite": 104308,
      "prevalent use": 90362,
      "difficult scale": 31659,
      "effectively enrich": 34808,
      "question taxonomy": 95227,
      "terms consistency": 120297,
      "consistency existing": 21358,
      "new term": 79214,
      "heterogeneous information": 49860,
      "supervision existing": 115884,
      "propose distantly": 92628,
      "extraction employ": 42322,
      "employ graph": 36262,
      "different online": 31308,
      "russian texts": 103505,
      "texts convey": 121485,
      "opinions attitudes": 82117,
      "attitudes entities": 10137,
      "negative effects": 76925,
      "entities described": 37764,
      "situations paper": 109899,
      "paper described": 83807,
      "words expressions": 133954,
      "structures syntactic": 114106,
      "brief survey": 14617,
      "survey topic": 116204,
      "targeted syntactic": 117773,
      "analysis article": 5503,
      "article briefly": 9113,
      "representative models": 100013,
      "rich semantics": 102783,
      "corpus development": 23756,
      "command line": 18789,
      "prediction despite": 89049,
      "early 90s": 34386,
      "systems steep": 117161,
      "steep learning": 113226,
      "instead usual": 55693,
      "models techniques": 74170,
      "techniques employed": 119877,
      "model leveraging": 71457,
      "leveraging continuous": 63669,
      "employed model": 36305,
      "dynamic way": 34333,
      "state model": 112507,
      "accuracy surpassing": 2296,
      "techniques adaptive": 119829,
      "arabic document": 8510,
      "imagebased character": 51804,
      "arabic text": 8559,
      "depend complex": 29045,
      "handcrafted feature": 49336,
      "features propose": 43672,
      "framework arabic": 45428,
      "created published": 24682,
      "published datasets": 94352,
      "wikipedia title": 132683,
      "dataset arabic": 26744,
      "arabic colloquial": 8497,
      "colloquial arabic": 18522,
      "classical arabic": 17085,
      "micro macro": 70038,
      "sentiment attitude": 106687,
      "attentionbased context": 10060,
      "outperform ones": 82721,
      "ones trained": 81707,
      "induction shown": 54034,
      "usually ignored": 130426,
      "labels essential": 58595,
      "experiments work": 41211,
      "using labeled": 129778,
      "human memory": 50916,
      "largescale eventuality": 61435,
      "eventuality knowledge": 39617,
      "computational cognitive": 20364,
      "states events": 113040,
      "events crucial": 39572,
      "humans understand": 51115,
      "propose scalable": 93024,
      "result construct": 101367,
      "robust adversarial": 103015,
      "idea minimize": 51333,
      "adversarial sentences": 4163,
      "translation benchmarks": 124671,
      "backtranslation using": 11343,
      "using extra": 129664,
      "applications areas": 6896,
      "employ semantic": 36278,
      "solve challenging": 110590,
      "classification information": 17233,
      "taxonomic relation": 119656,
      "relations large": 98198,
      "wordnet knowledge": 133772,
      "relations knowledge": 98193,
      "granularity levels": 48750,
      "fewshot classification": 43898,
      "domains performance": 33830,
      "performance conventional": 86258,
      "work tackles": 134841,
      "assuming access": 9656,
      "set test": 107605,
      "paper set": 84422,
      "tackle issue": 117290,
      "build representation": 14804,
      "representation extractor": 99237,
      "domain test": 33673,
      "classifier order": 17567,
      "propose clustering": 92582,
      "features target": 43752,
      "domain similarity": 33640,
      "entropy minimization": 38167,
      "minimization adversarial": 70193,
      "absolute gains": 1731,
      "accuracy 495": 2064,
      "fewshot settings": 43915,
      "combining neural": 18738,
      "word recently": 133428,
      "proposed generates": 93303,
      "generates possible": 47169,
      "word particular": 133399,
      "particular context": 85396,
      "models clusters": 72912,
      "sparse bagofwords": 111222,
      "work apply": 134374,
      "ways propose": 132170,
      "right contexts": 102831,
      "contexts resulting": 22426,
      "generated second": 47118,
      "established new": 38605,
      "improving current": 53081,
      "datasets large": 27540,
      "structured text": 114048,
      "widelyused datasets": 132604,
      "software platform": 110522,
      "pair build": 83422,
      "build evaluate": 14765,
      "models seven": 74016,
      "copy mechanisms": 23290,
      "pair dataset": 83426,
      "translation settings": 125246,
      "settings experiments": 107800,
      "learning translate": 63126,
      "using copy": 129503,
      "networks sentiment": 77758,
      "supervision sentiment": 115915,
      "documents experiments": 33229,
      "learning repeated": 62958,
      "speed process": 111875,
      "vocabulary increasing": 131877,
      "production open": 92059,
      "practices paper": 88741,
      "discuss properties": 32280,
      "learners english": 62298,
      "using acoustic": 129316,
      "prosodic lexical": 93646,
      "results developed": 101721,
      "framework preprocessing": 45654,
      "local knowledge": 64927,
      "base previous": 11469,
      "studies semantic": 114276,
      "abbreviations acronyms": 1467,
      "semantic context": 105020,
      "performance use": 86826,
      "lesk algorithm": 63395,
      "algorithm facilitate": 4740,
      "networks framework": 77604,
      "extract sentiments": 42116,
      "sentiments tweets": 106835,
      "reveals advantage": 102513,
      "use local": 128129,
      "introduce effective": 56417,
      "learning stages": 63051,
      "stages involved": 112159,
      "involved learning": 56883,
      "trained learn": 123188,
      "response respectively": 101235,
      "trained chinese": 123086,
      "effectiveness superiority": 34957,
      "evaluations achieving": 39451,
      "strategic planning": 113444,
      "perspectives new": 87178,
      "language sources": 60106,
      "designed texts": 29629,
      "model manually": 71499,
      "corrected annotated": 24132,
      "annotated texts": 6246,
      "texts large": 121539,
      "editing propose": 34553,
      "editing task": 34555,
      "task referred": 118619,
      "task important": 118276,
      "common requirement": 18916,
      "dataset research": 27161,
      "method public": 69086,
      "instances respectively": 55635,
      "facts using": 42925,
      "employ encoderdecoder": 36261,
      "outperforms encoderdecoder": 82882,
      "augmentation language": 10283,
      "augmentation pretrained": 10297,
      "episodic memory": 38214,
      "memory grows": 68317,
      "zero shot": 135481,
      "coreference task": 23379,
      "attracted growing": 10148,
      "growing attention": 49164,
      "attention use": 10028,
      "use sequencetosequence": 128270,
      "seq2seq architecture": 106886,
      "architecture generate": 8666,
      "safe responses": 103534,
      "guidance generate": 49226,
      "topic keywords": 122533,
      "final response": 44122,
      "focuses building": 44892,
      "comprehension systems": 20234,
      "requiring manually": 100349,
      "translated squad": 124565,
      "data release": 26332,
      "bert xlmroberta": 13260,
      "models main": 73542,
      "focus lies": 44782,
      "transfer models": 124154,
      "xlmroberta model": 135208,
      "data evaluated": 25909,
      "data result": 26356,
      "considering fact": 21324,
      "approach flexible": 7576,
      "comprehension language": 20187,
      "mrc tasks": 74931,
      "growing corpora": 49170,
      "mono multilingual": 74546,
      "multilingual lms": 75272,
      "struggle obtain": 114126,
      "bigger datasets": 14001,
      "datasets typical": 27769,
      "case implies": 15589,
      "corpora target": 23596,
      "demonstrate careful": 28685,
      "datasets user": 27781,
      "levels experimental": 63533,
      "terms exact": 120319,
      "match score": 67364,
      "classifier robust": 17576,
      "tweet data": 126010,
      "data enhance": 25894,
      "enhance coverage": 37465,
      "domain bias": 33474,
      "bias train": 13850,
      "dataset pretrained": 27105,
      "leverage different": 63583,
      "weighted sum": 132357,
      "approach achieved": 7299,
      "achieved positive": 2667,
      "place 26": 87599,
      "26 teams": 843,
      "dataset order": 27072,
      "benefit various": 12997,
      "systems information": 116955,
      "ability measure": 1541,
      "gap introduce": 46458,
      "pairs collected": 83496,
      "research dataset": 100457,
      "million sentence": 70105,
      "different sampling": 31395,
      "sampling strategies": 103612,
      "pairs provide": 83618,
      "balanced dataset": 11404,
      "dataset respect": 27162,
      "amt crowd": 5406,
      "crowd workers": 25098,
      "tasks verify": 119596,
      "collected annotations": 18401,
      "annotations final": 6433,
      "coverage lexicon": 24512,
      "linguistic behavior": 64431,
      "ontological concepts": 81827,
      "lexicon ontology": 63900,
      "evaluated accuracy": 38949,
      "obtain high": 81289,
      "available sentiment": 11101,
      "detection bengali": 29890,
      "reviews using": 102636,
      "recently sentiment": 96762,
      "detection increased": 29973,
      "customers opinions": 25452,
      "reviews online": 102620,
      "online platform": 81788,
      "introduces machine": 56618,
      "assess effectiveness": 9475,
      "corpus 2000": 23632,
      "analysis various": 5933,
      "taking consideration": 117545,
      "bigram trigram": 14007,
      "trigram features": 125795,
      "respectively experimental": 101136,
      "unigram feature": 127156,
      "outperforms techniques": 83036,
      "costly process": 24389,
      "especially dealing": 38447,
      "major source": 66586,
      "structures end": 114072,
      "extracting semantic": 42236,
      "exploiting semantic": 41477,
      "heterogeneous sources": 49866,
      "ontology annotation": 81835,
      "questionanswer pair": 95245,
      "declarative sentence": 27973,
      "results webnlg": 102325,
      "research annotation": 100404,
      "annotation rules": 6365,
      "recognition algorithm": 96803,
      "based phrase": 11927,
      "phrase window": 87378,
      "parsing mainly": 85151,
      "mainly uses": 66496,
      "learning main": 62702,
      "method firstly": 68843,
      "rules complex": 103387,
      "complex data": 19806,
      "data difficult": 25847,
      "difficult label": 31637,
      "language components": 58901,
      "components order": 20040,
      "labeling rule": 58534,
      "uses phrases": 129261,
      "divides sentences": 32913,
      "grammatical dependencies": 48694,
      "target area": 117583,
      "start end": 112398,
      "end positions": 36821,
      "nested phrases": 77110,
      "dependencies results": 29109,
      "use ambiguity": 127893,
      "algorithm experiments": 4731,
      "cpwd dataset": 24579,
      "dataset improve": 26971,
      "accuracy endtoend": 2145,
      "method point": 69053,
      "point corresponding": 87797,
      "corresponding method": 24293,
      "competition place": 19624,
      "syntactic agreement": 116363,
      "conducted numerous": 20937,
      "conducted english": 20919,
      "end design": 36803,
      "language experimental": 59027,
      "perform agreement": 85943,
      "tasks identify": 119162,
      "push limits": 94455,
      "manually selecting": 67076,
      "citations paper": 16993,
      "paper increasingly": 83982,
      "challenging timeconsuming": 16341,
      "approaches automated": 8071,
      "reason paper": 96201,
      "leverages deep": 63643,
      "knowledge approach": 57756,
      "approach combine": 7425,
      "combine deep": 18622,
      "deep representations": 28416,
      "submodular selection": 114784,
      "recommendation experiments": 97086,
      "corpus evaluated": 23782,
      "approach coreference": 7462,
      "task coreference": 118025,
      "resolution important": 100764,
      "important consider": 52131,
      "technique help": 119789,
      "longdistance relations": 65150,
      "relations presented": 98238,
      "report ongoing": 99019,
      "resolve coreference": 100799,
      "coreference mentions": 23350,
      "given mention": 48061,
      "used encoding": 128510,
      "encoding documents": 36688,
      "proposed span": 93555,
      "position embedding": 88284,
      "mentions document": 68450,
      "conll 2012": 21075,
      "2012 dataset": 565,
      "approaches novel": 8260,
      "classification opinion": 17316,
      "opinion classification": 82084,
      "classification challenging": 17142,
      "challenging problems": 16299,
      "modern deep": 74383,
      "capsule networks": 15249,
      "classification shown": 17404,
      "advantages convolutional": 4072,
      "validity domain": 130749,
      "text explored": 120940,
      "explored paper": 41630,
      "novel hybrid": 80594,
      "units bigru": 127234,
      "extraction layer": 42369,
      "preprocessing steps": 89335,
      "consists layers": 21480,
      "based glove": 11742,
      "bigru based": 14011,
      "layer order": 61738,
      "conducted extensive": 20924,
      "datasets ranging": 27658,
      "including movie": 53325,
      "dataset yelp": 27282,
      "review polarity": 102561,
      "benchmarks cover": 12894,
      "classification multilabel": 17286,
      "methods help": 69530,
      "help external": 49720,
      "positive sentiment": 88343,
      "negative sentiment": 76954,
      "way human": 132088,
      "classifying large": 17670,
      "task social": 118722,
      "networks twitter": 77805,
      "gather information": 46526,
      "sentiments opinions": 106833,
      "fact people": 42829,
      "people spend": 85893,
      "daily social": 25513,
      "share opinion": 107957,
      "helps analyze": 49811,
      "companies using": 19118,
      "tools provide": 122468,
      "provide various": 93955,
      "polarity given": 87911,
      "order understand": 82423,
      "aims perform": 4554,
      "analysis significantly": 5855,
      "tfidf word2vec": 121764,
      "quality features": 94667,
      "features considering": 43421,
      "considering contextual": 21321,
      "contextual semantics": 22496,
      "text improving": 121047,
      "analysis discourse": 5579,
      "argue new": 8925,
      "associated knowledge": 9598,
      "reasoning conversational": 96239,
      "conversational strategy": 23035,
      "communication domain": 19031,
      "constraints discourse": 21596,
      "learned probabilistic": 62245,
      "learning greedy": 62624,
      "parsing previous": 85199,
      "representing 1best": 100041,
      "representation entire": 99226,
      "results greedy": 101819,
      "dataset argue": 26745,
      "serve benchmark": 107292,
      "benchmark provide": 12851,
      "groups research": 49154,
      "work possible": 134671,
      "dataset narrative": 27048,
      "narrative structure": 76177,
      "narrative structures": 76178,
      "works utilized": 134992,
      "narrative theory": 76181,
      "narrative elements": 76171,
      "personal stories": 87124,
      "texts instead": 121531,
      "focus news": 44800,
      "shaping public": 107943,
      "process dataset": 91453,
      "success used": 115138,
      "scheme annotate": 103911,
      "articles containing": 9178,
      "score 07": 104045,
      "suggesting promising": 115350,
      "work learn": 134605,
      "domain related": 33624,
      "data separate": 26396,
      "correlated task": 24209,
      "research attention": 100418,
      "attention machine": 9873,
      "task usually": 118834,
      "associated multiple": 9602,
      "different terms": 31485,
      "step predict": 113295,
      "translation conduct": 124705,
      "different latency": 31220,
      "stock trend": 113366,
      "exploration better": 41496,
      "tasks boost": 118970,
      "performance main": 86515,
      "task jointly": 118325,
      "optimization experiments": 82191,
      "transfer data": 124042,
      "transfer crosslingual": 124041,
      "data prior": 26277,
      "transfer based": 124038,
      "language generally": 59062,
      "data translation": 26583,
      "rarely leverage": 95760,
      "collected potentially": 18431,
      "approach termed": 7967,
      "targetlanguage data": 117780,
      "distillation evaluate": 32505,
      "models consisting": 72968,
      "consisting millions": 21456,
      "models expensive": 73195,
      "layer pretrained": 61741,
      "finetuning entire": 44456,
      "entire model": 37709,
      "xlmr tasks": 135206,
      "sharing training": 108143,
      "changes training": 16394,
      "easy access": 34472,
      "particularly lowresource": 85487,
      "framework crosslingual": 45477,
      "present informationtheoretic": 89516,
      "helps better": 49812,
      "crosslingual representations": 25000,
      "based contrastive": 11603,
      "contrastive learning": 22731,
      "representations similar": 99893,
      "examples leveraging": 39853,
      "leveraging monolingual": 63688,
      "improve crosslingual": 52366,
      "results benchmarks": 101547,
      "benchmarks approach": 12885,
      "reasoning machine": 96264,
      "task testing": 118785,
      "related human": 97867,
      "human cognition": 50776,
      "human performances": 50931,
      "challenging machine": 16273,
      "proposed various": 93588,
      "various challenges": 131057,
      "build comprehensive": 14756,
      "comprehensive dataset": 20255,
      "instances covering": 55623,
      "covering multiple": 24545,
      "perform far": 86006,
      "human ceiling": 50774,
      "dataset serve": 27190,
      "nlp setting": 79687,
      "dataset freely": 26942,
      "hierarchical interaction": 49970,
      "networks rethinking": 77741,
      "analysis documentlevel": 5582,
      "summarization achieved": 115474,
      "ignoring inherent": 51690,
      "points document": 87853,
      "study effectively": 114362,
      "generate discriminative": 46929,
      "discriminative representation": 32215,
      "proposed explore": 93277,
      "explore bidirectional": 41526,
      "bidirectional interactions": 13929,
      "document multiple": 33042,
      "classification furthermore": 17216,
      "models public": 73841,
      "consistently demonstrate": 21414,
      "family tree": 43058,
      "tree indoeuropean": 125596,
      "indoeuropean languages": 53954,
      "crosslinguistic transfer": 25056,
      "languages consistent": 60467,
      "way paper": 132115,
      "paper validate": 84491,
      "parsing grammar": 85119,
      "mixed features": 70397,
      "structure resulting": 113945,
      "analysis persian": 5749,
      "using discourse": 129582,
      "external semantic": 42041,
      "attempts identify": 9770,
      "affective states": 4241,
      "text audio": 120669,
      "proposed extract": 93279,
      "written natural": 135138,
      "majority approaches": 66593,
      "approaches focused": 8165,
      "resourcelean languages": 100900,
      "lack research": 58741,
      "based classifier": 11574,
      "classifier combination": 17530,
      "benefits word": 13027,
      "advantage local": 4049,
      "local discourse": 64916,
      "levels word": 63569,
      "aspect sentence": 9346,
      "approach persian": 7800,
      "related works": 97914,
      "language essential": 59014,
      "representation scheme": 99400,
      "phenomena encountered": 87222,
      "images videos": 51816,
      "spatial representation": 111270,
      "languages increases": 60637,
      "increases coverage": 53639,
      "coverage semantic": 24523,
      "semantic aspects": 104994,
      "text world": 121416,
      "spatial relation": 111268,
      "concepts crucial": 20610,
      "dynamic spatial": 34327,
      "schema present": 103901,
      "diverse datasets": 32805,
      "datasets extend": 27478,
      "sentences discourse": 106281,
      "discourse representations": 32085,
      "correction neural": 24160,
      "training previous": 123787,
      "discrepancy training": 32158,
      "affects translation": 4250,
      "generated tokens": 47140,
      "tokens better": 122300,
      "twostream selfattention": 126177,
      "used correct": 128461,
      "previous predicted": 90443,
      "predicted tokens": 88969,
      "sampling simulate": 103609,
      "errors training": 38415,
      "iwslt translation": 57166,
      "wmt translation": 132806,
      "analyses verify": 5472,
      "approach offensive": 7749,
      "offensive language": 81542,
      "transformer encoders": 124323,
      "task offensive": 118474,
      "layers applied": 61761,
      "perform multihead": 86033,
      "multihead attentions": 75128,
      "representations taken": 99926,
      "concatenated fed": 20534,
      "make final": 66672,
      "models shows": 74034,
      "set achieves": 107348,
      "high performing": 50100,
      "performing systems": 86972,
      "participants subtask": 85326,
      "subtask shared": 114953,
      "shows ensemble": 108576,
      "accuracy development": 2137,
      "set improvement": 107462,
      "verbal multiword": 131536,
      "discriminating features": 32198,
      "features automatic": 43372,
      "leads various": 61974,
      "optimal set": 82169,
      "classification setting": 17402,
      "using optimal": 129992,
      "features outperforms": 43643,
      "best systems": 13461,
      "recent shared": 96512,
      "effects language": 34988,
      "relatedness crosslingual": 97918,
      "characterbased language": 16491,
      "training times": 123932,
      "character multicharacter": 16451,
      "suffer data": 115212,
      "performance allowing": 86134,
      "allowing information": 5112,
      "asr applying": 9424,
      "relatedness source": 97928,
      "language plays": 59814,
      "pretraining character": 90238,
      "asr tasks": 9450,
      "tasks target": 119547,
      "languages finnish": 60592,
      "finnish english": 44555,
      "swedish danish": 116263,
      "difference using": 30948,
      "pretraining related": 90313,
      "asr performance": 9444,
      "does pretrained": 33379,
      "model know": 71393,
      "information selfattention": 54961,
      "experiments analyses": 40764,
      "emphasis selection": 36137,
      "meaning present": 67663,
      "submitted semeval": 114769,
      "semeval 2020": 105508,
      "2020 task": 733,
      "graded word": 48590,
      "similarity context": 109212,
      "existing semantically": 40278,
      "propose approximate": 92552,
      "song lyrics": 110660,
      "convey meaningful": 23116,
      "meaningful story": 67729,
      "rhythmic patterns": 102722,
      "works explore": 134946,
      "script propose": 104260,
      "techniques capture": 119849,
      "dataset codes": 26792,
      "available publicly": 11086,
      "speech identification": 111691,
      "media paper": 68138,
      "approach utilize": 8008,
      "multilingual offensive": 75318,
      "identification shared": 51441,
      "task offenseval": 118473,
      "offenseval 2020": 81538,
      "combining cnn": 18711,
      "cnn bert": 17990,
      "bert better": 13077,
      "utilizing pretrained": 130574,
      "ranked 4th": 95638,
      "macro averaged": 66354,
      "averaged f1score": 11224,
      "set pretrained": 107541,
      "models arabic": 72766,
      "years growing": 135267,
      "platforms facebook": 87664,
      "social text": 110460,
      "text involves": 121065,
      "visual cues": 131790,
      "cues language": 25212,
      "analyze sentiment": 6006,
      "identifying categorizing": 51585,
      "dataset according": 26717,
      "according language": 1985,
      "adapt finetune": 3179,
      "finetune bidirectional": 44404,
      "cross domain": 24858,
      "baseline algorithm": 12184,
      "place subtask": 87608,
      "subtask sentiment": 114951,
      "classification code": 17147,
      "research language": 100543,
      "user interfaces": 129008,
      "approaches recent": 8306,
      "attempts build": 9765,
      "using modern": 129892,
      "benchmarks emphasize": 12900,
      "data cost": 25799,
      "research findings": 100516,
      "benchmarks relevant": 12934,
      "common ways": 18947,
      "recent benchmarks": 96435,
      "lastly offer": 61557,
      "work online": 134661,
      "paper work": 84494,
      "established metrics": 38604,
      "metrics including": 69972,
      "including metrics": 53323,
      "research excellence": 100492,
      "excellence framework": 39911,
      "framework ref": 45669,
      "ref 2014": 97492,
      "societal impact": 110468,
      "mentioned news": 68438,
      "graphbased ranking": 48908,
      "news used": 79393,
      "diachronic evolution": 30489,
      "classification stance": 17418,
      "studies highlight": 114234,
      "users opinion": 129150,
      "furthermore stance": 46214,
      "proposing novel": 93628,
      "approach annotation": 7353,
      "main aim": 66395,
      "investigating role": 56858,
      "stance evolution": 112172,
      "useful clues": 128868,
      "detecting stance": 29852,
      "train text": 123044,
      "multilingual representation": 75348,
      "present semisupervised": 89687,
      "input perturbations": 55392,
      "training allowing": 123355,
      "compared number": 19400,
      "baselines observe": 12436,
      "effectiveness document": 34884,
      "classification diverse": 17189,
      "languages leverage": 60685,
      "data abstractive": 25554,
      "speech summarization": 111809,
      "neural abstractive": 77824,
      "costly build": 24384,
      "audio recordings": 10235,
      "obtain automatic": 81257,
      "manual transcription": 67015,
      "examples suitable": 39883,
      "access large": 1914,
      "unaligned data": 126577,
      "professionally written": 92089,
      "straightforward context": 113431,
      "using approaches": 129343,
      "pretraining using": 90351,
      "using targetside": 130268,
      "summarization process": 115551,
      "process learning": 91521,
      "order align": 82276,
      "transcription use": 124002,
      "trained aligned": 123067,
      "task commonly": 117976,
      "languages mrls": 60741,
      "boundaries named": 14517,
      "need answer": 76781,
      "realistic settings": 96105,
      "ner benchmark": 77024,
      "modern hebrew": 74391,
      "improved ner": 52618,
      "performance novel": 86579,
      "architecture ner": 8706,
      "greatly outperforms": 49059,
      "outperforms standard": 82989,
      "setting new": 107762,
      "hebrew morphological": 49672,
      "translates source": 124582,
      "java python": 57220,
      "missing important": 70326,
      "define use": 28496,
      "use classes": 127936,
      "events related": 39598,
      "related knowledge": 97870,
      "sources various": 110928,
      "countries english": 24448,
      "information constructing": 54442,
      "bases enable": 12497,
      "social political": 110447,
      "source annotation": 110711,
      "using active": 129317,
      "samples annotated": 103579,
      "political scientists": 88000,
      "annotation errors": 6303,
      "develop benchmark": 30179,
      "crosscontext setting": 24872,
      "corpus reported": 23968,
      "set currently": 107410,
      "common ground": 18883,
      "network relation": 77405,
      "tools build": 122442,
      "reliance linguistic": 98631,
      "network sgcn": 77431,
      "graph using": 48881,
      "mechanism using": 68046,
      "encoded using": 36485,
      "tacred dataset": 117330,
      "result experiments": 101372,
      "task semeval2020": 118681,
      "characteristic human": 16515,
      "task provides": 118586,
      "provides benchmark": 94001,
      "determine given": 30126,
      "sentence counterfactual": 105816,
      "evaluation period": 39321,
      "received 27": 96355,
      "27 submissions": 853,
      "data baseline": 25690,
      "baseline code": 12203,
      "text come": 120800,
      "years thanks": 135309,
      "increasing availability": 53664,
      "availability parallel": 10923,
      "challenging modern": 16277,
      "challenging examples": 16250,
      "adding corpus": 3374,
      "resulting benefits": 101431,
      "benefits use": 13025,
      "multiple texts": 75728,
      "languages test": 60916,
      "given topic": 48161,
      "investigate hypothesis": 56762,
      "linguistic bias": 64432,
      "deals question": 27877,
      "linguistic relativity": 64543,
      "model information": 71342,
      "typical input": 126400,
      "model example": 71118,
      "article builds": 9114,
      "research educational": 100480,
      "research computational": 100448,
      "linguistics computational": 64609,
      "key terms": 57611,
      "linguistic perspective": 64520,
      "concepts instances": 20623,
      "specialized vocabularies": 111395,
      "used specialized": 128772,
      "learning environment": 62552,
      "combine lines": 18630,
      "provide computational": 93790,
      "different resources": 31388,
      "resources including": 100990,
      "online resources": 81796,
      "wikipedia largest": 132664,
      "largest widely": 61541,
      "used online": 128672,
      "structure kind": 113888,
      "kinds resources": 57691,
      "learning resource": 62965,
      "resource efficient": 100840,
      "presents efficient": 89841,
      "model adapted": 70625,
      "satisfy constraints": 103658,
      "marginal probabilities": 67210,
      "ngram model": 79445,
      "method originally": 69001,
      "entropy language": 38164,
      "corpus vocabulary": 24064,
      "algorithm large": 4762,
      "slightly worse": 110029,
      "rate results": 95804,
      "retrieval tasks": 102432,
      "generally steps": 46877,
      "query candidates": 94951,
      "based extracted": 11710,
      "extracted features": 42157,
      "features major": 43602,
      "major challenges": 66546,
      "tens milliseconds": 120177,
      "poses significant": 88274,
      "designing sophisticated": 29648,
      "poor quality": 88061,
      "quality contextual": 94616,
      "information fully": 54626,
      "fully utilized": 45981,
      "systems heavily": 116929,
      "search logs": 104325,
      "sufficient semantic": 115286,
      "effective context": 34651,
      "information possible": 54852,
      "generate relevant": 46998,
      "proposed effectively": 93262,
      "deep semantics": 28424,
      "queries approach": 94912,
      "approach presents": 7812,
      "better ranking": 13688,
      "ranking performance": 95680,
      "methods reduces": 69712,
      "latency compared": 61564,
      "compared neural": 19397,
      "good balance": 48462,
      "accurate neural": 2348,
      "field deep": 43947,
      "vast applications": 131275,
      "tasks works": 119615,
      "complexity inefficiency": 19917,
      "inefficiency insideoutside": 54071,
      "insideoutside algorithm": 55519,
      "algorithm work": 4817,
      "batchify inside": 12561,
      "direct large": 31789,
      "outside algorithm": 83194,
      "efficient backpropagation": 35061,
      "new scoring": 79119,
      "performance settings": 86708,
      "release code": 98436,
      "code httpsgithubcomyzhangcscrfpar": 18107,
      "usually needs": 130440,
      "largescale highquality": 61441,
      "base version": 11489,
      "quality dataset": 94627,
      "built based": 14913,
      "based set": 12037,
      "trained manually": 123192,
      "cleaned dataset": 17707,
      "dataset pretraining": 27106,
      "bert recent": 13214,
      "corpora large": 23514,
      "data long": 26095,
      "long time": 65141,
      "important capture": 52109,
      "capture languagespecific": 15323,
      "phenomena multilingual": 87233,
      "span tokens": 111137,
      "subcharacter level": 114651,
      "performed comparably": 86915,
      "parallel evaluation": 84666,
      "software documentation": 110520,
      "structure annotation": 113809,
      "paper accompanies": 83695,
      "set machine": 107485,
      "set data": 107411,
      "data originating": 26196,
      "evaluate machine": 38849,
      "additional metadata": 3549,
      "context provide": 22237,
      "scale multilingual": 103738,
      "benchmarks paper": 12931,
      "framework obtain": 45632,
      "obtain corpus": 81272,
      "systems pipeline": 117066,
      "pipeline consists": 87534,
      "alignment module": 4974,
      "module used": 74515,
      "press releases": 89954,
      "uses pipeline": 129262,
      "iteratively improve": 57140,
      "improve components": 52353,
      "components work": 20054,
      "size work": 109954,
      "work addition": 134353,
      "addition providing": 3469,
      "results generating": 101811,
      "corpus compared": 23700,
      "available indian": 11014,
      "2020 present": 727,
      "api focus": 6778,
      "art model": 9060,
      "training efficient": 123601,
      "cpu decoding": 24575,
      "metric scores": 69900,
      "scores shorter": 104203,
      "annotation guideline": 6321,
      "corpus report": 23967,
      "corpus multiple": 23891,
      "layers linguistic": 61787,
      "corpus offers": 23915,
      "annotated 15": 6135,
      "dataset considered": 26817,
      "joint neural": 57302,
      "linguistically rich": 64597,
      "interactive visualizations": 56036,
      "visualization understanding": 131832,
      "understanding nlp": 126905,
      "focus core": 44748,
      "model behavior": 70755,
      "change input": 16362,
      "aggregate analysis": 4345,
      "counterfactual generation": 24426,
      "including exploring": 53294,
      "measuring gender": 67910,
      "bias coreference": 13798,
      "coreference systems": 23378,
      "classification seq2seq": 17396,
      "highly extensible": 50322,
      "automatic quality": 10631,
      "new issue": 78964,
      "great value": 49034,
      "created users": 24688,
      "diverse content": 32800,
      "large semantic": 61267,
      "requirements solve": 100237,
      "challenges establish": 16155,
      "effectively learns": 34828,
      "classification major": 17262,
      "classification mltc": 17270,
      "label correlations": 58386,
      "correlations paper": 24261,
      "mltc task": 70453,
      "method significant": 69132,
      "significant advantages": 108722,
      "code released": 18144,
      "released facilitate": 98483,
      "higherlevel semantic": 50216,
      "semantic language": 105093,
      "models selfsupervised": 73988,
      "success wide": 115144,
      "tasks questions": 119431,
      "supervised parsing": 115820,
      "transformer biaffine": 124311,
      "finetune model": 44409,
      "parsing training": 85267,
      "parsing explicit": 85111,
      "effect downstream": 34595,
      "representation spaces": 99416,
      "parsing make": 85153,
      "make significant": 66725,
      "significant step": 108869,
      "providing answers": 94103,
      "highlevel semantic": 50249,
      "cointeractive relation": 18344,
      "relation network": 98056,
      "act recognition": 3067,
      "classification dialog": 17184,
      "speakers intentions": 111319,
      "act sentiment": 3069,
      "separately existing": 106872,
      "systems treat": 117208,
      "way explicitly": 132082,
      "modeling mutual": 72485,
      "mutual interaction": 75984,
      "interaction relation": 55966,
      "explicitly consider": 41364,
      "consider crossimpact": 21195,
      "tasks introducing": 119223,
      "introducing cointeractive": 56629,
      "proposed relation": 93535,
      "layer stacked": 61750,
      "mutual knowledge": 75986,
      "thoroughly study": 121907,
      "different relation": 31377,
      "34 terms": 953,
      "classification respectively": 17382,
      "verifies effectiveness": 131577,
      "modeling relation": 72526,
      "interaction mechanism": 55952,
      "mechanism finally": 67986,
      "finally employ": 44176,
      "employ bidirectional": 36253,
      "representation transformer": 99448,
      "framework boost": 45442,
      "tasks adding": 118919,
      "context size": 22271,
      "task standard": 118739,
      "method nlp": 68989,
      "models impressive": 73364,
      "models latest": 73472,
      "method applying": 68641,
      "applying pretrained": 7266,
      "memory requirement": 68380,
      "time additional": 121980,
      "additional benefit": 3493,
      "gpt2 language": 48562,
      "attains better": 9729,
      "computation memory": 20350,
      "order classical": 82293,
      "classical logic": 17088,
      "parser developed": 84935,
      "automated textual": 10474,
      "step automated": 113250,
      "extraction quality": 42456,
      "quality syntactic": 94806,
      "describes work": 29446,
      "team semeval": 119720,
      "commonsense validation": 19012,
      "validation explanation": 130740,
      "subtasks subtasks": 114984,
      "subtasks submissions": 114982,
      "subtask submission": 114955,
      "pretrained sequencetosequence": 90187,
      "model bart": 70721,
      "score ranking": 104113,
      "correlation bleu": 24225,
      "propose additional": 92530,
      "model subtask": 72105,
      "subtasks present": 114976,
      "media influence": 68122,
      "certain demographics": 15939,
      "largescale analysis": 61397,
      "movies method": 74914,
      "method develop": 68773,
      "used scripts": 128738,
      "method applicable": 68638,
      "complementary previous": 19751,
      "roles victim": 103238,
      "female characters": 43872,
      "characters appear": 16595,
      "tools assist": 122435,
      "opendomain chatbots": 81957,
      "standardized evaluation": 112347,
      "metrics model": 69988,
      "chatbot challenges": 16663,
      "challenges recently": 16200,
      "english resources": 37263,
      "collection efforts": 18474,
      "efforts create": 35192,
      "topics people": 122647,
      "retrievalbased evaluation": 102443,
      "corpora perform": 23556,
      "better chance": 13532,
      "metrics humans": 69971,
      "humans task": 51112,
      "responses questions": 101292,
      "set generated": 107452,
      "provides challenging": 94008,
      "challenging evaluation": 16249,
      "similarity combination": 109208,
      "describes submitted": 29435,
      "effect context": 34588,
      "models order": 73669,
      "order compute": 82298,
      "words hypothesis": 133985,
      "evidence additional": 39634,
      "highest scoring": 50237,
      "bases entity": 12498,
      "setting allows": 107731,
      "language paraphrases": 59801,
      "number facts": 80890,
      "representations allow": 99498,
      "millions entities": 70116,
      "models serve": 74012,
      "trees help": 125693,
      "help pretrained": 49765,
      "incorporating syntax": 53566,
      "taskspecific transformer": 119644,
      "models effect": 73115,
      "encode syntax": 36459,
      "syntax work": 116570,
      "systematically study": 116698,
      "study utility": 114544,
      "utility incorporating": 130481,
      "trees pretrained": 125704,
      "representative information": 100011,
      "labeling srl": 58543,
      "approach applies": 7358,
      "applies graph": 7153,
      "output transformer": 83141,
      "transformer joint": 124329,
      "structure transformer": 113977,
      "performance empirical": 86334,
      "transformers obtain": 124457,
      "contingent availability": 22596,
      "raises important": 95461,
      "important questions": 52222,
      "questions regarding": 95352,
      "t5 model": 117249,
      "need resources": 76852,
      "used stateoftheart": 128782,
      "stateoftheart research": 112888,
      "corpus extensive": 23796,
      "extensive collection": 41863,
      "collection web": 18498,
      "portuguese evaluate": 88198,
      "performance portuguese": 86608,
      "t5 models": 117250,
      "impact using": 51900,
      "using portuguese": 130033,
      "documentlevel event": 33141,
      "event entity": 39511,
      "problem documentlevel": 91016,
      "introduce generative": 56430,
      "level make": 63473,
      "make extraction": 66668,
      "extraction decisions": 42297,
      "phrase coreference": 87342,
      "performs substantially": 87030,
      "better prior": 13683,
      "contribute model": 22756,
      "capturing linguistic": 15476,
      "variational approach": 130907,
      "approach unsupervised": 7993,
      "unsupervised sentiment": 127720,
      "truth provided": 125884,
      "provided domain": 93963,
      "experts use": 41249,
      "use targetopinion": 128314,
      "targetopinion word": 117783,
      "parsers simple": 85044,
      "rules objective": 103423,
      "predict opinion": 88910,
      "opinion word": 82107,
      "word ultimate": 133613,
      "goal learn": 48364,
      "learn sentiment": 62146,
      "introducing latent": 56641,
      "variable sentiment": 130847,
      "polarity objective": 87917,
      "function inject": 46027,
      "inject sentiment": 55260,
      "classifier objective": 17562,
      "bound learn": 14508,
      "classifier optimizing": 17564,
      "optimizing lower": 82243,
      "opinion words": 82111,
      "regularization encourages": 97748,
      "documents similar": 33293,
      "words sentiment": 134202,
      "similar different": 109083,
      "method outperform": 69003,
      "baselines sentiment": 12460,
      "domains method": 33817,
      "method obtains": 68997,
      "obtains comparable": 81459,
      "method hundreds": 68874,
      "hundreds labels": 51148,
      "labels aspect": 58581,
      "methods clinical": 69369,
      "domain end": 33519,
      "humans computers": 51073,
      "networkbased model": 77487,
      "dialogue uses": 30811,
      "uses sequencetosequence": 129276,
      "sequencetosequence architecture": 107153,
      "architecture produce": 8731,
      "textual representation": 121731,
      "step use": 113312,
      "architecture instead": 8676,
      "rnnbased architecture": 102941,
      "architecture works": 8773,
      "accurate estimation": 2334,
      "popular efficient": 88091,
      "used extensively": 128541,
      "structure given": 113870,
      "tables representation": 117267,
      "similarity train": 109323,
      "method created": 68749,
      "extracted biomedical": 42143,
      "degree similarity": 28576,
      "similar task": 109155,
      "measures average": 67853,
      "translation sentiment": 125240,
      "required build": 100213,
      "propose employing": 92647,
      "employing machine": 36316,
      "mt translate": 75003,
      "sentences positive": 106448,
      "classifier furthermore": 17545,
      "performing direct": 86948,
      "translation possible": 125125,
      "performance automatic": 86158,
      "classification discuss": 17187,
      "advantages drawbacks": 4078,
      "using pointergenerator": 130022,
      "extraction crucial": 42293,
      "dictionary lookup": 30884,
      "propose generic": 92705,
      "extractive model": 42554,
      "pointergenerator network": 87832,
      "leveraging bidirectional": 63665,
      "corpus compiled": 23703,
      "outcome experiments": 82616,
      "furthermore model": 46191,
      "model readily": 71855,
      "gricean quantity": 49085,
      "information way": 55091,
      "representation information": 99263,
      "probabilistic treatment": 90835,
      "interpretation use": 56276,
      "rsa framework": 103333,
      "uncertainty model": 126621,
      "chinese clinical": 16745,
      "clinical named": 17773,
      "recognition clinical": 96834,
      "recognition cner": 96838,
      "cner aims": 17981,
      "aims automatically": 4504,
      "fundamental crucial": 46094,
      "step clinical": 113255,
      "clinical research": 17786,
      "labels labeling": 58614,
      "especially chinese": 38438,
      "timeconsuming expensive": 122145,
      "assumes single": 9654,
      "low overall": 65377,
      "annotation quality": 6363,
      "assign correct": 9540,
      "labels specific": 58645,
      "specific instances": 111451,
      "quality lower": 94708,
      "results ccks2017": 101563,
      "ccks2017 task": 15864,
      "knowledge efficient": 57885,
      "annotation timeconsuming": 6390,
      "methods knowledge": 69567,
      "efficient make": 35089,
      "annotation low": 6332,
      "settings various": 107845,
      "classical methods": 17091,
      "methods modern": 69625,
      "framework incorporating": 45573,
      "incorporating prior": 53552,
      "knowledge deep": 57853,
      "supervision second": 115914,
      "model assist": 70694,
      "models correct": 72992,
      "correct evidence": 24100,
      "evidence sentences": 39668,
      "sentences support": 106510,
      "transfer techniques": 124195,
      "techniques multilingual": 119934,
      "setting proposed": 107774,
      "improve pretrained": 52499,
      "tree models": 125603,
      "algorithms exploit": 4852,
      "algorithms computing": 4837,
      "known algorithms": 58286,
      "algorithms including": 4860,
      "approach matches": 7704,
      "existing algorithms": 40043,
      "algorithms cases": 4834,
      "implementation framework": 51945,
      "runtime experiments": 103481,
      "15 times": 352,
      "faster previous": 43188,
      "previous algorithms": 90377,
      "shannon entropy": 107936,
      "extraction free": 42344,
      "extraction conducted": 42287,
      "considered work": 21314,
      "definitions glosses": 28538,
      "appear explicit": 6796,
      "complex linguistic": 19828,
      "extraction adapt": 42253,
      "enhanced dependency": 37505,
      "2020 shared": 729,
      "17 languages": 402,
      "pipeline approach": 87523,
      "using udpipe": 130333,
      "provide initial": 93856,
      "levels annotation": 63527,
      "parser built": 84926,
      "built basic": 14914,
      "set heuristics": 107457,
      "languages semantic": 60859,
      "connected graph": 21125,
      "scores significantly": 104204,
      "submission ranked": 114733,
      "fourth competition": 45358,
      "competition evaluation": 19623,
      "checkthat lab": 16705,
      "task evaluated": 118156,
      "roberta models": 103001,
      "identify claims": 51485,
      "claims social": 17019,
      "finetuned roberta": 44435,
      "roberta model": 103000,
      "model added": 70627,
      "added extra": 3362,
      "mean pooling": 67602,
      "layer dropout": 61710,
      "layer enhance": 61713,
      "unseen text": 127551,
      "text arabic": 120656,
      "task finetuned": 118217,
      "balance dataset": 11396,
      "dataset work": 27280,
      "place english": 87602,
      "english track": 37309,
      "1st 2nd": 483,
      "2nd 3rd": 888,
      "place arabic": 87601,
      "label distribution": 58391,
      "distribution learning": 32666,
      "presents submission": 89905,
      "text various": 121402,
      "experiment choice": 40455,
      "choice model": 16890,
      "layers different": 61768,
      "different contextual": 31062,
      "performing architecture": 86943,
      "architecture ensemble": 8658,
      "ensemble different": 37593,
      "achieved overall": 2664,
      "terms parts": 120355,
      "sentence lengths": 105923,
      "upb semeval2020": 127783,
      "task joint": 118324,
      "joint textual": 57326,
      "analysis users": 5928,
      "main purpose": 66454,
      "ideas using": 51352,
      "images texts": 51813,
      "related various": 97910,
      "various situations": 131196,
      "paper developed": 83867,
      "developed team": 30312,
      "team semeval2020": 119722,
      "analysis specifically": 5867,
      "multimodal multitask": 75444,
      "performance subtasks": 86770,
      "macro f1score": 66360,
      "1st subtask": 486,
      "3rd subtask": 999,
      "similarity benchmark": 109203,
      "translation api": 124639,
      "api paper": 6779,
      "despite obvious": 29711,
      "resulting dataset": 101441,
      "use benchmark": 127920,
      "benchmark compare": 12742,
      "currently existing": 25404,
      "swedish text": 116265,
      "native models": 76229,
      "outperform multilingual": 82719,
      "multilingual ones": 75321,
      "simple bag": 109370,
      "performs remarkably": 87019,
      "selection pretrained": 104817,
      "describes designed": 29395,
      "team achieved": 119716,
      "place semeval2020": 87606,
      "leverage unsupervised": 63633,
      "model finetune": 71189,
      "following models": 44974,
      "roberta albert": 102993,
      "ranking loss": 95672,
      "performance best": 86179,
      "metrics quantifying": 70000,
      "design better": 29522,
      "communication platforms": 19039,
      "step using": 113314,
      "past conversations": 85635,
      "challenges particular": 16190,
      "particular domain": 85406,
      "illustrate potential": 51742,
      "lowresource polysynthetic": 65555,
      "polysynthetic languages": 88033,
      "translation tools": 125383,
      "yupik language": 135466,
      "polysynthetic language": 88032,
      "english developed": 37111,
      "developed morphological": 30289,
      "rules trained": 103450,
      "trained seq2seq": 123272,
      "english compared": 37095,
      "compared influence": 19378,
      "influence different": 54308,
      "methods rulebased": 69736,
      "did best": 30911,
      "reduced vocabulary": 97377,
      "models crossdomain": 73004,
      "systematic comparative": 116662,
      "study models": 114441,
      "models variety": 74320,
      "models explored": 73212,
      "study presented": 114476,
      "bert electra": 13107,
      "experiments finetuning": 40947,
      "finetuning models": 44479,
      "models performances": 73737,
      "capabilities models": 15170,
      "agnostic approach": 4380,
      "approach experimental": 7548,
      "results establish": 101764,
      "yelp 2013": 135319,
      "sentiment detection": 106729,
      "accuracy 882": 2079,
      "finally study": 44235,
      "greatly assist": 49046,
      "dealing large": 27868,
      "hours data": 50682,
      "selecting optimal": 104753,
      "optimal parameters": 82167,
      "times order": 122184,
      "save time": 103664,
      "parameters making": 84769,
      "obtain nearly": 81301,
      "accuracy given": 2174,
      "extraction documentlevel": 42308,
      "aims detect": 4515,
      "relation entities": 97974,
      "multiparty dialogue": 75482,
      "increasingly abundant": 53690,
      "additional context": 3500,
      "present graph": 89509,
      "networkbased method": 77485,
      "graph contains": 48774,
      "constructed graph": 21662,
      "graph fed": 48812,
      "network context": 77212,
      "context propagation": 22231,
      "different entity": 31128,
      "entity pairs": 37981,
      "pairs dialogue": 83516,
      "approaches significant": 8339,
      "language internet": 59223,
      "lack available": 58685,
      "resource training": 100883,
      "levels complexity": 63529,
      "task diversity": 118109,
      "large clean": 61047,
      "indonesian dataset": 53991,
      "available sources": 11107,
      "blogs news": 14373,
      "tasks framework": 119138,
      "framework benchmark": 45439,
      "benchmark performances": 12848,
      "solving arithmetic": 110639,
      "arithmetic word": 9009,
      "word problems": 133421,
      "assessing language": 9512,
      "understanding reasoning": 126933,
      "capabilities nlp": 15172,
      "use automatic": 127908,
      "extraction ranking": 42460,
      "problems work": 91383,
      "using established": 129643,
      "sequential representations": 107251,
      "expression tree": 41745,
      "treelstm architectures": 125676,
      "architectures experimental": 8806,
      "stateoftheart 15": 112567,
      "require complex": 100121,
      "sequential lstms": 107234,
      "lstms accuracy": 65697,
      "character vector": 16474,
      "nmt configurations": 79848,
      "input vector": 55470,
      "representation using": 99460,
      "operations addition": 82067,
      "models concatenation": 72949,
      "higher baseline": 50162,
      "model ranging": 71842,
      "cnn addition": 17985,
      "contextresponse matching": 22375,
      "model selfsupervised": 71962,
      "selfsupervised tasks": 104951,
      "ability select": 1569,
      "focus building": 44741,
      "single response": 109788,
      "response prediction": 101227,
      "task approaches": 117889,
      "approaches overlook": 8272,
      "overlook potential": 83344,
      "potential training": 88591,
      "data beneficial": 25694,
      "features response": 43704,
      "systems supervised": 117174,
      "challenges including": 16167,
      "inconsistency address": 53438,
      "learning contextresponse": 62458,
      "utterance restoration": 130614,
      "multitask manner": 75874,
      "means auxiliary": 67757,
      "better local": 13622,
      "benchmarks indicate": 12908,
      "proposed auxiliary": 93234,
      "improvement multiturn": 52730,
      "analysis representation": 5805,
      "technology assisted": 120036,
      "application text": 6890,
      "common type": 18942,
      "language ambiguity": 58826,
      "document difficult": 32990,
      "addressed using": 3791,
      "using common": 129453,
      "approach bagofwords": 7376,
      "bagofwords bow": 11371,
      "bow model": 14530,
      "situation paper": 109890,
      "provide semantic": 93914,
      "information addresses": 54368,
      "addresses issues": 3797,
      "development dataset": 30379,
      "dataset deep": 26847,
      "learning baseline": 62400,
      "entity recognizer": 38086,
      "bhojpuri maithili": 13780,
      "nouns named": 80443,
      "disease entities": 32336,
      "entities ner": 37827,
      "overcoming problem": 83308,
      "entities separately": 37867,
      "developed translate": 30315,
      "entity labels": 37949,
      "annotation labels": 6328,
      "tagset used": 117492,
      "datasets report": 27670,
      "ranking training": 95692,
      "training largescale": 123680,
      "largescale human": 61443,
      "human feedback": 50861,
      "existing opendomain": 40249,
      "generally trained": 46878,
      "responses human": 101275,
      "models increasingly": 73391,
      "capable producing": 15214,
      "order produce": 82387,
      "number replies": 80954,
      "prediction alleviate": 89028,
      "gpt2 based": 48559,
      "baselines particularly": 12440,
      "perplexity baseline": 87063,
      "crowdsourced human": 25111,
      "ranking method": 95673,
      "correlates better": 24214,
      "training dialogue": 123585,
      "responses dialogue": 101263,
      "capture onetomany": 15353,
      "objective dialogue": 81070,
      "divergence kld": 32783,
      "prevents model": 90374,
      "powerful pretrained": 88675,
      "data provides": 26302,
      "variational models": 130930,
      "results automated": 101528,
      "evaluation human": 39232,
      "baselines release": 12452,
      "detection languages": 29983,
      "languages task": 60912,
      "central information": 15901,
      "architecture effectively": 8648,
      "effectively model": 34831,
      "level training": 63512,
      "significantly boosts": 108895,
      "performance ace": 86111,
      "ace 2005": 2397,
      "average gain": 11193,
      "absolute f1": 1726,
      "principled method": 90682,
      "parsing high": 85122,
      "type models": 126213,
      "parsing speed": 85241,
      "maintaining improving": 66520,
      "outputs input": 83171,
      "differ substantially": 30933,
      "propose union": 93131,
      "built bert": 14915,
      "humanwritten stories": 51132,
      "stories propose": 113400,
      "approach constructing": 7454,
      "commonly observed": 18961,
      "existing nlg": 40236,
      "experiments story": 41155,
      "generated stories": 47129,
      "better human": 13600,
      "stateoftheart metrics": 112718,
      "metrics knowledge": 69976,
      "approaches proven": 8298,
      "effective significantly": 34745,
      "solutions nlp": 110579,
      "difficult challenge": 31612,
      "challenge rarely": 16093,
      "rarely seen": 95762,
      "common nouns": 18904,
      "known named": 58313,
      "work successfully": 134829,
      "successfully exploited": 115186,
      "nmt work": 80010,
      "works studied": 134980,
      "respective tasks": 101116,
      "use kgs": 128102,
      "entities consequently": 37758,
      "substantial attention": 114852,
      "traditional pipeline": 122859,
      "focused neural": 44868,
      "convert set": 23097,
      "triples text": 125820,
      "model verbalizes": 72305,
      "decoder approach": 27993,
      "fact knowledge": 42827,
      "generate multilingual": 46970,
      "settings standard": 107838,
      "approaches english": 8145,
      "consistent results": 21399,
      "multilingual scenario": 75355,
      "reducing latency": 97424,
      "networks great": 77616,
      "progress recently": 92181,
      "hand previous": 49323,
      "make heavy": 66682,
      "transformer fully": 124326,
      "inference algorithm": 54111,
      "ende enfr": 36844,
      "task fully": 118227,
      "transformer achieves": 124288,
      "inputs question": 55492,
      "answering recent": 6685,
      "context question": 22240,
      "competing hypotheses": 19617,
      "calibration neural": 15046,
      "dataset bias": 26770,
      "training seed": 123833,
      "seed model": 104504,
      "pretraining training": 90344,
      "demonstrate remarkable": 28859,
      "achieving significantly": 2975,
      "short queries": 108230,
      "adversarial robustness": 4160,
      "robustness results": 103128,
      "generalization capacity": 46771,
      "capacity models": 15234,
      "models hope": 73339,
      "human interpretable": 50870,
      "answering problem": 6663,
      "openly released": 82000,
      "discourse discussion": 32054,
      "specifically aim": 111525,
      "corpus minimal": 23882,
      "wikipedia entries": 132655,
      "relevant datasets": 98537,
      "domainspecific automatic": 33889,
      "automatic term": 10690,
      "problem problem": 91173,
      "bertbased multitask": 13277,
      "nlp present": 79667,
      "processing chinese": 91637,
      "segmentation cws": 104581,
      "ner dependency": 77038,
      "bert uses": 13247,
      "layers bert": 61763,
      "bert provide": 13213,
      "evaluated 13": 38948,
      "near stateoftheart": 76716,
      "sota performance": 110693,
      "ner achieving": 77018,
      "achieving sota": 2978,
      "performance cws": 86270,
      "strong performing": 113699,
      "performing better": 86945,
      "segmentation tools": 104644,
      "corpus better": 23683,
      "allow users": 5098,
      "convenient use": 22870,
      "released github": 98486,
      "understanding propose": 126929,
      "consists blocks": 21472,
      "sentence generating": 105887,
      "encoding decoding": 36686,
      "transformers work": 124471,
      "generates sentence": 47171,
      "transformers use": 124466,
      "attention transformers": 10026,
      "measure compare": 67784,
      "space set": 111060,
      "report experimental": 98996,
      "increase language": 53602,
      "particular english": 85410,
      "english multilingual": 37213,
      "models compromise": 72946,
      "compromise performance": 20338,
      "individual languages": 53916,
      "corpus discuss": 23761,
      "discuss corpus": 32243,
      "datasets open": 27604,
      "contains information": 21824,
      "information obtain": 54810,
      "corpus finetune": 23804,
      "examples fully": 39838,
      "inference dataset": 54132,
      "dataset natural": 27049,
      "fundamental aspects": 46084,
      "great effort": 49002,
      "middle east": 70069,
      "language indexed": 59166,
      "format useful": 45210,
      "samples generated": 103585,
      "dataset carefully": 26778,
      "multistep process": 75803,
      "results traditional": 102277,
      "provide solid": 93921,
      "research best": 100430,
      "obtained test": 81421,
      "room improving": 103268,
      "methods useful": 69833,
      "realworld nlp": 96171,
      "applications different": 6917,
      "languages investigate": 60650,
      "extent models": 41978,
      "known dataset": 58296,
      "dataset biases": 26771,
      "set easy": 107427,
      "biased models": 13858,
      "framework training": 45719,
      "leverages recent": 63657,
      "resulting highly": 101447,
      "direct assessments": 31784,
      "translation edit": 124773,
      "edit rate": 34536,
      "metrics models": 69989,
      "performance wmt": 86855,
      "highperforming systems": 50363,
      "small network": 110174,
      "generalize way": 46824,
      "network recent": 77401,
      "generator experiments": 47770,
      "dialogue augmentation": 30641,
      "using unpaired": 130342,
      "collecting largescale": 18450,
      "data usually": 26612,
      "training opendomain": 123763,
      "data specifically": 26491,
      "distillation process": 32520,
      "data ranking": 26315,
      "ranking module": 95677,
      "module employed": 74488,
      "process employed": 91469,
      "highquality paired": 50404,
      "paired data": 83472,
      "data augmented": 25660,
      "models affected": 72732,
      "affected noise": 4230,
      "data automatic": 25662,
      "contents proposed": 21991,
      "distillation improve": 32507,
      "network fewshot": 77256,
      "rapid domain": 95718,
      "propose vector": 93156,
      "target label": 117627,
      "projection based": 92231,
      "significantly surpass": 109053,
      "ner method": 77051,
      "score respectively": 104117,
      "respectively code": 101130,
      "using disentangled": 129587,
      "networks map": 77656,
      "input single": 55437,
      "small perturbations": 110185,
      "mapping single": 67146,
      "representation proposed": 99386,
      "strategy maps": 113528,
      "representations mapped": 99748,
      "mapped different": 67122,
      "framework variational": 45738,
      "information bottleneck": 54401,
      "identification opendomain": 51413,
      "maintaining consistent": 66512,
      "naturally converse": 76641,
      "humans existing": 51079,
      "mainly explored": 66485,
      "create largescale": 24620,
      "largescale humanannotated": 61445,
      "humanannotated dataset": 50997,
      "baselines evaluations": 12392,
      "evaluations downstream": 39460,
      "improving dialogue": 53088,
      "treebank universal": 125648,
      "method converting": 68744,
      "experiments manual": 40995,
      "manual linguistic": 67009,
      "vocabulary data": 131872,
      "data brings": 25710,
      "daily events": 25508,
      "understanding coherence": 126813,
      "important subtask": 52260,
      "network furthermore": 77260,
      "network enhance": 77244,
      "18 f1": 417,
      "task distributed": 118106,
      "universal dialogue": 127303,
      "taskoriented spoken": 118904,
      "assist human": 9568,
      "belief state": 12689,
      "respond user": 101181,
      "sequential decisionmaking": 107221,
      "markov decision": 67256,
      "decision process": 27940,
      "process pomdp": 91547,
      "policy optimization": 87963,
      "learning drl": 62527,
      "algorithms use": 4886,
      "hierarchical dialogue": 49953,
      "restaurant reservation": 101322,
      "practically important": 88728,
      "problem recently": 91199,
      "multilayer perceptrons": 75182,
      "input predicting": 55396,
      "predicting actions": 88975,
      "approach suffers": 7945,
      "suffers low": 115261,
      "problems utilizing": 91381,
      "efficiency robust": 35044,
      "multidomain endtoend": 75092,
      "neural taskoriented": 78699,
      "dialog recent": 30581,
      "previous strong": 90487,
      "strong models": 113688,
      "fully differentiable": 45954,
      "training framework": 123638,
      "dialog success": 30593,
      "learning dialog": 62498,
      "different dialog": 31094,
      "consequently model": 21181,
      "evaluation provided": 39344,
      "challenge best": 16001,
      "work comprehensive": 134422,
      "learning component": 62447,
      "techniques words": 120015,
      "words represented": 134173,
      "represented vectors": 100038,
      "useful applications": 128859,
      "meaning terms": 67699,
      "vectors present": 131462,
      "vectors based": 131409,
      "analysis assumes": 5504,
      "abstraction text": 1797,
      "contexts occurs": 22412,
      "field note": 43974,
      "partial ordering": 85294,
      "representations framework": 99661,
      "framework context": 45473,
      "framework implementation": 45564,
      "semantics context": 105402,
      "weighted sums": 132358,
      "framework analyse": 45422,
      "representations meaning": 99749,
      "new algebraic": 78766,
      "descriptions link": 29487,
      "datasets leverage": 27547,
      "model individual": 71335,
      "individual instances": 53914,
      "presence ambiguous": 89348,
      "outofdistribution generalization": 82643,
      "data easy": 25875,
      "easy learn": 34481,
      "role model": 103202,
      "model optimization": 71616,
      "optimization finally": 82192,
      "instances model": 55630,
      "model finds": 71187,
      "hard learn": 49466,
      "shift focus": 108176,
      "evaluation does": 39192,
      "translations approach": 125452,
      "candidate translation": 15128,
      "translation largescale": 124894,
      "evaluation references": 39356,
      "wmt19 metrics": 132853,
      "compression sentence": 20322,
      "sentences preserving": 106452,
      "benefit fields": 12969,
      "language education": 58990,
      "methods largely": 69580,
      "suffer poor": 115235,
      "finetuned bert": 44419,
      "model times": 72171,
      "adversarial dataset": 4109,
      "increasing focus": 53672,
      "assign high": 9541,
      "irrelevant responses": 56950,
      "context data": 22047,
      "response multiple": 101222,
      "randomly selected": 95533,
      "selected responses": 104729,
      "metrics introduce": 69975,
      "context ii": 22126,
      "correct references": 24117,
      "drops substantially": 34190,
      "pretraining help": 90263,
      "reddit conversations": 97298,
      "finetuned dataset": 44422,
      "performance random": 86647,
      "examples dataset": 39821,
      "information flows": 54620,
      "consider multipass": 21216,
      "process information": 91511,
      "later layers": 61647,
      "encoder stack": 36576,
      "architecture search": 8738,
      "search best": 104302,
      "study proposed": 114490,
      "base transformer": 11487,
      "enfr datasets": 37010,
      "case optimal": 15604,
      "chinese domain": 16761,
      "poem generation": 87779,
      "studies usually": 114297,
      "difficult obtain": 31649,
      "obtain largescale": 81298,
      "use easily": 128022,
      "easily available": 34438,
      "architecture bert": 8628,
      "bert trained": 13240,
      "models emerged": 73133,
      "finetuning new": 44485,
      "domains tasks": 33871,
      "used compositional": 128440,
      "step propose": 113298,
      "information structured": 55009,
      "does depend": 33341,
      "model conventional": 70915,
      "conventional language": 22880,
      "crossdomain settings": 24903,
      "adaptation approaches": 3209,
      "approaches analysis": 8058,
      "efficiency model": 35032,
      "model accurate": 70530,
      "tasks lexical": 119254,
      "independent model": 53774,
      "model advantage": 70642,
      "capturing shared": 15493,
      "addition knowledge": 3445,
      "distillation method": 32510,
      "model introduced": 71370,
      "model surpass": 72121,
      "surpass singletask": 116095,
      "provide collection": 93777,
      "multihop reasoning": 75142,
      "reasoning commonsense": 96234,
      "tasks suffer": 119535,
      "approaches integrate": 8195,
      "knowledge generative": 57951,
      "argue exploiting": 8919,
      "enables pretrained": 36399,
      "extracted external": 42155,
      "external commonsense": 41994,
      "graph empirically": 48800,
      "knowledge demonstrate": 57855,
      "reasoning module": 96274,
      "reasoning paths": 96290,
      "generation focused": 47410,
      "variation work": 130904,
      "difficulty generating": 31699,
      "text according": 120626,
      "according given": 1978,
      "lack diversity": 58702,
      "issues learning": 57055,
      "latent spaces": 61611,
      "time generating": 122033,
      "generating fluent": 47223,
      "performance assessed": 86152,
      "assessed automatic": 9499,
      "given post": 48084,
      "promising way": 92314,
      "way introduce": 132101,
      "finegrained wordlevel": 44395,
      "specifically firstly": 111550,
      "corresponding target": 24307,
      "target response": 117696,
      "response post": 101226,
      "words introducing": 134015,
      "signal propose": 108691,
      "models developing": 73061,
      "help humans": 49732,
      "developing recent": 30358,
      "actual information": 3150,
      "information adding": 54363,
      "based rnn": 11998,
      "framework lstm": 45604,
      "chatbot focuses": 16664,
      "text key": 121069,
      "nlp providing": 79675,
      "answering text": 6711,
      "summarization information": 115521,
      "scenarios range": 103863,
      "range simple": 95601,
      "pieces text": 87500,
      "text approaches": 120655,
      "recognizing text": 97076,
      "explain semantic": 41260,
      "language toolkit": 60171,
      "datasets significantly": 27712,
      "achieve 95": 2409,
      "10 training": 153,
      "features context": 43423,
      "context utterances": 22318,
      "topic shift": 122575,
      "exploiting local": 41471,
      "local topic": 64939,
      "capture essential": 15299,
      "effectively track": 34851,
      "topic segment": 122571,
      "dual crossattention": 34231,
      "way experimental": 132078,
      "dataset multitask": 27044,
      "main information": 66431,
      "subtasks named": 114971,
      "entities tackle": 37876,
      "issue proposing": 57027,
      "proposing new": 93627,
      "metric takes": 69905,
      "require models": 100181,
      "transfer information": 124062,
      "located different": 64958,
      "parts document": 85571,
      "document different": 32989,
      "mention spans": 68432,
      "spans experiments": 111201,
      "55 f1": 1111,
      "f1 percentage": 42665,
      "incorporating neural": 53549,
      "neural graph": 77925,
      "graph propagation": 48857,
      "stimulate research": 113345,
      "generating stylized": 47266,
      "systems task": 117186,
      "particular style": 85453,
      "especially target": 38517,
      "target style": 117716,
      "texts specifically": 121620,
      "responses coherent": 101258,
      "predict possible": 88915,
      "texts pseudo": 121588,
      "features decoder": 43438,
      "seen success": 104542,
      "success crosslingual": 115065,
      "xlm xlmr": 135198,
      "addition investigate": 3444,
      "investigate information": 56769,
      "layers models": 61791,
      "interesting differences": 56071,
      "differences encoding": 30956,
      "different pretraining": 31353,
      "pretraining strategies": 90328,
      "reasoning long": 96263,
      "problem lies": 91107,
      "learning sentencelevel": 63004,
      "propose multihop": 92795,
      "approach incorporates": 7633,
      "reasoning knowledge": 96258,
      "module responsible": 74509,
      "human writing": 50991,
      "module generates": 74492,
      "sentence unlike": 106117,
      "blackbox models": 14251,
      "provides explanatory": 94028,
      "experiments representative": 41117,
      "representative tasks": 100022,
      "review generation": 102551,
      "generation automatic": 47315,
      "knowledgeenhanced models": 58270,
      "models aspects": 72775,
      "knowledge rulebased": 58161,
      "translate text": 124547,
      "text source": 121314,
      "training corpusbased": 123408,
      "corpusbased machine": 24077,
      "kinds information": 57689,
      "performance systematically": 86780,
      "approaches dealing": 8117,
      "phenomena results": 87245,
      "certain cases": 15932,
      "novel parallel": 80681,
      "parallel interactive": 84669,
      "rnns based": 102961,
      "existing rnnbased": 40273,
      "jointly modeled": 57365,
      "correlation information": 24237,
      "studies attempt": 114183,
      "attempt capture": 9734,
      "model mutual": 71552,
      "selfattentive encoder": 104909,
      "module developed": 74484,
      "datasets snips": 27718,
      "snips atis": 110310,
      "competitive result": 19677,
      "bert method": 13165,
      "stateoftheart comparison": 112613,
      "embeddings named": 35820,
      "binary format": 14130,
      "binary representation": 14136,
      "include word": 53241,
      "formats tools": 45220,
      "dialogue understanding": 30810,
      "understanding requires": 126942,
      "requires context": 100256,
      "tasks approaches": 118934,
      "aspects dialogue": 9379,
      "dialogue emotion": 30669,
      "emotion intent": 36106,
      "act identification": 3064,
      "methods baselines": 69338,
      "baselines specifically": 12468,
      "tasks baselines": 118955,
      "provides insights": 94042,
      "provide support": 93938,
      "work available": 134393,
      "annotators propose": 6497,
      "types bias": 126256,
      "data meet": 26125,
      "incorporating document": 53531,
      "graph graph": 48816,
      "capture relationships": 15368,
      "documents words": 33316,
      "introducing graph": 56635,
      "neighboring nodes": 76993,
      "conducted results": 20944,
      "survey semantic": 116193,
      "compositionality different": 20142,
      "parsers handle": 85026,
      "field semantic": 43979,
      "using grammar": 129715,
      "consider semantic": 21227,
      "different benchmark": 31026,
      "et al2016": 38743,
      "evaluate semantic": 38920,
      "ability answer": 1485,
      "answer complex": 6519,
      "complex questions": 19866,
      "methods multilingual": 69628,
      "study multilingual": 114443,
      "mbert shown": 67568,
      "mbert pretrained": 67566,
      "mbert crosslingual": 67563,
      "crosslingual signal": 25015,
      "analyse different": 5432,
      "setting specifically": 107779,
      "different alignment": 30987,
      "tasks entity": 119099,
      "recognition semantic": 96991,
      "dissimilar languages": 32399,
      "supervision parallel": 115906,
      "make recommendations": 66716,
      "strategies present": 113484,
      "humans make": 51091,
      "design annotation": 29515,
      "lead successful": 61875,
      "recommendations based": 97091,
      "largescale resource": 61505,
      "resource text": 100882,
      "sources wikipedia": 110930,
      "stack exchange": 112119,
      "manual curation": 66986,
      "build general": 14767,
      "purpose text": 94441,
      "evaluate suite": 38930,
      "work benchmark": 134397,
      "embeddings increased": 35757,
      "built large": 14925,
      "pretrained contextual": 90003,
      "generate representations": 47000,
      "improvements afforded": 52803,
      "come high": 18764,
      "cost work": 24377,
      "baseline negligible": 12268,
      "run time": 103465,
      "multiple pretrained": 75643,
      "concatenation technique": 20543,
      "technique works": 119823,
      "works tasks": 134983,
      "analyze aspects": 5951,
      "works provide": 134971,
      "provide open": 93885,
      "source implementations": 110764,
      "tensorflow pytorch": 120197,
      "dictionary task": 30895,
      "task proper": 118572,
      "word description": 132987,
      "paper tried": 84478,
      "incorporate bert": 53452,
      "subword encoding": 115019,
      "make bert": 66624,
      "bert generate": 13133,
      "description propose": 29472,
      "language previous": 59825,
      "embeddings learn": 35780,
      "align embeddings": 4899,
      "subword embedding": 115017,
      "embedding alignment": 35371,
      "achieve remarkable": 2521,
      "input trained": 55458,
      "set dialogue": 107416,
      "creation datasets": 24715,
      "laborintensive timeconsuming": 58667,
      "domain ontologies": 33600,
      "nlg new": 79524,
      "ontology using": 81853,
      "example dataset": 39778,
      "method identifies": 68877,
      "automatically constructs": 10745,
      "input form": 55342,
      "new instances": 78959,
      "performance absolute": 86108,
      "model report": 71903,
      "documents serving": 33290,
      "related current": 97850,
      "propose compare": 92586,
      "aggregate document": 4347,
      "document information": 33024,
      "generation designed": 47363,
      "designed different": 29588,
      "reduce noise": 97342,
      "propose metrics": 92782,
      "metrics evaluating": 69958,
      "cmudog dataset": 17978,
      "approach strong": 7935,
      "sentiment changes": 106690,
      "changes introduced": 16384,
      "observed global": 81223,
      "time analysing": 121982,
      "method holds": 68873,
      "holds potential": 50595,
      "discontinuous constituent": 32029,
      "discontinuous parsing": 32032,
      "labeling shows": 58541,
      "parsing labeling": 85136,
      "fills gap": 44067,
      "models adds": 72724,
      "control text": 22821,
      "generation incorporating": 47439,
      "text generator": 121024,
      "coherent stories": 18325,
      "higher diversity": 50179,
      "roc story": 103140,
      "story dataset": 113413,
      "generate stories": 47020,
      "scaling model": 103775,
      "124 million": 270,
      "syntax representation": 116553,
      "analysis essential": 5601,
      "intelligence systems": 55838,
      "evaluating syntactic": 39105,
      "information included": 54676,
      "architectures mainly": 8822,
      "english monolingual": 37210,
      "systems multilingual": 117007,
      "transfer syntactic": 124189,
      "tasks human": 119161,
      "task high": 118257,
      "having switch": 49572,
      "webbased annotation": 132274,
      "addresses problems": 3804,
      "image data": 51779,
      "data usage": 26598,
      "state machine": 112505,
      "tool makes": 122393,
      "tasks integrate": 119218,
      "integrate machine": 55762,
      "learning human": 62635,
      "learning complex": 62446,
      "identification complex": 51372,
      "different areas": 31007,
      "speakers usually": 111332,
      "usually difficult": 130415,
      "evidence proposed": 39663,
      "learn characteristics": 61999,
      "surpasses stateoftheart": 116112,
      "crosslingual results": 25002,
      "scenario time": 103833,
      "time model": 122049,
      "morphological agreement": 74675,
      "step language": 113276,
      "automating process": 10843,
      "text concise": 120811,
      "focus extracting": 44764,
      "extracting rules": 42235,
      "languages apply": 60405,
      "apply framework": 7176,
      "annotations language": 6440,
      "language framework": 59054,
      "framework extracts": 45543,
      "created large": 24672,
      "data confirm": 25773,
      "rules framework": 103404,
      "framework produces": 45658,
      "accuracy 78": 2072,
      "release interface": 98453,
      "extracted rules": 42177,
      "financial reports": 44258,
      "reports using": 99090,
      "language interactions": 59216,
      "financial services": 44259,
      "services industry": 107329,
      "created based": 24656,
      "virtual assistant": 131741,
      "documents modeling": 33253,
      "modeling natural": 72488,
      "skills using": 109976,
      "continuously learn": 22653,
      "present evidence": 89479,
      "simulated users": 109644,
      "knowledge natural": 58075,
      "need labeled": 76825,
      "tasks contains": 119009,
      "category hierarchies": 15785,
      "baselines bert": 12362,
      "transferring models": 124242,
      "phrases extracted": 87425,
      "bases wordnet": 12509,
      "languages pretraining": 60810,
      "pretraining improves": 90266,
      "multilevel text": 75193,
      "text alignment": 120636,
      "crossdocument attention": 24880,
      "level learn": 63466,
      "texts example": 121511,
      "document levels": 33034,
      "levels propose": 63559,
      "representing documents": 100048,
      "documents crossdocument": 33215,
      "document pairs": 33048,
      "pairs align": 83486,
      "align multiple": 4903,
      "evaluation predicting": 39329,
      "relationships tasks": 98333,
      "structural correspondence": 113762,
      "documents interpretable": 33242,
      "interpretable reasoning": 56250,
      "reasoning paragraph": 96286,
      "paragraph effects": 84563,
      "requires model": 100299,
      "model understand": 72243,
      "cause effect": 15823,
      "apply knowledge": 7178,
      "knowledge novel": 58087,
      "novel situation": 80730,
      "model inspired": 71349,
      "cognitive processes": 18276,
      "processes paper": 91608,
      "propose sequential": 93044,
      "task explicitly": 118182,
      "leads interpretable": 61942,
      "agents performance": 4326,
      "improvement pretrained": 52744,
      "simply concatenate": 109623,
      "turns dialogue": 125992,
      "largely ignore": 61339,
      "algorithm transform": 4810,
      "threads based": 121933,
      "generation syntactic": 47649,
      "controlled text": 22846,
      "recurrent structures": 97262,
      "dependency problem": 29217,
      "transformer better": 124310,
      "trees method": 125699,
      "generation effectiveness": 47382,
      "new attention": 78804,
      "syntax guidance": 116536,
      "paraphrasing task": 84850,
      "best baselines": 13311,
      "adversarially trained": 4184,
      "contextualised embedding": 22529,
      "domain target": 33667,
      "training signal": 123860,
      "tokens paper": 122324,
      "models mlms": 73572,
      "objective leads": 81090,
      "adaptation tasks": 3273,
      "recognition method": 96907,
      "strongly outperforms": 113748,
      "outperforms random": 82969,
      "score improvements": 104086,
      "accuracy parsing": 2230,
      "accuracy use": 2313,
      "gold tags": 48445,
      "tags impact": 117476,
      "meaningful linguistic": 67723,
      "human dialogues": 50793,
      "specific scenario": 111488,
      "conversation scenarios": 22971,
      "future conversations": 46265,
      "extensive evaluations": 41878,
      "evaluations approach": 39452,
      "semantic gaps": 105063,
      "language clear": 58879,
      "efficiently encoded": 35142,
      "ambiguity word": 5287,
      "investigate case": 56729,
      "word entropy": 133266,
      "requires human": 100274,
      "readily applicable": 95964,
      "significant correlations": 108750,
      "selftraining improves": 104957,
      "improves pretraining": 53028,
      "pretraining led": 90277,
      "way leverage": 132104,
      "data semisupervised": 26393,
      "learning obtain": 62859,
      "obtain additional": 81256,
      "data specific": 26490,
      "method computes": 68720,
      "unlabeled sentences": 127408,
      "previous semisupervised": 90456,
      "methods approach": 69317,
      "improvements 26": 52796,
      "use facilitate": 128048,
      "facilitate users": 42798,
      "new public": 79090,
      "dataset supports": 27226,
      "aforementioned problem": 4269,
      "art deep": 9047,
      "ir models": 56930,
      "additionally analyze": 3598,
      "complexity hybrid": 19916,
      "provides best": 94003,
      "guiding attention": 49271,
      "allow efficient": 5080,
      "learning bidirectional": 62410,
      "bidirectional transformers": 13981,
      "motivated recent": 74860,
      "studies demonstrating": 114207,
      "selfattention patterns": 104900,
      "models contain": 72976,
      "propose computationally": 92591,
      "function guide": 46021,
      "guide attention": 49229,
      "patterns method": 85759,
      "objective results": 81114,
      "achieving state": 2980,
      "results lowresource": 101895,
      "modeling performance": 72507,
      "unsupervised constituency": 127613,
      "recent unsupervised": 96559,
      "sentences introduce": 106359,
      "introduce strong": 56544,
      "training existing": 123625,
      "50 examples": 1074,
      "improved simple": 52642,
      "method selftraining": 69118,
      "hyperparameter tuning": 51232,
      "ii use": 51717,
      "examples possible": 39865,
      "possible model": 88413,
      "trained labeled": 123166,
      "text crowdsourced": 120845,
      "crowdsourced dataset": 25108,
      "events occur": 39588,
      "entailment discourse": 37660,
      "relations modeling": 98211,
      "scale labeled": 103728,
      "grounded text": 49113,
      "prior text": 90738,
      "corpus introduce": 23845,
      "standard classification": 112210,
      "generation generative": 47420,
      "generative task": 47760,
      "aimed testing": 4487,
      "ability reason": 1567,
      "generation results": 47604,
      "results finetuning": 101795,
      "cause online": 15826,
      "online public": 81793,
      "increasingly use": 53714,
      "aligning entity": 4936,
      "entity particular": 37985,
      "present text": 89739,
      "volume messages": 131941,
      "ratings based": 95823,
      "low high": 65365,
      "identify potentially": 51537,
      "humans use": 51116,
      "language accomplish": 58806,
      "emotional support": 36124,
      "implicitly understanding": 52036,
      "better grasp": 13597,
      "grasp language": 48987,
      "ability identify": 1530,
      "increase efficiency": 53597,
      "english reddit": 37261,
      "showing pretrained": 108423,
      "identification challenging": 51369,
      "identify directions": 51494,
      "emnlp 2020": 36085,
      "xlmroberta xlmr": 135209,
      "encoders effectively": 36645,
      "effectively learn": 34826,
      "explicit alignment": 41305,
      "objectives based": 81131,
      "previous alignment": 90378,
      "randomly sampled": 95532,
      "opus collection": 82252,
      "results single": 102191,
      "dataset single": 27205,
      "model run": 71945,
      "multiple runs": 75676,
      "tasks extensive": 119128,
      "objective outperforms": 81104,
      "gains using": 46412,
      "using better": 129394,
      "underlying model": 126688,
      "alignment training": 4995,
      "care evaluating": 15504,
      "evaluating methods": 39073,
      "methods suggest": 69782,
      "limitations applying": 64169,
      "need manually": 76834,
      "leveraging labeled": 63684,
      "languages existing": 60564,
      "approaches transferring": 8382,
      "supervision languages": 115892,
      "expensive crosslingual": 40417,
      "crosslingual resources": 25001,
      "classifiers target": 17636,
      "documents work": 33317,
      "clts generates": 17910,
      "supervision target": 115920,
      "translations given": 125468,
      "given limited": 48058,
      "limited translation": 64295,
      "seed words": 104509,
      "based translated": 12133,
      "outperforms teacher": 83035,
      "just 20": 57461,
      "tasks lowresource": 119275,
      "using just": 129773,
      "desired style": 29672,
      "great potentials": 49012,
      "lacking parallel": 58771,
      "explore challenging": 41529,
      "models brought": 72866,
      "breakthrough various": 14584,
      "tasks end": 119094,
      "style classifier": 114570,
      "step order": 113290,
      "studies public": 114268,
      "datasets indicate": 27521,
      "way address": 132055,
      "address different": 3674,
      "dataset building": 26773,
      "building model": 14863,
      "single machine": 109756,
      "handle diverse": 49381,
      "new problems": 79086,
      "taskspecific annotations": 119617,
      "problems current": 91312,
      "rich annotations": 102727,
      "particularly target": 85499,
      "target nlp": 117680,
      "annotations universal": 6476,
      "achieved different": 2617,
      "entailment domains": 37662,
      "effectiveness unified": 34970,
      "span identification": 111124,
      "performance prediction": 86611,
      "short span": 108239,
      "chunking ner": 16964,
      "codeswitching detection": 18234,
      "ask models": 9297,
      "identify classify": 51486,
      "relevant spans": 98586,
      "structure little": 113901,
      "model families": 71172,
      "tasks inform": 119199,
      "largescale experiment": 61437,
      "data building": 25714,
      "insights model": 55540,
      "performance span": 86727,
      "annotation important": 6323,
      "task involved": 118311,
      "task novel": 118464,
      "including popular": 53350,
      "github repo": 47977,
      "models modern": 73579,
      "parameters leads": 84763,
      "response problem": 101228,
      "improve baseline": 52342,
      "model 08": 70493,
      "ability modeling": 1544,
      "modeling lowlevel": 72469,
      "lowlevel lexical": 65478,
      "search answer": 104297,
      "posteriori map": 88490,
      "decoding neural": 28110,
      "attained using": 9725,
      "objective does": 81071,
      "does express": 33349,
      "decoding objective": 28115,
      "objective order": 81103,
      "order gain": 82327,
      "uniform information": 127137,
      "information density": 54470,
      "motivated cognitive": 74844,
      "exact decoding": 39716,
      "produced using": 91981,
      "multiinstance multilabel": 75146,
      "multilabel learning": 75154,
      "aspectcategory sentiment": 9361,
      "analysis acsa": 5485,
      "acsa aims": 3056,
      "aims predict": 4555,
      "predict sentiment": 88932,
      "given aspect": 47986,
      "aspect categories": 9319,
      "sentiment particular": 106766,
      "aspect category": 9321,
      "generate aspect": 46906,
      "representation aspect": 99173,
      "category predict": 15789,
      "methods ignore": 69539,
      "ignore fact": 51666,
      "aggregation sentiments": 4368,
      "words indicating": 133998,
      "indicating aspect": 53869,
      "sentence leads": 105919,
      "learning network": 62833,
      "words instances": 134009,
      "instances words": 55641,
      "sentence aspect": 105775,
      "categories mentioned": 15742,
      "predicts sentiments": 89222,
      "categories finally": 15738,
      "models lot": 73528,
      "demonstrate linguistic": 28770,
      "captured representations": 15437,
      "neuronlevel analysis": 78745,
      "semantics pretrained": 105449,
      "parts network": 85579,
      "focused information": 44861,
      "various architectures": 131042,
      "properties small": 92483,
      "tasks lower": 119274,
      "study reveals": 114506,
      "reveals interesting": 102516,
      "predicting properties": 89002,
      "compared bert": 19337,
      "vary depending": 131242,
      "author genre": 10364,
      "behaviour paper": 12681,
      "used formalise": 128559,
      "embedding syntactic": 35509,
      "distance able": 32403,
      "distance points": 32431,
      "networks tasks": 77792,
      "processing involve": 91686,
      "structured outputs": 114022,
      "outputs sequence": 83180,
      "labeling semantic": 58536,
      "labeling parsing": 58520,
      "problems structured": 91376,
      "usually quite": 130446,
      "labeling including": 58503,
      "terms drawing": 120314,
      "gimpel 2018": 47969,
      "difficulties training": 31690,
      "models empirically": 73139,
      "data conditions": 25770,
      "qa pairs": 94511,
      "pairs representation": 83629,
      "relations propositions": 98242,
      "integral natural": 55740,
      "annotating discourse": 6265,
      "relations typically": 98273,
      "typically requires": 126457,
      "requires expert": 100268,
      "recently different": 96680,
      "annotated discourse": 6186,
      "answers based": 6726,
      "based proposed": 11957,
      "collect novel": 18389,
      "diverse sentences": 32841,
      "produce semantically": 91927,
      "sentences perform": 106444,
      "semantic sentence": 105269,
      "highly correlate": 50306,
      "causes effects": 15837,
      "effects demonstrate": 34980,
      "iterated learning": 57116,
      "language drift": 58988,
      "models interaction": 73418,
      "agents trained": 4329,
      "completing task": 19790,
      "jointly trains": 57400,
      "evaluated human": 38979,
      "introduce supervised": 56546,
      "attention transformer": 10024,
      "crosslingual relation": 24997,
      "progress crosslingual": 92144,
      "extraction use": 42527,
      "languageagnostic sentence": 60354,
      "language applied": 58836,
      "directly connected": 31869,
      "propose utilize": 93149,
      "syntactic distances": 116398,
      "outperforms recently": 82973,
      "margin detailed": 67190,
      "representations facilitate": 99652,
      "facilitate transfer": 42797,
      "simple knowledge": 109450,
      "distillation technique": 32522,
      "intermediate layers": 56135,
      "problem common": 90966,
      "distill knowledge": 32499,
      "compact student": 19113,
      "distillation kd": 32508,
      "cases study": 15668,
      "deep nmt": 28388,
      "novel alternative": 80460,
      "alternative model": 5237,
      "layerlevel supervision": 61757,
      "supervision paper": 115904,
      "settings evaluate": 107799,
      "proposed unified": 93578,
      "compare semantic": 19293,
      "understanding different": 126831,
      "systems survey": 117177,
      "dataset focus": 26939,
      "systems finegrained": 116907,
      "datasets focus": 27487,
      "classification question": 17361,
      "deploying models": 29315,
      "models real": 73861,
      "real systems": 96079,
      "systems highly": 116935,
      "hinton et": 50506,
      "faster inference": 43179,
      "inference existing": 54139,
      "works optimized": 134966,
      "texts consequently": 121480,
      "subpar performance": 114798,
      "pair tasks": 83458,
      "tasks coupled": 119019,
      "qualityspeed tradeoffs": 94834,
      "realworld ecommerce": 96167,
      "quality drop": 94636,
      "analysis open": 5726,
      "neural open": 78621,
      "extraction openie": 42419,
      "encoding partial": 36716,
      "comes significant": 18781,
      "significant computational": 108740,
      "cost hand": 24356,
      "labeling approaches": 58486,
      "paper bridge": 83754,
      "2d grid": 880,
      "task improve": 118278,
      "performance applying": 86144,
      "soft constraints": 110491,
      "incorporates new": 53508,
      "handle complicated": 49378,
      "improvement f1": 52708,
      "adversarial attacks": 4102,
      "implicit bias": 52003,
      "making robust": 66862,
      "preserving utility": 89943,
      "pronouns important": 92356,
      "usually formulate": 130422,
      "approaches shown": 8337,
      "utterance independently": 130601,
      "framework combines": 45456,
      "conversation datasets": 22947,
      "recovery models": 97146,
      "descriptions using": 29500,
      "summarization study": 115571,
      "visual entities": 131793,
      "classes specifically": 17077,
      "freetext descriptions": 45796,
      "learn classify": 62001,
      "focusing learning": 44927,
      "knowledge visual": 58231,
      "parts text": 85592,
      "leverage similarity": 63624,
      "extractive summaries": 42562,
      "summaries focus": 115439,
      "images propose": 51811,
      "critical importance": 24812,
      "texts zeroshot": 121656,
      "high performances": 50099,
      "learn generalizable": 62054,
      "general data": 46649,
      "data patterns": 26229,
      "debiasing methods": 27901,
      "framework improving": 45569,
      "outofdomain datasets": 82651,
      "data framework": 25958,
      "example based": 39775,
      "weights training": 132383,
      "objective model": 81094,
      "examples high": 39843,
      "framework extractive": 45542,
      "extractive question": 42556,
      "answering training": 6716,
      "biases different": 13870,
      "different strengths": 31454,
      "settings model": 107821,
      "domain multiple": 33586,
      "settings compared": 107791,
      "team participated": 119717,
      "method solving": 69159,
      "method focuses": 68845,
      "focuses learning": 44906,
      "level document": 63436,
      "approach including": 7631,
      "including multilingual": 53326,
      "30 languages": 902,
      "accuracy gaps": 2170,
      "systems languages": 116968,
      "pragmatic inference": 88751,
      "problem dialog": 91011,
      "questions humans": 95316,
      "humans interpret": 51083,
      "systems allowing": 116729,
      "release largescale": 98456,
      "largescale english": 61433,
      "answer pairs": 6551,
      "progress task": 92183,
      "responses present": 101289,
      "bertbased neural": 13278,
      "pair transfer": 83462,
      "performance sufficient": 86773,
      "models reach": 73857,
      "classes learning": 17065,
      "answering stateoftheart": 6703,
      "qa relies": 94517,
      "relies large": 98653,
      "data labeling": 26065,
      "annotating qa": 6271,
      "reduces human": 97385,
      "leverages underlying": 63662,
      "suggest potential": 115326,
      "improve future": 52386,
      "minimal annotation": 70155,
      "compare framework": 19249,
      "approach reduce": 7847,
      "cost improving": 24360,
      "tweets specific": 126060,
      "specific text": 111501,
      "compared general": 19374,
      "tweets popular": 126048,
      "english difficult": 37114,
      "huge annotated": 50716,
      "recent rise": 96511,
      "processing allows": 91613,
      "allows achieve": 5125,
      "model pretrain": 71771,
      "experiments french": 40951,
      "results transformers": 102285,
      "suffer issues": 115226,
      "directly use": 31915,
      "trained decode": 123110,
      "updates based": 127805,
      "wide class": 132491,
      "requires minimal": 100298,
      "translation benchmark": 124670,
      "summarization benchmarks": 115485,
      "headlines using": 49601,
      "using mutual": 129926,
      "attentive semantic": 10127,
      "texts useful": 121644,
      "misleading news": 70301,
      "works proposed": 134970,
      "proposed machine": 93327,
      "based solutions": 12059,
      "headline body": 49593,
      "body text": 14399,
      "fail perform": 42969,
      "news headline": 79340,
      "hand recent": 49324,
      "guided attention": 49254,
      "contextual representation": 22490,
      "representation news": 99357,
      "body result": 14398,
      "generated headlines": 47083,
      "words original": 134096,
      "pretraining sequence": 90322,
      "sequence encoder": 106937,
      "sequence embeddings": 106935,
      "embeddings largescale": 35776,
      "answering instead": 6638,
      "sentences train": 106524,
      "set short": 107579,
      "select useful": 104718,
      "predicting masked": 88993,
      "masked words": 67310,
      "entailment tasks": 37678,
      "demonstrate pretrained": 28824,
      "continuous sentence": 22632,
      "baselines proposed": 12446,
      "hotpotqa fullwiki": 50673,
      "fullwiki setting": 45939,
      "setting improving": 107755,
      "detecting finegrained": 29836,
      "supervision learning": 115893,
      "conveyed different": 23122,
      "crosslingual nlp": 24986,
      "nlp multilingual": 79649,
      "corpora analysis": 23407,
      "problem annotation": 90932,
      "annotation expensive": 6305,
      "expensive hard": 40421,
      "scale work": 103763,
      "improves prediction": 53026,
      "introduce training": 56555,
      "rank synthetic": 95629,
      "granularity evaluate": 48746,
      "annotated semantic": 6227,
      "helps detect": 49814,
      "finegrained sentencelevel": 44380,
      "tokenlevel predictions": 122291,
      "coarse finegrained": 18051,
      "critical roles": 24832,
      "function maps": 46032,
      "weighted attention": 132336,
      "potential enhance": 88551,
      "enhance output": 37482,
      "ignored existing": 51674,
      "existing attention": 40068,
      "optimal paper": 82165,
      "mechanism incorporating": 67996,
      "interactions propose": 55997,
      "queryaware attention": 94996,
      "attention values": 10032,
      "lample et": 58796,
      "enables improve": 36388,
      "accuracy increasing": 2186,
      "overhead empirical": 83325,
      "application limited": 6857,
      "incorporate large": 53478,
      "new memory": 79005,
      "memory slots": 68382,
      "effective solutions": 34748,
      "initialization model": 55237,
      "pretrained weights": 90211,
      "ability developing": 1506,
      "explicitly considering": 41365,
      "finegrained topic": 44388,
      "transition dynamics": 124488,
      "consider graph": 21203,
      "logic natural": 64992,
      "representations finegrained": 99657,
      "evaluate dialogue": 38818,
      "graphs enhanced": 48949,
      "commonsense graph": 18986,
      "diverse dialogue": 32807,
      "pearson spearman": 85837,
      "benchmark facilitate": 12823,
      "research automatic": 100425,
      "data textual": 26561,
      "annotating large": 6268,
      "model iteratively": 71377,
      "labels order": 58625,
      "data simulation": 26455,
      "experiment performed": 40482,
      "shown reduce": 108519,
      "number annotations": 80837,
      "inference improving": 54152,
      "labels based": 58582,
      "better address": 13504,
      "need retraining": 76855,
      "retraining experiments": 102373,
      "method nlu": 68990,
      "potential practical": 88578,
      "practical usage": 88719,
      "search nas": 104330,
      "individual neural": 53923,
      "detection semantic": 30044,
      "similarity use": 109326,
      "models esim": 73164,
      "sets embeddings": 107666,
      "contrast prior": 22707,
      "results mixed": 101931,
      "lstms perform": 65715,
      "role style": 103220,
      "parsing speech": 85240,
      "treebanked text": 125656,
      "language mismatch": 59282,
      "prosodic cues": 93640,
      "understood paper": 127026,
      "speech context": 111663,
      "using written": 130392,
      "text improve": 121045,
      "speech prosody": 111733,
      "parsers analysis": 85014,
      "childrens speech": 16730,
      "shown studies": 108532,
      "speech especially": 111684,
      "reveals significant": 102521,
      "switchboard callhome": 116279,
      "switchboard speech": 116290,
      "score 10": 104046,
      "dataset callhome": 26776,
      "dramatically improved": 34092,
      "describes contribution": 29388,
      "wmt 2020": 132778,
      "task main": 118369,
      "benchmark automatic": 12732,
      "translation make": 124926,
      "based transfer": 12125,
      "learning extend": 62570,
      "english evaluate": 37129,
      "finetuning data": 44452,
      "labelled examples": 58561,
      "examples additionally": 39807,
      "focus english": 44757,
      "use alternative": 127892,
      "task indicating": 118291,
      "2020 edition": 725,
      "processing named": 91719,
      "recast sequence": 96350,
      "tokens model": 122318,
      "assigns labels": 9566,
      "group tokens": 49144,
      "remedy introduce": 98858,
      "introduce opensource": 56522,
      "spans represented": 111209,
      "adaptive training": 3338,
      "phenomenon natural": 87255,
      "different tokens": 31501,
      "appear different": 6795,
      "different frequencies": 31157,
      "vanilla nmt": 130817,
      "model usually": 72286,
      "usually adopts": 130406,
      "token distribution": 122250,
      "tokens carry": 122301,
      "token frequencies": 122252,
      "assign appropriate": 9539,
      "words assigned": 133827,
      "ende translation": 36846,
      "especially sentences": 38506,
      "contain lowfrequency": 21745,
      "analyses method": 5459,
      "improve lexical": 52406,
      "diversity translation": 32898,
      "translation case": 124686,
      "nlp focused": 79617,
      "challenges understanding": 16218,
      "mitigating bias": 70382,
      "language building": 58863,
      "better design": 13562,
      "little discussion": 64803,
      "nlp particular": 79659,
      "index measures": 53805,
      "extend original": 41791,
      "measures external": 67863,
      "language switching": 60137,
      "classifiers results": 17631,
      "low correlation": 65352,
      "method set": 69128,
      "humanhuman dialog": 51029,
      "agent training": 4314,
      "agents given": 4322,
      "terms utterances": 120403,
      "fashion demonstrate": 43115,
      "users specify": 129175,
      "roles people": 103233,
      "artificial agents": 9244,
      "unable understand": 126571,
      "work exploit": 134505,
      "learning agents": 62316,
      "agent learn": 4305,
      "learn understand": 62175,
      "understand generate": 126747,
      "existing language": 40152,
      "roles training": 103236,
      "lexicon using": 63912,
      "assessment quality": 9525,
      "errors source": 38407,
      "candidate sentence": 15121,
      "pairwise similarity": 83684,
      "similarity response": 109298,
      "score proposed": 104112,
      "zeroshot models": 135514,
      "qe model": 94532,
      "model explicit": 71139,
      "model improvement": 71313,
      "documents existing": 33226,
      "error analyses": 38279,
      "directions task": 31854,
      "papers papers": 84507,
      "accurate use": 2372,
      "use realworld": 128224,
      "perform indepth": 86010,
      "major causes": 66542,
      "errors based": 38358,
      "evaluate standard": 38928,
      "alternative evaluation": 5228,
      "outperforms leading": 82914,
      "points respectively": 87872,
      "note performance": 80386,
      "structure features": 113866,
      "discuss remaining": 32282,
      "remaining challenges": 98779,
      "challenges documentlevel": 16150,
      "ideas improvements": 51350,
      "potential issues": 88568,
      "text planning": 121180,
      "task despite": 118072,
      "success contextualized": 115064,
      "textual coherence": 121678,
      "say making": 103672,
      "learn highlevel": 62064,
      "paragraph contains": 84562,
      "contains various": 21842,
      "motivated work": 74866,
      "masked sentences": 67298,
      "sentences paragraph": 106431,
      "task suffers": 118759,
      "topical content": 122592,
      "realization using": 96118,
      "using predicted": 130039,
      "models paragraph": 73712,
      "content keywords": 21895,
      "quality increases": 94686,
      "information ancient": 54376,
      "critical step": 24833,
      "approach studying": 7938,
      "turns conversations": 125991,
      "methods providing": 69700,
      "tool use": 122404,
      "analysis type": 5914,
      "taking place": 117550,
      "anger fear": 6109,
      "fear sadness": 43232,
      "effective healthcare": 34687,
      "dynamics language": 34352,
      "language culture": 58930,
      "yelp restaurant": 135320,
      "systems successfully": 117170,
      "applied documents": 7064,
      "coverage recall": 24522,
      "recall considering": 96335,
      "spanish chinese": 111149,
      "expensive require": 40430,
      "documents new": 33259,
      "train multilingual": 122968,
      "using annotations": 129338,
      "reviews recent": 102625,
      "recent zeroshot": 96646,
      "zeroshot approaches": 135490,
      "aspects sentiment": 9408,
      "approaches effective": 8136,
      "effective capturing": 34640,
      "capturing nuances": 15483,
      "annotations create": 6419,
      "training documents": 123589,
      "translation train": 125387,
      "language furthermore": 59057,
      "languages leads": 60682,
      "classifiers identify": 17612,
      "reviews yelp": 102642,
      "accurate reliable": 2356,
      "critical making": 24816,
      "work existing": 134501,
      "measurement experiments": 67844,
      "models question": 73848,
      "key takeaway": 57602,
      "sentencelevel direct": 106155,
      "assessment shared": 9528,
      "baseline used": 12331,
      "fine tune": 44324,
      "framework performing": 45652,
      "performing ensemble": 86951,
      "ensemble data": 37588,
      "winning solution": 132722,
      "wmt20 biomedical": 132865,
      "task 2020": 117824,
      "task meaning": 118381,
      "data distinct": 25858,
      "distinct style": 32544,
      "style vocabulary": 114610,
      "vocabulary models": 131884,
      "bias effects": 13801,
      "translations result": 125489,
      "using robust": 130124,
      "contrast approach": 22688,
      "results directions": 101730,
      "models ensembling": 73161,
      "lms shown": 64889,
      "significant linguistic": 108822,
      "knowledge form": 57939,
      "general commonsense": 46643,
      "factors affecting": 42881,
      "affecting performance": 4232,
      "performance simple": 86716,
      "snli mnli": 110315,
      "advances modeling": 3997,
      "lack reliable": 58740,
      "largescale nli": 61488,
      "nli dataset": 79539,
      "unlike recent": 127450,
      "annotation instead": 6325,
      "protocol used": 93676,
      "new strategies": 79186,
      "chinese best": 16740,
      "performance 12": 86102,
      "12 absolute": 246,
      "gap making": 46465,
      "corpus nonenglish": 23905,
      "using adaptive": 129320,
      "scarcity parallel": 103812,
      "significant hurdle": 108763,
      "highquality neural": 50400,
      "approach transfer": 7980,
      "offers best": 81583,
      "furthermore different": 46163,
      "models complementary": 72941,
      "semantic andor": 104984,
      "andor syntactic": 6099,
      "distillation propose": 32521,
      "propose distill": 92629,
      "knowledge ensemble": 57901,
      "ensemble teacher": 37620,
      "models single": 74057,
      "models varies": 74319,
      "distillation approach": 32502,
      "approach dynamically": 7510,
      "process experiments": 91481,
      "approach achieving": 7327,
      "compositional generalization": 20114,
      "interpretation natural": 56269,
      "compositionality meaning": 20144,
      "meaning complex": 67622,
      "abilities language": 1478,
      "architectures introduce": 8810,
      "fragment english": 45372,
      "english evaluation": 37133,
      "familiar words": 43050,
      "structures experiments": 114075,
      "transformers lstms": 124453,
      "indistribution accuracy": 53896,
      "generalization accuracy": 46766,
      "measure progress": 67813,
      "evaluating factuality": 39054,
      "progress text": 92184,
      "tendency produce": 120168,
      "produce text": 91942,
      "factually inconsistent": 42943,
      "input recent": 55411,
      "work studied": 134820,
      "identify factual": 51501,
      "factual errors": 42932,
      "sentencelevel entailment": 106157,
      "trained solve": 123287,
      "different problem": 31360,
      "dependency arcs": 29133,
      "individual dependency": 53905,
      "generated output": 47103,
      "supported input": 116022,
      "obtain propose": 81311,
      "automatically create": 10748,
      "create data": 24608,
      "paraphrase corpora": 84805,
      "data identify": 26011,
      "factual inconsistencies": 42933,
      "summarization better": 115486,
      "based question": 11964,
      "generation additionally": 47294,
      "datadriven techniques": 26681,
      "processing mainly": 91708,
      "select efficient": 104697,
      "achieve greater": 2471,
      "fewer labeled": 43888,
      "instances crosslingual": 55624,
      "definitions english": 28537,
      "based publicly": 11958,
      "directly applied": 31862,
      "languages trained": 60925,
      "generation experiments": 47403,
      "manual analyses": 66973,
      "lexical complexity": 63742,
      "generated reference": 47113,
      "project report": 92225,
      "online corpus": 81756,
      "result work": 101421,
      "theoretical underpinnings": 121801,
      "discuss key": 32261,
      "contribution make": 22771,
      "different individuals": 31176,
      "user groups": 128994,
      "refer entity": 97496,
      "documents present": 33269,
      "labeling english": 58498,
      "english train": 37310,
      "model coherence": 70840,
      "shallow semantic": 107930,
      "using resulting": 130114,
      "semantic analyzer": 104983,
      "learning encourage": 62544,
      "global coherence": 48227,
      "coherence document": 18296,
      "document semantic": 33073,
      "encoders different": 36643,
      "different expressivity": 31141,
      "approach semantics": 7876,
      "exhibited excellent": 40008,
      "terms used": 120401,
      "attributes context": 10193,
      "attention domainspecific": 9825,
      "terms relevance": 120374,
      "dialogues domain": 30828,
      "likelihood objective": 64122,
      "context analysis": 22005,
      "answering deep": 6621,
      "cases example": 15641,
      "multiplechoice qa": 75754,
      "input question": 55410,
      "predict correct": 88879,
      "using causal": 129432,
      "specifically leverage": 111565,
      "multibranch architecture": 75038,
      "correlations training": 24263,
      "inference methods": 54166,
      "qa benchmarks": 94497,
      "annotation corrections": 6292,
      "number publicly": 80946,
      "state annotations": 112422,
      "leading suboptimal": 61912,
      "annotations dialogue": 6423,
      "dataset ensure": 26894,
      "ensure consistency": 37637,
      "consistency dialogue": 21357,
      "versions multiwoz": 131623,
      "methods recent": 69707,
      "explanation model": 41284,
      "goal used": 48393,
      "input tokens": 55456,
      "tokens relevant": 122329,
      "goal user": 48394,
      "paper hope": 83972,
      "recent focus": 96463,
      "focus attention": 44736,
      "meaning variation": 67704,
      "allows encode": 5150,
      "model meaning": 71507,
      "discriminate word": 32194,
      "datasets paired": 27608,
      "method best": 68676,
      "bases kb": 12502,
      "requires properly": 100312,
      "represent entity": 99107,
      "issue explore": 56997,
      "achieve propose": 2518,
      "enhanced transformer": 37523,
      "transformer framework": 124325,
      "novel memory": 80627,
      "entity focus": 37941,
      "focus relevant": 44810,
      "framework achieve": 45406,
      "kurdish language": 58363,
      "translating texts": 124609,
      "major tasks": 66587,
      "indoeuropean language": 53953,
      "paper addressing": 83718,
      "addressing main": 3813,
      "main issues": 66434,
      "sorani dialect": 110679,
      "translation discuss": 124759,
      "discuss major": 32265,
      "challenges kurdish": 16171,
      "fundamental text": 46126,
      "performance words": 86862,
      "metrics language": 69977,
      "generation rely": 47593,
      "compared reference": 19433,
      "models behavior": 72825,
      "properties words": 92494,
      "practice furthermore": 88733,
      "accuracy including": 2183,
      "language majority": 59268,
      "performance vary": 86848,
      "words practice": 134130,
      "framework simple": 45689,
      "picture language": 87489,
      "differences performance": 30967,
      "attentive graph": 10110,
      "regarded important": 97653,
      "effective modeling": 34712,
      "powerful encoders": 88665,
      "enhance neural": 37477,
      "neural ccg": 77855,
      "leveraging contextual": 63667,
      "ngrams extracted": 79463,
      "extracted lexicon": 42161,
      "apply attention": 7161,
      "graph different": 48793,
      "model facilitate": 71167,
      "studies terms": 114290,
      "analyses illustrate": 5455,
      "illustrate effectiveness": 51739,
      "component approach": 19968,
      "based claim": 11572,
      "generation argument": 47313,
      "argument generation": 8944,
      "considering potential": 21328,
      "impact social": 51890,
      "dissemination information": 32394,
      "pipeline based": 87530,
      "types claims": 126259,
      "using array": 129345,
      "information networks": 54798,
      "unstructured realworld": 127580,
      "attention natural": 9940,
      "performance mitigate": 86533,
      "issue define": 56993,
      "set composed": 107396,
      "entities incorporate": 37796,
      "missing information": 70328,
      "network hin": 77274,
      "method verify": 69221,
      "method conduct": 68723,
      "datasets compare": 27360,
      "methods entity": 69477,
      "trigger extraction": 125783,
      "translation challenge": 124689,
      "resource multilingual": 100866,
      "multilingual mt": 75295,
      "describes development": 29397,
      "translation provides": 125159,
      "thousands language": 121923,
      "pairs covering": 83506,
      "languages tools": 60922,
      "models collection": 72920,
      "models broader": 72864,
      "broader coverage": 14700,
      "comprehensive collection": 20251,
      "collection diverse": 18471,
      "language script": 60068,
      "data splits": 26497,
      "benchmarks data": 12896,
      "models individual": 73396,
      "individual language": 53915,
      "language groups": 59114,
      "use transfer": 128339,
      "learning tl": 63107,
      "tl techniques": 122228,
      "effective lowresource": 34700,
      "faces challenges": 42752,
      "effective alternative": 34625,
      "paper contributions": 83789,
      "second benchmark": 104394,
      "speech classification": 111654,
      "task performs": 118531,
      "sexual harassment": 107894,
      "providing support": 94141,
      "users examine": 129112,
      "gender racialethnic": 46609,
      "white women": 132476,
      "black women": 14249,
      "women men": 132894,
      "public figures": 94259,
      "work digital": 134477,
      "article discusses": 9125,
      "multilingual benchmark": 75200,
      "correctly model": 24179,
      "distinct meanings": 32539,
      "word crucial": 132980,
      "effectiveness semantic": 34949,
      "techniques existing": 119883,
      "usually wordnet": 130467,
      "wordincontext dataset": 133718,
      "problem limited": 91108,
      "languages varied": 60961,
      "set performance": 107527,
      "baselines recent": 12450,
      "recent contextualized": 96440,
      "instances available": 55619,
      "compositional structures": 20135,
      "larger numbers": 61374,
      "compositional reasoning": 20127,
      "reasoning key": 96257,
      "key natural": 57588,
      "tasks tested": 119553,
      "reasoning numerical": 96280,
      "language number": 59722,
      "comparison classification": 19533,
      "evidence information": 39650,
      "encoded pretrained": 36480,
      "models embeddings": 73132,
      "grammaticality judgments": 48728,
      "comparisons analyze": 19589,
      "possible reasons": 88421,
      "opensource resources": 82029,
      "dialects overview": 30539,
      "overview program": 83370,
      "designed address": 29577,
      "need developing": 76798,
      "available speech": 11109,
      "speech resources": 111796,
      "underrepresented languages": 126724,
      "datasets building": 27339,
      "texttospeech automatic": 121658,
      "recognition applications": 96807,
      "used developing": 128489,
      "language communities": 58894,
      "using latin": 129802,
      "latin script": 61665,
      "additional challenge": 3496,
      "work addressed": 134358,
      "resources training": 101059,
      "tunisian dialect": 125946,
      "used social": 128767,
      "media previous": 68155,
      "techniques combined": 119852,
      "features recently": 43689,
      "recently deep": 96675,
      "especially english": 38456,
      "word2vec bert": 133669,
      "networks bidirectional": 77523,
      "memory using": 68395,
      "features experimental": 43496,
      "datasets showed": 27706,
      "performances languages": 86882,
      "seen increase": 104532,
      "language japanese": 59227,
      "english documentation": 37118,
      "transformer transformer": 124385,
      "transformer achieved": 124286,
      "composing various": 20085,
      "various advanced": 131034,
      "used methods": 128631,
      "contexts paper": 22414,
      "distances tokens": 32440,
      "selfattention weights": 104903,
      "relevance attention": 98503,
      "query key": 94969,
      "different selfattention": 31406,
      "selfattention heads": 104878,
      "preferences long": 89245,
      "distance information": 32419,
      "outperform vanilla": 82770,
      "neural narrative": 78221,
      "narrative generation": 76172,
      "generation narrative": 47496,
      "generates story": 47178,
      "given prompt": 48094,
      "generation chatbots": 47333,
      "despite similarity": 29729,
      "gap applying": 46441,
      "applying evaluating": 7244,
      "particular employ": 85408,
      "analyzing results": 6048,
      "metric new": 69892,
      "theory underlying": 121829,
      "core component": 23314,
      "component generation": 19981,
      "generation summaries": 47647,
      "consistent semantics": 21400,
      "critical recent": 24826,
      "language generative": 59104,
      "typically represent": 126454,
      "capture factual": 15301,
      "according statistical": 2004,
      "propose graphbased": 92707,
      "entity graph": 37943,
      "representations graph": 99671,
      "representations composed": 99554,
      "representation making": 99322,
      "modeled results": 72362,
      "base models": 11464,
      "model analysis": 70661,
      "analysis indicates": 5655,
      "multiple input": 75578,
      "input sources": 55441,
      "achieved outstanding": 2662,
      "tasks investigating": 119226,
      "sources work": 110933,
      "adapted pretrained": 3287,
      "fuse multiple": 46225,
      "corresponding different": 24282,
      "sources experimental": 110888,
      "fusion methods": 46242,
      "simple fusion": 109437,
      "induction propose": 54030,
      "behavioral data": 12672,
      "induced grammar": 54007,
      "ii complex": 51699,
      "iii set": 51720,
      "combine types": 18644,
      "deploy framework": 29301,
      "acceptability judgment": 1890,
      "dataset focused": 26940,
      "syntactic distribution": 116399,
      "sentence scoring": 106056,
      "scoring tasks": 104234,
      "crossencoders perform": 24910,
      "input pair": 55386,
      "pair biencoders": 83421,
      "dense vector": 29019,
      "require substantial": 100201,
      "finetuning target": 44519,
      "strategy called": 113507,
      "process selecting": 91566,
      "crucial success": 25174,
      "points indomain": 87861,
      "37 points": 973,
      "learning negation": 62831,
      "classification majority": 17264,
      "finding better": 44267,
      "better methods": 13627,
      "paper models": 84048,
      "datasets evaluate": 27455,
      "performance challenge": 86198,
      "improvement release": 52756,
      "release datasets": 98451,
      "datasets source": 27724,
      "language online": 59725,
      "student reviews": 114145,
      "make reference": 66717,
      "focus used": 44838,
      "used design": 128484,
      "design feature": 29539,
      "metoo timesup": 69865,
      "timesup movements": 122205,
      "ensemble classifiers": 37582,
      "use resulting": 128248,
      "augmentation natural": 10288,
      "model generalization": 71226,
      "challenging paper": 16288,
      "augmentation framework": 10276,
      "framework dubbed": 45507,
      "integrating multiple": 55798,
      "contrastive regularization": 22738,
      "introduced capture": 56575,
      "memory bank": 68290,
      "framework apply": 45426,
      "models wide": 74337,
      "model importantly": 71305,
      "including lowresource": 53318,
      "performance highlighting": 86429,
      "wide applicability": 132487,
      "framework incorporate": 45571,
      "preserved output": 89930,
      "text words": 121412,
      "tend convey": 120145,
      "semantic core": 105022,
      "approach named": 7723,
      "leverages power": 63650,
      "identify semantic": 51549,
      "words experimental": 133947,
      "dataset rich": 27173,
      "modeling order": 72498,
      "data capture": 25716,
      "multidomain chinese": 75085,
      "annotated dialogues": 6183,
      "datasets single": 27716,
      "multidomain dialogues": 75091,
      "form natural": 45107,
      "states acts": 113036,
      "useful dialogue": 128877,
      "ellipsis resolution": 35306,
      "resolution tasks": 100790,
      "filling dialogue": 44036,
      "comparison future": 19545,
      "knowledgegrounded dialogue": 58275,
      "constraint propose": 21582,
      "generation defined": 47358,
      "model knowledge": 71394,
      "selection module": 104805,
      "aim introducing": 4463,
      "model documentlevel": 71017,
      "strong documentlevel": 113667,
      "translation interestingly": 124873,
      "interestingly observe": 56095,
      "appropriate training": 8442,
      "training techniques": 123911,
      "techniques achieve": 119827,
      "translation length": 124903,
      "approaches documentlevel": 8134,
      "documentlevel datasets": 33136,
      "sentencelevel datasets": 106154,
      "experiments documentlevel": 40921,
      "outperforms sentencelevel": 82977,
      "ones previous": 81703,
      "set metrics": 107494,
      "including bleu": 53267,
      "linguistic indicators": 64487,
      "indicators human": 53881,
      "corpus hindienglish": 23824,
      "hindienglish code": 50480,
      "code mixed": 18110,
      "normalization present": 80345,
      "hindienglish codemixed": 50484,
      "codemixed sentences": 18180,
      "normalization task": 80348,
      "sentence corpus": 105813,
      "corresponding human": 24288,
      "attempt direction": 9739,
      "metric evaluation": 69880,
      "meteor score": 68572,
      "terminology constraints": 120256,
      "use specific": 128290,
      "constraints inference": 21601,
      "ape approaches": 6771,
      "nonautoregressive models": 80158,
      "lexically constrained": 63856,
      "englishgerman benchmarks": 37371,
      "output approach": 83048,
      "modeling contextualized": 72405,
      "temporal graph": 120105,
      "graph generation": 48815,
      "models automated": 72788,
      "tasks potential": 119388,
      "temporal reasoning": 120116,
      "event graphs": 39521,
      "sufficiently explored": 115293,
      "explored reason": 41633,
      "obtaining large": 81443,
      "temporal links": 120111,
      "generate large": 46961,
      "documentgraph pairs": 33129,
      "novel formulation": 80572,
      "problem sequencetosequence": 91222,
      "mapping task": 67152,
      "effective generating": 34683,
      "generating structurally": 47265,
      "method large": 68931,
      "linear contextfree": 64341,
      "contextfree rewriting": 22364,
      "rewriting systems": 102698,
      "parsers accuracy": 85012,
      "approach efficient": 7515,
      "transformation lexical": 124264,
      "evaluated english": 38973,
      "negra tiger": 76981,
      "bert wordlevel": 13256,
      "domain suitable": 33666,
      "building models": 14864,
      "models specialized": 74075,
      "domains medical": 33815,
      "domain adopting": 33464,
      "shifts focus": 108191,
      "focus word": 44841,
      "level subword": 63507,
      "represent entire": 99106,
      "bert variety": 13251,
      "domain tasks": 33670,
      "tasks time": 119564,
      "time producing": 122078,
      "datasets makes": 27558,
      "makes common": 66764,
      "step train": 113305,
      "method generalizes": 68852,
      "effects existing": 34985,
      "state key": 112502,
      "key good": 57573,
      "parsing recurrent": 85215,
      "global state": 48271,
      "amr parsing": 5393,
      "tasks particularly": 119372,
      "data multilingual": 26143,
      "contextual affective": 22441,
      "affective analysis": 4235,
      "differ languages": 30929,
      "difficulty generalizing": 31698,
      "usefulness method": 128959,
      "method analyzing": 68635,
      "analyzing wikipedia": 6057,
      "russian spanish": 103501,
      "spanish results": 111172,
      "systematic differences": 116668,
      "contain content": 21735,
      "representation particular": 99373,
      "social groups": 110342,
      "mainly based": 66479,
      "learning small": 63034,
      "handlabeled training": 49370,
      "data lacking": 26068,
      "causal commonsense": 15803,
      "automatically labeled": 10792,
      "data gender": 25970,
      "based vietnamese": 12154,
      "individual human": 53913,
      "human work": 50989,
      "chinese languages": 16780,
      "vietnamese far": 131677,
      "far propose": 43097,
      "dataset comprises": 26814,
      "website research": 132296,
      "purposes addition": 94446,
      "addition paper": 3458,
      "tree random": 125608,
      "regression deep": 97700,
      "impact component": 51861,
      "detecting gender": 29837,
      "result best": 101361,
      "best f1score": 13334,
      "f1score achieved": 42713,
      "model analyzing": 70663,
      "predictions neural": 89183,
      "understand internal": 126753,
      "internal workings": 56179,
      "workings nmt": 134910,
      "generation decision": 47356,
      "conduct analysis": 20824,
      "analysis nmt": 5721,
      "relative contributions": 98347,
      "varying training": 131269,
      "objective training": 81124,
      "process models": 91530,
      "data tend": 26550,
      "tend rely": 120161,
      "rely source": 98741,
      "media focused": 68118,
      "taskspecific neural": 119636,
      "models adapting": 72716,
      "models combinations": 72922,
      "information topics": 55048,
      "topics sentiment": 122655,
      "evaluate battery": 38802,
      "networks subsequently": 77783,
      "information experiments": 54551,
      "achieving macro": 2958,
      "macro f1": 66358,
      "selfsupervised objectives": 104948,
      "replaced token": 98937,
      "token detection": 122249,
      "force model": 45012,
      "generating target": 47274,
      "benchmark demonstrate": 12811,
      "using phonetic": 130011,
      "characteristics pose": 16532,
      "challenges scripts": 16209,
      "segmented words": 104663,
      "known language": 58308,
      "based international": 11778,
      "international phonetic": 56184,
      "phonetic alphabet": 87287,
      "alphabet ipa": 5206,
      "generative framework": 47727,
      "jointly models": 57367,
      "experiments incorporating": 40961,
      "leads clear": 61925,
      "propose measure": 92758,
      "correctly identifies": 24176,
      "does strong": 33403,
      "networks huge": 77620,
      "textual conversations": 121682,
      "conversations lead": 23059,
      "ongoing discussions": 81734,
      "messages conversation": 68506,
      "aims separate": 4573,
      "features dataset": 43436,
      "translation sense": 125237,
      "translation source": 125277,
      "ends words": 36865,
      "far ideal": 43085,
      "inference biomedical": 54119,
      "biomedical entity": 14180,
      "labelled training": 58562,
      "linking models": 64663,
      "able link": 1663,
      "link mentions": 64636,
      "unseen entities": 127523,
      "entities previous": 37841,
      "mentions relations": 68467,
      "linking mentions": 64661,
      "mentions biomedical": 68446,
      "text linking": 121093,
      "linking decisions": 64654,
      "highly specialized": 50349,
      "base entity": 11455,
      "multiple mentions": 75614,
      "linking predictions": 64667,
      "predictions experiments": 89170,
      "available biomedical": 10955,
      "biomedical dataset": 14171,
      "independent prediction": 53777,
      "30 points": 905,
      "points accuracy": 87844,
      "23 points": 791,
      "yields unsatisfactory": 135449,
      "pooling paper": 88048,
      "demonstrate bayesian": 28676,
      "mean word": 67615,
      "theoretically prove": 121805,
      "reflect true": 97612,
      "true sentence": 125854,
      "similarity addition": 109191,
      "provides explanations": 94027,
      "fast training": 43154,
      "having small": 49568,
      "memory footprints": 68316,
      "copy network": 23291,
      "network dialogue": 77225,
      "generation past": 47539,
      "audiences different": 10222,
      "different fields": 31149,
      "pointer generator": 87821,
      "content generation": 21886,
      "generation content": 47343,
      "fluency accuracy": 44692,
      "information particular": 54829,
      "particular domains": 85407,
      "court debate": 24469,
      "observed different": 81219,
      "information provide": 54886,
      "context similar": 22268,
      "structure simultaneously": 113957,
      "simultaneously experiments": 109677,
      "proved proposed": 93714,
      "superior existing": 115683,
      "introduction area": 56653,
      "solved using": 110632,
      "tasks commonly": 118987,
      "learning methodology": 62733,
      "computer linguistics": 20483,
      "preprocessing tasks": 89336,
      "text search": 121272,
      "accuracy task": 2298,
      "task implications": 118275,
      "tasks area": 118940,
      "development research": 30417,
      "texts early": 121504,
      "early stage": 34393,
      "effectiveness detection": 34882,
      "detection features": 29954,
      "features certain": 43391,
      "language taking": 60145,
      "model chosen": 70825,
      "texts utilizing": 121648,
      "dependencies means": 29094,
      "corpus ukrainian": 24053,
      "performed different": 86919,
      "metrics method": 69987,
      "used noun": 128665,
      "accuracy increase": 2184,
      "increase method": 53603,
      "advances language": 3992,
      "generation style": 47644,
      "specific target": 111495,
      "approach generate": 7586,
      "conduct quantitative": 20890,
      "baselines illustrate": 12411,
      "illustrate benefits": 51738,
      "benefits proposed": 13019,
      "automatic corpus": 10504,
      "represent stateoftheart": 99144,
      "languages pretrained": 60809,
      "true performance": 125853,
      "present contributions": 89421,
      "contributions propose": 22792,
      "methodology automatically": 69248,
      "automatically producing": 10817,
      "nli benchmark": 79535,
      "articles create": 9179,
      "dataset lowresource": 27004,
      "produce new": 91913,
      "new pretrained": 79079,
      "alleviate resource": 5050,
      "resource scarcity": 100876,
      "benchmarking dataset": 12877,
      "domains use": 33881,
      "analysis simple": 5857,
      "compare data": 19234,
      "tokenlevel sequence": 122294,
      "experiments data": 40863,
      "efficient pretraining": 35102,
      "objective novel": 81102,
      "novel pairwise": 80680,
      "using reddit": 130097,
      "data aligned": 25588,
      "layers pretrained": 61799,
      "domains data": 33755,
      "largest gains": 61533,
      "fewshot setups": 43916,
      "cost efficient": 24352,
      "efficient finetuning": 35076,
      "sequencelabeling tasks": 107105,
      "dataset transfer": 27245,
      "base queries": 11473,
      "designed facilitate": 29591,
      "facilitate task": 42795,
      "datasets quality": 27656,
      "explicit description": 41314,
      "tasks generalize": 119142,
      "known unknown": 58330,
      "effectiveness models": 34916,
      "zeroshot generalization": 135503,
      "multilingual question": 75346,
      "tasks typically": 119579,
      "typically assume": 126412,
      "answer content": 6521,
      "content language": 21896,
      "language construct": 58913,
      "built questions": 14936,
      "questions tydi": 95370,
      "tydi qa": 126180,
      "answers task": 6746,
      "questions diverse": 95303,
      "systems crosslingual": 116825,
      "qa challenging": 94498,
      "task facilitate": 118200,
      "answering data": 6617,
      "transfer transformer": 124203,
      "multilingual variant": 75398,
      "t5 pretrained": 117251,
      "modified training": 74449,
      "translation zeroshot": 125438,
      "setting generative": 107752,
      "model checkpoints": 70820,
      "representations emerging": 99620,
      "emerging neural": 36069,
      "models spoken": 74083,
      "employed various": 36310,
      "various spoken": 131202,
      "identification speech": 51445,
      "speech signals": 111803,
      "objective measures": 81093,
      "shows language": 108591,
      "captures language": 15450,
      "learning implicit": 62637,
      "feedback improve": 43831,
      "understanding largescale": 126876,
      "largescale conversational": 61411,
      "digital assistant": 31715,
      "understanding user": 127005,
      "user interaction": 129003,
      "interaction data": 55942,
      "data dialog": 25837,
      "context rich": 22254,
      "information embedded": 54519,
      "supervision data": 115877,
      "applying framework": 7246,
      "10 domains": 131,
      "art countless": 9044,
      "tasks available": 118951,
      "pretrained seq2seq": 90186,
      "french based": 45800,
      "based bart": 11543,
      "generative tasks": 47761,
      "novel summarization": 80740,
      "stateoftheart bertbased": 112602,
      "pretraining multilingual": 90290,
      "multilingual bart": 75198,
      "corpus resulting": 23976,
      "available unsupervised": 11139,
      "crosslingual adaptation": 24924,
      "mainly consists": 66481,
      "works zeroshot": 134995,
      "translationbased approach": 125442,
      "approach studied": 7936,
      "studied extensively": 114160,
      "verify efficacy": 131586,
      "performances finegrained": 86881,
      "strengths drawbacks": 113574,
      "performance instead": 86465,
      "simply augmenting": 109620,
      "data adaptation": 25569,
      "transfer performed": 124177,
      "tools knowledge": 122452,
      "annotations crowdsourcing": 6420,
      "ner tools": 77093,
      "ways integrating": 132164,
      "parsing allows": 85069,
      "linear modeling": 64356,
      "modeling perspective": 72509,
      "spider challenging": 111930,
      "challenging zeroshot": 16348,
      "zeroshot semantic": 135522,
      "speedup training": 111888,
      "compared semantic": 19437,
      "autoregressive decoding": 10862,
      "695 exact": 1201,
      "processing chains": 91634,
      "crosslingual eventcentric": 24954,
      "eventcentric knowledge": 39558,
      "presents strategy": 89903,
      "consisting tokenization": 21464,
      "parsing including": 85129,
      "knowledge processing": 58125,
      "terms availability": 120277,
      "availability language": 10913,
      "wellresourced languages": 132430,
      "new modules": 79038,
      "order classify": 82294,
      "languages terms": 60915,
      "models mainstream": 73544,
      "helping users": 49805,
      "identify distinct": 51495,
      "topic providing": 122563,
      "evaluate joint": 38844,
      "results conclude": 101617,
      "judgments results": 57450,
      "stance topic": 112176,
      "interpreting output": 56290,
      "output topic": 83139,
      "learn biases": 61997,
      "turn limits": 125974,
      "robustness existing": 103097,
      "objective models": 81095,
      "learn biased": 61996,
      "biased examples": 13857,
      "performance adversarial": 86130,
      "adversarial evaluation": 4114,
      "bias bias": 13794,
      "higherlevel abstraction": 50214,
      "meaning help": 67638,
      "important parts": 52202,
      "addition models": 3452,
      "models vulnerable": 74333,
      "bias training": 13851,
      "contain bias": 21733,
      "alignment propose": 4984,
      "method align": 68626,
      "improved zeroshot": 52658,
      "method aligns": 68627,
      "proposed translation": 93577,
      "creative language": 24728,
      "ability rank": 1566,
      "important general": 52166,
      "deep ranking": 28401,
      "substantial amounts": 114851,
      "fully explored": 45960,
      "gaussian process": 46549,
      "process preference": 91548,
      "learning gppl": 62620,
      "work sparse": 134811,
      "remains effective": 98795,
      "settings leveraging": 107818,
      "leveraging complementary": 63666,
      "combined approach": 18647,
      "stateoftheart humor": 112672,
      "tasks increasing": 119196,
      "spearmans rho": 111347,
      "triples natural": 125815,
      "text focused": 120957,
      "entire english": 37706,
      "kg like": 57658,
      "architectures developed": 8800,
      "developed integrate": 30280,
      "sources approach": 110878,
      "text allowing": 120637,
      "improvements knowledge": 52863,
      "knowledge intensive": 58021,
      "tasks open": 119350,
      "knowledge probe": 58123,
      "introduction pretrained": 56657,
      "pretrained transformerbased": 90201,
      "differ various": 30935,
      "various dimensions": 131080,
      "dimensions including": 31768,
      "including choice": 53269,
      "lstm layers": 65635,
      "new modular": 79037,
      "series analyses": 107268,
      "robust choice": 103024,
      "embeddings multitask": 35816,
      "training setup": 123859,
      "taking insights": 117548,
      "insights propose": 55543,
      "models provides": 73837,
      "provides multiple": 94052,
      "pretraining data": 90244,
      "data statistics": 26504,
      "statistics making": 113205,
      "making hard": 66844,
      "hand robust": 49325,
      "robust misspellings": 103046,
      "table pretrained": 117256,
      "characters subword": 16623,
      "vocabulary used": 131917,
      "used dropin": 128503,
      "robust characterlevel": 103022,
      "integrate bert": 55745,
      "providing practical": 94130,
      "practical method": 88706,
      "performance social": 86725,
      "linguistic codeswitching": 64438,
      "codeswitching evaluation": 18235,
      "evaluation lince": 39255,
      "generative conversational": 47719,
      "systems systematic": 117178,
      "recent models": 96473,
      "using paired": 129997,
      "datasets findings": 27485,
      "findings demonstrate": 44291,
      "sets finally": 107672,
      "finally make": 44203,
      "available researchers": 11094,
      "models temporal": 74171,
      "reasoning implicit": 96254,
      "reasoning dataset": 96241,
      "dataset evaluates": 26897,
      "degree systems": 28577,
      "systems understand": 117215,
      "events events": 39578,
      "mentioned explicitly": 68436,
      "text inferred": 121051,
      "reasoning research": 96305,
      "events human": 39581,
      "human readers": 50947,
      "better reasoning": 13690,
      "struggle predicting": 114127,
      "predicting temporal": 89016,
      "temporal relationships": 120126,
      "implicit explicit": 52008,
      "events address": 39562,
      "propose neurosymbolic": 92830,
      "largescale text": 61512,
      "reasoning tasks": 96316,
      "dialogues existing": 30830,
      "chatbots aim": 16668,
      "propose integrate": 92725,
      "types systems": 126364,
      "goal making": 48366,
      "chitchat responses": 16883,
      "schemaguided dialogue": 103904,
      "multiwoz 21": 75954,
      "evaluation lastly": 39253,
      "lastly propose": 61558,
      "models adding": 72717,
      "explicitly trained": 41396,
      "trained predict": 123241,
      "responses automatic": 101254,
      "evaluations compared": 39455,
      "stateoftheart taskoriented": 113000,
      "users speak": 129174,
      "hard impossible": 49463,
      "mapping users": 67153,
      "free form": 45757,
      "utterance ontology": 130605,
      "model user": 72260,
      "knowledge users": 58228,
      "conversational recommendation": 23025,
      "user responses": 129038,
      "showing usefulness": 108435,
      "challenges problem": 16197,
      "serve valuable": 107309,
      "faithful rationales": 43005,
      "reflect models": 97609,
      "subset input": 114831,
      "investigate lessstudied": 56776,
      "rationales demonstrate": 95842,
      "mechanisms model": 68063,
      "test property": 120481,
      "questionanswering natural": 95253,
      "potential producing": 88580,
      "language way": 60338,
      "design computer": 29527,
      "step separate": 113300,
      "parts classification": 85570,
      "information organization": 54816,
      "architecture including": 8669,
      "hierarchy information": 50028,
      "paper achieved": 83696,
      "goal enabling": 48350,
      "enabling machines": 36411,
      "understand information": 126752,
      "basic conditions": 12518,
      "practical theoretical": 88718,
      "theoretical basis": 121791,
      "research methods": 100556,
      "nlu applied": 79805,
      "applied largescale": 7086,
      "wordnet using": 133782,
      "approach composed": 7441,
      "composed modules": 20068,
      "module produces": 74506,
      "likelihood scores": 64127,
      "relation scores": 98068,
      "task graph": 118248,
      "graph optimization": 48851,
      "outputs maximum": 83175,
      "maximum spanning": 67551,
      "graph train": 48876,
      "task constructing": 118016,
      "ancestor f1": 6080,
      "used transfer": 128827,
      "ability encode": 1513,
      "crosslingual similarity": 25017,
      "known models": 58311,
      "method study": 69166,
      "models mbert": 73557,
      "mbert xlmr": 67573,
      "insight information": 55522,
      "information sharing": 54978,
      "suggest linguistic": 115316,
      "paradigm neural": 84545,
      "oneshot generation": 81714,
      "produced single": 91967,
      "constraints user": 21616,
      "text dynamic": 120894,
      "generation setting": 47617,
      "user interacts": 129005,
      "baselines obtains": 12437,
      "analyses models": 5460,
      "new norm": 79059,
      "nlp remains": 79680,
      "applied unseen": 7137,
      "languages covered": 60472,
      "covered available": 24528,
      "available largescale": 11029,
      "generally available": 46849,
      "work comparing": 134420,
      "multilingual monolingual": 75294,
      "benefit transfer": 12993,
      "behave similarly": 12644,
      "improves significantly": 53044,
      "inference speedup": 54222,
      "autoregressive machine": 10866,
      "work hypothesize": 134556,
      "empirically verify": 36248,
      "encoders capture": 36636,
      "propose adopt": 92533,
      "specifically model": 111570,
      "performance experimental": 86356,
      "englishromanian datasets": 37423,
      "furthermore performance": 46201,
      "performance largescale": 86492,
      "englishgerman datasets": 37375,
      "method addition": 68616,
      "demonstrate multitask": 28804,
      "complementary knowledge": 19747,
      "standard knowledge": 112248,
      "sanskrit language": 103633,
      "describes neural": 29417,
      "approaches process": 8290,
      "process formation": 91490,
      "formation splitting": 45215,
      "analysis sanskrit": 5827,
      "sanskrit texts": 103634,
      "rules sandhi": 103439,
      "nature words": 76674,
      "technique demonstrate": 119781,
      "accuracy better": 2109,
      "methods multiple": 69630,
      "multiple standard": 75702,
      "additional lexical": 3541,
      "resources code": 100956,
      "knowledge adversarial": 57748,
      "knowledge critical": 57845,
      "knowledge remains": 58141,
      "remains limited": 98807,
      "used commonsense": 128434,
      "points machine": 87864,
      "comprehension models": 20198,
      "solutions investigate": 110576,
      "investigate machine": 56778,
      "answer dataset": 6523,
      "questions generated": 95310,
      "extracted wordnet": 42197,
      "identify synonyms": 51565,
      "achieves stateofart": 2878,
      "20 20": 488,
      "did perform": 30918,
      "knowledge essential": 57905,
      "comprehension contextualized": 20168,
      "embeddings encode": 35687,
      "humanlike word": 51058,
      "variation word": 130902,
      "key aspect": 57540,
      "senses polysemy": 105719,
      "polysemy homonymy": 88029,
      "relatedness multiple": 97924,
      "participants judgments": 85320,
      "bert embedding": 13109,
      "point potential": 87810,
      "potential utility": 88597,
      "efficient transformer": 35126,
      "nlu applications": 79804,
      "applications models": 6968,
      "better various": 13767,
      "tasks transformer": 119574,
      "computationally challenging": 20447,
      "various nlu": 131156,
      "utilizing knowledge": 130570,
      "structured pruning": 114032,
      "inference efficiency": 54136,
      "provide effective": 93810,
      "practitioners choose": 88745,
      "superglue benchmark": 115674,
      "benchmark achieve": 12729,
      "speedup compared": 111885,
      "presented methods": 89789,
      "reduce cost": 97320,
      "graph transformer": 48878,
      "networks syntactic": 77790,
      "event argument": 39496,
      "extraction eae": 42314,
      "works syntactic": 134982,
      "learn effective": 62037,
      "bias based": 13793,
      "performed demonstrate": 86918,
      "model leading": 71423,
      "leading stateoftheart": 61910,
      "arabic reviews": 8544,
      "improvement quality": 52752,
      "lacks accuracy": 58774,
      "major errors": 66557,
      "need extensive": 76806,
      "investigate challenges": 56731,
      "challenges involved": 16170,
      "involved translating": 56885,
      "errors lead": 38380,
      "lead incorrect": 61859,
      "incorrect translation": 53580,
      "study points": 114470,
      "special characteristics": 111355,
      "characteristics arabic": 16519,
      "sentiment transfer": 106809,
      "model respect": 71917,
      "significantly help": 108916,
      "correcting sentiment": 24138,
      "bert bert": 13076,
      "architectures like": 8814,
      "popularity recent": 88152,
      "use unlabeled": 128348,
      "data exploring": 25934,
      "taskagnostic bert": 118871,
      "bert multiple": 13188,
      "multiple settings": 75693,
      "include domain": 53223,
      "domain task": 33669,
      "lighter model": 64004,
      "achieves similar": 2866,
      "studies event": 114219,
      "vectors graphbased": 131440,
      "candidate words": 15130,
      "addition current": 3417,
      "overall contextual": 83222,
      "contextual importance": 22467,
      "importance scores": 52078,
      "obtained dependency": 81359,
      "novel gating": 80579,
      "filter noisy": 44073,
      "noisy information": 80105,
      "information hidden": 54656,
      "gcn models": 46567,
      "novel mechanisms": 80626,
      "applications nlp": 6979,
      "actions based": 3089,
      "trees existing": 125690,
      "based shiftreduce": 12041,
      "tree based": 125582,
      "transition develop": 124486,
      "parser step": 84997,
      "parser penn": 84982,
      "treebank ptb": 125645,
      "ptb chinese": 94223,
      "chinese treebank": 16831,
      "constituency trees": 21532,
      "trees performs": 125702,
      "parser establishes": 84949,
      "wnut2020 task": 132885,
      "task overview": 118498,
      "wet lab": 132462,
      "lab protocols": 58380,
      "task wnut": 118849,
      "wnut 2020": 132881,
      "outline task": 82635,
      "process corpus": 91445,
      "sub task": 114644,
      "semantics models": 105434,
      "evaluate neural": 38875,
      "features modeling": 43612,
      "results consistent": 101631,
      "child language": 16718,
      "difficulty learning": 31703,
      "distributional data": 32698,
      "data discuss": 25855,
      "facilitating research": 42810,
      "humanannotated corpora": 50995,
      "techniques require": 119973,
      "explore unsupervised": 41604,
      "easier obtain": 34420,
      "builds recent": 14908,
      "augmentation training": 10313,
      "amounts highquality": 5335,
      "conventional training": 22907,
      "propose conversation": 92602,
      "graphbased representation": 48911,
      "data volume": 26623,
      "volume diversity": 131939,
      "task metrics": 118391,
      "reference test": 97537,
      "sets previous": 107696,
      "novel ranking": 80696,
      "predictions documentlevel": 89164,
      "score overall": 104100,
      "overall systems": 83261,
      "sets cases": 107653,
      "using named": 129931,
      "automated speech": 10468,
      "recognition including": 96893,
      "common problems": 18911,
      "new annotations": 78779,
      "annotations training": 6473,
      "training custom": 123416,
      "contextual string": 22501,
      "embeddings applying": 35554,
      "results novel": 101998,
      "tackling natural": 117324,
      "method common": 68706,
      "instead automatically": 55654,
      "effort development": 35170,
      "make step": 66731,
      "overcome shortcomings": 83300,
      "best explanation": 13332,
      "problem words": 91287,
      "words instead": 134010,
      "sentencelevel inference": 106162,
      "inference relations": 54208,
      "help lexical": 49746,
      "account sentencelevel": 2039,
      "relations implement": 98185,
      "dataset 14": 26708,
      "systems sentiment": 117135,
      "urdu text": 127842,
      "form image": 45093,
      "text voice": 121409,
      "huge growth": 50722,
      "usage social": 127869,
      "users communicate": 129102,
      "shared content": 107969,
      "rumor detection": 103457,
      "developed models": 30288,
      "learning vanilla": 63154,
      "detection achieved": 29869,
      "low precision": 65379,
      "precision time": 88806,
      "extract deep": 42076,
      "quantities text": 94897,
      "function helps": 46023,
      "helps learn": 49822,
      "ukrainian text": 126551,
      "text growing": 121031,
      "articles quality": 9212,
      "quality approach": 94595,
      "automated assessment": 10431,
      "language analyzed": 58829,
      "improvement method": 52723,
      "pretraining neural": 90297,
      "sentences experimental": 106303,
      "modifications training": 74444,
      "previously retrieved": 90621,
      "discrimination task": 32203,
      "task accordingly": 117829,
      "classification critical": 17164,
      "recently graph": 96692,
      "received increasing": 96362,
      "attention research": 9993,
      "handle large": 49387,
      "propose principled": 92998,
      "learning extensive": 62571,
      "various benchmark": 131051,
      "attention joint": 9862,
      "levels natural": 63554,
      "languages words": 60973,
      "difficult acquire": 31607,
      "acquire automatically": 3031,
      "linguistic components": 64444,
      "components evaluate": 20026,
      "using multihead": 129907,
      "representations single": 99894,
      "like sequence": 64092,
      "lower wordlevel": 65452,
      "simulated multiple": 109639,
      "reference training": 97540,
      "training smrt": 123869,
      "2020 use": 738,
      "measured human": 67837,
      "evaluation quality": 39348,
      "quality outperforms": 94737,
      "diversity requiring": 32893,
      "performance illustrate": 86436,
      "sentence analogies": 105770,
      "conversational question": 23021,
      "systems users": 117222,
      "poses exciting": 88264,
      "little evidence": 64806,
      "evidence provided": 39664,
      "users able": 129089,
      "answer able": 6516,
      "correct incorrect": 24103,
      "feedback paper": 43833,
      "based importance": 11765,
      "importance sampling": 52077,
      "supervised using": 115853,
      "experiments document": 40920,
      "classification development": 17183,
      "datasets like": 27548,
      "annotations results": 6460,
      "indomain experiments": 53967,
      "exploit interactions": 41422,
      "users improve": 129132,
      "visual dialog": 131792,
      "dialog neural": 30575,
      "shown exhibit": 108469,
      "exhibit strong": 40005,
      "performance singleturn": 86720,
      "tasks extending": 119127,
      "vqa models": 131961,
      "retain information": 102358,
      "previous dialog": 90406,
      "turns strong": 125995,
      "network decomposes": 77220,
      "reasoning steps": 96313,
      "singleturn question": 109855,
      "referring past": 97579,
      "contextaware attention": 22337,
      "current question": 25328,
      "dataset beating": 26762,
      "performance particularly": 86603,
      "model indonesian": 71337,
      "indonesian nlp": 53994,
      "indonesian language": 53992,
      "200 million": 534,
      "language world": 60350,
      "datasets sparsity": 27726,
      "sparsity language": 111255,
      "seven tasks": 107872,
      "language spanning": 60109,
      "discourse additionally": 32038,
      "resources experiments": 100974,
      "influence patterns": 54314,
      "bert attention": 13067,
      "attention need": 9943,
      "attentionbased transformer": 10102,
      "superior information": 115684,
      "flows input": 44688,
      "tokens output": 122323,
      "output predictions": 83106,
      "introduce influence": 56436,
      "model nodes": 71588,
      "portion information": 88185,
      "far model": 43093,
      "previous attentionbased": 90388,
      "methods contextaware": 69403,
      "nonautoregressive translation": 80161,
      "significantly accelerates": 108876,
      "accelerates inference": 1874,
      "paper reveal": 84408,
      "adequately capture": 3838,
      "representative datasets": 100009,
      "analyses demonstrate": 5449,
      "source contexts": 110737,
      "does bert": 33330,
      "bert look": 13158,
      "clustering analysis": 17935,
      "analysis berts": 5516,
      "berts attention": 13286,
      "focus probing": 44805,
      "analytical methods": 5947,
      "analysis specific": 5866,
      "remains primitive": 98820,
      "patterns unsupervised": 85790,
      "set proposed": 107547,
      "previous observations": 90440,
      "heads transformer": 49611,
      "adopt stateoftheart": 3897,
      "stateoftheart data": 112627,
      "reported work": 99077,
      "including text": 53394,
      "results japanese": 101868,
      "japanese zar": 57204,
      "cost reduction": 24372,
      "augmented training": 10338,
      "american english": 5309,
      "knowledge learn": 58045,
      "complex phenomenon": 19858,
      "based naturalistic": 11881,
      "naturalistic dataset": 76630,
      "dataset probing": 27109,
      "shows models": 108603,
      "especially semantic": 38505,
      "comparison evaluation": 19542,
      "performance linguistic": 86498,
      "evaluations combining": 39454,
      "event semantics": 39540,
      "modifiers using": 74454,
      "logicbased nli": 65026,
      "knowledge evaluate": 57906,
      "datasets comparison": 27364,
      "comparison previous": 19563,
      "deeplearningbased systems": 28462,
      "systems suggests": 117173,
      "suggests semantic": 115382,
      "various combinations": 131060,
      "advantage framework": 4045,
      "transformerbased methods": 124423,
      "embeddings derived": 35640,
      "shown stateoftheart": 108527,
      "years extensive": 135264,
      "extensive work": 41962,
      "represent abstract": 99098,
      "granular level": 48741,
      "level representation": 63497,
      "problem textual": 91263,
      "level requiring": 63499,
      "finegrained attributes": 44339,
      "text abstract": 120624,
      "abstract level": 1766,
      "capture overall": 15355,
      "despite high": 29692,
      "outperformed simple": 82790,
      "like tfidf": 64103,
      "use contextual": 127960,
      "embeddings achieving": 35542,
      "multiview learning": 75934,
      "translation limited": 124911,
      "representation directly": 99211,
      "layers existing": 61776,
      "introducing additional": 56626,
      "learning solve": 63037,
      "change model": 16367,
      "view input": 131706,
      "sentence way": 106126,
      "shared decoder": 107972,
      "prediction consistency": 89044,
      "consistency regularization": 21365,
      "used encourage": 128511,
      "results translation": 102286,
      "improvements multiple": 52877,
      "multiple strong": 75706,
      "speed original": 111871,
      "compute language": 20466,
      "mbert representations": 67567,
      "evaluate language": 38846,
      "language trees": 60194,
      "analysis finding": 5624,
      "structural factors": 113768,
      "novel measure": 80624,
      "lists based": 64714,
      "linguistic approaches": 64427,
      "results contribute": 101637,
      "usergenerated contents": 129071,
      "text news": 121146,
      "suggest nmt": 115323,
      "crosscultural communication": 24875,
      "gap performance": 46471,
      "question present": 95202,
      "evaluating robustness": 39093,
      "used offtheshelf": 128671,
      "systems greatly": 116925,
      "bertbased dual": 13270,
      "dual embedding": 34233,
      "fixed phrases": 44603,
      "encode contextual": 36423,
      "words learn": 134030,
      "representations tokens": 99935,
      "context pooling": 22218,
      "use separate": 128267,
      "matching experiments": 67403,
      "experiments recently": 41111,
      "existing state": 40288,
      "experiments context": 40855,
      "contribute improvement": 22754,
      "utilizing bert": 130566,
      "learns multimodal": 63223,
      "multimodal embeddings": 75427,
      "embeddings text": 35979,
      "learns text": 63240,
      "text embeddings": 120902,
      "features images": 43547,
      "combining features": 18725,
      "results produced": 102057,
      "resnet bert": 100747,
      "features helpful": 43536,
      "learning efficient": 62531,
      "statistics leads": 113204,
      "lexical properties": 63792,
      "notion context": 80405,
      "defined training": 28505,
      "similar vectors": 109169,
      "layer deep": 61707,
      "solve downstream": 110596,
      "combine multiple": 18633,
      "differently trained": 31598,
      "use just": 128100,
      "combine source": 18643,
      "embeddings according": 35539,
      "combine input": 18628,
      "second version": 104466,
      "prize 2018": 90788,
      "2018 introduce": 663,
      "ontologybased topic": 81856,
      "topic hierarchy": 122528,
      "intent allows": 55892,
      "create unique": 24650,
      "2020 competition": 723,
      "novel contribution": 80515,
      "innovative approach": 55285,
      "allows utilize": 5198,
      "knowledge expressed": 57922,
      "conversational structures": 23037,
      "user inputs": 128998,
      "results individual": 101862,
      "individual parts": 53926,
      "multilingual multidomain": 75296,
      "multidomain dataset": 75086,
      "generation news": 47510,
      "image captions": 51774,
      "different audiences": 31018,
      "entities provided": 37852,
      "written human": 135130,
      "spans languages": 111205,
      "news carry": 79313,
      "various modeling": 131134,
      "challenges poses": 16194,
      "results obtain": 102003,
      "plenty room": 87749,
      "components transformer": 20050,
      "transformer stateoftheart": 124381,
      "model studied": 72096,
      "challenges designing": 16147,
      "architectures work": 8860,
      "gap evaluating": 46452,
      "trained transformer": 123315,
      "perspectives experimental": 87175,
      "training strategies": 123887,
      "strategies model": 113476,
      "model capacities": 70804,
      "report number": 99017,
      "number interesting": 80899,
      "interesting findings": 56073,
      "better analyze": 13508,
      "improve transformer": 52566,
      "strategy improves": 113522,
      "correct natural": 24110,
      "text response": 121262,
      "using templatebased": 130277,
      "developed research": 30298,
      "provide benefits": 93768,
      "modelbased solutions": 72353,
      "challenging high": 16256,
      "high data": 50056,
      "needs paper": 76897,
      "systems production": 117080,
      "quality lightweight": 94706,
      "lightweight neural": 64014,
      "task segmenting": 118669,
      "segmenting long": 104667,
      "novels using": 80786,
      "neural inference": 77929,
      "data ground": 25989,
      "cues present": 25216,
      "reveal interesting": 102497,
      "nmt aims": 79830,
      "shows enhanced": 108575,
      "allow flexible": 5082,
      "bert representation": 13217,
      "layers dynamically": 61770,
      "attention different": 9819,
      "linear unit": 64385,
      "unit glu": 127204,
      "scores multiple": 104188,
      "multiple translation": 75735,
      "models realworld": 73867,
      "data resources": 26354,
      "resources based": 100952,
      "language dialect": 58958,
      "methods dealing": 69424,
      "dealing lowresource": 27869,
      "end study": 36831,
      "sequential transfer": 107259,
      "learning various": 63156,
      "various lowresource": 131123,
      "adapt neural": 3186,
      "models explore": 73211,
      "tasks minimal": 119297,
      "enhanced reader": 37515,
      "task asks": 117896,
      "machine answer": 65740,
      "context chinese": 22027,
      "chinese mrc": 16787,
      "characteristics chinese": 16520,
      "studies tend": 114289,
      "treat idioms": 125550,
      "fully exploiting": 45958,
      "literal meaning": 64729,
      "measure consistency": 67786,
      "meanings chinese": 67741,
      "relationship propose": 98306,
      "embeddings incorporate": 35755,
      "mechanism encode": 67975,
      "graph experimental": 48806,
      "comprehension dataset": 20175,
      "component dialogue": 19974,
      "detect dialogue": 29798,
      "component training": 20008,
      "use active": 127879,
      "dialogue including": 30690,
      "expressions natural": 41757,
      "inference context": 54128,
      "nli fundamental": 79543,
      "texts popular": 121579,
      "popular nli": 88113,
      "datasets present": 27625,
      "present task": 89736,
      "task sentencelevel": 118690,
      "testing semantic": 120602,
      "fall short": 43028,
      "natural human": 76258,
      "human inference": 50863,
      "process introduce": 91513,
      "contexthypothesis pairs": 22369,
      "previous nli": 90438,
      "types empirical": 126276,
      "factual correctness": 42931,
      "distributions latent": 32763,
      "latent features": 61588,
      "workings neural": 134909,
      "trustworthiness models": 125876,
      "network interpretability": 77288,
      "typically faces": 126429,
      "models constrained": 72972,
      "useful solutions": 128931,
      "novel strategy": 80737,
      "strategy achieving": 113502,
      "success using": 115139,
      "instance attention": 55593,
      "unordered set": 127483,
      "usefulness approach": 128956,
      "provides interpretable": 94044,
      "slightly outperforms": 110028,
      "classical cnn": 17087,
      "cnn bilstm": 17994,
      "datasets unsupervised": 27773,
      "translation adversarial": 124630,
      "adversarial text": 4167,
      "selfattention based": 104874,
      "representation unsupervised": 99458,
      "generate distributed": 46930,
      "based decoder": 11624,
      "appropriate decoder": 8416,
      "based adversarial": 11499,
      "loss text": 65301,
      "generation demonstrate": 47359,
      "monolingual baselines": 74550,
      "effectively using": 34860,
      "assessment text": 9530,
      "estimation paper": 38659,
      "estimation method": 38653,
      "user understanding": 129050,
      "understanding evaluation": 126837,
      "process applied": 91422,
      "set english": 107428,
      "arabic texts": 8561,
      "texts exploiting": 121516,
      "gold syntax": 48444,
      "historical languages": 50526,
      "languages generic": 60608,
      "explores possibility": 41649,
      "possibility improving": 88370,
      "different related": 31376,
      "data exploited": 25931,
      "previous experiments": 90409,
      "regard ability": 97649,
      "ability tackle": 1572,
      "joint partofspeech": 57306,
      "parsing shown": 85237,
      "dependency ud": 29251,
      "treebanks including": 125661,
      "dialogues grounded": 30836,
      "construct dialogue": 21627,
      "discourse units": 32099,
      "conversational utterances": 23044,
      "utterances dataset": 130633,
      "datasets dataset": 27391,
      "dataset covers": 26834,
      "informationseeking conversations": 55121,
      "introduce multiple": 56463,
      "multiple dialogue": 75538,
      "processing model": 91712,
      "detection language": 29982,
      "modelling partofspeech": 72614,
      "utterance segmentation": 130615,
      "provide positive": 93894,
      "delivers competitive": 28623,
      "beneficial future": 12947,
      "use conversational": 127966,
      "collection written": 18500,
      "teachers learners": 119701,
      "linguistic abilities": 64417,
      "mandarin english": 66914,
      "offer computational": 81562,
      "work corpus": 134446,
      "corpus referring": 23963,
      "offer indepth": 81564,
      "discuss shortcomings": 32289,
      "surprising results": 116130,
      "english morphologically": 37212,
      "wordlevel translation": 133759,
      "structured way": 114052,
      "basic linguistic": 12528,
      "linguistic intuition": 64497,
      "lexical unit": 63836,
      "additional syntactic": 3578,
      "leads substantial": 61971,
      "accuracy language": 2192,
      "highlight issues": 50263,
      "issues associated": 57038,
      "improving task": 53175,
      "dataset multidomain": 27040,
      "focused summaries": 44877,
      "summaries based": 115430,
      "specific points": 111477,
      "large differences": 61079,
      "aspects different": 9380,
      "domains sentiment": 33855,
      "summarization specifically": 115568,
      "using section": 130134,
      "section titles": 104490,
      "annotation propose": 6360,
      "existing summarization": 40304,
      "models face": 73228,
      "web texts": 132265,
      "texts obtained": 121563,
      "rules sentence": 103440,
      "texts perform": 121576,
      "annotation sentence": 6373,
      "obtains highest": 81467,
      "dataset sentence": 27186,
      "mentioned texts": 68440,
      "high variation": 50146,
      "names used": 76152,
      "names occur": 76149,
      "paper ask": 83741,
      "truth data": 125880,
      "data future": 25966,
      "challenge baseline": 15999,
      "evaluation strategy": 39407,
      "baseline suggest": 12318,
      "suggest improvements": 115314,
      "improvements future": 52858,
      "theories cognitive": 121809,
      "inference graph": 54149,
      "feature mapping": 43295,
      "representing lexical": 100054,
      "base paper": 11465,
      "contribute research": 22759,
      "experiments systematically": 41168,
      "featurebased approach": 43333,
      "questions hierarchical": 95314,
      "systems generative": 116921,
      "seq2seq problem": 106908,
      "train dialog": 122925,
      "meaningful utterance": 67732,
      "conversation level": 22959,
      "2016 proposed": 615,
      "based dialog": 11644,
      "question ask": 95129,
      "propose generalized": 92699,
      "generalized framework": 46828,
      "framework hierarchical": 45559,
      "hierarchical encoder": 49959,
      "encoder including": 36533,
      "demonstrate hierarchical": 28752,
      "systems wide": 117234,
      "range experiments": 95568,
      "experiments answer": 40773,
      "answer identification": 6536,
      "approach answer": 7354,
      "different locations": 31238,
      "questions critical": 95296,
      "critical work": 24842,
      "work efficiency": 134485,
      "realtime search": 96138,
      "structural lexical": 113778,
      "model fits": 71193,
      "density estimation": 29033,
      "tagging empirical": 117388,
      "solution outperforms": 110560,
      "dataset tracking": 27239,
      "entities open": 37831,
      "text arbitrary": 120658,
      "arbitrary domains": 8580,
      "example text": 39798,
      "formulations task": 45298,
      "entity attribute": 37906,
      "state values": 112539,
      "judged humans": 57432,
      "stateoftheart generation": 112663,
      "bleu metric": 14293,
      "leaving room": 63268,
      "architectures learning": 8813,
      "training thousands": 123921,
      "humans solve": 51111,
      "introduce framework": 56424,
      "developing nlp": 30355,
      "framework new": 45628,
      "unseen tasks": 127548,
      "descriptions questions": 29491,
      "systematic generalization": 116679,
      "achieves score": 2851,
      "leaving significant": 63270,
      "multivocab pretraining": 75938,
      "pretraining despite": 90252,
      "tasks vocabulary": 119599,
      "vocabulary chinese": 131868,
      "bert citedevlin2018bert": 13088,
      "based chinese": 11571,
      "characters second": 16620,
      "form vocabulary": 45145,
      "bert help": 13139,
      "experiments compared": 40835,
      "performance especially": 86343,
      "largescale pretraining": 61498,
      "learning utilize": 63152,
      "unlabeled indomain": 127397,
      "context largescale": 22162,
      "lm pretraining": 64870,
      "pretraining make": 90279,
      "make best": 66625,
      "best use": 13467,
      "poorly understood": 88072,
      "semisupervised model": 105616,
      "model actually": 70622,
      "comprehensive studies": 20271,
      "classification context": 17158,
      "indomain pretraining": 53979,
      "pretraining strategy": 90329,
      "performance boosts": 86190,
      "better smaller": 13722,
      "largest performance": 61537,
      "labels original": 58626,
      "original dataset": 82510,
      "accuracy 50": 2065,
      "50 training": 1077,
      "dataset competitive": 26808,
      "performance 966": 86106,
      "understanding behavior": 126803,
      "study detecting": 114353,
      "factors factors": 42888,
      "english literature": 37196,
      "set patterns": 107525,
      "methods embedding": 69465,
      "entities typically": 37885,
      "evaluated downstream": 38972,
      "intrinsically using": 56374,
      "interpret differences": 56207,
      "task structure": 118745,
      "evaluations look": 39467,
      "models address": 72721,
      "issues evaluating": 57046,
      "entities learn": 37811,
      "learn type": 62174,
      "framework entity": 45523,
      "linking tasks": 64674,
      "datasets predictions": 27624,
      "investigate helpful": 56761,
      "pretraining step": 90327,
      "making best": 66825,
      "produced users": 91980,
      "predict pseudo": 88920,
      "use pseudo": 128220,
      "labels supervise": 58646,
      "training step": 123885,
      "dataset relatively": 27152,
      "pretraining selftraining": 90319,
      "wmt20 paper": 132867,
      "submission wmt20": 114744,
      "multilingual approach": 75197,
      "model agglutinative": 70645,
      "tokenization quality": 122280,
      "evaluation line": 39256,
      "spaces variational": 111109,
      "variational autoencoding": 130920,
      "making process": 66860,
      "entire vocabulary": 37723,
      "deal large": 27861,
      "latent distribution": 61584,
      "explore ways": 41616,
      "task captures": 117952,
      "data labels": 26066,
      "labels approach": 58580,
      "topology word": 122687,
      "space argue": 110972,
      "based topology": 12122,
      "motivated solution": 74861,
      "semeval2010 task": 105522,
      "simple scalable": 109512,
      "learning platform": 62894,
      "applications literature": 6957,
      "success leveraging": 115092,
      "algorithms wide": 4891,
      "develop deep": 30189,
      "algorithms make": 4867,
      "strategies combined": 113454,
      "compiler optimization": 19720,
      "including mainstream": 53321,
      "features various": 43781,
      "applications toolkit": 7027,
      "convenient users": 22871,
      "evaluation online": 39315,
      "business scenarios": 14982,
      "scenarios including": 103853,
      "item recommendation": 57106,
      "answering extensive": 6628,
      "datasets online": 27603,
      "express opinion": 41703,
      "emotion sentiment": 36113,
      "tweets facebook": 126031,
      "analysis deals": 5559,
      "sentiment reviews": 106790,
      "service providers": 107325,
      "classify reviews": 17658,
      "negative polarities": 76940,
      "reviews constructed": 102601,
      "technique machine": 119800,
      "models instead": 73407,
      "stateoftheart benchmarks": 112600,
      "try understand": 125900,
      "knowledge stored": 58184,
      "stored parameters": 113391,
      "generation extensive": 47406,
      "containing knowledge": 21797,
      "constraints report": 21610,
      "spoken conversation": 111969,
      "based general": 11734,
      "response generators": 101216,
      "sources knowledge": 110904,
      "retrieval methods": 102408,
      "architecture perform": 8725,
      "perform analysis": 85944,
      "participated 2019": 85334,
      "studies carried": 114191,
      "philosophy language": 87268,
      "literary studies": 64737,
      "quality challenging": 94609,
      "ongoing efforts": 81736,
      "analysis literary": 5683,
      "literary analysis": 64734,
      "requires analysis": 100245,
      "embeddings various": 36014,
      "performed human": 86925,
      "vocabulary learning": 131882,
      "linguistic methods": 64508,
      "exercise generation": 39985,
      "models answer": 72754,
      "model concepts": 70879,
      "practice using": 88736,
      "generated book": 47058,
      "approach offers": 7753,
      "adaptation pretrained": 3250,
      "pretrained crosslingual": 90019,
      "pretraining crosslingual": 90243,
      "various crosslingual": 131068,
      "crosslingual lowresource": 24971,
      "texts crosslingual": 121488,
      "effective leveraging": 34698,
      "leveraging highresource": 63680,
      "crosslingual crossdomain": 24935,
      "setting pretrained": 107772,
      "domains specifically": 33860,
      "decomposition method": 28158,
      "domainspecific features": 33900,
      "features domaininvariant": 43465,
      "texts source": 121615,
      "tasks studies": 119528,
      "creation lexical": 24719,
      "languages create": 60473,
      "create novel": 24633,
      "technique creating": 119778,
      "augment replace": 10264,
      "ability present": 1560,
      "aims improving": 4541,
      "interactive ebooks": 56018,
      "aims develop": 4518,
      "develop illustrate": 30206,
      "aim predict": 4469,
      "predict message": 88901,
      "feature subsets": 43320,
      "attain better": 9720,
      "performance focusing": 86390,
      "information topic": 55047,
      "topic received": 122564,
      "sequences patterns": 107132,
      "tags specific": 117483,
      "specific patterns": 111474,
      "enhancing deep": 37535,
      "nlp ability": 79558,
      "extract informative": 42089,
      "techniques designed": 119867,
      "applied english": 7071,
      "stems words": 113241,
      "approaches explicitly": 8156,
      "word morphology": 133366,
      "features lstm": 43600,
      "lstm bert": 65602,
      "testbed use": 120558,
      "parsing dp": 85104,
      "baselines involving": 12417,
      "tags universal": 117486,
      "universal features": 127307,
      "features compare": 43411,
      "task features": 118206,
      "lstmbased models": 65683,
      "models ner": 73623,
      "benefit performance": 12986,
      "task bertbased": 117932,
      "quality showing": 94790,
      "improvements pronounced": 52902,
      "bert variants": 13250,
      "variants compared": 130874,
      "datasets manually": 27559,
      "informationseeking conversation": 55119,
      "applications especially": 6922,
      "ecommerce companies": 34504,
      "retrieve appropriate": 102450,
      "responses users": 101302,
      "compute matching": 20467,
      "matching degrees": 67402,
      "users queries": 129161,
      "historical dialogue": 50519,
      "dialogue utterances": 30814,
      "based heuristic": 11750,
      "heuristic rules": 49876,
      "require heavy": 100143,
      "suitable solving": 115408,
      "task alleviate": 117871,
      "problem treat": 91271,
      "propose reinforced": 93014,
      "annotations specifically": 6466,
      "reinforced selector": 97788,
      "performance ranker": 86651,
      "serves reward": 107318,
      "datasets prove": 27648,
      "term selector": 120243,
      "methods case": 69363,
      "capable selecting": 15219,
      "production ecommerce": 92055,
      "form basis": 45070,
      "reasoning decisionmaking": 96243,
      "exploit large": 41425,
      "available today": 11128,
      "emerged significant": 36053,
      "challenge recent": 16094,
      "years existing": 135261,
      "approaches realm": 8305,
      "relations individual": 98189,
      "inference causal": 54121,
      "end leverage": 36815,
      "superiority approach": 115699,
      "approaches mainly": 8226,
      "based extraction": 11711,
      "data did": 25839,
      "training overcome": 123768,
      "learning address": 62312,
      "address tasks": 3773,
      "tasks adaptive": 118918,
      "integrates types": 55787,
      "types causality": 126258,
      "interview data": 56328,
      "data computational": 25763,
      "topical clustering": 122590,
      "provide convenient": 93793,
      "generate test": 47031,
      "compare textual": 19304,
      "studies social": 114283,
      "analysis especially": 5600,
      "research requires": 100616,
      "study built": 114331,
      "traditional natural": 122848,
      "processing mechanisms": 91710,
      "resolution application": 100753,
      "investigate challenge": 56730,
      "challenge processing": 16087,
      "data comprehensively": 25760,
      "investigation language": 56863,
      "model interpretability": 71362,
      "plms like": 87756,
      "bert used": 13246,
      "remain largely": 98770,
      "editing dataset": 34544,
      "extracted model": 42165,
      "model rationales": 71848,
      "new testbed": 79218,
      "systematic investigation": 116680,
      "methods different": 69447,
      "generates new": 47167,
      "understanding attention": 126797,
      "tasks benefited": 118964,
      "encode knowledge": 36436,
      "commonsense factual": 18984,
      "questions covering": 95295,
      "linguistic world": 64582,
      "knowledge chinese": 57825,
      "pretrained chinese": 90002,
      "build machine": 14783,
      "literary domain": 64735,
      "nmt transformer": 79997,
      "assess extent": 9477,
      "domainspecific systems": 33916,
      "systems recurrent": 117104,
      "required make": 100219,
      "outperformed systems": 82794,
      "evaluations conducted": 39456,
      "cases large": 15653,
      "aspectbased sentiment": 9355,
      "dataset bengali": 26765,
      "baseline evaluation": 12214,
      "online product": 81791,
      "reviews comments": 102599,
      "increase domain": 53596,
      "considerable number": 21254,
      "number datasets": 80863,
      "analysis absa": 5481,
      "associated sentiment": 9607,
      "speakers dataset": 111311,
      "neutral data": 78755,
      "news portals": 79356,
      "addition conducted": 3412,
      "aspect term": 9348,
      "extraction accuracy": 42247,
      "dataset cnn": 26786,
      "better terms": 13741,
      "outperforms cnn": 82864,
      "average f1score": 11191,
      "largescale generative": 61440,
      "tasks challenging": 118974,
      "challenging training": 16344,
      "data largest": 26078,
      "largest chinese": 61527,
      "facilitate downstream": 42767,
      "essay generation": 38529,
      "settings fewshot": 107801,
      "denoising pretraining": 29001,
      "growth popularity": 49194,
      "way store": 132134,
      "facts large": 42918,
      "representation makes": 99321,
      "difficult humans": 31631,
      "challenge aims": 15993,
      "score seen": 104119,
      "seen categories": 104530,
      "entities unseen": 37886,
      "unseen categories": 127517,
      "know language": 57732,
      "models know": 73438,
      "shown language": 108488,
      "knowledge regarding": 58137,
      "fail provide": 42971,
      "appropriate answers": 8412,
      "ask question": 9298,
      "confidence answer": 20983,
      "models t5": 74151,
      "finding answer": 44265,
      "scores correlate": 104163,
      "range datasets": 95553,
      "systems aim": 116726,
      "aim understand": 4474,
      "naturally adapt": 76636,
      "interactions motivated": 55991,
      "concepts directly": 20613,
      "users using": 129181,
      "uses models": 129246,
      "identify gaps": 51505,
      "gaps understanding": 46489,
      "interactions learn": 55990,
      "interpretations unknown": 56279,
      "specifically tailored": 111592,
      "propose stateoftheart": 93081,
      "models accuracy": 72654,
      "components demonstrate": 20023,
      "leading way": 61915,
      "way build": 132062,
      "personalized language": 87144,
      "treestructured decoding": 125722,
      "accuracy standard": 2284,
      "rare complex": 95738,
      "types long": 126314,
      "structure including": 113879,
      "prediction best": 89038,
      "best tagger": 13462,
      "capable recovering": 15217,
      "prior state": 90730,
      "approaches generalize": 8170,
      "generalize outofdomain": 46817,
      "intercultural communication": 56053,
      "communication languages": 19037,
      "continuously changing": 22650,
      "changing words": 16403,
      "words shift": 134213,
      "linguistics paper": 64620,
      "cognate sets": 18259,
      "languages extended": 60577,
      "requiring large": 100344,
      "corpora involved": 23506,
      "involved languages": 56882,
      "dictionary pair": 30887,
      "method facilitates": 68837,
      "pairs provides": 83619,
      "false friends": 43039,
      "friends second": 45892,
      "second contribution": 104397,
      "notion soft": 80411,
      "pair additionally": 83419,
      "enhanced event": 37506,
      "types event": 126281,
      "wordtrigger mismatch": 134344,
      "injecting word": 55265,
      "word information": 133320,
      "information characterlevel": 54413,
      "ignore semantic": 51672,
      "transform sentence": 124255,
      "sentence graph": 105895,
      "margin loss": 67197,
      "range competitive": 95549,
      "methods fewshot": 69510,
      "fewshot event": 43899,
      "novel event": 80565,
      "tagging problem": 117439,
      "fewshot scenario": 43912,
      "distribution introduced": 32661,
      "insufficient data": 55720,
      "results unified": 102292,
      "unified models": 127120,
      "dataset interpersonal": 26984,
      "ability human": 1527,
      "interpersonal relationship": 56199,
      "relationship language": 98299,
      "focuses relation": 44912,
      "task relation": 118623,
      "relation labels": 98046,
      "challenging existing": 16251,
      "dataefficient methods": 26684,
      "everyday life": 39629,
      "systems requires": 117116,
      "smaller data": 110230,
      "methods end": 69474,
      "outline steps": 82634,
      "steps obtain": 113329,
      "transfer network": 124163,
      "dialogue representations": 30731,
      "representations hybrid": 99680,
      "task address": 117847,
      "input present": 55397,
      "focus social": 44820,
      "3rd place": 998,
      "2017 2018": 625,
      "dialogue length": 30703,
      "ranking objective": 95678,
      "counterpart terms": 24437,
      "form unstructured": 45140,
      "text notes": 121151,
      "set tags": 107598,
      "using world": 130390,
      "based keywords": 11785,
      "concepts present": 20633,
      "tags used": 117487,
      "used summarize": 128793,
      "desired information": 29665,
      "information stored": 55005,
      "stored form": 113388,
      "text proposed": 121220,
      "ondevice mobile": 81668,
      "conceptnet resource": 20599,
      "tags given": 117475,
      "parsing codeswitching": 85084,
      "world especially": 135028,
      "like india": 64049,
      "languages number": 60767,
      "bilingual speakers": 14060,
      "codeswitched data": 18216,
      "respect various": 101110,
      "structure english": 113851,
      "data apart": 25608,
      "datasets showcase": 27705,
      "showcase capabilities": 108361,
      "generated codeswitched": 47060,
      "using qualitative": 130071,
      "metrics providing": 69998,
      "tasks modeling": 119304,
      "recommendation dialogue": 97085,
      "systems expected": 116888,
      "humans machines": 51090,
      "estimate users": 38637,
      "result paper": 101390,
      "design response": 29562,
      "changes using": 16396,
      "fully endtoend": 45955,
      "dialog session": 30584,
      "finetuning large": 44471,
      "unidirectional language": 127094,
      "realistic setting": 96104,
      "access user": 1925,
      "combined score": 18667,
      "generation policy": 47546,
      "optimization endtoend": 82190,
      "training sequence": 123842,
      "provide visualization": 93956,
      "complicate task": 19953,
      "copy single": 23292,
      "handwritten text": 49432,
      "investigation data": 56860,
      "progress datadriven": 92145,
      "moving away": 74920,
      "performance dialogue": 86292,
      "interaction task": 55970,
      "task collected": 117971,
      "shared task1": 108109,
      "classification relation": 17374,
      "processing introduce": 91685,
      "contextualized knowledge": 22552,
      "graph completion": 48769,
      "classify relations": 17657,
      "text environment": 120918,
      "able effectively": 1633,
      "effectively extract": 34813,
      "generating semantic": 47258,
      "semantic maps": 105104,
      "multidimensional scaling": 75070,
      "linguistic applications": 64426,
      "theory paper": 121823,
      "scaling mds": 103774,
      "techniques create": 119856,
      "maps linguistic": 67165,
      "combination parallel": 18574,
      "variation introduce": 130894,
      "past research": 85650,
      "frameworks finally": 45745,
      "dialects spoken": 30540,
      "normalization method": 80337,
      "method different": 68776,
      "dialects covering": 30536,
      "reported earlier": 99061,
      "earlier research": 34379,
      "gave best": 46557,
      "available model": 11048,
      "study provides": 114494,
      "baselines study": 12472,
      "dependencies natural": 29097,
      "unannotated corpora": 126584,
      "syntactic abstractions": 116362,
      "abstractions representations": 1799,
      "hand lack": 49321,
      "apply novel": 7201,
      "graph encoders": 48802,
      "explicitly incorporate": 41373,
      "benchmark approach": 12731,
      "translation encoderdecoder": 124787,
      "nmt methods": 79892,
      "multiple passes": 75636,
      "issue present": 57020,
      "involves multiple": 56899,
      "passes pass": 85624,
      "rewriting process": 102695,
      "improve general": 52387,
      "previous baselines": 90389,
      "make models": 66700,
      "successful english": 115159,
      "computational limitations": 20389,
      "method overcome": 69037,
      "languages specifically": 60888,
      "adaptation english": 3221,
      "layers result": 61807,
      "result obtain": 101387,
      "embeddings additionally": 35545,
      "scale complexity": 103706,
      "gpt2 small": 48571,
      "training prevents": 123786,
      "losing information": 65242,
      "gpt2 models": 48565,
      "generate realistic": 46996,
      "generated gpt2": 47081,
      "trained scratch": 123267,
      "effectiveness language": 34900,
      "token representations": 122263,
      "information handle": 54651,
      "currently common": 25402,
      "common language": 18888,
      "predicts word": 89226,
      "framework decomposes": 45481,
      "creation annotated": 24711,
      "describes simple": 29431,
      "translation design": 124744,
      "algorithm generating": 4746,
      "generating artificial": 47200,
      "artificial parallel": 9265,
      "learning sequencetosequence": 63014,
      "model annotate": 70664,
      "sentences diverse": 106284,
      "set domains": 107423,
      "shows higher": 108585,
      "compared offtheshelf": 19403,
      "scenarios training": 103868,
      "mt reached": 74987,
      "languages documentlevel": 60513,
      "evaluate main": 38850,
      "evaluation address": 39115,
      "mt fails": 74966,
      "fails produce": 42983,
      "produce adequate": 91873,
      "translations lack": 125471,
      "lack context": 58695,
      "systems train": 117201,
      "context leads": 22163,
      "improving zero": 53183,
      "shot learning": 108356,
      "baselines commonsense": 12370,
      "disjoint set": 32366,
      "consisting human": 21452,
      "used facilitate": 128547,
      "visual semantic": 131810,
      "relations nodes": 98222,
      "graph generate": 48814,
      "embeddings class": 35593,
      "surpass strong": 116097,
      "embeddings existing": 35705,
      "matters multilingual": 67492,
      "modeling analysis": 72374,
      "extend studies": 41799,
      "compile larger": 19712,
      "missing typological": 70336,
      "typological data": 126483,
      "strategies like": 113474,
      "finitestate transducers": 44551,
      "strategies yield": 113496,
      "languages morphology": 60740,
      "task consisting": 118009,
      "consisting assigning": 21447,
      "categories textual": 15756,
      "despite increasing": 29697,
      "increasing use": 53688,
      "exploited word": 41456,
      "embeddings inherently": 35763,
      "words contexts": 133882,
      "distributions words": 32777,
      "categories classification": 15731,
      "embeddings incorporating": 35756,
      "class distributions": 17035,
      "novel weighting": 80776,
      "performance scores": 86692,
      "scores existing": 104172,
      "gap increases": 46457,
      "data grows": 25992,
      "alignment aims": 4942,
      "aims align": 4498,
      "equivalent words": 38257,
      "sentences plays": 106445,
      "current unsupervised": 25385,
      "neural alignment": 77828,
      "does leverage": 33365,
      "leverage context": 63579,
      "sequence paper": 107026,
      "target model": 117674,
      "twostep process": 126174,
      "process based": 91429,
      "assumption source": 9665,
      "aligned introduce": 4917,
      "attention variant": 10033,
      "previous unsupervised": 90513,
      "architecture achieved": 8604,
      "hurts generalization": 51170,
      "generalization unseen": 46799,
      "unseen inputs": 127527,
      "pairs negative": 83589,
      "negative pairs": 76939,
      "easily distinguishable": 34446,
      "correct output": 24113,
      "especially models": 38478,
      "corpora generating": 23490,
      "positive examples": 88320,
      "requires domainspecific": 100265,
      "heuristics generalize": 49883,
      "domains tackle": 33868,
      "learning seq2seq": 63009,
      "specifically generate": 111556,
      "examples adding": 39806,
      "adding small": 3393,
      "perturbations input": 87196,
      "enforcing high": 37008,
      "guides model": 49269,
      "better distinguish": 13565,
      "form sentences": 45122,
      "understanding dialog": 126826,
      "context difficult": 22062,
      "context generate": 22114,
      "response paper": 101224,
      "answer reading": 6570,
      "comprehension question": 20221,
      "omitted information": 81654,
      "information dialog": 54481,
      "scheme propose": 103935,
      "joint framework": 57277,
      "framework unifies": 45728,
      "features better": 43383,
      "information question": 54892,
      "question dialog": 95149,
      "encoding propose": 36718,
      "architecture memory": 8693,
      "history dialog": 50551,
      "annotators write": 6501,
      "conducted dataset": 20917,
      "brings substantial": 14653,
      "reasoning help": 96251,
      "intent slot": 55909,
      "labels noisy": 58624,
      "including spoken": 53380,
      "sensitive quality": 105744,
      "consistency training": 21369,
      "inference conditions": 54127,
      "performance slu": 86723,
      "training resources": 123819,
      "types natural": 126325,
      "conventional finetuning": 22878,
      "method modelagnostic": 68968,
      "performance variation": 86835,
      "present fake": 89490,
      "networks nlp": 77678,
      "presents solutions": 89900,
      "analyze tweets": 6011,
      "related covid19": 97849,
      "different solutions": 31437,
      "covid19 related": 24573,
      "treat task": 125553,
      "task ternary": 118782,
      "f1score 606": 42711,
      "set respectively": 107566,
      "obtained average": 81347,
      "set sentiment": 107577,
      "analysis bengali": 5515,
      "bert sentiment": 13225,
      "indoaryan languages": 53951,
      "forms noun": 45254,
      "24 different": 799,
      "harder paper": 49488,
      "sa datasets": 103519,
      "demonstrate multilingual": 28803,
      "model relevant": 71896,
      "trained approach": 123069,
      "learning novel": 62852,
      "novel datasets": 80531,
      "accuracy 68": 2071,
      "accuracy 60": 2068,
      "model analyze": 70662,
      "article comments": 9117,
      "blackbox nature": 14252,
      "nature neural": 76662,
      "research aims": 100402,
      "reason natural": 96199,
      "predictions relying": 89188,
      "investigate multiple": 56785,
      "neural knowledge": 77933,
      "tasks train": 119568,
      "train generative": 122938,
      "unseen instances": 127528,
      "task inference": 118294,
      "introduced model": 56584,
      "likely given": 64138,
      "jointly predicting": 57382,
      "suggesting important": 115347,
      "direction writing": 31831,
      "makes comparison": 66765,
      "similarities different": 109178,
      "writings existing": 135111,
      "task writing": 118857,
      "polish texts": 87985,
      "containing million": 21800,
      "context experimental": 22092,
      "filling blanks": 44034,
      "methods address": 69296,
      "costly work": 24395,
      "formulate novel": 45277,
      "problem design": 91005,
      "problem high": 91074,
      "relevant descriptions": 98538,
      "conducted realworld": 20938,
      "score 9090": 104050,
      "complex patterns": 19857,
      "according recent": 1997,
      "frequently cooccurring": 45870,
      "information argue": 54380,
      "cooccurring patterns": 23247,
      "model lead": 71422,
      "add regularization": 3352,
      "model rely": 71898,
      "information efficiently": 54515,
      "sentence frequently": 105881,
      "frequently cooccur": 45869,
      "performances downstream": 86879,
      "deep transformerbased": 28432,
      "face difficulties": 42731,
      "performance cost": 86261,
      "sequence text": 107087,
      "par deep": 84520,
      "lower cost": 65427,
      "tradeoff performance": 122779,
      "performance gain": 86400,
      "cost models": 24366,
      "especially underresourced": 38521,
      "complex interaction": 19822,
      "convey information": 23114,
      "environment study": 38192,
      "efficacy different": 35011,
      "forms communication": 45235,
      "ones language": 81692,
      "implications human": 51993,
      "interactive question": 56028,
      "asking questions": 9308,
      "questions common": 95286,
      "human interaction": 50868,
      "intents user": 55927,
      "propose reinforcement": 93015,
      "problem select": 91211,
      "phrases user": 87462,
      "user query": 129026,
      "policy network": 87962,
      "realworld user": 96189,
      "different experiments": 31139,
      "better interaction": 13608,
      "interaction users": 55974,
      "say write": 103673,
      "successful interaction": 115160,
      "interaction user": 55973,
      "framework consisting": 45466,
      "humancomputer interaction": 51006,
      "benchmark binary": 12736,
      "architectures best": 8786,
      "small manually": 110170,
      "architectures used": 8856,
      "performance retrievalbased": 86684,
      "recently previous": 96735,
      "matching degree": 67401,
      "decisions models": 27962,
      "models access": 72650,
      "comparison information": 19550,
      "module retrievalbased": 74510,
      "extensive experiment": 41879,
      "module effectively": 74486,
      "effectively boosts": 34792,
      "developed participation": 30293,
      "consists subtasks": 21497,
      "subtasks task": 114986,
      "task identify": 118263,
      "domain given": 33541,
      "second task": 104459,
      "science domain": 103974,
      "developed perform": 30294,
      "model obtained": 71601,
      "obtained f1": 81370,
      "bilstm approach": 14090,
      "achieved higher": 2639,
      "voice assistant": 131923,
      "rewriting qr": 102697,
      "used reduce": 128720,
      "caused errors": 15831,
      "pipeline underlying": 87558,
      "pairs hard": 83557,
      "propose augmentation": 92564,
      "patterns existing": 85742,
      "training pairs": 123769,
      "approach goes": 7592,
      "methods constrained": 69399,
      "predefined patterns": 88828,
      "effectiveness compared": 34877,
      "compared fully": 19373,
      "baseline demonstrate": 12209,
      "subword sampling": 115031,
      "area nlp": 8877,
      "designed high": 29598,
      "millions parallel": 70119,
      "sentences available": 106225,
      "thousands sentences": 121929,
      "aggregation different": 4357,
      "different granularities": 31168,
      "wordlevel alignment": 133726,
      "propose iterative": 92730,
      "outperforms wordlevel": 83045,
      "languages supervision": 60907,
      "results observe": 102001,
      "existing wordlevel": 40335,
      "methods pretraining": 69682,
      "study intrinsic": 114410,
      "intrinsic nature": 56366,
      "different transformerbased": 31513,
      "transformerbased masked": 124421,
      "certain features": 15943,
      "finetune language": 44406,
      "models glue": 73304,
      "unstructured data": 127570,
      "trained directly": 123118,
      "model acquire": 70620,
      "transferred natural": 124231,
      "data gives": 25982,
      "close performance": 17826,
      "present datadriven": 89429,
      "datadriven endtoend": 26661,
      "discourse features": 32056,
      "qualitative human": 94564,
      "cost effective": 24351,
      "actions trained": 3095,
      "set models": 107499,
      "analysis clearly": 5529,
      "human ability": 50734,
      "draw meaningful": 34117,
      "meaningful conclusions": 67713,
      "incomplete inconsistent": 53421,
      "convert sentences": 23096,
      "representation additional": 99162,
      "systems does": 116857,
      "questions instead": 95319,
      "knowledge derived": 57856,
      "paper effective": 83886,
      "effective procedure": 34733,
      "general question": 46701,
      "demonstrate practical": 28822,
      "practical usefulness": 88722,
      "information contextual": 54449,
      "interaction information": 55948,
      "information key": 54706,
      "consider important": 21204,
      "simultaneously paper": 109687,
      "propose cointeractive": 92583,
      "jointly perform": 57377,
      "tasks core": 119015,
      "proposed cointeractive": 93240,
      "graph interaction": 48823,
      "iteratively updated": 57149,
      "consider types": 21236,
      "successfully captures": 115181,
      "information achieve": 54355,
      "contributions contextual": 22785,
      "representations bert": 99520,
      "roberta xlnet": 103002,
      "tamil paper": 117571,
      "tools resources": 122472,
      "phases improve": 87215,
      "morphological annotations": 74689,
      "based stanza": 12067,
      "data shows": 26447,
      "assigned score": 9550,
      "best achieved": 13301,
      "viable approach": 131642,
      "label smoothing": 58415,
      "second nmt": 104433,
      "word end": 133261,
      "types target": 126365,
      "transformer recently": 124372,
      "improvements neural": 52880,
      "paper natural": 84051,
      "learning strong": 63061,
      "model shallow": 71999,
      "shallow model": 107922,
      "benchmarks validate": 12939,
      "model loss": 71477,
      "method randomly": 69091,
      "embarrassingly simple": 35345,
      "extraction predict": 42437,
      "predict relation": 88923,
      "dialogue paper": 30717,
      "dialogue novel": 30715,
      "input format": 55343,
      "bert relation": 13216,
      "token sequence": 122266,
      "possible relations": 88423,
      "different pairs": 31316,
      "manner experiments": 66949,
      "projection knowledge": 92232,
      "coupled training": 24460,
      "tries mimic": 125780,
      "results setting": 102163,
      "final predictions": 44120,
      "components motivated": 20034,
      "student teacher": 114146,
      "relies attention": 98642,
      "glue tasks": 48318,
      "results combinatorial": 101586,
      "retrieval existing": 102392,
      "suffer shortcomings": 115244,
      "task setup": 118704,
      "english question": 37256,
      "answering model": 6646,
      "analyze methods": 5985,
      "queries english": 94919,
      "answers target": 6745,
      "product search": 92046,
      "strongest baselines": 113740,
      "demonstrates particular": 28952,
      "challenging settings": 16312,
      "languages extensive": 60578,
      "retrieval analysis": 102380,
      "answer generation": 6532,
      "colleagues proposed": 18370,
      "multiple scales": 75677,
      "framework set": 45679,
      "attention scholars": 9999,
      "methods machine": 69605,
      "document analysis": 32950,
      "used scientific": 128736,
      "challenges remain": 16203,
      "work offers": 134660,
      "learning conditional": 62453,
      "conditional masked": 20764,
      "method conditional": 68722,
      "unlabeled corpora": 127376,
      "sentences english": 106292,
      "supervised signals": 115839,
      "domains multilingual": 33822,
      "margin 10": 67183,
      "explore language": 41553,
      "language bias": 58862,
      "approach remove": 7855,
      "language identifying": 59151,
      "semantics understanding": 105476,
      "understanding improving": 126860,
      "improving lexical": 53109,
      "essential training": 38572,
      "models reducing": 73898,
      "reducing complexity": 97408,
      "model alleviate": 70656,
      "data end": 25891,
      "divergence term": 32785,
      "model embedded": 71052,
      "architectures demonstrate": 8798,
      "effectiveness universality": 34971,
      "universality proposed": 127339,
      "reducing lexical": 97425,
      "respectively source": 101166,
      "models cascade": 72888,
      "dynamic early": 34304,
      "early exiting": 34389,
      "accelerate inference": 1868,
      "empirically analyze": 36215,
      "working mechanism": 134901,
      "mechanism dynamic": 67970,
      "performance bottleneck": 86191,
      "representations shallow": 99886,
      "shallow layers": 107920,
      "predictions hand": 89174,
      "manner providing": 66960,
      "providing comprehensive": 94107,
      "mechanism experimental": 67981,
      "15 improvement": 348,
      "4times speedup": 1068,
      "methods classification": 69368,
      "code summarization": 18154,
      "area recent": 8879,
      "learn structural": 62157,
      "lstm shown": 65667,
      "shown performance": 108507,
      "code release": 18143,
      "mechanism extensive": 67984,
      "finding models": 44278,
      "trained random": 123255,
      "scenarios propose": 103860,
      "framework trains": 45720,
      "mismatching information": 70313,
      "studies benchmark": 114188,
      "way annotate": 132058,
      "semantics texts": 105474,
      "role labelling": 103196,
      "languages added": 60389,
      "new interesting": 78961,
      "interesting challenges": 56069,
      "dataset generating": 26954,
      "sections datasets": 104492,
      "typically focus": 126430,
      "work cast": 134409,
      "wikipedia sections": 132674,
      "task create": 118034,
      "contains millions": 21830,
      "shows best": 108555,
      "quality texts": 94811,
      "potential dataset": 88545,
      "inspire future": 55561,
      "important text": 52279,
      "processing pipelines": 91771,
      "automatic language": 10573,
      "hindi magahi": 50467,
      "magahi maithili": 66372,
      "languages lexical": 60687,
      "level finally": 63449,
      "identification systems": 51448,
      "qa research": 94519,
      "explaining predictions": 41276,
      "gaining momentum": 46377,
      "studies employ": 114216,
      "studies measure": 114253,
      "accept reject": 1886,
      "systems answer": 116731,
      "answer unlike": 6587,
      "retrieved evidence": 102460,
      "evidence passages": 39657,
      "calibrated confidence": 15044,
      "failure cases": 42986,
      "endtoend evaluation": 36899,
      "modalities different": 70478,
      "nlp significant": 79689,
      "prominent approaches": 92250,
      "linear projection": 64364,
      "structures work": 114119,
      "type structural": 126228,
      "relevant dimensions": 98540,
      "method novel": 68993,
      "sentence jointly": 105911,
      "makes structural": 66808,
      "probes vulnerable": 90886,
      "use beam": 127918,
      "search neural": 104331,
      "model synthetic": 72131,
      "synthetic real": 116641,
      "event temporal": 39545,
      "models ptlms": 73839,
      "tasks struggle": 119526,
      "reasoning essential": 96248,
      "pretraining approach": 90233,
      "information yielding": 55112,
      "yielding enhanced": 135377,
      "framework event": 45531,
      "performances relation": 86892,
      "design principles": 29560,
      "seemingly simple": 104528,
      "process text": 91575,
      "challenges order": 16188,
      "account specifics": 2040,
      "develop realistic": 30228,
      "datatotext applications": 27810,
      "texts sentences": 121603,
      "applications available": 6899,
      "representation pretrained": 99382,
      "capturing implicit": 15469,
      "implicit language": 52013,
      "features pretraining": 43666,
      "approaches focus": 8164,
      "employs multiple": 36331,
      "multiple sentencelevel": 75687,
      "strategies order": 113481,
      "reasons make": 96330,
      "numerous experiments": 81019,
      "observe different": 81193,
      "pretraining lead": 90275,
      "multiple existing": 75562,
      "models extractive": 73224,
      "nl explanations": 79508,
      "explanations model": 41295,
      "large blackbox": 61041,
      "fact verification": 42837,
      "sequence seq2seq": 107049,
      "models shortcomings": 74022,
      "incorrect predictions": 53579,
      "predictions difficult": 89163,
      "difficult adapt": 31608,
      "input documents": 55324,
      "training requires": 123812,
      "intermediate finetuning": 56131,
      "improves prior": 53031,
      "settings neural": 107824,
      "resources tools": 101056,
      "processing aims": 91610,
      "aims translate": 4581,
      "computers recent": 20512,
      "years endtoend": 135257,
      "practical mt": 88707,
      "methods nmt": 69643,
      "useful researchers": 128925,
      "finally conclude": 44157,
      "sampleefficient pretraining": 103575,
      "task efficiently": 118132,
      "efficiently learning": 35151,
      "token replacements": 122262,
      "distinguish true": 32578,
      "true input": 125850,
      "tokens replaced": 122330,
      "replaced generator": 98934,
      "generator network": 47776,
      "hand current": 49317,
      "pretraining masked": 90280,
      "develop arabic": 30175,
      "multiple arabic": 75495,
      "stateoftheart arabic": 112585,
      "modeling understanding": 72577,
      "given trained": 48162,
      "generation arabic": 47312,
      "nlp advances": 79559,
      "primarily lack": 90642,
      "corpus internet": 23844,
      "parameters makes": 84768,
      "largest arabic": 61525,
      "success different": 115077,
      "including synthetic": 53391,
      "synthetic news": 116634,
      "generation zeroshot": 47703,
      "zeroshot question": 135519,
      "achieves perplexity": 2833,
      "conducted human": 20930,
      "significant success": 108871,
      "articles difficult": 9184,
      "98 percent": 1448,
      "percent accuracy": 85918,
      "matching best": 67399,
      "length propose": 63376,
      "method inspired": 68911,
      "splitting text": 111957,
      "pass experimental": 85605,
      "text average": 120679,
      "text tokenization": 121368,
      "nonautoregressive generation": 80154,
      "autoregressive ar": 10859,
      "nonautoregressive nar": 80159,
      "structure largescale": 113893,
      "overall scores": 83255,
      "scores squad": 104207,
      "respectively addition": 101121,
      "systems adopt": 116724,
      "work inspired": 134572,
      "success language": 115083,
      "pretraining present": 90304,
      "improvements wmt": 52939,
      "wmt dataset": 132783,
      "dataset 10": 26703,
      "10 language": 137,
      "surprisingly method": 116143,
      "monolingual performance": 74605,
      "models versus": 74330,
      "set typologically": 107622,
      "pretrained monolingual": 90158,
      "models set": 74013,
      "aim establish": 4454,
      "gap multilingual": 46468,
      "language exists": 59026,
      "performance difference": 86293,
      "new monolingual": 79039,
      "monolingually multilingually": 74634,
      "role downstream": 103172,
      "performance decreases": 86275,
      "aligning crosslingual": 4935,
      "semantics monolingual": 105435,
      "achieve impressive": 2487,
      "tasks improvement": 119172,
      "corpora especially": 23470,
      "method encourages": 68805,
      "model align": 70652,
      "representation multiple": 99346,
      "pretraining process": 90307,
      "process generate": 91494,
      "pseudoparallel sentence": 94193,
      "corpus enable": 23768,
      "semantic alignments": 104974,
      "enhancing semantic": 37539,
      "semantic modeling": 105112,
      "existing crosslingual": 40097,
      "delivers new": 28624,
      "increasing memory": 53675,
      "memory time": 68389,
      "time consumption": 122000,
      "problem lead": 91101,
      "lead inferior": 61860,
      "modeling capability": 72385,
      "comparable model": 19150,
      "documentlevel language": 33148,
      "mechanism enhanced": 67978,
      "mechanism enable": 67973,
      "context length": 22165,
      "explicitly learn": 41375,
      "improved stateoftheart": 52647,
      "answering understanding": 6719,
      "contextualized discourse": 22540,
      "discourse processing": 32073,
      "events arguments": 39563,
      "contexts various": 22434,
      "fairly consistent": 42995,
      "spite recent": 111937,
      "advances pretrained": 4011,
      "representations designed": 99591,
      "propose compositional": 92589,
      "reader model": 95954,
      "capture leverage": 15326,
      "information generate": 54635,
      "generate effective": 46936,
      "process documents": 91462,
      "documents generate": 33233,
      "analysis representations": 5806,
      "sequences models": 107129,
      "capture temporal": 15410,
      "relationships events": 98321,
      "propose single": 93071,
      "model addresses": 70635,
      "events order": 39589,
      "predicting new": 88997,
      "new events": 78912,
      "tasks space": 119509,
      "events attempt": 39564,
      "recover original": 97141,
      "event sequence": 39541,
      "make inferences": 66687,
      "knowledge events": 57908,
      "data outperforming": 26198,
      "pairwise model": 83676,
      "model bertbased": 70767,
      "network event": 77247,
      "compared gpt2": 19375,
      "story completion": 113409,
      "completion models": 19793,
      "annotation manual": 6334,
      "ucca abend": 126523,
      "abend rappoport": 1475,
      "rappoport 2013": 95735,
      "typological linguistic": 126488,
      "languages ease": 60519,
      "english new": 37220,
      "2020 using": 739,
      "transformers particular": 124458,
      "module employ": 74487,
      "number students": 80973,
      "exploit interaction": 41421,
      "interaction knowledge": 55949,
      "learned transformer": 62270,
      "thoroughly examine": 121902,
      "models just": 73435,
      "just relying": 57469,
      "increasing input": 53674,
      "modeling transformers": 72575,
      "efficiency improvements": 35027,
      "improvements new": 52883,
      "initially training": 55249,
      "model short": 72004,
      "time surprisingly": 122116,
      "models condition": 72952,
      "tokens generating": 122311,
      "length transformer": 63380,
      "simple alternative": 109358,
      "embeddings efficiently": 35677,
      "short input": 108213,
      "dataset diverse": 26877,
      "diverse text": 32856,
      "modeling recent": 72521,
      "increased training": 53633,
      "mind present": 70145,
      "existing newly": 40235,
      "academic professional": 1861,
      "academic writing": 1864,
      "improve significantly": 52539,
      "downstream evaluations": 33998,
      "available code": 10960,
      "discourse work": 32102,
      "work modelling": 134637,
      "modelling spoken": 72624,
      "speech paper": 111719,
      "expert human": 41222,
      "varying language": 131260,
      "proficiency levels": 92096,
      "covid19 fake": 24562,
      "detection recent": 30034,
      "led great": 63275,
      "false information": 43040,
      "information fake": 54612,
      "reduce spread": 97360,
      "information shared": 54977,
      "bert albert": 13062,
      "evaluated context": 38964,
      "2021 shared": 741,
      "task covid19": 118030,
      "english obtained": 37228,
      "ranked 5th": 95639,
      "inference question": 54205,
      "answering benchmark": 6605,
      "diverse nlu": 32829,
      "covering text": 24547,
      "understanding benchmark": 126804,
      "models making": 73551,
      "improving models": 53117,
      "examples useful": 39896,
      "useful analysis": 128858,
      "analysis training": 5906,
      "current generation": 25285,
      "manual labor": 67008,
      "paraphrases word": 84844,
      "allows control": 5136,
      "trained finetuning": 123145,
      "finetuning gpt2": 44462,
      "augmenting stateoftheart": 10345,
      "based ner": 11883,
      "work represent": 134780,
      "lexicon information": 63889,
      "information chinese": 54415,
      "natural manner": 76612,
      "using special": 130203,
      "require word": 100208,
      "experiments ontonotes": 41047,
      "words does": 133925,
      "words stored": 134241,
      "dynamic global": 34310,
      "global attention": 48226,
      "entire input": 37708,
      "input predict": 55395,
      "predict masked": 88900,
      "model multidocument": 71534,
      "multidocument setting": 75076,
      "structured nlp": 114020,
      "classification explicitly": 17205,
      "based constituency": 11598,
      "augmented dataset": 10325,
      "performance training": 86808,
      "training original": 123767,
      "performance investigated": 86471,
      "seed dataset": 104499,
      "nonenglish data": 80195,
      "arabic varieties": 8566,
      "targeting different": 117778,
      "different task": 31476,
      "rich conditions": 102731,
      "majority tasks": 66609,
      "xlmr large": 135204,
      "zeroshot domain": 135497,
      "exploits different": 41490,
      "domain labels": 33564,
      "use particular": 128194,
      "exploit knowledge": 41424,
      "encoded different": 36474,
      "task formulations": 118226,
      "stateoftheart english": 112656,
      "used way": 128847,
      "resources nlp": 101022,
      "parsing objective": 85178,
      "design based": 29521,
      "reduce dependency": 97323,
      "nlp provides": 79674,
      "finally create": 44163,
      "label words": 58421,
      "learning fewshot": 62580,
      "rc task": 95874,
      "relation information": 98041,
      "text generate": 120973,
      "generate structured": 47021,
      "enable neural": 36359,
      "new knowledge": 78970,
      "like humans": 64048,
      "humans work": 51120,
      "classifier generalize": 17546,
      "number samples": 80960,
      "better feature": 13586,
      "instance propose": 55605,
      "class prototype": 17047,
      "adaptive mixture": 3334,
      "mixture mechanism": 70422,
      "measure distances": 67792,
      "introduce loss": 56451,
      "learning encode": 62542,
      "manner extensive": 66950,
      "conducted fewrel": 20928,
      "different fewshot": 31148,
      "learning achieved": 62305,
      "achieved significant": 2693,
      "accuracy increased": 2185,
      "rc models": 95872,
      "detection shared": 30050,
      "task poses": 118537,
      "challenge classifying": 16007,
      "fake real": 43021,
      "features ngrams": 43630,
      "experiment various": 40519,
      "various steps": 131207,
      "stop word": 113372,
      "based linear": 11806,
      "linear svm": 64378,
      "soft keyboard": 110493,
      "detecting language": 29842,
      "respective language": 101114,
      "present fast": 89492,
      "fast lightweight": 43142,
      "intended language": 55867,
      "regression based": 97693,
      "identify language": 51515,
      "present unique": 89752,
      "method reducing": 69097,
      "reducing inference": 97422,
      "parameter reduction": 84718,
      "resolve ambiguity": 100798,
      "languages latin": 60680,
      "spite advances": 111935,
      "advances domain": 3988,
      "word completion": 132959,
      "prediction nwp": 89093,
      "strategies generate": 113466,
      "linear respect": 64369,
      "improvement language": 52717,
      "languages computational": 60461,
      "available development": 10981,
      "tools paper": 122464,
      "reviews collected": 102598,
      "ecommerce platforms": 34507,
      "lowlevel features": 65473,
      "features character": 43392,
      "ngrams based": 79461,
      "highlevel features": 50245,
      "obtaining better": 81436,
      "generated clusters": 47059,
      "embeddings closer": 35596,
      "law distribution": 61682,
      "language demonstrate": 58941,
      "demonstrate generalization": 28747,
      "capacity using": 15237,
      "commonsense causal": 18982,
      "requires complex": 100250,
      "complex inference": 19819,
      "inference ability": 54109,
      "train large": 122945,
      "task scarce": 118662,
      "instability model": 55589,
      "presents number": 89886,
      "techniques making": 119929,
      "firstly perform": 44565,
      "training generating": 123640,
      "theory discourse": 121816,
      "perform data": 85977,
      "discourse parser": 32069,
      "generating distractors": 47214,
      "boost model": 14432,
      "dataset balanced": 26756,
      "balanced copa": 11402,
      "version original": 131614,
      "datasets small": 27717,
      "points text": 87879,
      "aims increase": 4542,
      "coverage input": 24510,
      "task trains": 118802,
      "original examples": 82522,
      "augmented samples": 10334,
      "common data": 18864,
      "paper devoted": 83871,
      "detection challenge": 29892,
      "problem fake": 91054,
      "necessary create": 76754,
      "tools allow": 122427,
      "allow identify": 5083,
      "create model": 24627,
      "real fake": 96064,
      "news context": 79322,
      "context covid19": 22044,
      "allowed achieve": 5104,
      "dataset targeting": 27232,
      "obtained translating": 81425,
      "having parallel": 49564,
      "parallel inference": 84668,
      "allows compare": 5132,
      "perform baseline": 85951,
      "stress tests": 113582,
      "main finding": 66421,
      "challenging english": 16247,
      "results stress": 102214,
      "models dont": 73108,
      "systematic studies": 116686,
      "mechanism machine": 68010,
      "relations context": 98129,
      "models insufficient": 73408,
      "utilizing dependency": 130569,
      "realworld benchmark": 96153,
      "tracking using": 122766,
      "using metalearning": 129877,
      "forms core": 45239,
      "systems designed": 116841,
      "increasing need": 53678,
      "knowledge resourcerich": 58153,
      "domains minimal": 33819,
      "dst problem": 34209,
      "problem extensive": 91049,
      "evidence benefits": 39637,
      "domains methods": 33818,
      "datasets significant": 27710,
      "lowdata setting": 65407,
      "proposed metalearner": 93332,
      "strategy makes": 113527,
      "makes good": 66785,
      "lots attention": 65336,
      "learning ability": 62301,
      "examples work": 39899,
      "investigate effective": 56747,
      "effective strategies": 34751,
      "random sampling": 95507,
      "augment largescale": 10260,
      "random baseline": 95473,
      "datasets yield": 27804,
      "gains observed": 46401,
      "tasks tabletotext": 119543,
      "nq dataset": 80806,
      "hope investigation": 50629,
      "investigation help": 56861,
      "understand behaviors": 126737,
      "enhance fewshot": 37469,
      "networks transformers": 77802,
      "corpus fail": 23800,
      "fail represent": 42972,
      "represent linguistic": 99118,
      "monolingual hindi": 74590,
      "conducted text": 20952,
      "datasets hindi": 27504,
      "hindi dependency": 50463,
      "performance contextualized": 86255,
      "stateoftheart publicly": 112875,
      "like text": 64102,
      "language automatic": 58851,
      "punctuation restoration": 94396,
      "used benchmark": 128408,
      "annotating training": 6274,
      "learning open": 62860,
      "annotation budget": 6282,
      "thoroughly investigate": 121903,
      "uncertainty estimation": 126618,
      "estimation methods": 38654,
      "models active": 72709,
      "best combinations": 13320,
      "distilled version": 32526,
      "better computational": 13546,
      "computational performance": 20418,
      "performance reduces": 86662,
      "deep active": 28195,
      "learning practice": 62898,
      "practice word": 88737,
      "embeddings parallel": 35854,
      "alignment parallel": 4979,
      "lexicons crosslingual": 63923,
      "translation outputs": 125083,
      "great majority": 49008,
      "performing unsupervised": 86979,
      "recently work": 96780,
      "attractive alternative": 10169,
      "absence explicit": 1716,
      "training parallel": 123774,
      "approaches leveraging": 8217,
      "designed improve": 29602,
      "proposing methods": 93623,
      "obtain robust": 81316,
      "encoders pretrained": 36661,
      "systems aims": 116727,
      "realworld settings": 96179,
      "dst systems": 34210,
      "systems generalize": 116916,
      "domains unseen": 33880,
      "types existing": 126284,
      "new slot": 79139,
      "queries unseen": 94941,
      "based generative": 11735,
      "questionanswering using": 95259,
      "improves joint": 52988,
      "accuracy zeroshot": 2321,
      "adaptation settings": 3266,
      "21 dataset": 759,
      "dataset multilingual": 27041,
      "transformers convolutional": 124447,
      "present transfer": 89746,
      "identification multilingual": 51401,
      "uses transformer": 129293,
      "models allowed": 72744,
      "identification ranked": 51429,
      "domain terms": 33672,
      "terms extraction": 120322,
      "extraction subtask": 42498,
      "phrases given": 87429,
      "focuses unsupervised": 44917,
      "domainspecific terms": 33918,
      "modeling unstructured": 72579,
      "unstructured knowledge": 127573,
      "knowledge access": 57738,
      "systems restricted": 117121,
      "restricted limited": 101341,
      "coverage domain": 24502,
      "define tasks": 28495,
      "different largescale": 31218,
      "model detecting": 70980,
      "internet social": 56194,
      "developed machine": 30285,
      "texts generate": 121522,
      "blocks text": 14365,
      "writing prompts": 135095,
      "look like": 65215,
      "text line": 121092,
      "able detect": 1629,
      "arabic sentence": 8549,
      "sentence written": 106133,
      "based tweets": 12142,
      "using twitter": 130326,
      "twitter api": 126095,
      "arabic sentences": 8550,
      "compared different": 19357,
      "different recurrent": 31374,
      "lstm bilstm": 65605,
      "gru bigru": 49199,
      "obtained accuracy": 81340,
      "texts training": 121638,
      "bytelevel subwords": 15002,
      "fundamental components": 46092,
      "components pretrained": 20044,
      "languages technical": 60914,
      "practices training": 88742,
      "bytelevel bpe": 15001,
      "outperforms google": 82908,
      "margin multilingual": 67198,
      "tasks release": 119452,
      "building tools": 14894,
      "dialog length": 30569,
      "matter response": 67490,
      "release bert": 98435,
      "bert multilingual": 13184,
      "based architectures": 11522,
      "architectures little": 8816,
      "tasks dialog": 119056,
      "tasks inherently": 119208,
      "novel twostage": 80758,
      "sentences news": 106411,
      "framework extract": 45540,
      "probability sentence": 90868,
      "sentences treated": 106531,
      "set manually": 107487,
      "method actually": 68613,
      "underlying structure": 126704,
      "article extracting": 9131,
      "supporting facts": 116030,
      "opinions article": 82116,
      "spoken communication": 111964,
      "environmental noise": 38196,
      "variability speakers": 130833,
      "robust spoken": 103074,
      "evaluate candidate": 38805,
      "candidate interpretations": 15106,
      "probabilistic generative": 90806,
      "human linguistic": 50908,
      "spoken utterances": 112021,
      "models yielded": 74368,
      "abstract representations": 1780,
      "best predict": 13413,
      "words utterance": 134299,
      "terminology integration": 120259,
      "work bilingual": 134400,
      "apriori specified": 8479,
      "forms unrealistic": 45262,
      "practical applicability": 88699,
      "work work": 134882,
      "words target": 134256,
      "trained augmented": 123072,
      "uralic languages": 127837,
      "evaluation indicate": 39242,
      "term translation": 120246,
      "develop process": 30227,
      "novel interactive": 80603,
      "keeps track": 57523,
      "models finding": 73249,
      "popular way": 88142,
      "documents rich": 33288,
      "rich visual": 102800,
      "visual markup": 131798,
      "presents challenge": 89829,
      "annotating documents": 6266,
      "documents difficult": 33220,
      "tool designed": 122378,
      "scenarios annotators": 103840,
      "extended context": 41805,
      "training multimodal": 123722,
      "power language": 88632,
      "data understanding": 26593,
      "policy changes": 87949,
      "allocation topic": 5071,
      "twitter dataset": 126107,
      "dataset study": 27221,
      "shows possible": 108608,
      "split conversation": 111940,
      "moving average": 74919,
      "gives promising": 48204,
      "mean squared": 67610,
      "squared error": 112083,
      "data respectively": 26355,
      "respectively understanding": 101174,
      "methods density": 69431,
      "gru based": 49198,
      "models aspectbased": 72773,
      "aspect level": 9337,
      "focuses english": 44896,
      "work arabic": 134377,
      "arabic based": 8493,
      "based regular": 11986,
      "tools analyzing": 122430,
      "advantage word": 4064,
      "representations combining": 99548,
      "cnn conditional": 17996,
      "opinionated aspects": 82113,
      "interactive attention": 56013,
      "identify sentiment": 51550,
      "extracted aspects": 42140,
      "baseline research": 12295,
      "research tasks": 100637,
      "combining deep": 18719,
      "areas nlp": 8897,
      "potentially effective": 88609,
      "effective tackling": 34756,
      "data shortage": 26444,
      "pipeline document": 87536,
      "baselines semisupervised": 12459,
      "semisupervised classification": 105595,
      "settings languages": 107815,
      "nlp annotations": 79564,
      "easily distributed": 34447,
      "27 million": 852,
      "healthcare organizations": 49633,
      "translation documentlevel": 124763,
      "approaches literature": 8221,
      "phenomena evaluation": 87223,
      "backtranslating monolingual": 11329,
      "data performs": 26233,
      "evaluation deep": 39173,
      "features multilingual": 43614,
      "investigate multilingual": 56784,
      "alignment different": 4953,
      "languages define": 60488,
      "mbert embeddings": 67565,
      "embeddings transitive": 35991,
      "evaluate zeroshot": 38946,
      "resulting classifier": 101433,
      "rely features": 98695,
      "classification decisions": 17174,
      "mbert does": 67564,
      "does encode": 33344,
      "embedding continuous": 35390,
      "dependent semantic": 29265,
      "discourse factors": 32055,
      "factors proposed": 42899,
      "literature results": 64770,
      "entity alignment": 37904,
      "model xlmroberta": 72333,
      "entities detected": 37766,
      "english parallel": 37235,
      "approach target": 7961,
      "compared recent": 19430,
      "benchmark chinese": 12739,
      "analyses language": 5457,
      "improvement models": 52727,
      "semiautomatically generated": 105574,
      "phenomena models": 87231,
      "fillergap dependencies": 44029,
      "dependencies overall": 29102,
      "chance level": 16352,
      "level neural": 63478,
      "graphs sentence": 48972,
      "list sentences": 64700,
      "order different": 82311,
      "different distances": 31098,
      "encoded graph": 36476,
      "networks fused": 77605,
      "representations finally": 99656,
      "determined using": 30139,
      "networks integrate": 77628,
      "sentence content": 105808,
      "parsing lowresource": 85147,
      "supervision simple": 115918,
      "direct transfer": 31807,
      "transfer learned": 124073,
      "benchmark paper": 12846,
      "transfer systems": 124190,
      "implicit supervision": 52021,
      "unlabelled text": 127418,
      "language method": 59279,
      "method assumes": 68648,
      "parser assuming": 84921,
      "access source": 1922,
      "language significant": 60091,
      "despite conceptually": 29680,
      "conceptually simpler": 20682,
      "provide analyses": 93756,
      "choice source": 16905,
      "transfer advantage": 124034,
      "model nonnative": 71592,
      "word processing": 133422,
      "processing studies": 91805,
      "encoding words": 36736,
      "lexical memory": 63783,
      "train computational": 122912,
      "model exhibits": 71119,
      "discrimination tasks": 32204,
      "model spoken": 72077,
      "effects observed": 35000,
      "additional analysis": 3488,
      "lexical representation": 63797,
      "space showing": 111061,
      "showing training": 108433,
      "languages fully": 60602,
      "languages bilingual": 60430,
      "lags far": 58786,
      "lack highquality": 58716,
      "introduces task": 56624,
      "dst module": 34208,
      "settings including": 107810,
      "efficient chinese": 35063,
      "chinese named": 16788,
      "base bert": 11446,
      "bert syntactic": 13237,
      "use results": 128250,
      "error paper": 38312,
      "paper uses": 84487,
      "process nlp": 91539,
      "nlp tool": 79783,
      "annotate largescale": 6130,
      "reduce dependence": 97322,
      "new nlp": 79058,
      "finally model": 44206,
      "evaluated based": 38955,
      "improves test": 53053,
      "based attention": 11530,
      "classification area": 17124,
      "relatively rare": 98409,
      "number network": 80921,
      "texts great": 121524,
      "great practical": 49013,
      "disaster relief": 32004,
      "fields use": 44007,
      "function graph": 46019,
      "construct chinese": 21621,
      "classification multiple": 17293,
      "language transmission": 60190,
      "using artificial": 129347,
      "process using": 91586,
      "explicit reasoning": 41341,
      "various question": 131179,
      "need strong": 76860,
      "neural symbolic": 78693,
      "endtoend reinforcement": 36959,
      "framework discrete": 45499,
      "answer matching": 6541,
      "model genbert": 71223,
      "trained comparable": 123096,
      "weak supervised": 132182,
      "discrete reasoning": 32173,
      "use composition": 127947,
      "functions make": 46077,
      "improvements parsing": 52893,
      "concerns small": 20693,
      "relations including": 98188,
      "fighting covid19": 44012,
      "outbreak covid19": 82613,
      "available people": 11073,
      "transformerbased pretrained": 124435,
      "additional layers": 3540,
      "ensemble classifier": 37581,
      "f1score test": 42719,
      "large ontology": 61196,
      "extraction given": 42350,
      "document input": 33025,
      "textual entity": 121704,
      "events doing": 39576,
      "achieve use": 2586,
      "xlmr conneau": 135200,
      "leveraging multilingual": 63690,
      "combining pretrained": 18745,
      "nlp benchmarks": 79582,
      "information semantics": 54963,
      "variety approaches": 130964,
      "knowledge current": 57847,
      "structured unstructured": 114049,
      "resolution cdcr": 100755,
      "concepts text": 20645,
      "type news": 126215,
      "desirable perform": 29659,
      "concepts scientific": 20642,
      "mainstream media": 66499,
      "media propose": 68157,
      "dataset crossdocument": 26839,
      "identify links": 51518,
      "links entities": 64680,
      "heterogeneous document": 49854,
      "document types": 33100,
      "crossdocument setting": 24883,
      "set annotation": 107360,
      "development science": 30418,
      "subject matter": 114679,
      "peer reviews": 85844,
      "papers generate": 84506,
      "comprehensive experimental": 20261,
      "reviews tend": 102633,
      "aspects paper": 9400,
      "reviews generated": 102610,
      "text suffer": 121340,
      "suffer lower": 115230,
      "factually correct": 42942,
      "potential solutions": 88584,
      "code dataset": 18089,
      "data certain": 25722,
      "certain number": 15949,
      "manually collect": 67043,
      "human correction": 50788,
      "know correct": 57729,
      "evaluation achieves": 39114,
      "sentiment annotations": 106683,
      "annotations finegrained": 6434,
      "attempts extract": 9769,
      "holders targets": 50591,
      "polar expressions": 87899,
      "task focusing": 118224,
      "sentiment targets": 106802,
      "improve target": 52560,
      "extraction classification": 42280,
      "datasets conclude": 27370,
      "target polarity": 117690,
      "augmenting input": 10341,
      "text gold": 121029,
      "sentiment datasets": 106728,
      "datasets time": 27761,
      "benefit information": 12973,
      "information practice": 54853,
      "lack adequate": 58677,
      "difficult access": 31606,
      "community members": 19081,
      "processing use": 91848,
      "aligned normalized": 4918,
      "classification generation": 17217,
      "problem information": 91084,
      "uses lstm": 129242,
      "lstm obtain": 65657,
      "leakage problem": 61980,
      "problem happens": 91073,
      "problem showed": 91229,
      "techniques useful": 120010,
      "iii additional": 51719,
      "features effective": 43472,
      "transformer consider": 124317,
      "protests news": 93672,
      "articles dataset": 9180,
      "semiautomated data": 105567,
      "local news": 64931,
      "news sources": 79369,
      "events structured": 39601,
      "data understand": 26592,
      "understand temporal": 126779,
      "known challenges": 58293,
      "challenges nlp": 16184,
      "domain detection": 33503,
      "news stories": 79372,
      "dataset news": 27056,
      "dataset benchmark": 26763,
      "processing news": 91724,
      "articles based": 9173,
      "paragraphs sentences": 84579,
      "count number": 24414,
      "improving consistency": 53076,
      "consistency model": 21363,
      "property natural": 92502,
      "study question": 114497,
      "knowledge end": 57894,
      "end create": 36800,
      "contains total": 21840,
      "relations analysis": 98110,
      "structure currently": 113831,
      "suitable representing": 115407,
      "raw audio": 95848,
      "learning acoustic": 62306,
      "acoustic linguistic": 3012,
      "audio text": 10239,
      "text labels": 121075,
      "labels set": 58641,
      "representations acoustic": 99491,
      "consisting discrete": 21450,
      "speech encoder": 111682,
      "trained supervision": 123300,
      "wav2vec 20": 132047,
      "approach textbased": 7974,
      "study conversational": 114342,
      "conversational dialog": 23003,
      "conversation turns": 22982,
      "make multiple": 66701,
      "study dialog": 114356,
      "setting order": 107765,
      "based bipartite": 11560,
      "diverse references": 32835,
      "study multiple": 114444,
      "text human": 121036,
      "human text": 50973,
      "generation measuring": 47469,
      "language remains": 60035,
      "comparison measure": 19553,
      "directly compares": 31867,
      "distribution text": 32683,
      "model distribution": 71012,
      "modern text": 74423,
      "models computing": 72948,
      "space extensive": 111004,
      "properties generated": 92453,
      "existing distributional": 40117,
      "meaning relating": 67666,
      "syntactic contextual": 116388,
      "perform rigorous": 86057,
      "rigorous evaluations": 102854,
      "interpret proposed": 56215,
      "senses words": 105723,
      "work creates": 134448,
      "generation interpretation": 47445,
      "emotion recognition": 36110,
      "model modern": 71526,
      "hebrew text": 49674,
      "outperform alternative": 82686,
      "analysis suggested": 5879,
      "particularly appropriate": 85469,
      "analyzing multiple": 6045,
      "multiple bert": 75507,
      "model complexity": 70867,
      "better captures": 13529,
      "sentiment entire": 106737,
      "bertbased language": 13272,
      "detect polarity": 29814,
      "polarity extract": 87910,
      "extract emotions": 42078,
      "collected annotated": 18400,
      "study data": 114344,
      "collection annotation": 18459,
      "yields high": 135411,
      "classification emotion": 17194,
      "various target": 131212,
      "target emotions": 117613,
      "model failed": 71170,
      "bestreported performance": 13484,
      "performance englishlanguage": 86341,
      "models emotion": 73134,
      "learning select": 62988,
      "answer user": 6588,
      "resources relevant": 101034,
      "queries work": 94946,
      "advanced techniques": 3960,
      "schema guided": 103898,
      "approach ranked": 7834,
      "best human": 13338,
      "bias transfer": 13853,
      "classification classification": 17144,
      "vision cv": 131759,
      "highquality labeled": 50387,
      "training new": 123745,
      "expensive acquire": 40412,
      "feature distribution": 43264,
      "approaches detecting": 8123,
      "including traditional": 53398,
      "traditional deep": 122810,
      "approach bridge": 7397,
      "imbalance issue": 51825,
      "advances transformerbased": 4024,
      "applications dialogue": 6916,
      "humanlike responses": 51056,
      "trained predicting": 123244,
      "relevant informative": 98559,
      "conditioned predicted": 20803,
      "document user": 33104,
      "user interested": 129006,
      "dialogue benchmark": 30645,
      "dialogues propose": 30840,
      "model keeping": 71391,
      "engagement user": 37016,
      "sensitive attributes": 105733,
      "close humans": 17820,
      "model errors": 71093,
      "output beam": 83054,
      "search relies": 104345,
      "limits potential": 64320,
      "restricts applicability": 101351,
      "search paper": 104334,
      "beam decoding": 12597,
      "considered instead": 21293,
      "leads improvement": 61939,
      "points vanilla": 87883,
      "e2e webnlg": 34365,
      "respectively proposed": 101158,
      "method outperformed": 69004,
      "outperformed strong": 82793,
      "e2e challenge": 34357,
      "webnlg dataset": 132284,
      "does fit": 33351,
      "fit finding": 44579,
      "optimal subword": 82172,
      "fasttext models": 43205,
      "languages unsupervised": 60943,
      "disambiguation semantic": 31985,
      "german word": 47927,
      "turkish russian": 125964,
      "coverage model": 24516,
      "14 improvement": 314,
      "replaced simple": 98935,
      "verbal communication": 131533,
      "little resemblance": 64828,
      "recognition understand": 97035,
      "communicative success": 19056,
      "children adults": 16724,
      "discuss consequences": 32242,
      "commonlyused methods": 18977,
      "models acquire": 72707,
      "knowledge pretraining": 58119,
      "tasks finetuning": 119135,
      "acquired language": 3039,
      "questions existing": 95306,
      "focus knowledge": 44777,
      "entities introduce": 37798,
      "probing evaluate": 90890,
      "match words": 67373,
      "poorly new": 88070,
      "new probing": 79084,
      "aims review": 4571,
      "present recent": 89660,
      "nlp challenges": 79584,
      "paragraphs methods": 84578,
      "topological data": 122680,
      "analysis test": 5893,
      "textclassification datasets": 121433,
      "size complexity": 109915,
      "alternative stateoftheart": 5245,
      "processing detect": 91662,
      "explored use": 41636,
      "roberta language": 102998,
      "detection sentencelevel": 30046,
      "paper performed": 84070,
      "performed task": 86934,
      "able slightly": 1689,
      "humanlabeled data": 51047,
      "translation crucial": 124725,
      "crucial machine": 25147,
      "translation needed": 125004,
      "estimation based": 38649,
      "costly human": 24386,
      "human labelled": 50892,
      "data alternative": 25595,
      "technique does": 119783,
      "uses synthetic": 129289,
      "trained humanannotated": 123157,
      "sentence wordlevel": 106130,
      "using dual": 129609,
      "ontology alignment": 81834,
      "inefficient work": 54076,
      "mechanism compute": 67957,
      "able exploit": 1637,
      "exploit syntactic": 41445,
      "languages establish": 60554,
      "detailed ablation": 29750,
      "used available": 128404,
      "efficient retrieval": 35106,
      "retrieval augmented": 102382,
      "ninth dialog": 79491,
      "challenge dstc": 16025,
      "divided subtasks": 32911,
      "selection generation": 104784,
      "compute efficient": 20464,
      "hierarchical classification": 49943,
      "dense knowledge": 29015,
      "knowledge retrieval": 58159,
      "documents method": 33251,
      "reduces computation": 97379,
      "time factor": 122030,
      "use retrieval": 128251,
      "selected snippets": 104734,
      "used finetune": 128556,
      "finetune trained": 44417,
      "building multiturn": 14867,
      "aims simplify": 4574,
      "rely massive": 98723,
      "laborintensive annotate": 58666,
      "contextual query": 22488,
      "tackle issues": 117292,
      "cqr dataset": 24584,
      "annotation introduce": 6327,
      "detection intent": 29974,
      "data visualization": 26622,
      "plan representations": 87630,
      "method contributes": 68741,
      "study used": 114541,
      "synthetic text": 116648,
      "evaluate causal": 38806,
      "data requires": 26349,
      "making assumptions": 66821,
      "research typically": 100657,
      "structured medical": 114016,
      "methods evaluations": 69485,
      "evaluations rely": 39483,
      "rely synthetic": 98750,
      "generation widely": 47699,
      "immediately applicable": 51847,
      "develop framework": 30198,
      "produce synthetic": 91941,
      "methods estimating": 69480,
      "effects text": 35006,
      "propose sequencetosequence": 93042,
      "entity masking": 37967,
      "conditional training": 20791,
      "words pos": 134126,
      "tag generate": 117335,
      "generate syntactically": 47024,
      "syntactically diverse": 116512,
      "indicate method": 53839,
      "strong lstm": 113686,
      "seq2seq baselines": 106892,
      "little pretraining": 64825,
      "goes long": 48425,
      "study dependency": 114349,
      "massive labeled": 67331,
      "data limits": 26092,
      "limits effectiveness": 64317,
      "task morphological": 118407,
      "morphological disambiguation": 74697,
      "lack powerful": 58735,
      "proposed pretraining": 93523,
      "observe average": 81188,
      "distribution natural": 32668,
      "predicted label": 88959,
      "majority label": 66602,
      "introduce small": 56541,
      "small examples": 110149,
      "standard practice": 112291,
      "single reference": 109785,
      "analyses comparing": 5443,
      "pruning results": 94169,
      "models significant": 74035,
      "accuracy cost": 2122,
      "targeted test": 117775,
      "context hierarchical": 22123,
      "global perspective": 48259,
      "redundant information": 97479,
      "context challenge": 22025,
      "words utterances": 134300,
      "context selection": 22255,
      "selection based": 104769,
      "based state": 12068,
      "global guidance": 48240,
      "detect relevant": 29815,
      "translation survey": 125304,
      "massive amounts": 67323,
      "expand current": 40355,
      "current dataset": 25270,
      "dataset generalize": 26950,
      "translate language": 124528,
      "language translations": 60189,
      "metrics text": 70013,
      "embeddings emotion": 35680,
      "years emotion": 135255,
      "text popular": 121181,
      "wide ranging": 132516,
      "great attention": 48994,
      "hinglish dataset": 50499,
      "detecting emotions": 29832,
      "mixed tweets": 70403,
      "derived fasttext": 29349,
      "fasttext word2vec": 43209,
      "word2vec approaches": 133667,
      "including cnns": 53271,
      "lstms bidirectional": 65703,
      "lstms attention": 65702,
      "transformers like": 124452,
      "models giving": 73302,
      "giving best": 48213,
      "learning cl": 62433,
      "works effectiveness": 134941,
      "indepth study": 53801,
      "include task": 53236,
      "tasks lead": 119246,
      "tasks higher": 119158,
      "performance curriculum": 86269,
      "curriculum strategies": 25426,
      "strategies suffer": 113489,
      "suffer catastrophic": 115208,
      "forgetting attention": 45059,
      "task easier": 118126,
      "ensemble learning": 37597,
      "novel ideas": 80597,
      "provide generic": 93837,
      "based area": 11523,
      "submitted shared": 114772,
      "capable classifying": 15195,
      "abstracts scientific": 1826,
      "10 model": 143,
      "tfidf features": 121761,
      "using majority": 129853,
      "gives f1": 48193,
      "sota model": 110691,
      "score validation": 104140,
      "web machine": 132238,
      "industrial setting": 54060,
      "trained general": 123149,
      "data derived": 25830,
      "typically driven": 126424,
      "average domains": 11184,
      "specialized domain": 111387,
      "domain typically": 33685,
      "selecting data": 104749,
      "data similar": 26452,
      "similar target": 109154,
      "document classifiers": 32971,
      "monolingual target": 74620,
      "approach benchmark": 7388,
      "task news": 118457,
      "news domains": 79336,
      "sentence reranking": 106050,
      "selection as2": 104767,
      "as2 models": 9280,
      "modern question": 74412,
      "resource rich": 100875,
      "paper training": 84476,
      "answers multiple": 6733,
      "models select": 73985,
      "select answers": 104689,
      "just language": 57465,
      "multilingual qa": 75345,
      "experiments validate": 41197,
      "documents achieved": 33174,
      "extraction challenging": 42276,
      "finance area": 44250,
      "extraction financial": 42339,
      "documents proposed": 33277,
      "table header": 117255,
      "financial documents": 44254,
      "dataset main": 27009,
      "paper proposing": 84372,
      "extraction pdf": 42430,
      "unseen lexical": 127532,
      "difficult neural": 31647,
      "context context": 22037,
      "meaning conveyed": 67629,
      "used known": 128600,
      "instances context": 55622,
      "unseen cases": 127516,
      "cases learning": 15654,
      "task sufficient": 118760,
      "learning patterns": 62882,
      "focus understanding": 44837,
      "understanding certain": 126809,
      "contexts neural": 22411,
      "models design": 73048,
      "augmentation models": 10287,
      "instance general": 55600,
      "understanding problem": 126927,
      "understand meaning": 126759,
      "provided context": 93959,
      "language necessity": 59709,
      "thesis neural": 121855,
      "contexts enhance": 22390,
      "enhance learning": 37474,
      "impact data": 51864,
      "helps highlight": 49815,
      "networks provides": 77718,
      "designing robust": 29646,
      "models subword": 74118,
      "multiple subwords": 75713,
      "word paper": 133397,
      "morphological probing": 74720,
      "probing pos": 90900,
      "strategies perform": 113483,
      "strategy works": 113546,
      "using fasttext": 129667,
      "urdu urdu": 127843,
      "models exist": 73190,
      "exist english": 40021,
      "language high": 59121,
      "neglected long": 76971,
      "time create": 122007,
      "create efficient": 24611,
      "languages good": 60612,
      "using skipgram": 130184,
      "built corpus": 14916,
      "techniques bilingual": 119846,
      "bilingual language": 14041,
      "modeling transfer": 72573,
      "huge challenge": 50717,
      "challenge multilingual": 16056,
      "property languages": 92500,
      "perform crosslingual": 85976,
      "learning corresponding": 62467,
      "technique termed": 119818,
      "present collection": 89407,
      "sites twitter": 109878,
      "twitter train": 126128,
      "bilingual models": 14048,
      "bilingual model": 14047,
      "23 accuracy": 786,
      "mlm task": 70446,
      "translation dialogue": 124749,
      "translate texts": 124548,
      "level ignoring": 63456,
      "like previous": 64084,
      "gender speaker": 46614,
      "relatively new": 98407,
      "difficult develop": 31620,
      "develop better": 30181,
      "subtitles languages": 114994,
      "systems improving": 116948,
      "improving machine": 53113,
      "automatic deidentification": 10509,
      "different note": 31304,
      "note types": 80387,
      "patient care": 85698,
      "potential facilitate": 88554,
      "patient privacy": 85702,
      "limit use": 64158,
      "use clinical": 127940,
      "information documented": 54506,
      "identifying protected": 51613,
      "protected health": 93660,
      "health information": 49619,
      "information phi": 54839,
      "way improving": 132092,
      "improving access": 53068,
      "deidentification systems": 28585,
      "able consistently": 1624,
      "sources medical": 110908,
      "medical specialties": 68225,
      "present performance": 89642,
      "data external": 25936,
      "available additionally": 10936,
      "need create": 76790,
      "domains combining": 33744,
      "improved training": 52649,
      "model multiwoz": 71551,
      "difficulties annotating": 31677,
      "transcribed spoken": 123986,
      "data fame": 25943,
      "difficulties dataset": 31681,
      "informal nature": 54339,
      "data codeswitching": 25738,
      "codeswitching nonstandard": 18238,
      "annotators annotated": 6487,
      "las points": 61551,
      "solutions proposed": 110582,
      "shown multilingual": 108495,
      "models underperform": 74264,
      "fact training": 42835,
      "costly timeconsuming": 24391,
      "timeconsuming process": 122147,
      "dataset scraped": 27179,
      "tweets containing": 126026,
      "sentences additionally": 106202,
      "model created": 70938,
      "additional pretraining": 3558,
      "bert experiments": 13122,
      "bert produces": 13210,
      "notable performance": 80376,
      "translation reinforcement": 125208,
      "discrepancy loss": 32156,
      "functions used": 46081,
      "final evaluation": 44103,
      "time applied": 121983,
      "function training": 46048,
      "tend sparse": 120162,
      "used reference": 128721,
      "function explore": 46017,
      "function balance": 45998,
      "reward tends": 102679,
      "better outofdomain": 13640,
      "new emerging": 78897,
      "aims categorize": 4509,
      "collected web": 18441,
      "documents document": 33221,
      "specifically jointly": 111561,
      "train modules": 122966,
      "different inductive": 31177,
      "analysis module": 5705,
      "module text": 74511,
      "network learning": 77304,
      "learning module": 62800,
      "datasets challenging": 27343,
      "ecommerce product": 34508,
      "accuracy 92": 2083,
      "automated quality": 10463,
      "typically adopts": 126410,
      "specific dimensions": 111430,
      "complex nature": 19842,
      "highly depend": 50310,
      "robust recognition": 103064,
      "quality assurance": 94598,
      "bertbased model": 13275,
      "called cognitive": 15053,
      "therapy cbt": 121833,
      "rating scale": 95820,
      "leading consistent": 61889,
      "score equal": 104068,
      "task low": 118362,
      "hope speech": 50639,
      "using logistic": 129828,
      "regression random": 97710,
      "forest svm": 45053,
      "svm lstm": 116242,
      "voting ensemble": 131953,
      "obtained finetuning": 81373,
      "layer second": 61745,
      "english tamil": 37299,
      "tamil malayalam": 117569,
      "weighted f1": 132345,
      "ranked english": 95643,
      "information vietnamese": 55088,
      "vietnamese chinese": 131671,
      "vietnamese natural": 131684,
      "reality people": 96111,
      "people tend": 85895,
      "performing word": 86980,
      "segmentation speech": 104635,
      "tagging solely": 117448,
      "process word": 91589,
      "word left": 133335,
      "implement idea": 51934,
      "language employing": 59001,
      "parser neural": 84972,
      "label indicating": 58399,
      "chinese benchmark": 16737,
      "performances previous": 86891,
      "works languages": 134956,
      "text usually": 121401,
      "identification plays": 51422,
      "score test": 104133,
      "outofvocabulary word": 82681,
      "number occurrences": 80928,
      "learn embedding": 62039,
      "gradient steps": 48606,
      "used parameter": 128678,
      "algorithm leverages": 4767,
      "probing classifiers": 90888,
      "analyzing deep": 6037,
      "simple classifier": 109382,
      "linguistic property": 64534,
      "examine wide": 39760,
      "variety models": 130998,
      "intelligence work": 55839,
      "ai machine": 4422,
      "implement word": 51937,
      "word association": 132927,
      "shifts language": 108192,
      "word associations": 132929,
      "studies explored": 114225,
      "methods provide": 69699,
      "pretrained text": 90193,
      "useful multilingual": 128906,
      "nonnative learners": 80246,
      "research past": 100580,
      "multidimensional nature": 75068,
      "single overall": 109778,
      "approaches considers": 8107,
      "classification report": 17378,
      "model seven": 71998,
      "finetuned embeddings": 44427,
      "consistently best": 21410,
      "supplementary material": 115943,
      "learning rich": 62972,
      "applicable multiple": 6826,
      "base completion": 11448,
      "context include": 22134,
      "text related": 121235,
      "related entity": 97859,
      "enables new": 36395,
      "approaches learn": 8211,
      "learn jointly": 62074,
      "strategies compare": 113455,
      "prediction user": 89149,
      "little finetuning": 64807,
      "datasets pretrained": 27626,
      "reviews mcauley": 102618,
      "gender age": 46580,
      "age categories": 4280,
      "examines gender": 39766,
      "features examine": 43494,
      "identify range": 51540,
      "set categories": 107389,
      "female speakers": 43873,
      "male speakers": 66881,
      "speakers feature": 111316,
      "typetoken ratio": 126392,
      "swear words": 116260,
      "young speakers": 135458,
      "data contribute": 25792,
      "classification bidirectional": 17138,
      "cui et": 25223,
      "pretrain language": 89974,
      "model adopted": 70640,
      "chinese texts": 16828,
      "texts categories": 121470,
      "categories containing": 15732,
      "containing descriptions": 21785,
      "improved models": 52617,
      "set derived": 107415,
      "chinese public": 16810,
      "models failed": 73234,
      "brought great": 14711,
      "deal models": 27862,
      "regarding languages": 97657,
      "enhanced set": 37522,
      "14 probing": 318,
      "combination complementary": 18554,
      "probing methods": 90893,
      "methods explore": 69497,
      "understanding linguistic": 126878,
      "knowledge represented": 58148,
      "demonstrate properties": 28830,
      "properties learned": 92458,
      "similar manner": 109109,
      "manner despite": 66942,
      "covid19 infodemic": 24566,
      "using explainable": 129654,
      "extremely high": 42598,
      "combat misinformation": 18543,
      "transformers achieved": 124443,
      "misinformation paper": 70298,
      "based distilbert": 11649,
      "augmenting data": 10340,
      "datasets better": 27336,
      "order boost": 82288,
      "public trust": 94275,
      "model explainability": 71134,
      "colorless green": 18534,
      "english indic": 37173,
      "grammatical genders": 48709,
      "highly inflectional": 50328,
      "morphology paper": 74797,
      "tasks probe": 119400,
      "models varying": 74326,
      "codeswitching setting": 18239,
      "indic language": 53828,
      "languages efficiently": 60522,
      "efficiently highly": 35146,
      "probing experiments": 90891,
      "middle layers": 70071,
      "approach estimate": 7537,
      "assist users": 9570,
      "methods effectiveness": 69463,
      "twitter corpus": 126100,
      "hateful offensive": 49545,
      "join tasks": 57248,
      "detection stance": 30062,
      "opinion expression": 82086,
      "enables analyze": 36374,
      "donald trump": 33941,
      "baseline classifier": 12201,
      "classifier shows": 17580,
      "f1 respectively": 42673,
      "novel resource": 80708,
      "resource computational": 100831,
      "dialogues paper": 30839,
      "supporting development": 116026,
      "informative relevant": 55150,
      "grammar systems": 48659,
      "humanrobot interaction": 51066,
      "knowledge necessary": 58078,
      "necessary condition": 76753,
      "perform pretraining": 86050,
      "good scores": 48501,
      "modeling objectives": 72497,
      "experiments synthetic": 41167,
      "vision recently": 131766,
      "recently used": 96773,
      "methods slot": 69760,
      "method extends": 68832,
      "relation networks": 98057,
      "applications general": 6932,
      "results snips": 102196,
      "translating natural": 124598,
      "power natural": 88638,
      "models transform": 74240,
      "metrics data": 69947,
      "solutions lessons": 110577,
      "related topics": 97908,
      "transition multiple": 124493,
      "normal human": 80328,
      "models comparative": 72933,
      "dataset current": 26841,
      "provide significant": 93917,
      "introducing background": 56628,
      "good benchmark": 48464,
      "benchmark research": 12853,
      "evaluate validity": 38942,
      "systems dataset": 116830,
      "progressive selfsupervised": 92191,
      "selfsupervised attention": 104934,
      "attention learning": 9869,
      "equipped attention": 38239,
      "prediction mechanism": 89076,
      "suffers drawback": 115254,
      "taken consideration": 117506,
      "deal issue": 27859,
      "propose progressive": 93003,
      "absa models": 1713,
      "iteratively perform": 57142,
      "prediction training": 89145,
      "useful attention": 128861,
      "attention supervision": 10019,
      "supervision information": 115891,
      "information meantime": 54760,
      "iteration context": 57120,
      "impact sentiment": 51889,
      "activemisleading influence": 3126,
      "influence correctincorrect": 54305,
      "correctincorrect prediction": 24134,
      "prediction instance": 89070,
      "masked subsequent": 67304,
      "subsequent iterations": 114813,
      "augment conventional": 10251,
      "objective regularization": 81111,
      "extracted active": 42135,
      "active context": 3106,
      "weights misleading": 132379,
      "misleading words": 70302,
      "integrate proposed": 55765,
      "approach stateoftheart": 7933,
      "analyses approach": 5439,
      "enhances performance": 37533,
      "models release": 73914,
      "code trained": 18158,
      "level paper": 63484,
      "approach utilizes": 8010,
      "hybrid neuralsymbolic": 51191,
      "expressed emotion": 41711,
      "affective labels": 4238,
      "labels natural": 58622,
      "leverage pretrained": 63613,
      "processing dependency": 91660,
      "dependency treelstm": 29242,
      "symbolic rules": 116322,
      "rules natural": 103421,
      "language making": 59271,
      "provides higher": 94040,
      "task overcoming": 118497,
      "models depend": 73037,
      "words rarely": 134155,
      "challenging natural": 16279,
      "explore model": 41561,
      "provided natural": 93976,
      "models understanding": 74269,
      "using completely": 129459,
      "emergent communication": 36061,
      "transformations applied": 124277,
      "grammars result": 48681,
      "grammars neural": 48678,
      "learn easily": 62036,
      "current metrics": 25299,
      "networks generalize": 77610,
      "unseen examples": 127526,
      "measure aspects": 67779,
      "networks similar": 77770,
      "data enabling": 25890,
      "arabic dialogue": 8509,
      "building humanlike": 14845,
      "humanlike conversational": 51054,
      "processing seen": 91789,
      "seen significant": 104541,
      "models arabert": 72765,
      "models primarily": 73798,
      "lack arabic": 58683,
      "overcome issue": 83286,
      "propose transformerbased": 93121,
      "weights encoder": 132376,
      "performance response": 86676,
      "generation enable": 47384,
      "empathetic response": 36133,
      "relevant fluent": 98551,
      "arabic msa": 8529,
      "compare use": 19310,
      "features pos": 43660,
      "reaches f1": 95913,
      "bidirectional context": 13903,
      "constrained decoding": 21566,
      "evaluations confirm": 39457,
      "generation allows": 47306,
      "allows explicit": 5152,
      "minimal loss": 70168,
      "humangenerated text": 51024,
      "text methods": 121116,
      "methods flexibly": 69513,
      "demo available": 28639,
      "generation human": 47429,
      "containing information": 21796,
      "used pretrained": 128694,
      "model additional": 70631,
      "additional token": 3584,
      "domainspecific vocabulary": 33921,
      "using rouge": 130126,
      "rouge metric": 103296,
      "comparison human": 19549,
      "human machinegenerated": 50914,
      "contrast original": 22699,
      "research proposes": 100596,
      "promising method": 92280,
      "generation fewshot": 47408,
      "pretraining synthetic": 90330,
      "major issue": 66568,
      "component endtoend": 19976,
      "explicit disentanglement": 41318,
      "traditional twostep": 122882,
      "decoder outperforms": 28039,
      "approach compare": 7431,
      "syntactic ngrams": 116432,
      "generalized phrases": 46831,
      "stateoftheart embeddingbased": 112650,
      "embeddingbased methods": 35535,
      "learning asr": 62372,
      "asr models": 9437,
      "models degrading": 73031,
      "quality lowresource": 94709,
      "lowresource setups": 65566,
      "contrastive representation": 22739,
      "vision speech": 131767,
      "applications recently": 7003,
      "visual objects": 131800,
      "directly predicting": 31899,
      "task transcribing": 118803,
      "media videos": 68181,
      "standard crossentropy": 112218,
      "used annotate": 128388,
      "generalizes better": 46835,
      "progress research": 92182,
      "singleturn dialogue": 109853,
      "performance multiturn": 86555,
      "existing multiturn": 40218,
      "methods low": 69603,
      "predicted results": 88967,
      "predicting current": 88981,
      "framework slu": 45691,
      "obtain contextual": 81271,
      "information multiturn": 54781,
      "history current": 50549,
      "prediction experimental": 89058,
      "models baselines": 72820,
      "slu tasks": 110125,
      "relations biomedical": 98117,
      "knowledgebased approach": 58249,
      "approach combination": 7424,
      "approach outperformed": 7763,
      "outperformed existing": 82781,
      "existing technique": 40310,
      "sentences selfsupervised": 106487,
      "knowledge target": 58201,
      "objective taskspecific": 81121,
      "taskspecific objective": 119637,
      "order tackle": 82417,
      "texttosql task": 121667,
      "table contents": 117254,
      "task unlabeled": 118821,
      "able transfer": 1700,
      "knowledge supervised": 58197,
      "supervised texttosql": 115844,
      "training annotated": 123359,
      "annotated samples": 6226,
      "model leverage": 71455,
      "knowledge better": 57818,
      "code work": 18162,
      "category sentiment": 15792,
      "prediction sentiment": 89123,
      "underlying user": 126708,
      "essential tasks": 38568,
      "potential improve": 88563,
      "usually employed": 130418,
      "employed jointly": 36301,
      "ecommerce scenarios": 34509,
      "scenarios public": 103861,
      "restaurant review": 101323,
      "genuine reviews": 47842,
      "predefined aspect": 88821,
      "hope release": 50637,
      "dataset shed": 27196,
      "propose intuitive": 92729,
      "size task": 109946,
      "task type": 118814,
      "language variants": 60320,
      "finetuning task": 44521,
      "build pretrained": 14801,
      "models variants": 74317,
      "arabic modern": 8525,
      "msa dialectal": 74942,
      "importance pretraining": 52072,
      "size building": 109913,
      "additional models": 3551,
      "studied tasks": 114173,
      "occurring given": 81518,
      "syntactic position": 116450,
      "position different": 88283,
      "universal character": 127275,
      "turn relations": 125975,
      "significant subset": 108870,
      "subset natural": 114833,
      "relations structure": 98262,
      "lowresource multilingual": 65536,
      "translation mnmt": 124936,
      "pairs aid": 83485,
      "search based": 104301,
      "attempt learn": 9749,
      "scratch jointly": 104241,
      "dataset learned": 26998,
      "starting points": 112412,
      "aims classify": 4511,
      "categories given": 15739,
      "category paper": 15788,
      "explores data": 41644,
      "technique particularly": 119807,
      "particularly suitable": 85498,
      "simple training": 109533,
      "dataset paraphrase": 27090,
      "conversation knowledge": 22958,
      "graphs kg": 48961,
      "kg dataset": 57655,
      "using semiautomated": 130143,
      "answering kgs": 6642,
      "focus question": 44807,
      "contains 5000": 21816,
      "minimum maximum": 70210,
      "dataset baseline": 26758,
      "illustrate advantage": 51736,
      "adaptation research": 3258,
      "intelligent tutoring": 55858,
      "student answers": 114137,
      "gains automatic": 46383,
      "neural discourse": 77896,
      "discourse segmentation": 32086,
      "graph discourse": 48794,
      "answers use": 6747,
      "feedback generated": 43828,
      "results highquality": 101828,
      "feedback significantly": 43836,
      "systematic review": 116683,
      "science nlp": 103979,
      "reproducibility results": 100087,
      "area field": 8872,
      "field far": 43952,
      "focused contribution": 44851,
      "categories previous": 15744,
      "usually predict": 130444,
      "using multiclass": 129906,
      "classification despite": 17179,
      "better handle": 13599,
      "correct sequence": 24120,
      "annotations different": 6424,
      "sentence contexts": 105810,
      "generator able": 47764,
      "955 accuracy": 1424,
      "outofdomain texts": 82665,
      "analyses evaluation": 5453,
      "automatic simultaneous": 10654,
      "simultaneous speech": 109662,
      "speechtotext translation": 111858,
      "gained momentum": 46368,
      "intelligence especially": 55829,
      "domains speech": 33861,
      "context evaluation": 22085,
      "evaluation campaigns": 39141,
      "output systems": 83131,
      "performances similar": 86897,
      "quality realtime": 94770,
      "realtime speech": 96141,
      "perform manual": 86030,
      "machine performances": 65845,
      "terms informativeness": 120339,
      "limitations study": 64192,
      "intrinsic limitations": 56364,
      "limitations use": 64196,
      "methodology evaluating": 69260,
      "translation multiview": 124997,
      "subword regularization": 115029,
      "generally rely": 46874,
      "rely subword": 98748,
      "multilingual vocabulary": 75400,
      "algorithms lead": 4862,
      "suboptimal segmentation": 114793,
      "problem demonstrate": 91003,
      "demonstrate empirically": 28736,
      "applying existing": 7245,
      "existing subword": 40303,
      "representations improves": 99685,
      "second advantage": 104382,
      "advantage different": 4040,
      "propose multiview": 92812,
      "consistency predictions": 21364,
      "predictions using": 89196,
      "using inputs": 129760,
      "xtreme multilingual": 135223,
      "brings consistent": 14645,
      "points using": 87882,
      "challenging aspects": 16226,
      "logic formalism": 64989,
      "learned approach": 62199,
      "growing set": 49182,
      "examples results": 39872,
      "task available": 117920,
      "robustly optimized": 103087,
      "learning mtl": 62805,
      "performance generalization": 86412,
      "learns effectively": 63208,
      "effectively multiple": 34834,
      "electra model": 35242,
      "heads conjuncts": 49605,
      "english treebanks": 37318,
      "using heuristic": 129731,
      "heuristic rulebased": 49875,
      "rulebased converter": 103361,
      "core arguments": 23311,
      "graphs identify": 48958,
      "propose propagate": 93007,
      "high interannotator": 50079,
      "new manually": 79000,
      "comparison rulebased": 19571,
      "currently predominant": 25412,
      "interactions label": 55989,
      "joint information": 57282,
      "extraction graph": 42351,
      "extraction mainly": 42383,
      "main tasks": 66468,
      "tasks separately": 119493,
      "detection argument": 29884,
      "jointly performing": 57379,
      "features novel": 43632,
      "level introduce": 63463,
      "interaction graph": 55946,
      "propose dependency": 92622,
      "types expressed": 126288,
      "expressed input": 41715,
      "new regularization": 79101,
      "mechanism introduced": 68000,
      "improve representation": 52522,
      "performance joint": 86474,
      "multilingual learning": 75270,
      "semeval2021 task": 105556,
      "task toxic": 118795,
      "toxic spans": 122711,
      "spans detection": 111198,
      "availability digital": 10906,
      "toxic content": 122710,
      "detection type": 30091,
      "type language": 126208,
      "language paramount": 59800,
      "paramount importance": 84802,
      "traditional bagofwords": 122799,
      "word characters": 132947,
      "wellknown bert": 132417,
      "bagofwords method": 11376,
      "uses new": 129256,
      "neologisms frequently": 77011,
      "observe influence": 81198,
      "influence english": 54309,
      "tool building": 122375,
      "social relationships": 110450,
      "target sequences": 117707,
      "led strong": 63289,
      "length bias": 63354,
      "models high": 73325,
      "cat got": 15686,
      "got tongue": 48535,
      "present possible": 89644,
      "wordlevel tasks": 133758,
      "effectively solve": 34848,
      "model error": 71092,
      "error neural": 38310,
      "regularization technique": 97756,
      "loss models": 65283,
      "art multilingual": 9063,
      "multilingual graphemetophoneme": 75257,
      "crosslingual morphological": 24979,
      "structure classification": 113817,
      "relation classifications": 97960,
      "existing annotations": 40046,
      "properties capturing": 92438,
      "capturing finegrained": 15467,
      "finegrained aspects": 44337,
      "structure events": 113855,
      "india multilingual": 53814,
      "2020 despite": 724,
      "multilingual systems": 75379,
      "leading small": 61909,
      "substantially effective": 114886,
      "effective resourcelean": 34739,
      "resourcelean scenarios": 100901,
      "various nuances": 131159,
      "text transliterated": 121379,
      "codemixed english": 18173,
      "english especially": 37127,
      "especially informal": 38462,
      "example social": 39796,
      "address aforementioned": 3647,
      "gaps propose": 46488,
      "built languages": 14924,
      "monolingual text": 74624,
      "crosslingual signals": 25016,
      "signals training": 108708,
      "outperforms multilingual": 82920,
      "challenging crosslingual": 16237,
      "xtreme benchmark": 135222,
      "data conceptual": 25767,
      "communicative need": 19055,
      "experimental study": 40733,
      "language crosslinguistic": 58929,
      "test recent": 120485,
      "general tendency": 46719,
      "patterns approach": 85730,
      "artificial language": 9258,
      "language communication": 58893,
      "communication game": 19032,
      "need distinguish": 76800,
      "similar pairs": 109117,
      "speakers adjust": 111307,
      "communicative efficiency": 19050,
      "similar meanings": 109111,
      "controlling text": 22861,
      "benefit training": 12992,
      "fluent diverse": 44705,
      "generation takes": 47653,
      "target attributes": 117586,
      "method controlling": 68742,
      "learn alignment": 61987,
      "texts target": 121631,
      "target attribute": 117585,
      "fluency diversity": 44697,
      "quality indepth": 94687,
      "taskoriented visual": 118907,
      "dialogues training": 30845,
      "chosen based": 16939,
      "success popular": 115112,
      "choice prevents": 16898,
      "takes longer": 117536,
      "task comparing": 117987,
      "models playing": 73747,
      "different games": 31159,
      "model taskagnostic": 72145,
      "investigate better": 56728,
      "models increase": 73388,
      "encode decode": 36425,
      "decode words": 27987,
      "frequently training": 45878,
      "set monolingual": 107500,
      "resources corpus": 100958,
      "score statistical": 104127,
      "corpus freely": 23808,
      "available noncommercial": 11057,
      "prediction recurrent": 89116,
      "networks popular": 77698,
      "popular strategy": 88133,
      "strategy train": 113543,
      "inputs training": 55497,
      "impairs ability": 51911,
      "capture longterm": 15341,
      "probability sequence": 90869,
      "words predicted": 134132,
      "convex hull": 23111,
      "possibly unseen": 88450,
      "search experiments": 104315,
      "especially sequencelevel": 38507,
      "sequencelevel metrics": 107109,
      "annotation chinese": 6285,
      "verbal expression": 131534,
      "plays role": 87740,
      "identifying predicate": 51612,
      "critical understanding": 24841,
      "understanding sentence": 126954,
      "leading role": 61903,
      "syntactic elements": 116401,
      "sentence including": 105904,
      "indicate words": 53855,
      "words grammatical": 133976,
      "chinese sentence": 16817,
      "sentence contains": 105807,
      "structure provide": 113938,
      "involves significant": 56902,
      "challenges chinese": 16140,
      "recognition generally": 96885,
      "work important": 134561,
      "attempt develop": 9738,
      "develop annotation": 30173,
      "proposed achieve": 93165,
      "simpler syntactic": 109563,
      "unit type": 127217,
      "type based": 126189,
      "proposed annotation": 93178,
      "community critical": 19069,
      "resource lacking": 100852,
      "leveraging sentence": 63703,
      "make people": 66705,
      "users age": 129095,
      "visual content": 131787,
      "offensive messages": 81553,
      "emergence social": 36057,
      "media effective": 68109,
      "effective analysis": 34626,
      "solve set": 110618,
      "competition propose": 19625,
      "attentionbased deep": 10065,
      "visual modality": 131799,
      "image various": 51800,
      "various textual": 131224,
      "textual segments": 121733,
      "extract finegrained": 42085,
      "finegrained feature": 44357,
      "representations classification": 99545,
      "shows sota": 108632,
      "sota performances": 110694,
      "unlike baselines": 127427,
      "baselines perform": 12441,
      "tasks average": 118952,
      "annotated test": 6243,
      "observe consistent": 81191,
      "finally establish": 44178,
      "strategies paper": 113482,
      "complaint classification": 19724,
      "present qualitative": 89656,
      "analysis behavior": 5512,
      "behavior models": 12664,
      "process czech": 91451,
      "pretrain models": 89978,
      "models 11": 72634,
      "11 datasets": 217,
      "datasets end": 27447,
      "publish pretrained": 94345,
      "models freely": 73262,
      "context generation": 22116,
      "generation improves": 47435,
      "improves zeroshot": 53065,
      "improve reasoning": 52513,
      "ability pretrained": 1561,
      "tasks context": 119010,
      "context problem": 22229,
      "generated language": 47090,
      "main result": 66459,
      "predicting answer": 88976,
      "learning relative": 62955,
      "specific problem": 111478,
      "problem characteristics": 90959,
      "corresponding problem": 24299,
      "problem particular": 91162,
      "faithful original": 43004,
      "original problem": 82537,
      "big languages": 13994,
      "nlp resources": 79685,
      "resources models": 101014,
      "resources scarce": 101041,
      "embeddings finetuned": 35724,
      "match semantic": 67365,
      "resulting crosslingual": 101437,
      "crosslingual embeddings": 24948,
      "skolt sami": 110002,
      "embeddings evaluation": 35702,
      "conducted shows": 20949,
      "model released": 71894,
      "paraphrase corpus": 84806,
      "especially questions": 38497,
      "corpus construction": 23720,
      "simultaneously considers": 109672,
      "language utilizing": 60317,
      "generated natural": 47096,
      "expand corpus": 40353,
      "informal sentences": 54341,
      "applicability approach": 6819,
      "classification inference": 17232,
      "performance fit": 86389,
      "approaches time": 8376,
      "formality transfer": 45192,
      "transfer task": 124191,
      "bert sentence": 13224,
      "applies deep": 7151,
      "tokens multiple": 122319,
      "listwise ranking": 64724,
      "facilitate optimization": 42785,
      "follow recent": 44939,
      "recent trend": 96553,
      "shows feasible": 108578,
      "relatively lowresource": 98406,
      "languages performing": 60799,
      "using 12": 129312,
      "million tokens": 70109,
      "tokens single": 122333,
      "exhaustive evaluation": 39992,
      "need include": 76821,
      "include additional": 53220,
      "output classification": 83060,
      "classification layer": 17249,
      "label experiments": 58396,
      "discourseaware graph": 32105,
      "graph network": 48837,
      "reasoning recent": 96303,
      "reasoning questions": 96302,
      "focus sentencelevel": 44815,
      "reasoning qa": 96301,
      "qa using": 94525,
      "structure texts": 113972,
      "texts model": 121554,
      "elementary discourse": 35272,
      "units edus": 127239,
      "features graph": 43530,
      "downstream qa": 34022,
      "results source": 102198,
      "unsupervised document": 127624,
      "representations unsupervised": 99957,
      "algorithms used": 4887,
      "nlp pretraining": 79668,
      "original document": 82512,
      "learning frameworks": 62606,
      "improve embedding": 52379,
      "quality unsupervised": 94819,
      "generated simple": 47123,
      "wordlevel manipulation": 133744,
      "method classifier": 68693,
      "classifier compare": 17532,
      "rate 64": 95782,
      "task matching": 118379,
      "outperform humans": 82712,
      "interesting nlp": 56080,
      "problem graph": 91072,
      "adopts hierarchical": 3913,
      "hierarchical graph": 49965,
      "learn connection": 62005,
      "method newly": 68988,
      "toefl dataset": 122241,
      "dataset verify": 27271,
      "using crossdomain": 129515,
      "crossdomain learning": 24893,
      "average human": 11194,
      "greek language": 49077,
      "submission eacl": 114728,
      "context open": 22202,
      "twofold build": 126140,
      "architecture second": 8740,
      "input nlp": 55376,
      "evaluation nmt": 39312,
      "people reading": 85888,
      "reading natural": 96027,
      "interpret human": 56208,
      "natural reading": 76616,
      "reaction times": 95925,
      "relative baseline": 98343,
      "embedding layers": 35425,
      "attack victim": 9706,
      "maintain competitive": 66501,
      "word inserted": 133323,
      "certain degree": 15938,
      "degree data": 28569,
      "dataset users": 27260,
      "datasets similar": 27714,
      "data poisoning": 26242,
      "sentencepair classification": 106184,
      "method efficient": 68794,
      "languages africa": 60394,
      "data case": 25718,
      "end training": 36834,
      "small train": 110215,
      "normalization data": 80334,
      "set lowresource": 107484,
      "experiments training": 41183,
      "models african": 72733,
      "toolkit nltk": 122415,
      "systems knowledge": 116962,
      "decoding pretrained": 28125,
      "transformers generating": 124448,
      "grounded responses": 49111,
      "abstraction real": 1795,
      "potentially facilitate": 88611,
      "facilitate dialogue": 42766,
      "responses integrating": 101277,
      "process endtoend": 91471,
      "architecture integrating": 8677,
      "process training": 91580,
      "training bert": 123374,
      "learns answer": 63193,
      "kg entities": 57657,
      "relations multitask": 98214,
      "subgraph kg": 114670,
      "models goal": 73305,
      "toolkit used": 122422,
      "models web": 74335,
      "elmo models": 35317,
      "text showing": 121291,
      "speech module": 111710,
      "static embedding": 113062,
      "web services": 132257,
      "implemented new": 51970,
      "new functionality": 78931,
      "pretrained elmo": 90025,
      "english evaluating": 37132,
      "texts text": 121634,
      "systems ubiquitous": 117214,
      "systems remains": 117108,
      "evaluate morphosyntactic": 38872,
      "present way": 89763,
      "extract various": 42131,
      "outputs text": 83185,
      "simple methodology": 109469,
      "metric task": 69907,
      "systems translating": 117207,
      "translating morphologicallyrich": 124596,
      "need adapt": 76776,
      "maintain performance": 66504,
      "teach pretrained": 119692,
      "novel vocabulary": 80773,
      "translations ii": 125469,
      "different finetuning": 31150,
      "possible combining": 88392,
      "combining data": 18718,
      "sentences leads": 106374,
      "examples model": 39855,
      "scores reference": 104198,
      "parallel examples": 84667,
      "joint extraction": 57276,
      "extraction concepts": 42286,
      "concepts relations": 20639,
      "valuable knowledge": 130765,
      "paper submissions": 84448,
      "relations furthermore": 98178,
      "provide ablation": 93748,
      "results error": 101761,
      "analysis evaluating": 5605,
      "sanskrit computational": 103632,
      "computational linguists": 20399,
      "embedding helps": 35413,
      "helps transfer": 49832,
      "unlabelled data": 127417,
      "resources essential": 100970,
      "study word": 114552,
      "broad categories": 14668,
      "investigate efficacy": 56750,
      "embeddings approaches": 35556,
      "proposed languages": 93319,
      "challenges posed": 16193,
      "crucial tasks": 25176,
      "finding boundaries": 44268,
      "conditional sentence": 20788,
      "problem best": 90951,
      "using exact": 129648,
      "malicious users": 66887,
      "users multiple": 129144,
      "language interaction": 59215,
      "interaction modeling": 55954,
      "lowdimensional representations": 65412,
      "demonstrating efficacy": 28973,
      "canonical surface": 15156,
      "surface morphological": 116067,
      "segmentation involves": 104593,
      "units language": 127252,
      "language important": 59154,
      "morphologicallyrich agglutinative": 74776,
      "languages southern": 60884,
      "canonical segmentation": 15155,
      "obtaining average": 81433,
      "score 725": 104048,
      "obtain average": 81258,
      "characterlevel lstm": 16572,
      "model fails": 71171,
      "segmentation models": 104602,
      "development better": 30373,
      "better nlp": 13636,
      "south african": 110961,
      "generation research": 47602,
      "intrinsic performance": 56367,
      "languages extremely": 60579,
      "using bytepair": 129424,
      "morphology languages": 74790,
      "different variants": 31549,
      "rnns transformers": 102982,
      "smallscale datasets": 110250,
      "rnns best": 102962,
      "open new": 81917,
      "new avenues": 78808,
      "multilingual lowresource": 75273,
      "conversations dataset": 23052,
      "settings introduce": 107812,
      "dataset 10k": 26706,
      "unique sequences": 127194,
      "sequences actions": 107111,
      "success propose": 115118,
      "simpler models": 109561,
      "models considerable": 72963,
      "considerable gap": 21249,
      "inflection learning": 54288,
      "despite performance": 29712,
      "models makes": 73550,
      "difficult determine": 31619,
      "determine complex": 30123,
      "rote memorization": 103283,
      "experiments finnish": 40948,
      "semeval 2021": 105514,
      "2021 task": 743,
      "al 2021": 4653,
      "given table": 48143,
      "set contains": 107403,
      "examples generate": 39839,
      "generate artificial": 46905,
      "examples train": 39891,
      "majority baseline": 66594,
      "understand linguistic": 126757,
      "using digital": 129578,
      "diversity using": 32899,
      "distribution languages": 32664,
      "data goal": 25983,
      "inferences underlying": 54254,
      "digital corpora": 31718,
      "significant changes": 108739,
      "taken place": 117512,
      "measures paper": 67887,
      "digital sources": 31726,
      "sources web": 110929,
      "stable languages": 112113,
      "shows consistent": 108568,
      "consistently represent": 21439,
      "able quantify": 1677,
      "type representation": 126223,
      "segmentation problem": 104622,
      "length constraint": 63355,
      "measures identify": 67869,
      "paper takes": 84470,
      "approach segmentation": 7871,
      "languages shows": 60871,
      "varying lengths": 131262,
      "types representation": 126350,
      "scibert based": 103972,
      "scientific data": 103998,
      "task focused": 118222,
      "focused finding": 44857,
      "information including": 54677,
      "entities properties": 37846,
      "rank leaderboard": 95625,
      "cls token": 17906,
      "opinion summarization": 82100,
      "summarization recent": 115560,
      "advances text": 4023,
      "consistent text": 21405,
      "vectors successful": 131476,
      "unsupervised opinion": 127686,
      "generate summary": 47023,
      "summary decoding": 115637,
      "vectors inputs": 131443,
      "specifically perform": 111572,
      "simple average": 109369,
      "quality study": 94804,
      "used simple": 128763,
      "generated summaries": 47130,
      "summaries text": 115464,
      "generic summaries": 47810,
      "summary vector": 115652,
      "degeneration issue": 28547,
      "benchmarks code": 12889,
      "sexism detection": 107890,
      "detection corpus": 29912,
      "corpus algerian": 23650,
      "dialect codeswitching": 30518,
      "approach hate": 7597,
      "similar works": 109174,
      "english best": 37079,
      "including deep": 53283,
      "performance cnn": 86209,
      "f1score 86": 42712,
      "compared lstm": 19379,
      "embedding approach": 35373,
      "embedding sentence": 35488,
      "way valuable": 132142,
      "language matching": 59275,
      "retrieval problems": 102422,
      "thorough examination": 121885,
      "based unsupervised": 12148,
      "models conduct": 72956,
      "experiments seven": 41130,
      "seven datasets": 107869,
      "datasets regarding": 27665,
      "vector normalization": 131332,
      "consistently boosts": 21412,
      "layers transformer": 61812,
      "method motivated": 68972,
      "natural extension": 76255,
      "parameter efficient": 84713,
      "experiments wmt": 41208,
      "heterogeneous knowledge": 49864,
      "growing popularity": 49179,
      "mentions text": 68468,
      "use implicit": 128091,
      "number candidate": 80850,
      "query different": 94954,
      "results example": 101772,
      "product knowledge": 92039,
      "reranking model": 100374,
      "specific customer": 111425,
      "search model": 104329,
      "extracting linguistic": 42219,
      "information bert": 54396,
      "adding linguistic": 3386,
      "information syntax": 55024,
      "massive pretrained": 67334,
      "nmt making": 79891,
      "method incorporating": 68898,
      "variety training": 131028,
      "training contexts": 123395,
      "tables graphs": 117265,
      "data efficiently": 25883,
      "explain predictions": 41257,
      "task tackle": 118773,
      "problem fact": 91053,
      "comparison baselines": 19531,
      "finding form": 44273,
      "score 069": 104044,
      "features useful": 43775,
      "annotated spans": 6236,
      "annotation supports": 6382,
      "prone error": 92331,
      "compared complex": 19349,
      "rules defined": 103391,
      "model f1": 71165,
      "learning materials": 62712,
      "reading difficulty": 96017,
      "scoring automated": 104224,
      "tasks automated": 118946,
      "educational applications": 34572,
      "interpretability models": 56226,
      "ml algorithms": 70432,
      "based handcrafted": 11749,
      "features wide": 43784,
      "regression classification": 97695,
      "classification modern": 17285,
      "require deep": 100127,
      "offering flexible": 81579,
      "seamless integration": 104288,
      "engineering efforts": 37048,
      "model improving": 71321,
      "knowledge facilitate": 57932,
      "critical challenge": 24805,
      "incorporate new": 53480,
      "concepts previous": 20634,
      "ignoring dependencies": 51686,
      "example existing": 39782,
      "selfsupervised framework": 104938,
      "taxonomies experiments": 119659,
      "realworld dataset": 96157,
      "dataset validate": 27266,
      "systems leading": 116975,
      "computer technology": 20493,
      "parties understand": 85505,
      "question requires": 95213,
      "gap exists": 46453,
      "science provide": 103980,
      "development reliable": 30416,
      "alignment source": 4988,
      "attention behavior": 9799,
      "attention functions": 9844,
      "introduce monotonicity": 56460,
      "mechanisms test": 68071,
      "experiments achieve": 40757,
      "rnn baselines": 102909,
      "baselines general": 12402,
      "implementation english": 51942,
      "verb phrase": 131522,
      "english verb": 37330,
      "phrase text": 87376,
      "language collected": 58885,
      "domain lexical": 33570,
      "lexical translation": 63835,
      "languages realized": 60832,
      "realized using": 96122,
      "rewrite rules": 102690,
      "used expert": 128534,
      "shows performed": 108606,
      "google translation": 48526,
      "tagging does": 117387,
      "networks pretraining": 77706,
      "sentencelevel tagging": 106176,
      "achieved superior": 2710,
      "performance typical": 86819,
      "discussed topic": 32301,
      "information introduced": 54701,
      "approach tagging": 7955,
      "general strategy": 46717,
      "thorough comparative": 121878,
      "training experimentally": 123628,
      "thirteen datasets": 121871,
      "deepen understanding": 28439,
      "answer subjective": 6583,
      "conversational discourse": 23005,
      "discourse signals": 32089,
      "multiple valid": 75742,
      "valid interpretations": 130710,
      "present discourse": 89455,
      "discourse dataset": 32052,
      "analyze dataset": 5962,
      "create computational": 24605,
      "attribution technique": 10211,
      "making statement": 66864,
      "change models": 16368,
      "investigates different": 56850,
      "attribution techniques": 10212,
      "techniques align": 119830,
      "challenging test": 16338,
      "studied nlp": 114166,
      "represent reasoning": 99126,
      "settings best": 107790,
      "upb semeval2021": 127785,
      "multiturn question": 75920,
      "dataset participants": 27091,
      "quantity span": 94901,
      "identification relation": 51430,
      "jointly solve": 57390,
      "solve subtasks": 110621,
      "set attention": 107366,
      "models prlms": 73802,
      "size end": 109921,
      "general effective": 46655,
      "effective selfattention": 34742,
      "dropout method": 34178,
      "models elaborate": 73124,
      "training design": 123580,
      "achieve stronger": 2576,
      "verify universality": 131595,
      "extensive natural": 41944,
      "challenging entailment": 16248,
      "multiplayer game": 75487,
      "based evidence": 11694,
      "goal identify": 48356,
      "temporal inference": 120106,
      "entailment evidence": 37663,
      "evidence retrieval": 39666,
      "sanity check": 103629,
      "effect data": 34589,
      "allows assess": 5127,
      "dataset constitutes": 26820,
      "good testbed": 48502,
      "meaning understanding": 67701,
      "specifically apply": 111526,
      "entire word": 37725,
      "classes lead": 17064,
      "data remains": 26337,
      "likely contain": 64137,
      "models reasoning": 73870,
      "achieves great": 2795,
      "works alleviate": 134922,
      "superior performances": 115693,
      "techniques dropout": 119874,
      "greatly boost": 49049,
      "design paper": 29557,
      "feature dropout": 43266,
      "play different": 87693,
      "empirically conduct": 36221,
      "extensive results": 41954,
      "using strong": 130229,
      "strong pretrained": 113703,
      "pretrained roberta": 90179,
      "roberta backbone": 102994,
      "paraphrase pairs": 84829,
      "way learning": 132103,
      "naturally share": 76648,
      "learns disentangle": 63204,
      "paraphrasing based": 84846,
      "based source": 12060,
      "semantics target": 105471,
      "tasks additionally": 118922,
      "approach effectively": 7514,
      "embeddings leading": 35779,
      "robustness syntactic": 103132,
      "tasks assessing": 118943,
      "potential make": 88572,
      "easily new": 34460,
      "probabilities given": 90843,
      "modifications model": 74442,
      "demonstrate scaling": 28862,
      "various potential": 131171,
      "potential weaknesses": 88599,
      "weaknesses approach": 132206,
      "approach surprisingly": 7951,
      "performance broad": 86192,
      "broad spectrum": 14683,
      "use subjective": 128303,
      "subjective evaluations": 114692,
      "setting case": 107737,
      "subjective ratings": 114701,
      "learning goal": 62617,
      "generated dialogue": 47071,
      "training human": 123651,
      "development process": 30413,
      "common choice": 18862,
      "geometry pretrained": 47869,
      "indicates word": 53867,
      "method fix": 68844,
      "improvement 13": 52666,
      "method proven": 69081,
      "proven robust": 93729,
      "evaluating knowledge": 39063,
      "study indicates": 114406,
      "informative manner": 55142,
      "set topics": 107610,
      "build conversational": 14758,
      "leveraging deep": 63670,
      "methods train": 69809,
      "applied visual": 7143,
      "visual data": 131791,
      "score use": 104137,
      "sequences learn": 107128,
      "semisupervised setting": 105624,
      "attempts jointly": 9772,
      "typically comes": 126416,
      "performance tradeoff": 86804,
      "rich interactions": 102750,
      "interface explore": 56099,
      "multiple model": 75617,
      "fewshot intent": 43901,
      "important practical": 52208,
      "practical scenarios": 88712,
      "needs learn": 76895,
      "having retrain": 49565,
      "nontrivial apply": 80310,
      "methods tasks": 69794,
      "complex label": 19824,
      "propose spanlevel": 93080,
      "retrieval method": 102407,
      "learns similar": 63236,
      "representations spans": 99901,
      "labels retrieved": 58637,
      "systems various": 117230,
      "simplification medical": 109586,
      "medical texts": 68229,
      "manual simplification": 67013,
      "rapidly growing": 95732,
      "motivating need": 74875,
      "automated approaches": 10430,
      "largescale resources": 61506,
      "texts english": 121505,
      "lay summaries": 61696,
      "published evidence": 94355,
      "different clinical": 31043,
      "texts automated": 121460,
      "measure better": 67782,
      "better differentiates": 13563,
      "heuristics introduce": 49885,
      "evaluate baseline": 38799,
      "encoderdecoder transformer": 36628,
      "models simplification": 74052,
      "simplification propose": 109590,
      "augmentation explicitly": 10275,
      "decoder producing": 28048,
      "jargon terms": 57218,
      "terms readability": 120369,
      "used interpret": 128596,
      "network predictions": 77392,
      "methods disagree": 69450,
      "model cases": 70816,
      "question conduct": 95139,
      "comprehensive quantitative": 20267,
      "quantitative evaluation": 94867,
      "quality prediction": 94759,
      "annotation syntactic": 6383,
      "drawing insights": 34128,
      "encoded contextualized": 36473,
      "help explain": 49719,
      "nlp existing": 79612,
      "use accuracy": 127877,
      "work argue": 134379,
      "need different": 76799,
      "develop heuristic": 30202,
      "tasks contextualized": 119011,
      "space represents": 111053,
      "multitask transformer": 75899,
      "models primary": 73799,
      "represent input": 99114,
      "belongs different": 12725,
      "task trained": 118799,
      "heads exhibit": 49606,
      "trained target": 123305,
      "target question": 117694,
      "querybased summaries": 95000,
      "given question": 48096,
      "context answer": 22006,
      "answer extracted": 6530,
      "words instance": 134008,
      "2016 word": 621,
      "methods previously": 69685,
      "pimentel et": 87513,
      "bert finetuning": 13130,
      "finetuning paper": 44488,
      "crosslingual wordincontext": 25047,
      "wordincontext disambiguation": 133719,
      "disambiguation mclwic": 31968,
      "setting task": 107781,
      "experiment pretrained": 40483,
      "finetuning procedure": 44506,
      "bias metrics": 13817,
      "metrics texts": 70014,
      "approach recent": 7840,
      "years word": 135314,
      "embeddings widely": 36024,
      "variety biases": 130966,
      "lack transparency": 58766,
      "quantify biases": 94845,
      "simple interpretation": 109448,
      "interpretation terms": 56274,
      "odds ratio": 81531,
      "estimating confidence": 38646,
      "produces similar": 92009,
      "embeddings capturing": 35587,
      "world embedded": 135027,
      "generation study": 47643,
      "aims transfer": 4579,
      "psycholinguistic theories": 94208,
      "ones native": 81698,
      "given literal": 48059,
      "literal sentence": 64730,
      "sentence extracts": 105871,
      "span sentence": 111132,
      "effectively transfer": 34854,
      "ones furthermore": 81687,
      "outperforms series": 82981,
      "models plm": 73748,
      "tasks obtaining": 119348,
      "obtaining labeled": 81442,
      "investigate zeroshot": 56831,
      "arabic da": 8500,
      "data identifying": 26012,
      "context named": 22188,
      "ner partofspeech": 77063,
      "varieties results": 130961,
      "effectiveness selftraining": 34948,
      "improving zeroshot": 53184,
      "transfer large": 124071,
      "accuracy pos": 2237,
      "observed directly": 81220,
      "datasets develop": 27423,
      "tasks opensource": 119352,
      "event salience": 39537,
      "salience estimation": 103542,
      "term importance": 120224,
      "existing humanannotated": 40142,
      "humanannotated datasets": 50998,
      "datasets subjective": 27735,
      "protocol proposed": 93675,
      "proposed prior": 93527,
      "linking tool": 64675,
      "entity matching": 37968,
      "argument event": 8940,
      "furthermore conduct": 46155,
      "analysis popular": 5754,
      "standards present": 112353,
      "task reduce": 118618,
      "heterogeneous graphs": 49859,
      "graphs capture": 48941,
      "interactions entities": 55984,
      "previous sota": 90464,
      "learning learning": 62685,
      "seldom available": 104683,
      "points achieve": 87845,
      "outperforming approaches": 82797,
      "like masked": 64063,
      "shortcoming previous": 108276,
      "require domain": 100129,
      "gap evaluate": 46451,
      "datasets heterogeneous": 27502,
      "heterogeneous domains": 49855,
      "2008 2012": 557,
      "extraction emotion": 42319,
      "emotion distribution": 36101,
      "lucene index": 65723,
      "data supplied": 26521,
      "communication time": 19043,
      "order sentences": 82403,
      "sentences coherent": 106240,
      "text used": 121392,
      "causal temporal": 15814,
      "formulate sentence": 45283,
      "task conditional": 117999,
      "order given": 82330,
      "input generates": 55345,
      "generates sequence": 47173,
      "markers sentences": 67235,
      "kendalls tau": 57526,
      "generalize datasets": 46806,
      "additionally perform": 3617,
      "framework static": 45699,
      "efficient knowledge": 35084,
      "instead structural": 55684,
      "queries masked": 94925,
      "sentences paris": 106436,
      "paris capital": 84865,
      "capital mask": 15239,
      "mask used": 67284,
      "used probes": 128701,
      "study knowledge": 114419,
      "simple nearest": 109477,
      "neighbor matching": 76985,
      "matching using": 67439,
      "points better": 87851,
      "comparative performance": 19209,
      "bert exploits": 13123,
      "expensive ability": 40411,
      "learning explainable": 62568,
      "provide tools": 93942,
      "enable insights": 36355,
      "review process": 102563,
      "form linguistic": 45103,
      "second global": 104411,
      "generate causal": 46910,
      "main characteristics": 66403,
      "multigranularity representations": 75119,
      "sequence characters": 106922,
      "novel pretraining": 80688,
      "pretraining paradigm": 90303,
      "manner specifically": 66961,
      "graph characters": 48767,
      "feed text": 43819,
      "selfattention layers": 104880,
      "propose masked": 92757,
      "information inherent": 54687,
      "model bring": 70785,
      "clue benchmarks": 17913,
      "benchmarks analysis": 12884,
      "improvement comes": 52693,
      "representations code": 99546,
      "taskoriented semantic": 118901,
      "building seq2seq": 14886,
      "map utterances": 67115,
      "ontology tokens": 81851,
      "propose span": 93077,
      "slots model": 110090,
      "variability gold": 130831,
      "datasets notably": 27599,
      "parsers achieving": 85013,
      "indigenous languages": 53886,
      "languages transformer": 60930,
      "models fundamental": 73265,
      "components natural": 20035,
      "spanish present": 111165,
      "approaches translate": 8383,
      "translate spanish": 124546,
      "task open": 118478,
      "research hierarchical": 100529,
      "learning generation": 62614,
      "long source": 65131,
      "current sequence": 25335,
      "summarization document": 115498,
      "level machine": 63471,
      "attention transformerbased": 10025,
      "tasks furthermore": 119139,
      "stateoftheart rouge": 112938,
      "rouge scores": 103300,
      "scores summarization": 104211,
      "pubmed arxiv": 94382,
      "attention finally": 9839,
      "study hierarchical": 114397,
      "highquality machine": 50394,
      "form pairs": 45113,
      "pairs new": 83591,
      "transfer parsing": 124174,
      "knowledge additional": 57746,
      "generation auxiliary": 47320,
      "crosslingual latent": 24968,
      "parser performs": 84985,
      "baselines cases": 12366,
      "asking answering": 9306,
      "multimodal question": 75449,
      "systems considered": 116808,
      "considered tasks": 21306,
      "tasks seldom": 119480,
      "synthetic multimodal": 116631,
      "metric referenceless": 69898,
      "stateoftheart correlations": 112624,
      "benchmarks make": 12918,
      "new standard": 79145,
      "entirely clear": 37729,
      "representations built": 99528,
      "stacked layers": 112127,
      "layers network": 61792,
      "questions general": 95309,
      "cognitive neuroscience": 18273,
      "adjacent words": 3852,
      "suggest transformers": 115335,
      "structure plays": 113933,
      "role process": 103213,
      "broadly results": 14706,
      "structured input": 114004,
      "systems serve": 117137,
      "models masked": 73553,
      "language segmentation": 60073,
      "segmentation segmentation": 104631,
      "like morphemes": 64066,
      "continuous speech": 22636,
      "data meaningful": 26122,
      "pause words": 85801,
      "languages morphologically": 60738,
      "dataset gold": 26957,
      "lightly supervised": 64006,
      "segmentation tasks": 104641,
      "bidirectional masked": 13964,
      "outperforms recurrent": 82975,
      "performs similarly": 87025,
      "discussing different": 32316,
      "different challenges": 31032,
      "systems humanlike": 116940,
      "aims build": 4506,
      "conversations naturally": 23060,
      "naturally humans": 76644,
      "humans draw": 51076,
      "draw insights": 34114,
      "annotate humanhuman": 6128,
      "conversations switchboard": 23072,
      "switchboard dialog": 116285,
      "act corpus": 3062,
      "corpus examine": 23785,
      "provided new": 93977,
      "content introduce": 21894,
      "conversational history": 23007,
      "better captured": 13528,
      "use pointwise": 128197,
      "60 time": 1147,
      "mimicking human": 70138,
      "utility linguistic": 130482,
      "lexicalized tree adjoining": 63853,
      "tree adjoining grammar": 125576,
      "paper present new": 84118,
      "compared previous approaches": 19413,
      "tree adjoining grammars": 125578,
      "grammars paper describes": 48680,
      "tree adjoining languages": 125580,
      "parsing incremental parser": 85131,
      "network paper presents": 77381,
      "new statistical approach": 79184,
      "large corpus results": 61065,
      "syntactic structure language": 116484,
      "language modeling paper": 59451,
      "information word history": 55098,
      "automatic speech recognition": 10656,
      "speech recognition model": 111761,
      "language model paper": 59363,
      "model paper presents": 71695,
      "approach text generation": 7973,
      "quick adaptation new": 95385,
      "new tasks domains": 79209,
      "method especially designed": 68810,
      "using training data": 130304,
      "training data limited": 123486,
      "prohibitively expensive work": 92204,
      "work aims improve": 134367,
      "complex noun phrases": 19852,
      "labeled training data": 58474,
      "corpora results suggest": 23579,
      "models paper presents": 73705,
      "paper presents new": 84184,
      "presents new approach": 89873,
      "structures approach based": 114059,
      "based markov models": 11829,
      "development spoken language": 30428,
      "natural language processing": 76389,
      "language processing nlp": 59890,
      "processing nlp modules": 91745,
      "language processing present": 59933,
      "lexical knowledge base": 63777,
      "robust approach linking": 103017,
      "approach linking existing": 7691,
      "linking existing lexicalsemantic": 64659,
      "existing lexicalsemantic hierarchies": 40161,
      "constraint satisfaction algorithm": 21584,
      "satisfaction algorithm relaxation": 103650,
      "method achieves accuracy": 68591,
      "approach machine translation": 7699,
      "machine translation present": 66160,
      "translation present approach": 125130,
      "multilingual machine translation": 75275,
      "analysis recent work": 5796,
      "recent work presented": 96581,
      "formal framework linguistic": 45160,
      "framework linguistic annotation": 45603,
      "text speech data": 121323,
      "coreference named entity": 23354,
      "task paper examine": 118504,
      "learning machine translation": 62700,
      "machine translation paper": 66139,
      "translation paper present": 125090,
      "performance results indicate": 86681,
      "results indicate approach": 101855,
      "corpus annotation scheme": 23663,
      "corpus naturally occurring": 23895,
      "method automatically generating": 68658,
      "generation process method": 47567,
      "paper describes speech": 83847,
      "higher degree robustness": 50175,
      "extraction natural language": 42405,
      "natural language queries": 76516,
      "process natural language": 91534,
      "quality speech recognition": 94795,
      "extracting meaningful information": 42221,
      "given contextfree grammar": 48004,
      "using surface expressions": 130251,
      "surface expressions examples": 116058,
      "language understanding machine": 60236,
      "understanding machine translation": 126883,
      "processing paper describes": 91765,
      "using information obtained": 129756,
      "information obtained recall": 54812,
      "obtained recall rate": 81401,
      "bilingual corpora experiments": 14028,
      "does require handcraft": 33394,
      "meaning natural language": 67651,
      "natural language documents": 76290,
      "different levels abstraction": 31228,
      "inference natural language": 54172,
      "exploiting syntactic structure": 41484,
      "structure natural language": 113912,
      "natural language modeling": 76371,
      "language models speech": 59661,
      "models speech recognition": 74082,
      "parsing language modeling": 85139,
      "language modeling using": 59483,
      "wall street journal": 132002,
      "word error rate": 133268,
      "approach language modeling": 7665,
      "hierarchical syntactic structure": 50010,
      "vocabulary speech recognition": 131907,
      "model new language": 71581,
      "experiments switchboard corpus": 41164,
      "language modeling speech": 59467,
      "modeling speech recognition": 72548,
      "error rate wer": 38335,
      "text processing systems": 121212,
      "developing semantic parsers": 30361,
      "paper describes new": 83834,
      "describes new approach": 29419,
      "processing machine learning": 91705,
      "machine learning techniques": 65833,
      "conventional ngram language": 22894,
      "ngram language models": 79438,
      "language models usually": 59685,
      "rate paper investigate": 95800,
      "paper investigate use": 84024,
      "present approach estimating": 89378,
      "approach estimating relative": 7539,
      "reduction perplexity 10": 97456,
      "task named entity": 118430,
      "named entity identification": 76046,
      "sparse training data": 111238,
      "performance machine learning": 86512,
      "machine learning algorithms": 65767,
      "classifiers using different": 17645,
      "applied natural language": 7097,
      "natural language parsing": 76382,
      "exploiting diversity natural": 41465,
      "diversity natural language": 32886,
      "applying machine learning": 7252,
      "machine learning methods": 65799,
      "methods computational linguistics": 69388,
      "language processing systems": 59948,
      "parsing machine learning": 85150,
      "machine learning community": 65789,
      "spoken natural language": 112015,
      "natural language dialogue": 76285,
      "language dialogue interfaces": 58960,
      "dialogue interfaces semiautonomous": 30694,
      "interfaces semiautonomous systems": 56109,
      "knowledge domain knowledge": 57882,
      "executable program simple": 39964,
      "program simple scripting": 92113,
      "simple scripting language": 109514,
      "outputmetaoutput distinction permits": 83156,
      "language models developed": 59538,
      "surface natural language": 116069,
      "natural language generation": 76300,
      "language generation present": 59090,
      "syntactic dependency information": 116393,
      "phrase present experiments": 87364,
      "prince smolensky 1993": 90666,
      "broadcast news switchboard": 14689,
      "having different formats": 49556,
      "different formats levels": 31152,
      "using structural information": 130231,
      "present robust approach": 89682,
      "select set candidates": 104712,
      "set candidates node": 107384,
      "candidates node target": 15138,
      "node target taxonomy": 80027,
      "target taxonomy bests": 117726,
      "taxonomy bests matches": 119662,
      "bests matches node": 13487,
      "matches node source": 67382,
      "node source taxonomy": 80025,
      "wordnet 15 wordnet": 133764,
      "15 wordnet 16": 355,
      "features needed distinguish": 43622,
      "new general approach": 78934,
      "small training corpus": 110217,
      "statistical significance testing": 113163,
      "best published result": 13424,
      "improves parser accuracy": 53004,
      "useful natural language": 128908,
      "natural language applications": 76267,
      "word sense disambiguation": 133517,
      "second method uses": 104429,
      "development practical systems": 30412,
      "street journal corpus": 113563,
      "little attention paid": 64795,
      "accuracy statistical parsers": 2290,
      "using new corpus": 129962,
      "obtained precision rate": 81392,
      "rules paper describes": 103425,
      "using supervised learning": 130239,
      "new methods using": 79018,
      "information retrieval using": 54953,
      "model does use": 71023,
      "language information retrieval": 59206,
      "language processing approach": 59842,
      "present novel machine": 89611,
      "novel machine learning": 80623,
      "learning algorithm used": 62321,
      "background information data": 11301,
      "information data sets": 54465,
      "present general overview": 89502,
      "general overview systems": 46691,
      "overview systems taken": 83378,
      "speech recognition word": 111787,
      "recognition word error": 97044,
      "accuracy speech recognition": 2281,
      "posterior probabilities word": 88485,
      "probabilities word lattices": 90854,
      "extraction semantic relations": 42478,
      "lexical knowledge bases": 63778,
      "nominal verbal adjectival": 80142,
      "highly inflected languages": 50327,
      "world wide web": 135055,
      "texts paper propose": 121571,
      "paper propose method": 84260,
      "effectiveness method way": 34910,
      "method way experiments": 69223,
      "information retrieval information": 54933,
      "propose evaluation method": 92663,
      "data oriented parsing": 26194,
      "referential properties noun": 97565,
      "properties noun phrases": 92471,
      "noun phrases japanese": 80431,
      "phrases japanese language": 87433,
      "japanese language articles": 57194,
      "anaphora resolution japanese": 6068,
      "achieved good results": 2630,
      "problems natural language": 91346,
      "language generation nlg": 59081,
      "referring expression generation": 97573,
      "discourse context approach": 32048,
      "representations communicative intent": 99550,
      "syntax semantics pragmatics": 116560,
      "used machine translation": 128620,
      "machine translation using": 66293,
      "corpus machine translation": 23872,
      "learning algorithms automatically": 62325,
      "tagging parsing models": 117435,
      "statistical language models": 113100,
      "conditional probability distributions": 20773,
      "training data somewhat": 123541,
      "data somewhat surprisingly": 26469,
      "probabilistic topdown parser": 90831,
      "problem language modeling": 91099,
      "speech recognition paper": 111765,
      "recognition paper introduces": 96954,
      "previous approaches using": 90385,
      "approaches using syntactic": 8395,
      "lead improved performance": 61856,
      "language model uses": 59419,
      "error rate reduction": 38331,
      "question answering propose": 95075,
      "propose method generate": 92769,
      "training test data": 123913,
      "test data results": 120437,
      "results obtained using": 102009,
      "training data single": 123537,
      "application machine learning": 6860,
      "noun phrase np": 80429,
      "progress language modeling": 92155,
      "depending training data": 29277,
      "training data size": 123538,
      "size word error": 109952,
      "paper presents study": 84205,
      "street journal wsj": 113564,
      "relative reduction word": 98371,
      "reduction word error": 97467,
      "baseline word error": 12337,
      "sense disambiguation task": 105667,
      "paper investigates use": 84030,
      "present simple methods": 89708,
      "syntactic parse trees": 116437,
      "new model achieves": 79025,
      "model achieves improvement": 70573,
      "semantic syntactic structure": 105321,
      "language modeling structured": 59470,
      "speech recognition shown": 111773,
      "significantly improved word": 108938,
      "parsing model trained": 85160,
      "paper describes experiments": 83822,
      "describes experiments carried": 29404,
      "allow easy integration": 5079,
      "experimental results word": 40727,
      "paper systems developed": 84463,
      "english allwords task": 37066,
      "different feature sets": 31146,
      "data sets results": 26435,
      "results compared systems": 101608,
      "room improvement unsupervised": 103266,
      "morphologically related words": 74765,
      "japanese word segmentation": 57203,
      "crucial step processing": 25172,
      "data despite simplicity": 25834,
      "novel evaluation metrics": 80564,
      "evaluation metrics based": 39284,
      "paper presents evaluation": 84170,
      "lexical sample tasks": 63809,
      "spanish english lexical": 111154,
      "english lexical sample": 37193,
      "paper describes results": 83843,
      "given word use": 48174,
      "real world applications": 96089,
      "play important role": 87696,
      "language processing paper": 59928,
      "different data sets": 31074,
      "research natural language": 100564,
      "wellformed natural language": 132412,
      "based lexical semantic": 11803,
      "lexical semantic knowledge": 63812,
      "time propose method": 122080,
      "known natural language": 58315,
      "intermediate parsing results": 56138,
      "shared task languageindependent": 108060,
      "task languageindependent named": 118337,
      "languageindependent named entity": 60366,
      "named entity recognition": 76052,
      "entity recognition background": 37993,
      "recognition background information": 96822,
      "data sets evaluation": 26425,
      "evaluation method present": 39269,
      "method present general": 69064,
      "systems taken task": 117182,
      "taken task discuss": 117517,
      "task discuss performance": 118105,
      "passing turing test": 85627,
      "large collection text": 61051,
      "shed light nature": 108155,
      "challenge nlp applications": 16068,
      "nlp applications machine": 79569,
      "applications machine translation": 6963,
      "machine translation mt": 66063,
      "monolingual parallel corpora": 74604,
      "parallel corpora evaluate": 84598,
      "statistical mt systems": 113137,
      "history natural language": 50563,
      "language models task": 59667,
      "previous works mainly": 90573,
      "works mainly focus": 134961,
      "propose general framework": 92697,
      "used natural language": 128646,
      "text categorization based": 120698,
      "samples training set": 103596,
      "set preliminary experiments": 107539,
      "text categorization method": 120699,
      "present paper describes": 89637,
      "machine translation makes": 66042,
      "indian language machine": 53819,
      "equivalents target language": 38261,
      "source language target": 110779,
      "language target language": 60147,
      "produce good quality": 91895,
      "lexical resources paper": 63806,
      "english indian languages": 37172,
      "spoken language resources": 111990,
      "speech recognition propose": 111769,
      "dutch english german": 34279,
      "nondeterministic pushdown automata": 80189,
      "statistical machine translation": 113106,
      "machine translation smt": 66225,
      "translation smt systems": 125275,
      "finally discuss improvement": 44171,
      "word alignment model": 132908,
      "sentiment analysis using": 106677,
      "review thumbs thumbs": 102585,
      "state art evaluation": 112441,
      "approach paper propose": 7787,
      "paper propose new": 84273,
      "correspondence source target": 24272,
      "source target text": 110842,
      "make possible study": 66707,
      "machine translation output": 66137,
      "limitation paper propose": 64163,
      "paper propose use": 84340,
      "interpreting natural language": 56287,
      "natural language texts": 76545,
      "language texts paper": 60167,
      "available online research": 11062,
      "online research purposes": 81795,
      "13 million words": 286,
      "prior domain knowledge": 90702,
      "methods make use": 69613,
      "different word classes": 31566,
      "particular focus paper": 85415,
      "proposed approach does": 93192,
      "textual entailment rte": 121696,
      "entailment rte task": 37673,
      "representation natural language": 99348,
      "natural language semantics": 76530,
      "languages french portuguese": 60600,
      "paper discusses new": 83880,
      "resources natural language": 101016,
      "natural language parser": 76381,
      "methods resources used": 69731,
      "parse sentences containing": 84888,
      "integrate language model": 55758,
      "language model semantic": 59398,
      "semantic information word": 105080,
      "systems make use": 116994,
      "language models lm": 59589,
      "syntactic semantic information": 116467,
      "latent semantic analysis": 61602,
      "semantic analysis lsa": 104980,
      "words context present": 133880,
      "significant improvements compared": 108788,
      "semantic similarities paragraph": 105277,
      "similarities paragraph paragraph": 109185,
      "words w1 w2": 134311,
      "provide insight sentence": 93858,
      "grammar engineering paper": 48630,
      "engineering paper present": 37055,
      "using techniques borrowed": 130276,
      "syntactic semantic properties": 116473,
      "important tasks natural": 52275,
      "tasks natural language": 119326,
      "language processing information": 59871,
      "processing information retrieval": 91682,
      "information retrieval machine": 54937,
      "retrieval machine translation": 102405,
      "machine translation confidence": 65930,
      "machine translation problem": 66172,
      "machine translation based": 65909,
      "based mutual information": 11876,
      "ngram language model": 79436,
      "language model lexical": 59342,
      "features language model": 43583,
      "language model evaluate": 59317,
      "confidence measures based": 20990,
      "error rate low": 38330,
      "context multidocument summarization": 22183,
      "multidocument summarization evolving": 75080,
      "summarization evolving events": 115504,
      "used second stage": 128740,
      "multiword expressions mwes": 75946,
      "semantic regularities words": 105194,
      "model able discover": 70513,
      "task identifying textual": 118269,
      "detection coreference resolution": 29910,
      "coreference resolution task": 23373,
      "using local features": 129826,
      "term extraction using": 120213,
      "olac extension dravidian": 81640,
      "language resources paper": 60057,
      "effective natural language": 34718,
      "model reference resolution": 71879,
      "paper provide overview": 84379,
      "dependency relations extracted": 29222,
      "usual dependency tree": 130399,
      "evaluate proposed method": 38900,
      "manually constructed gold": 67048,
      "proposed method compared": 93353,
      "work word sense": 134880,
      "texts natural language": 121559,
      "using natural language": 129934,
      "proteinprotein interaction extraction": 93667,
      "extracting proteinprotein interactions": 42231,
      "proteinprotein interactions ppis": 93669,
      "current natural language": 25305,
      "biomedical literature paper": 14194,
      "sentence simplification automatic": 106078,
      "challenge natural language": 16060,
      "trained largescale corpora": 123186,
      "order improve performance": 82343,
      "improve performance syntactic": 52490,
      "sentences annotated syntactic": 106209,
      "17th century english": 412,
      "century 19th century": 15923,
      "19th century use": 474,
      "processing nlp like": 91740,
      "context propose method": 22233,
      "low dimensional embedding": 65357,
      "dimensionality reduction methods": 31759,
      "order achieve better": 82267,
      "paper describes details": 83816,
      "processing nlp applications": 91727,
      "simple english wikipedia": 109427,
      "recent work proposed": 96582,
      "work proposed method": 134752,
      "negative polarity items": 76943,
      "various natural language": 131141,
      "phrase structure parsing": 87372,
      "various decisionmaking processes": 131074,
      "opinion mining important": 82091,
      "values positive negative": 130800,
      "positive negative neutral": 88328,
      "naive bayesian classifiers": 76023,
      "controlled natural language": 22841,
      "controlled natural languages": 22843,
      "facilitate future research": 42770,
      "paper presents design": 84163,
      "closely related languages": 17862,
      "text mining natural": 121121,
      "mining natural language": 70244,
      "lexical syntactic ambiguity": 63827,
      "semantic relatedness measure": 105196,
      "explicit semantic analysis": 41345,
      "svm classifier trained": 116232,
      "language text speech": 60162,
      "text speech synthesis": 121324,
      "languages english spanish": 60549,
      "english spanish french": 37283,
      "propose new method": 92862,
      "errors ocr output": 38396,
      "graphical user interface": 48935,
      "biomedical information extraction": 14189,
      "model sentence simplification": 71975,
      "automatic discourse analysis": 10518,
      "sentence simplification improving": 106079,
      "constituent dependency structures": 21536,
      "finally case study": 44151,
      "translate natural language": 124532,
      "natural language sentences": 76532,
      "language sentences formulas": 60079,
      "semantic representation words": 105234,
      "words phrases sentences": 134119,
      "semantic meaning sentences": 105108,
      "meaning sentences directed": 67689,
      "corpora natural language": 23535,
      "natural language interfaces": 76365,
      "understand natural language": 126762,
      "natural language text": 76544,
      "text answer questions": 120648,
      "answer questions given": 6567,
      "questions given natural": 95312,
      "given natural language": 48066,
      "natural language respect": 76526,
      "meaning words sentence": 67708,
      "compare existing methods": 19247,
      "data proposed technique": 26297,
      "demonstrated promising results": 28927,
      "design automatic scoring": 29519,
      "feature sets based": 43314,
      "representations discourse structure": 99600,
      "majority class baseline": 66597,
      "unseen test data": 127550,
      "performing significantly better": 86970,
      "best performing models": 13404,
      "feature sets achieving": 43313,
      "forms contemporary english": 45237,
      "previous work using": 90563,
      "english natural language": 37216,
      "natural language understanding": 76551,
      "problem natural language": 91139,
      "knowledge domain consideration": 57879,
      "domain consideration social": 33484,
      "consideration social behavior": 21282,
      "language processing tools": 59979,
      "spoken dialogue systems": 111980,
      "linguistic knowledge automatically": 64500,
      "paper present evaluate": 84099,
      "models trained tested": 74233,
      "paper presents preliminary": 84198,
      "recognition named entities": 96913,
      "recognize named entities": 97056,
      "applications natural language": 6971,
      "paper presents work": 84211,
      "using transformation based": 130313,
      "achieving good results": 2950,
      "context free grammar": 22111,
      "using freely available": 129689,
      "semantic analysis method": 104981,
      "evaluate method text": 38855,
      "method text classification": 69188,
      "reasoning natural language": 96279,
      "language understanding based": 60204,
      "understanding based objectoriented": 126800,
      "based objectoriented semantics": 11906,
      "objectoriented semantics algorithms": 81145,
      "computer oriented input": 20485,
      "semantic processing text": 105185,
      "processing text information": 91838,
      "text information presented": 121054,
      "systems closely connected": 116783,
      "closely connected text": 17852,
      "connected text processing": 21130,
      "text processing criminology": 121205,
      "processing criminology operation": 91648,
      "criminology operation business": 24778,
      "operation business medicine": 82054,
      "business medicine document": 14977,
      "medicine document systems": 68236,
      "vector space representations": 131382,
      "tensor product representations": 120193,
      "sentences paper describes": 106424,
      "partofspeech tagging chunking": 85549,
      "use naive bayesian": 128161,
      "achieves good result": 2793,
      "machine translation crosslanguage": 65933,
      "machine translation literature": 66033,
      "machine translation systems": 66247,
      "rogets thesaurus semantic": 103151,
      "measures semantic similarity": 67896,
      "improvement state art": 52764,
      "similar sounding words": 109148,
      "test data set": 120438,
      "tagging paper describes": 117432,
      "use naive bayes": 128160,
      "syntactic ambiguities resolved": 116365,
      "paraphrasing text simplification": 84852,
      "machine translation word": 66302,
      "paper introduce new": 83998,
      "wsj section penn": 135180,
      "section penn treebank": 104489,
      "paper describes tool": 83856,
      "existing natural language": 40222,
      "language processing methods": 59886,
      "methods limited scope": 69599,
      "languages english french": 60534,
      "texts paper presents": 121570,
      "paper presents novel": 84190,
      "presents novel approach": 89881,
      "novel approach converting": 80474,
      "syntactic semantic lexical": 116471,
      "paper addresses problem": 83716,
      "problem mapping natural": 91123,
      "mapping natural language": 67141,
      "learning algorithm takes": 62320,
      "conditioned input sentence": 20802,
      "method task learning": 69179,
      "learning natural language": 62826,
      "language interfaces databases": 59220,
      "outperform previous methods": 82725,
      "clustering based approach": 17940,
      "average length words": 11200,
      "google books ngram": 48519,
      "languages natural language": 60753,
      "paper focus problem": 83950,
      "variation social media": 130900,
      "social media present": 110414,
      "offers new perspective": 81590,
      "transitionbased dependency parsing": 124505,
      "transitionbased dependency parsers": 124504,
      "dependency parsing variety": 29210,
      "size training set": 109949,
      "knowledge representation systems": 58145,
      "approach solve problem": 7923,
      "using external sources": 129662,
      "word alignment paper": 132909,
      "present new simple": 89590,
      "information machine translation": 54744,
      "trained small corpus": 123282,
      "results comparable obtained": 101592,
      "alignment error rate": 4957,
      "furthermore results obtained": 46213,
      "results obtained indicate": 102007,
      "sentences natural language": 106406,
      "natural language semantic": 76529,
      "neural probabilistic language": 78633,
      "probabilistic language model": 90813,
      "machine translation neural": 66081,
      "language model nplm": 59358,
      "better perplexity ngram": 13669,
      "language models paper": 59610,
      "models paper investigates": 73703,
      "language model experiment": 59319,
      "trained relatively small": 123264,
      "long training time": 65144,
      "human language acquisition": 50897,
      "rapid development natural": 95716,
      "development natural language": 30402,
      "language understanding module": 60239,
      "understanding natural language": 126897,
      "natural language interface": 76364,
      "paper describes submission": 83848,
      "using publicly available": 130067,
      "provided training data": 93989,
      "training data built": 123441,
      "translation model using": 124946,
      "significant improvement baseline": 108767,
      "information extraction natural": 54582,
      "propose probabilistic approach": 93001,
      "produced stateoftheart results": 91970,
      "obtain good results": 81288,
      "automatic lexical semantic": 10578,
      "lexical semantic information": 63811,
      "complexity task results": 19942,
      "work present results": 134687,
      "reducing human effort": 97420,
      "human effort required": 50799,
      "semantic classes using": 105008,
      "role argument structure": 103159,
      "extract semantic information": 42113,
      "sense disambiguation techniques": 105669,
      "syntactic semantic features": 116466,
      "multiple paraphrases given": 75635,
      "question answering systems": 95110,
      "novel probabilistic framework": 80690,
      "necessary sufficient condition": 76761,
      "human natural language": 50919,
      "natural language present": 76385,
      "foreign language learners": 45042,
      "common european framework": 18876,
      "machine translation partofspeech": 66151,
      "translation partofspeech tagging": 125102,
      "machine translation indian": 66008,
      "translation indian languages": 124867,
      "emerging research area": 36071,
      "source language text": 110782,
      "language text target": 60163,
      "text target language": 121360,
      "latent variable model": 61629,
      "analysis natural language": 5715,
      "paper describes approach": 83809,
      "better evaluation results": 13579,
      "results statistical machine": 102212,
      "machine translation tasks": 66261,
      "analysis tweets paper": 5911,
      "task detect sentiment": 118075,
      "application natural language": 6869,
      "language processing including": 59870,
      "including machine translation": 53320,
      "machine translation information": 66010,
      "translation information retrieval": 124871,
      "outperforming existing methods": 82809,
      "data shared task": 26441,
      "morphologically rich languages": 74769,
      "2013 shared task": 570,
      "used shared task": 128759,
      "provided shared task": 93982,
      "natural language tasks": 76542,
      "natural language analysis": 76266,
      "generates natural language": 47166,
      "task machine translation": 118367,
      "output natural language": 83100,
      "natural language human": 76324,
      "natural language input": 76359,
      "languages machine translation": 60713,
      "translation systems paper": 125314,
      "large monolingual data": 61158,
      "distributed representation words": 32618,
      "mapping vector spaces": 67157,
      "vector spaces languages": 131392,
      "despite simplicity method": 29732,
      "european union eu": 38773,
      "manually labelled data": 67072,
      "parallel training data": 84694,
      "training data languages": 123482,
      "tasks including crosslingual": 119176,
      "plagiarism detection sentence": 87620,
      "translation memory tm": 124932,
      "extraction named entity": 42402,
      "entity recognition ner": 38031,
      "introduce new resource": 56499,
      "multilingual named entity": 75300,
      "new freely available": 78928,
      "machine learning systems": 65829,
      "improve machine translation": 52412,
      "machine translation results": 66202,
      "improving quality mt": 53151,
      "quality mt output": 94723,
      "novel approach machine": 80483,
      "rule based approach": 103347,
      "entities proper names": 37844,
      "proper names location": 92419,
      "using statistical machine": 130222,
      "machine translation toolkit": 66270,
      "translation toolkit moses": 125382,
      "sentiment polarity classification": 106772,
      "classification widely studied": 17501,
      "positive negative opinion": 88329,
      "using movie review": 129903,
      "comparative study different": 19212,
      "bayes support vector": 12576,
      "support vector machine": 116000,
      "downstream natural language": 34006,
      "language applications question": 58834,
      "applications question answering": 6997,
      "question answering summarization": 95108,
      "summarization paper propose": 115546,
      "propose new task": 92885,
      "new task identifying": 79200,
      "number applications including": 80840,
      "semantic parsing framework": 105148,
      "formal representation meaning": 45174,
      "research machine translation": 100552,
      "time consuming expensive": 121999,
      "different mt engines": 31272,
      "human intervention required": 50872,
      "results human ranking": 101834,
      "learning semantic representations": 62995,
      "source target phrases": 110837,
      "latent semantic space": 61605,
      "multilayer neural network": 75178,
      "training data learning": 123485,
      "model significantly improves": 72021,
      "significantly improves performance": 108952,
      "improves performance stateoftheart": 53018,
      "phrasebased statistical machine": 87392,
      "natural language discourse": 76288,
      "discourse representation structures": 32083,
      "natural language syntax": 76540,
      "method used extract": 69208,
      "success rate high": 115123,
      "text mining applications": 121119,
      "information extraction text": 54609,
      "training data manually": 123492,
      "developing natural language": 30352,
      "language processing applications": 59836,
      "news articles day": 79294,
      "resources freely available": 100981,
      "previous work shown": 90549,
      "compare performance different": 19273,
      "information extraction paper": 54588,
      "paper presents scalable": 84202,
      "semantic relatedness words": 105201,
      "present new approach": 89580,
      "new approach measuring": 78791,
      "approach measuring semantic": 7708,
      "measure semantic relatedness": 67823,
      "method evaluate performance": 68814,
      "similarity relatedness synonym": 109296,
      "semantic relatedness tasks": 105199,
      "experimental evaluation shows": 40532,
      "evaluation shows proposed": 39400,
      "shows proposed method": 108617,
      "proposed method outperforms": 93384,
      "small parallel corpora": 110181,
      "parallel corpora paper": 84607,
      "corpora paper describes": 23552,
      "paper describes method": 83830,
      "experiments conducted using": 40849,
      "different language pairs": 31193,
      "present entirely unsupervised": 89475,
      "semantic dependency parsing": 105030,
      "predicateargument structure sentence": 88855,
      "structure sentence plays": 113949,
      "sentence plays critical": 105989,
      "plays critical role": 87723,
      "critical role deep": 24829,
      "role deep processing": 103165,
      "deep processing natural": 28397,
      "processing natural language": 91721,
      "rely pipeline framework": 98728,
      "maintaining competitive performance": 66511,
      "word pair classification": 133391,
      "pair classification problem": 83424,
      "using maximum entropy": 129865,
      "maximum entropy classifier": 67533,
      "achieves stateoftheart performance": 2881,
      "evaluation data set": 39168,
      "multiword expressions using": 75947,
      "using semantic clustering": 130140,
      "issues natural language": 57062,
      "language understanding generation": 60223,
      "languages like bengali": 60691,
      "vector space model": 131373,
      "pointwise mutual information": 87890,
      "mutual information pmi": 75983,
      "log likelihood ratio": 64980,
      "likelihood ratio llr": 64125,
      "propose novel unsupervised": 92974,
      "empirical evaluation shows": 36160,
      "et al 2000": 38682,
      "new open dataset": 79064,
      "available online used": 11064,
      "learning model explains": 62759,
      "american sign language": 5311,
      "sign language asl": 108678,
      "et al 2012": 38694,
      "automatic summarization documents": 10679,
      "semantic relationships terms": 105223,
      "language processing tasks": 59952,
      "approach natural language": 7728,
      "current stateoftheart approaches": 25351,
      "task currently preparation": 118040,
      "task involves identifying": 118313,
      "relations clinical text": 98123,
      "language natural language": 59705,
      "structural relationship words": 113790,
      "words given sentence": 133975,
      "machine translation major": 66041,
      "area natural language": 8875,
      "play key role": 87702,
      "using rule based": 130128,
      "machine learning approaches": 65773,
      "translation mt research": 124984,
      "proper transliteration entities": 92426,
      "paper address issue": 83703,
      "englishhindi language pair": 37408,
      "language pair experiments": 59729,
      "using precision recall": 130037,
      "plays crucial role": 87727,
      "translation quality various": 125189,
      "various automatic metrics": 131049,
      "languages like hindi": 60695,
      "free word order": 45772,
      "word order language": 133377,
      "multilingual natural language": 75303,
      "convolutional neural network": 23176,
      "central language understanding": 15903,
      "sentences network uses": 106409,
      "achieves excellent performance": 2784,
      "excellent performance tasks": 39918,
      "25 error reduction": 814,
      "language model combination": 59307,
      "introduce novel approach": 56510,
      "building language models": 14854,
      "language models based": 59508,
      "models special case": 74074,
      "traditional language models": 122824,
      "language models using": 59684,
      "domain specific corpus": 33645,
      "approach lies ability": 7687,
      "training data using": 123564,
      "small training data": 110218,
      "training data set": 123534,
      "parallel multilingual corpus": 84673,
      "approaches semantic parsing": 8332,
      "queries paper presents": 94932,
      "presents novel deep": 89883,
      "novel deep learning": 80536,
      "deep learning architecture": 28250,
      "neural models language": 78179,
      "queries natural language": 94930,
      "natural language statements": 76536,
      "using multilingual data": 129916,
      "data named entity": 26150,
      "entity recognition named": 38025,
      "languages share common": 60867,
      "improve named entity": 52428,
      "recognition ner using": 96946,
      "large scale text": 61263,
      "scale text corpora": 103756,
      "training statistical machine": 123883,
      "widely used training": 132594,
      "contrary previous approaches": 22683,
      "feature functions experiments": 43286,
      "better results obtained": 13705,
      "languages french german": 60596,
      "present natural language": 89568,
      "logic type theory": 65000,
      "semantic analysis esa": 104979,
      "machine translation model": 66047,
      "translation model based": 124938,
      "machine translation natural": 66078,
      "translation natural language": 125002,
      "training corpus statistical": 123404,
      "corpus statistical machine": 24019,
      "parallel sentences nonparallel": 84683,
      "large unlabeled corpus": 61313,
      "corpus experimental results": 23789,
      "learning method effectively": 62720,
      "improve performance machine": 52468,
      "performance machine translation": 86514,
      "main goal project": 66425,
      "finite state transducer": 44540,
      "terms based analysis": 120282,
      "topics information retrieval": 122633,
      "fully automatic multilanguage": 45948,
      "automatic multilanguage translation": 10614,
      "swiss avalanche bulletin": 116269,
      "avalanche bulletin swiss": 11155,
      "bulletin swiss avalanche": 14959,
      "avalanche bulletin produced": 11152,
      "bulletin produced twice": 14955,
      "produced twice day": 91977,
      "twice day languages": 126083,
      "day languages lack": 27842,
      "languages lack time": 60666,
      "lack time available": 58759,
      "time available manual": 121986,
      "available manual translation": 11042,
      "manual translation fully": 67017,
      "translation fully automated": 124830,
      "fully automated translation": 45943,
      "automated translation employed": 10477,
      "translation employed based": 124783,
      "employed based catalogue": 36293,
      "based catalogue predefined": 11563,
      "catalogue predefined phrases": 15703,
      "predefined phrases predetermined": 88830,
      "phrases predetermined rules": 87444,
      "predetermined rules phrases": 88842,
      "rules phrases combined": 103428,
      "phrases combined produce": 87413,
      "combined produce sentences": 18665,
      "able automatically translate": 1611,
      "automatically translate sentences": 10832,
      "translate sentences german": 124538,
      "sentences german target": 106332,
      "german target languages": 47916,
      "target languages french": 117668,
      "languages french italian": 60597,
      "french italian english": 45814,
      "italian english subsequent": 57089,
      "english subsequent proofreading": 37295,
      "subsequent proofreading correction": 114818,
      "catalogue phrases limited": 15696,
      "phrases limited small": 87437,
      "limited small sublanguage": 64281,
      "reduction daily translation": 97444,
      "daily translation costs": 25515,
      "translation costs expected": 124718,
      "costs expected offset": 24399,
      "expected offset initial": 40395,
      "offset initial development": 81619,
      "initial development costs": 55211,
      "development costs years": 30377,
      "operational winter seasons": 82059,
      "winter seasons assess": 132733,
      "seasons assess quality": 104377,
      "assess quality produced": 9489,
      "quality produced texts": 94762,
      "produced texts based": 91973,
      "origins catalogue phrases": 82581,
      "catalogue phrases versus": 15699,
      "phrases versus manually": 87465,
      "versus manually written": 131628,
      "manually written translated": 67084,
      "written translated texts": 135154,
      "translated texts mean": 124572,
      "texts mean recognition": 121549,
      "mean recognition rate": 67606,
      "recognition rate 55": 96979,
      "rate 55 users": 95779,
      "55 users hardly": 1113,
      "users hardly distinguish": 129129,
      "types texts similar": 126370,
      "texts similar ratings": 121608,
      "similar ratings respect": 109132,
      "ratings respect language": 95826,
      "respect language quality": 101083,
      "language quality overall": 60010,
      "quality overall output": 94741,
      "overall output catalogue": 83242,
      "output catalogue considered": 83057,
      "catalogue considered virtually": 15692,
      "considered virtually equivalent": 21311,
      "virtually equivalent text": 131750,
      "equivalent text written": 38254,
      "text written avalanche": 121418,
      "written avalanche forecasters": 135116,
      "avalanche forecasters manually": 11159,
      "forecasters manually translated": 45032,
      "manually translated professional": 67080,
      "translated professional translators": 124558,
      "professional translators furthermore": 92082,
      "translators furthermore forecasters": 125514,
      "furthermore forecasters declared": 46176,
      "forecasters declared relevant": 45028,
      "declared relevant situations": 27976,
      "relevant situations captured": 98582,
      "situations captured sufficient": 109896,
      "captured sufficient accuracy": 15439,
      "lexical acquisition paper": 63728,
      "yield better results": 135334,
      "paper reports work": 84406,
      "nlp tools contest": 79787,
      "punjabi tamil telugu": 94403,
      "hidden markov models": 49901,
      "based model used": 11850,
      "present novel framework": 89610,
      "novel framework learning": 80575,
      "generation present novel": 47551,
      "present novel algorithm": 89597,
      "languages south africa": 60883,
      "focus natural language": 44797,
      "specific use cases": 111514,
      "performing qualitative analysis": 86966,
      "qualitative analysis results": 94554,
      "method machine translation": 68948,
      "machine translation process": 66173,
      "automatically extract information": 10769,
      "information natural language": 54788,
      "methods text classification": 69800,
      "text classification text": 120781,
      "classification text classification": 17467,
      "text classification task": 120772,
      "task automatic classification": 117909,
      "text classification widely": 120786,
      "like natural language": 64073,
      "language processing data": 59859,
      "processing data mining": 91655,
      "data mining information": 26131,
      "information retrieval text": 54951,
      "text classification important": 120725,
      "performance text classification": 86799,
      "text classification improves": 120727,
      "uses syntactic information": 129288,
      "question answering information": 95051,
      "information extraction machine": 54577,
      "extraction machine translation": 42380,
      "machine translation text": 66268,
      "text classification paper": 120753,
      "paper present ongoing": 84128,
      "languages named entity": 60750,
      "processing tasks information": 91818,
      "tasks information extraction": 119201,
      "machine translation document": 65955,
      "document summarization paper": 33088,
      "rulebased machine learning": 103365,
      "entity recognition paper": 38055,
      "approaches named entity": 8244,
      "like indian languages": 64051,
      "requires large data": 100288,
      "large data sets": 61070,
      "nlp tasks analyze": 79706,
      "word embedding method": 133018,
      "word embeddings including": 133132,
      "representations named entity": 99764,
      "results proposed method": 102068,
      "proposed method achieves": 93339,
      "method achieves good": 68600,
      "better results compared": 13703,
      "word embeddings tasks": 133220,
      "achieves stateoftheart results": 2899,
      "word embeddings languages": 133142,
      "available public use": 11085,
      "entity recognition social": 38072,
      "recognition social media": 97001,
      "present novel approach": 89598,
      "entities named entities": 37826,
      "new entities arrive": 78903,
      "approach does require": 7507,
      "evaluation sets including": 39387,
      "natural language process": 76387,
      "information extraction systems": 54604,
      "training machine learning": 123696,
      "machine learning models": 65809,
      "extended periods time": 41810,
      "recent studies shown": 96525,
      "amazons mechanical turk": 5274,
      "ncbi disease corpus": 76700,
      "set gold standard": 107455,
      "evaluating semantic models": 39095,
      "genuine similarity estimation": 47844,
      "distributional semantic models": 32720,
      "wider range applications": 132610,
      "unsupervised keyword extraction": 127646,
      "comparative study neural": 19213,
      "different semantic spaces": 31408,
      "detection dialogue act": 29923,
      "dialogue act tagging": 30632,
      "neural word embeddings": 78730,
      "paper provides method": 84386,
      "evaluate performance using": 38891,
      "performance using incremental": 86831,
      "metrics propose new": 69995,
      "tree edit distance": 125587,
      "et al 2011": 38693,
      "provides empirical evidence": 94024,
      "twitter data twitter": 126105,
      "processing nlp tools": 91763,
      "commonly used english": 18969,
      "used english language": 128515,
      "biomedical texts paper": 14210,
      "terms precision recall": 120362,
      "precision recall fscore": 88802,
      "play vital role": 87708,
      "crfbased named entity": 24764,
      "shared task task": 108098,
      "nouns verbs adjectives": 80447,
      "open source morphological": 81935,
      "source morphological analyzer": 110795,
      "using open source": 129989,
      "unsupervised word alignment": 127751,
      "important natural language": 52187,
      "language processing task": 59950,
      "received considerable attention": 96361,
      "generative discriminative approaches": 47725,
      "experiments approach achieves": 40778,
      "approach achieves significant": 7316,
      "achieves significant improvements": 2857,
      "significant improvements stateoftheart": 108807,
      "stateoftheart unsupervised word": 113021,
      "alignment methods learning": 4970,
      "methods learning distributed": 69589,
      "distributed word representations": 32637,
      "word representations natural": 133464,
      "distributed semantic representations": 32624,
      "tasks remains open": 119455,
      "remains open question": 98815,
      "address question using": 3757,
      "neural networkbased models": 78427,
      "neural networks neural": 78524,
      "neural tensor networks": 78704,
      "experiments evaluate models": 40940,
      "models ability learn": 72640,
      "tweets social media": 126059,
      "social media texts": 110427,
      "tasks named entity": 119322,
      "usually perform poorly": 130442,
      "perform poorly applied": 86044,
      "media texts paper": 68175,
      "paper report experiments": 84400,
      "improving named entity": 53120,
      "annotated data sets": 6168,
      "entity recognition performance": 38059,
      "results different settings": 101727,
      "used starting point": 128778,
      "opinion mining text": 82093,
      "classified positive negative": 17512,
      "provide detailed evaluation": 93805,
      "generate natural language": 46973,
      "use natural language": 128163,
      "deep semantic structure": 28422,
      "natural language expressions": 76299,
      "semantic parsing significant": 105165,
      "introduce new techniques": 56504,
      "semantic parsing simple": 105166,
      "compositional distributional semantics": 20112,
      "logistic regression classifier": 65032,
      "advancement natural language": 3966,
      "topological analysis complex": 122678,
      "analysis complex networks": 5537,
      "finegrained entity type": 44350,
      "labels mention entity": 58619,
      "ling weld 2012": 64411,
      "improvements downstream tasks": 52846,
      "training data existing": 123458,
      "local context sentence": 64910,
      "baseline experimental results": 12218,
      "extraction important task": 42354,
      "important task natural": 52264,
      "task natural language": 118434,
      "language processing used": 59982,
      "applications previous studies": 6992,
      "real world scenarios": 96093,
      "learning based deep": 62388,
      "based deep neural": 11630,
      "deep neural network": 28354,
      "neural network model": 78328,
      "compare different approaches": 19237,
      "learning based neural": 62396,
      "based neural network": 11893,
      "learning word representations": 63167,
      "semantic relations words": 105219,
      "semantic relations propose": 105213,
      "propose method learn": 92773,
      "learn latent representation": 62078,
      "words proposed method": 134148,
      "words existing approaches": 133946,
      "existing approaches word": 40065,
      "word representation learning": 133438,
      "learning semantic relations": 62994,
      "word representations learnt": 133462,
      "representations learnt using": 99737,
      "using proposed method": 130061,
      "proposed method use": 93404,
      "learnt word representations": 63257,
      "problems experimental results": 91321,
      "experimental results possible": 40669,
      "word representations using": 133479,
      "small training datasets": 110220,
      "parsing semantic parsing": 85225,
      "word similarity neural": 133563,
      "neural machine translation": 77965,
      "neural language models": 77940,
      "language models learn": 59580,
      "models learn word": 73489,
      "learn word representations": 62194,
      "capture rich linguistic": 15372,
      "rich linguistic conceptual": 102765,
      "linguistic conceptual information": 64447,
      "conceptual information investigate": 20664,
      "information investigate embeddings": 54703,
      "investigate embeddings learned": 56752,
      "embeddings learned neural": 35782,
      "learned neural machine": 62236,
      "machine translation models": 66051,
      "neural language model": 77937,
      "language model embeddings": 59315,
      "embeddings translation models": 35993,
      "translation models outperform": 124961,
      "tasks require knowledge": 119461,
      "knowledge conceptual similarity": 57837,
      "translating english french": 124589,
      "english french english": 37141,
      "source target languages": 110836,
      "method training neural": 69194,
      "neural translation models": 78723,
      "quality embedding spaces": 94639,
      "neural language modelling": 77939,
      "language modelling machine": 59488,
      "modelling machine translation": 72610,
      "translation paper presents": 125091,
      "paper presents indepth": 84173,
      "presents indepth investigation": 89855,
      "language models translation": 59678,
      "language models difficult": 59539,
      "models difficult task": 73077,
      "realworld applications paper": 96151,
      "noise contrastive estimation": 80046,
      "backoff ngram models": 11316,
      "ngram models neural": 79447,
      "models neural models": 73630,
      "neural models make": 78181,
      "corresponding english translations": 24284,
      "nearest neighbor search": 76722,
      "distributional semantic model": 32719,
      "important information different": 52173,
      "translation lexicon induction": 124909,
      "information paper present": 54824,
      "based language model": 11792,
      "based language models": 11793,
      "commonly used word": 18974,
      "preliminary experimental results": 89267,
      "experimental results approach": 40550,
      "machine translation empirical": 65967,
      "challenge machine translation": 16051,
      "translation mt systems": 124986,
      "mt widely used": 75008,
      "widely used approach": 132561,
      "language model lm": 59344,
      "words speech recognition": 134236,
      "nbest list reranking": 76692,
      "machine translation challenging": 65921,
      "natural language sentence": 76531,
      "paper provides survey": 84388,
      "machine translation language": 66020,
      "language model important": 59331,
      "translation models phrasebased": 124963,
      "models generate target": 73287,
      "target language sentences": 117651,
      "paper tackle problem": 84465,
      "language model determine": 59312,
      "monolingual data order": 74579,
      "alleviate data sparsity": 5031,
      "data sparsity problem": 26486,
      "experiments chinesetoenglish translation": 40831,
      "language model significantly": 59400,
      "model significantly improve": 72020,
      "significantly improve translation": 108933,
      "improve translation quality": 52574,
      "absolute bleu score": 1724,
      "computational models language": 20412,
      "present comprehensive survey": 89415,
      "comprehensive survey word": 20279,
      "given language pair": 48055,
      "conduct qualitative analysis": 20889,
      "knowledge empirical results": 57890,
      "characteristics language pair": 16528,
      "word representations capture": 133446,
      "representations capture rich": 99530,
      "semantic relationships words": 105224,
      "word representations does": 133453,
      "model empirically demonstrate": 71060,
      "empirically demonstrate efficiency": 36226,
      "linguistics natural language": 64617,
      "language contains complex": 58917,
      "tools natural language": 122458,
      "language processing important": 59869,
      "applications downstream tasks": 6920,
      "downstream tasks information": 34040,
      "tasks information retrieval": 119203,
      "machine translation knowledge": 66018,
      "dependency parser based": 29160,
      "wsj test set": 135183,
      "models learn rich": 73488,
      "semantic word representations": 105356,
      "embedding models trained": 35458,
      "language model propose": 59384,
      "large amounts data": 61009,
      "significant reduction perplexity": 108855,
      "introduce method using": 56456,
      "specifically proposed method": 111588,
      "proposed method uses": 93405,
      "different similarity measures": 31428,
      "combinatory categorial grammar": 18612,
      "categorial grammar ccg": 15720,
      "model significantly outperforms": 72028,
      "significantly outperforms baseline": 109003,
      "outperforms baseline model": 82845,
      "model performs significantly": 71742,
      "performs significantly better": 87023,
      "gold standard data": 48439,
      "entity recognition neural": 38054,
      "stateoftheart named entity": 112764,
      "recognition ner systems": 96942,
      "rely handcrafted features": 98705,
      "nlp tasks partofspeech": 79749,
      "tasks partofspeech pos": 119374,
      "partofspeech pos tagging": 85534,
      "extensive number experiments": 41948,
      "number experiments using": 80887,
      "corpora different languages": 23459,
      "achieve stateoftheart results": 2566,
      "knowledge base construction": 57767,
      "distributional information sources": 32706,
      "dataset achieve stateoftheart": 26719,
      "vector representation words": 131346,
      "unsupervised learning approaches": 127655,
      "inspired deep learning": 55566,
      "framework significantly improve": 45683,
      "significantly improve quality": 108929,
      "words using syntactic": 134297,
      "conducted series experiments": 20947,
      "neural net models": 78230,
      "language model recently": 59387,
      "recently shown promising": 96766,
      "shown promising results": 108515,
      "promising results tasks": 92302,
      "results tasks machine": 102257,
      "tasks machine translation": 119277,
      "source words used": 110869,
      "image captioning models": 51770,
      "quality terms bleu": 94809,
      "input output vocabularies": 55384,
      "bidirectional long shortterm": 13938,
      "long shortterm memory": 65102,
      "shortterm memory lstm": 108324,
      "memory lstm neural": 68341,
      "lstm neural networks": 65656,
      "neural networks use": 78599,
      "experiments natural language": 41034,
      "comparable state art": 19176,
      "state art english": 112439,
      "achieve f1 score": 2459,
      "phrase based smt": 87335,
      "using neural networks": 129957,
      "neural networks vector": 78602,
      "networks vector representations": 77813,
      "vector representations paper": 131357,
      "representations paper propose": 99797,
      "phrasebased machine translation": 87383,
      "monolingual data method": 74575,
      "recently introduced neural": 96699,
      "introduced neural network": 56590,
      "vector representation models": 131344,
      "models used end": 74287,
      "improving statistical machine": 53171,
      "building recent success": 14880,
      "based neural networks": 11896,
      "neural networks model": 78514,
      "architecture neural network": 8710,
      "proposed methods results": 93416,
      "results significant improvements": 102180,
      "coherent long texts": 18320,
      "documents challenging problem": 33199,
      "models paper explore": 73699,
      "longshort term memory": 65188,
      "using standard metrics": 130212,
      "standard metrics like": 112265,
      "metrics like rouge": 69982,
      "showing neural models": 108419,
      "neural models able": 78160,
      "models described paper": 73047,
      "neural models nlp": 78186,
      "neural networks successfully": 78587,
      "models difficult interpret": 73076,
      "meanings words phrases": 67754,
      "using simple recurrent": 130178,
      "corpus annotated using": 23661,
      "annotated using amazon": 6258,
      "using amazon mechanical": 129333,
      "amazon mechanical turk": 5270,
      "relations empirical results": 98145,
      "various data sources": 131070,
      "people use language": 85901,
      "communicate natural language": 19019,
      "based largescale neural": 11796,
      "neural architecture developed": 77833,
      "neural network takes": 78402,
      "network takes input": 77450,
      "information working memory": 55109,
      "broad range language": 14676,
      "best knowledge time": 13354,
      "manually annotated corpus": 67026,
      "word vector representations": 133629,
      "representation learning words": 99313,
      "features downstream tasks": 43470,
      "present method constructing": 89548,
      "resources like wordnet": 101000,
      "paper present simple": 84133,
      "present simple approach": 89700,
      "proposed sequence sequence": 93549,
      "sequence sequence framework": 107055,
      "model trained endtoend": 72189,
      "trained endtoend requires": 123132,
      "preliminary results suggest": 89278,
      "common sense reasoning": 18924,
      "failure mode model": 42988,
      "complex network approach": 19844,
      "statistical methods widely": 113129,
      "methods widely employed": 69846,
      "language recent years": 60025,
      "recent years methods": 96618,
      "language models despite": 59537,
      "despite large studies": 29703,
      "large studies devoted": 61282,
      "studies devoted represent": 114210,
      "devoted represent texts": 30477,
      "represent texts physical": 99149,
      "texts physical models": 121578,
      "performance natural language": 86557,
      "processing tasks paper": 91822,
      "paper address problem": 83704,
      "able improve performance": 1653,
      "current statistical methods": 25371,
      "plays pivotal role": 87739,
      "semantic parsing based": 105145,
      "average f1 score": 11188,
      "word embeddings embeddings": 133094,
      "word embeddings input": 133137,
      "does need additional": 33373,
      "vector space word": 131389,
      "space word embeddings": 111085,
      "stateoftheart performance word": 112857,
      "performance word similarity": 86860,
      "word similarity word": 133573,
      "similarity word sense": 109336,
      "sense disambiguation tasks": 105668,
      "consider large number": 21211,
      "text data sets": 120853,
      "natural language cnl": 76273,
      "wide variety languages": 132523,
      "comprehensive survey existing": 20275,
      "natural languages english": 76608,
      "goal article provide": 48329,
      "article provide common": 9155,
      "starting point researchers": 112410,
      "help developers make": 49712,
      "experimental results tasks": 40709,
      "results tasks demonstrate": 102256,
      "tasks demonstrate effectiveness": 119036,
      "demonstrate effectiveness proposed": 28718,
      "word embedding training": 133039,
      "word embedding algorithms": 133004,
      "neural networks provide": 78547,
      "training word embeddings": 123953,
      "word embeddings discover": 133086,
      "corpus yields better": 24072,
      "yields better results": 135402,
      "used training corpus": 128824,
      "word embeddings using": 133236,
      "recent work explored": 96571,
      "continuous vector space": 22644,
      "space word representations": 111086,
      "introduce new approach": 56475,
      "continuous word representations": 22648,
      "word representations based": 133444,
      "approach performs significantly": 7798,
      "significantly better previous": 108888,
      "better previous approaches": 13681,
      "aspects natural language": 9398,
      "natural language systems": 76541,
      "obtained results suggest": 81406,
      "vector space models": 131374,
      "predict human judgments": 88891,
      "lexical semantic relations": 63814,
      "semantic relations word": 105218,
      "english italian german": 37177,
      "correlate human judgments": 24196,
      "correlations human judgments": 24260,
      "characters instead words": 16612,
      "transitionbased parser uses": 124512,
      "memory lstm recurrent": 68344,
      "lstm recurrent neural": 65661,
      "recurrent neural networks": 97226,
      "neural networks learn": 78502,
      "representations words using": 99994,
      "experiments morphologically rich": 41024,
      "languages parsing model": 60790,
      "words word sense": 134318,
      "sense disambiguation wsd": 105672,
      "languages world research": 60982,
      "state art performance": 112463,
      "bengali text corpus": 13035,
      "text corpus developed": 120835,
      "corpus developed tdil": 23753,
      "developed tdil project": 30309,
      "tdil project govt": 119679,
      "project govt india": 92218,
      "automatic text classification": 10693,
      "text classification machine": 120736,
      "classification machine learning": 17258,
      "extraction word sense": 42539,
      "automatic machine translation": 10582,
      "machine translation evaluation": 65975,
      "evaluation metric based": 39278,
      "dependency parsing model": 29182,
      "avoid problems propose": 11241,
      "problems propose novel": 91361,
      "propose novel automatic": 92904,
      "automatic evaluation metric": 10535,
      "machine translation orthographic": 66136,
      "especially closely related": 38440,
      "related language pairs": 97873,
      "maximum likelihood training": 67540,
      "training computationally expensive": 123393,
      "model address challenge": 70634,
      "perform approximate inference": 85947,
      "translation evaluation metrics": 124800,
      "automatic evaluation metrics": 10536,
      "metrics like bleu": 69980,
      "stateoftheart mt systems": 112759,
      "nbest lists using": 76695,
      "domain adaptation domain": 33432,
      "sets word embeddings": 107727,
      "word embeddings distributed": 133087,
      "embeddings distributed representations": 35661,
      "distributed representations words": 32622,
      "learning paper proposes": 62871,
      "word similarity analogy": 133553,
      "similarity analogy tasks": 109194,
      "tasks partofspeech tagging": 119376,
      "language models neural": 59607,
      "models neural networks": 73635,
      "neural networks shown": 78575,
      "shown improve performance": 108485,
      "method language modeling": 68929,
      "models machine translation": 73541,
      "machine translation decoder": 65944,
      "language processing models": 59888,
      "models natural language": 73607,
      "exploratory data analysis": 41512,
      "event extraction task": 39517,
      "computational linguistics cl": 20395,
      "provide comprehensive overview": 93788,
      "largescale datadriven methods": 61419,
      "methods widely used": 69847,
      "event detection event": 39509,
      "case study paper": 15623,
      "perform detailed analysis": 85980,
      "human machine translation": 50913,
      "machine translation specifically": 66235,
      "entity recognition classification": 38001,
      "classifier named entity": 17561,
      "person location organization": 87097,
      "essential challenging task": 38544,
      "aim work develop": 4476,
      "based multinomial naive": 11868,
      "multinomial naive bayes": 75474,
      "naive bayes mnb": 76019,
      "training corpus using": 123407,
      "term frequency inverse": 120216,
      "frequency inverse document": 45845,
      "inverse document frequency": 56698,
      "problematic natural language": 91293,
      "processing applications paper": 91620,
      "form word embeddings": 45149,
      "word embeddings enable": 133096,
      "present unsupervised approach": 89755,
      "semantic vector space": 105352,
      "splitting algorithm based": 111954,
      "english machine translation": 37204,
      "machine translation experiments": 65985,
      "received little attention": 96366,
      "little attention despite": 64792,
      "text classification problem": 120756,
      "naive bayes classifier": 76014,
      "extensive experimental evaluation": 41883,
      "compared state art": 19452,
      "state art methodologies": 112447,
      "text classification techniques": 120780,
      "realtime sentiment analysis": 96140,
      "state art benchmarks": 112430,
      "twitter sentiment analysis": 126123,
      "sentiment analysis consider": 106576,
      "dataset high quality": 26963,
      "performs slightly better": 87027,
      "perform significantly better": 86069,
      "task needs better": 118448,
      "parallel text corpora": 84689,
      "training language models": 123675,
      "metrics used evaluate": 70019,
      "additional sources information": 3574,
      "previous research shown": 90449,
      "developed method combines": 30287,
      "neural network models": 78342,
      "network models natural": 77348,
      "years neural networks": 135281,
      "yielding stateoftheart results": 135387,
      "recognition speech processing": 97006,
      "recently neural network": 96720,
      "perspective natural language": 87167,
      "language processing research": 59942,
      "feedforward networks convolutional": 43847,
      "networks convolutional networks": 77553,
      "convolutional networks recurrent": 23173,
      "networks recurrent networks": 77727,
      "important role shaping": 52246,
      "set train classifier": 107612,
      "sentence experimental results": 105863,
      "experimental results large": 40630,
      "neural networks multilingual": 78517,
      "multimodal representation learning": 75451,
      "representations learned using": 99733,
      "using parallel corpus": 130002,
      "parallel data available": 84634,
      "using parallel data": 130003,
      "using pivot language": 130019,
      "model achieves stateoftheart": 70596,
      "stateoftheart performance multilingual": 112831,
      "multilingual document classification": 75247,
      "publicly available multilingual": 94312,
      "promising results multilingual": 92299,
      "deep learning baselines": 28262,
      "paper presents results": 84201,
      "ubuntu dialog corpus": 126519,
      "largest publicly available": 61540,
      "evaluation using data": 39436,
      "evaluate performances various": 38893,
      "achieves stateoftheart result": 2898,
      "finally discuss future": 44170,
      "special case machine": 111351,
      "case machine translation": 15597,
      "machine translation machine": 66036,
      "translation machine translation": 124921,
      "translations systems perfect": 125497,
      "specific domains paper": 111434,
      "different training methods": 31507,
      "machine translation used": 66291,
      "different language models": 31192,
      "machine translation engine": 65971,
      "quality machine translation": 94711,
      "year nara institute": 135235,
      "nara institute science": 76158,
      "institute science technology": 55699,
      "workshop asian translation": 134999,
      "asian translation based": 9291,
      "evaluation measures bleu": 39265,
      "shortterm memory recurrent": 108346,
      "memory recurrent neural": 68372,
      "recurrent neural network": 97194,
      "neural network bidirectional": 78266,
      "network bidirectional long": 77170,
      "neural network blstmrnn": 78268,
      "network blstmrnn shown": 77176,
      "blstmrnn shown effective": 14383,
      "sequential data speech": 107218,
      "data speech utterances": 26494,
      "speech utterances handwritten": 111845,
      "utterances handwritten documents": 130643,
      "properties natural language": 92469,
      "study propose use": 114488,
      "propose use blstmrnn": 93140,
      "pos tagging task": 88238,
      "penn treebank wsj": 85866,
      "test set stateoftheart": 120520,
      "accuracy achieved using": 2094,
      "achieve good performance": 2469,
      "stanford pos tagger": 112376,
      "rational speech acts": 95834,
      "speech acts model": 111644,
      "speech acts rsa": 111645,
      "acts rsa model": 3147,
      "language use recursive": 60301,
      "use recursive process": 128235,
      "nonlinear activation function": 80222,
      "new application domains": 78782,
      "best performance achieved": 13391,
      "integer linear programming": 55735,
      "linear programming ilp": 64361,
      "orders magnitude faster": 82446,
      "results empirical study": 101750,
      "using word embedding": 130379,
      "word embeddings used": 133234,
      "word embeddings represent": 133199,
      "different machine learning": 31242,
      "methods sentiment classification": 69742,
      "sentiment classification including": 106700,
      "svm logistic regression": 116241,
      "logistic regression convolutional": 65033,
      "regression convolutional neural": 97698,
      "neural network cnn": 78278,
      "methods better performance": 69347,
      "better performance compared": 13653,
      "ngram models using": 79448,
      "using naive bayes": 129928,
      "naive bayes nb": 76020,
      "learning methods proposed": 62742,
      "precision recall f1": 88799,
      "british french novels": 14659,
      "using rulebased algorithm": 130131,
      "level lexical diversity": 63468,
      "models analysis data": 72749,
      "representations introduce new": 99705,
      "language models capture": 59525,
      "language modelling benchmarks": 59487,
      "task predicting syntactic": 118545,
      "range stateoftheart models": 95603,
      "stateoftheart neural language": 112776,
      "news articles achieve": 79288,
      "articles achieve stateoftheart": 9167,
      "achieve stateoftheart performance": 2558,
      "stateoftheart performance chinese": 112812,
      "information paper propose": 54825,
      "data paper describes": 26209,
      "ner shared task": 77079,
      "shared task goal": 108051,
      "social media text": 110421,
      "work propose novel": 134733,
      "propose novel method": 92937,
      "novel method incorporate": 80636,
      "discourse information language": 32061,
      "language model introduce": 59335,
      "recurrent language model": 97178,
      "language model based": 59294,
      "model based long": 70733,
      "based long shortterm": 11813,
      "shortterm memory units": 108350,
      "separate evaluation corpora": 106852,
      "proposed model improves": 93451,
      "evaluate proposed approach": 38897,
      "number context sentences": 80859,
      "language model improves": 59333,
      "range natural language": 95583,
      "application programming interface": 6876,
      "resources available open": 100951,
      "available open source": 11066,
      "characterbased neural machine": 16496,
      "machine translation introduce": 66013,
      "representations character sequences": 99540,
      "model target language": 72142,
      "target language translation": 117658,
      "word vectors word": 133652,
      "character level model": 16443,
      "present neural network": 89575,
      "neural network architecture": 78244,
      "based bidirectional lstms": 11553,
      "evaluate quality representations": 38907,
      "machine translation lexical": 66030,
      "lexical substitution task": 63825,
      "task obtain stateoftheart": 118471,
      "obtain stateoftheart results": 81329,
      "semantics previous works": 105451,
      "address problem previous": 3741,
      "represent words multiple": 99155,
      "embedded space paper": 35362,
      "space paper propose": 111038,
      "training experiments benchmarks": 123630,
      "effectiveness proposed model": 34936,
      "learning vector representations": 63158,
      "vector representations words": 131362,
      "neural network classifier": 78276,
      "used input features": 128594,
      "proposed approach demonstrated": 93190,
      "novel approach text": 80491,
      "text segmentation using": 121278,
      "information retrieval ir": 54935,
      "various nlp tasks": 131154,
      "tasks paper proposes": 119364,
      "hierarchical agglomerative clustering": 49924,
      "various levels granularity": 131120,
      "proposed method evaluated": 93366,
      "partofspeech tagging named": 85555,
      "tagging named entity": 117415,
      "data sparseness problem": 26479,
      "problem machine translation": 91115,
      "machine translation especially": 65973,
      "sourcetopivot pivottotarget translation": 110950,
      "pivottotarget translation models": 87595,
      "pivot phrase table": 87583,
      "method takes account": 69176,
      "experimental result shows": 40547,
      "method significantly outperforms": 69143,
      "significantly outperforms conventional": 109009,
      "using english pivot": 129628,
      "english pivot language": 37242,
      "modeling semantic relatedness": 72539,
      "semantic relatedness target": 105198,
      "context words sentence": 22331,
      "different context words": 31060,
      "sentiment polarity sentence": 106778,
      "target word context": 117743,
      "word context words": 132968,
      "memory lstm models": 68336,
      "benchmark dataset twitter": 12762,
      "models achieve stateoftheart": 72672,
      "achieve stateoftheart performances": 2564,
      "using syntactic parser": 130257,
      "syntactic parser external": 116439,
      "parser external sentiment": 84956,
      "external sentiment lexicons": 42043,
      "generate semantic representations": 47008,
      "concept space models": 20590,
      "performance benchmark datasets": 86171,
      "benchmark datasets measuring": 12791,
      "datasets measuring semantic": 27561,
      "results competitive performance": 101612,
      "compared prior stateoftheart": 19425,
      "prior stateoftheart methods": 90733,
      "statistical significance results": 113162,
      "results reported different": 102128,
      "methods measuring semantic": 69616,
      "hidden markov model": 49895,
      "markov model based": 67260,
      "extraction social media": 42490,
      "experiments carried jadavpur": 40813,
      "carried jadavpur university": 15543,
      "jadavpur university participation": 57176,
      "tool developed task": 122380,
      "developed task based": 30304,
      "task based trigram": 117922,
      "based trigram hidden": 12139,
      "trigram hidden markov": 125797,
      "markov model utilizes": 67263,
      "model utilizes information": 72290,
      "pos tag word": 88215,
      "tag word level": 117344,
      "word level features": 133340,
      "level features enhance": 63446,
      "features enhance observation": 43485,
      "enhance observation probabilities": 37479,
      "observation probabilities known": 81167,
      "probabilities known tokens": 90845,
      "known tokens unknown": 58325,
      "tokens unknown tokens": 122347,
      "unknown tokens submitted": 127361,
      "tokens submitted runs": 122336,
      "submitted runs english": 114768,
      "trained tested datasets": 123308,
      "tested datasets released": 120568,
      "attentionbased neural machine": 10085,
      "machine translation attentional": 65902,
      "translation attentional mechanism": 124654,
      "mechanism proven effective": 68033,
      "proven effective improving": 93720,
      "endtoend neural machine": 36939,
      "capture partial aspects": 15357,
      "partial aspects attentional": 85286,
      "aspects attentional regularities": 9374,
      "machine translation instead": 66012,
      "training sourcetotarget targettosource": 123873,
      "sourcetotarget targettosource translation": 110956,
      "targettosource translation models": 117808,
      "training data experiments": 123462,
      "englishfrench translation tasks": 37369,
      "training significantly improves": 123863,
      "generation using graphbased": 47693,
      "graphbased semisupervised learning": 48913,
      "semisupervised learning method": 105607,
      "learning method uses": 62732,
      "syntactic semantic relations": 116474,
      "features improve performance": 43554,
      "improve performance downstream": 52464,
      "performance downstream tasks": 86322,
      "tasks morphological tagging": 119312,
      "tagging dependency parsing": 117384,
      "attentionbased convolutional neural": 10062,
      "neural network modeling": 78341,
      "issue nlp tasks": 57014,
      "tasks answer selection": 118931,
      "textual entailment te": 121701,
      "based convolutional neural": 11609,
      "applied wide variety": 7145,
      "wide variety tasks": 132528,
      "loss semantic information": 65298,
      "methods achieve stateoftheart": 69284,
      "stateoftheart results tasks": 112922,
      "tasks sentiment classification": 119492,
      "embeddings methods proposed": 35808,
      "model able capture": 70510,
      "vectors experimental results": 131435,
      "shows model outperforms": 108600,
      "model outperforms previous": 71655,
      "deep learning models": 28285,
      "models source code": 74071,
      "new evaluation metric": 78909,
      "widely used metric": 132575,
      "evaluating language models": 39066,
      "shared vocabulary model": 108118,
      "address problem propose": 3742,
      "problem propose new": 91177,
      "language models like": 59586,
      "sentence level models": 105932,
      "neural network based": 78254,
      "network based language": 77160,
      "using neural encoderdecoder": 129947,
      "neural encoderdecoder framework": 77905,
      "attentionbased neural translation": 10092,
      "neural translation model": 78721,
      "learning word embeddings": 63166,
      "word embeddings provide": 133195,
      "extensive evaluation using": 41877,
      "using different types": 129577,
      "different types context": 31526,
      "skipgram word embeddings": 109995,
      "performance wide range": 86852,
      "extrinsic nlp tasks": 42623,
      "tasks results suggest": 119474,
      "word embeddings learned": 133147,
      "yield performance gains": 135347,
      "model learns word": 71454,
      "learns word embeddings": 63247,
      "word embeddings weighted": 133242,
      "embeddings weighted contexts": 36021,
      "neural encoderdecoder models": 77907,
      "machine translation achieved": 65884,
      "achieved impressive results": 2647,
      "languages low resource": 60706,
      "low resource setting": 65394,
      "semantic slot filling": 105300,
      "neural network rnn": 78385,
      "architectures long shortterm": 8818,
      "lstm widely used": 65675,
      "used sequence labeling": 128752,
      "sequence labeling paper": 106976,
      "sequence labeling explicitly": 106969,
      "explicitly model label": 41381,
      "encodes input sequence": 36675,
      "fixed length vector": 44599,
      "information input sequence": 54691,
      "slot filling task": 110060,
      "component natural language": 19990,
      "language understanding using": 60292,
      "crosslingual transfer learning": 25032,
      "transfer learning sequence": 124132,
      "target language training": 117657,
      "language training data": 60180,
      "training data consists": 123446,
      "languages labeled data": 60664,
      "training data noisy": 123503,
      "instead propose use": 55676,
      "use source language": 128286,
      "language models estimate": 59546,
      "significantly better performance": 108887,
      "online social networks": 81806,
      "online social network": 81805,
      "variety languages using": 130988,
      "using monolingual data": 129898,
      "languages using english": 60953,
      "achieve state art": 2554,
      "state art results": 112470,
      "crosslingual text classification": 25023,
      "word translation tasks": 133609,
      "stateoftheart nlp methods": 112790,
      "latent dirichlet allocation": 61581,
      "models language model": 73451,
      "language model takes": 59405,
      "evaluate quality generated": 38906,
      "experiments using new": 41196,
      "teaching large classes": 119711,
      "maximum likelihood estimation": 67537,
      "models easily overfit": 73114,
      "training data data": 123448,
      "alignment paper propose": 4977,
      "paper propose framework": 84242,
      "strategy according performance": 113499,
      "events text documents": 39605,
      "various machine learning": 131125,
      "applications event detection": 6924,
      "lexical syntactic semantic": 63830,
      "present new method": 89587,
      "prior work using": 90765,
      "using synthetic data": 130264,
      "sentiment analysis twitter": 106675,
      "analysis twitter data": 5913,
      "social networking sites": 110442,
      "like twitter facebook": 64107,
      "rapidly gaining popularity": 95731,
      "field sentiment analysis": 43982,
      "evaluation metrics using": 39297,
      "using various machine": 130355,
      "learning algorithms like": 62326,
      "entropy support vector": 38173,
      "challenges applications sentiment": 16129,
      "applications sentiment analysis": 7015,
      "task text segmentation": 118790,
      "model language generation": 71405,
      "computational text analysis": 20440,
      "different types features": 31529,
      "including support vector": 53388,
      "support vector machines": 116009,
      "vector machines svm": 131326,
      "linear discriminant analysis": 64346,
      "deep neural networks": 28366,
      "neural networks dnn": 78468,
      "mutual information matrix": 75978,
      "information matrix stochastic": 54754,
      "matrix stochastic gradient": 67483,
      "stochastic gradient descent": 113356,
      "learn rich representations": 62133,
      "recognition ner entity": 96926,
      "prior linguistic knowledge": 90717,
      "fully automated way": 45946,
      "tasks text mining": 119557,
      "search engine technology": 104311,
      "bidirectional neural network": 13967,
      "neural network language": 78313,
      "network language model": 77295,
      "conventional maximum likelihood": 22886,
      "models large scale": 73464,
      "text paper present": 121166,
      "neural network lstm": 78322,
      "nlp tasks word": 79777,
      "features improves performance": 43556,
      "example sentence selection": 39793,
      "sentence selection task": 106062,
      "natural language nl": 76376,
      "question answering sentence": 95101,
      "work propose model": 134727,
      "experimental results model": 40645,
      "performance answer sentence": 86139,
      "answer sentence selection": 6576,
      "achieves comparable result": 2763,
      "paraphrase identification task": 84827,
      "ultradense word embeddings": 126562,
      "useful nlp tasks": 128913,
      "nlp tasks paper": 79747,
      "tasks paper introduce": 119360,
      "reach state art": 95901,
      "sentiment analysis task": 106666,
      "order magnitude efficient": 82357,
      "match source text": 67367,
      "passages large corpus": 85616,
      "language processing machine": 59881,
      "case study approach": 15614,
      "use sequence alignment": 128269,
      "knowledgebased question answering": 58255,
      "annotated training data": 6249,
      "meaning representation methods": 67674,
      "relation extraction method": 97997,
      "question answering dataset": 95035,
      "dataset method achieves": 27019,
      "simple approach exploiting": 109362,
      "multiple word embeddings": 75747,
      "word embeddings sentence": 133205,
      "classification introduce novel": 17237,
      "convolution neural network": 23143,
      "network cnn architecture": 77188,
      "training time furthermore": 123924,
      "word embeddings dimensionality": 133083,
      "consistently outperforms baseline": 21430,
      "outperforms baseline models": 82846,
      "language generation spoken": 59095,
      "generation spoken dialogue": 47631,
      "domain adaptation paper": 33444,
      "recurrent neural networkbased": 97225,
      "small set indomain": 110198,
      "achieve competitive performance": 2446,
      "performance terms bleu": 86793,
      "terms bleu score": 120288,
      "new unseen domains": 79235,
      "named entities relations": 76036,
      "new entities relations": 78904,
      "best result obtained": 13433,
      "positive negative opinions": 88330,
      "information news articles": 54803,
      "articles written different": 9231,
      "written different languages": 135122,
      "social media posts": 110413,
      "pretrained word embeddings": 90216,
      "stateoftheart performance tasks": 112849,
      "disambiguation named entity": 31974,
      "named entity disambiguation": 76043,
      "entity disambiguation ned": 37923,
      "wordsense disambiguation wsd": 134330,
      "provide empirical evidence": 93813,
      "treestructured neural networks": 125730,
      "neural networks exploit": 78481,
      "syntactic parse information": 116435,
      "address issues introducing": 3702,
      "task significantly outperforms": 118717,
      "feature selection methods": 43308,
      "state art languages": 112443,
      "prepositional attachment disambiguation": 89310,
      "prepositional phrase pp": 89313,
      "like machine translation": 64061,
      "parallel data language": 84639,
      "model trained english": 72191,
      "trained english features": 123137,
      "features random forest": 43683,
      "et al 2016": 38702,
      "random forest algorithm": 95492,
      "10fold cross validation": 207,
      "achieves f1 score": 2786,
      "et al 2014": 38699,
      "levy et al": 63717,
      "et al 2015": 38701,
      "drastically reduce bias": 34104,
      "results competitive stateoftheart": 101615,
      "et al 2010": 38692,
      "models paper present": 73704,
      "memory lstm language": 68333,
      "lstm language model": 65631,
      "language model language": 59338,
      "language generation model": 59075,
      "method outperforms traditional": 69036,
      "language model perplexity": 59368,
      "neural attention models": 77845,
      "key term extraction": 57610,
      "neural network architectures": 78250,
      "attention mechanism neural": 9907,
      "neural attention model": 77844,
      "attention model shown": 9932,
      "shown promising performance": 108514,
      "tasks including speech": 119189,
      "including speech recognition": 53379,
      "speech recognition image": 111754,
      "image caption generation": 51761,
      "visual question answering": 131803,
      "sequence classification tasks": 106924,
      "sequence labeling tasks": 106985,
      "performance attention mechanism": 86154,
      "task experimental results": 118173,
      "experimental results attention": 40556,
      "results attention mechanism": 101527,
      "sequence labeling task": 106984,
      "neural networks natural": 78520,
      "networks natural language": 77669,
      "use multitask learning": 128158,
      "slot filling models": 110050,
      "training data needed": 123499,
      "new task proposed": 79205,
      "models generalize unseen": 73278,
      "generalize unseen words": 46823,
      "little training data": 64831,
      "training data used": 123562,
      "covering different domains": 24537,
      "characterlevel neural translation": 16578,
      "multilingual media monitoring": 75279,
      "nlp tasks like": 79733,
      "speech recognition asr": 111736,
      "recognition asr machine": 96811,
      "asr machine translation": 9434,
      "sequence sequence neural": 107062,
      "joint multitask learning": 57301,
      "learning multilingual neural": 62814,
      "multilingual neural translation": 75312,
      "translation morphologically rich": 124973,
      "results paper novel": 102024,
      "paper novel approach": 84055,
      "novel approach automatic": 80469,
      "adjoining grammar tag": 3867,
      "languages like tamil": 60697,
      "language paper present": 59792,
      "knowledge graph construction": 57960,
      "answering information retrieval": 6635,
      "retrieval information extraction": 102398,
      "shallow parsing analysis": 107929,
      "finally experimental results": 44183,
      "experimental results prove": 40683,
      "lstm language models": 65633,
      "growing body research": 49166,
      "recent recurrent neural": 96505,
      "using bidirectional lstm": 129401,
      "bidirectional longshort term": 13947,
      "model takes input": 72138,
      "vocabulary size training": 131904,
      "takes advantage explicit": 117526,
      "stateoftheart performance standard": 112846,
      "disfluency detection task": 32358,
      "task analysis shows": 117877,
      "analysis shows model": 5851,
      "tuning statistical machine": 125942,
      "address problem automatically": 3728,
      "multilingual partofspeech tagging": 75332,
      "shortterm memory bilstm": 108318,
      "memory bilstm networks": 68295,
      "sequence modeling tasks": 107015,
      "target languages data": 117665,
      "pos taggers languages": 88223,
      "auxiliary loss function": 10879,
      "model obtains stateoftheart": 71605,
      "obtains stateoftheart performance": 81478,
      "morphologically complex languages": 74761,
      "data sparsity issues": 26485,
      "obtain better results": 81265,
      "machine learning research": 65824,
      "humans learn language": 51087,
      "weston et al": 132459,
      "largescale question answering": 61502,
      "european framework reference": 38768,
      "work studies comparatively": 134822,
      "studies comparatively typical": 114195,
      "comparatively typical sentence": 19217,
      "sentence pair classification": 105965,
      "classification tasks textual": 17459,
      "tasks textual entailment": 119560,
      "entailment te answer": 37680,
      "te answer selection": 119684,
      "answer selection observing": 6574,
      "prior work limitations": 90752,
      "work limitations phrase": 134613,
      "limitations phrase generation": 64186,
      "phrase generation representation": 87350,
      "generation representation ii": 47597,
      "representation ii conducts": 99254,
      "ii conducts alignment": 51702,
      "conducts alignment word": 20968,
      "alignment word phrase": 5003,
      "word phrase levels": 133404,
      "phrase levels handcrafted": 87356,
      "levels handcrafted features": 63539,
      "handcrafted features iii": 49343,
      "features iii utilizes": 43545,
      "iii utilizes single": 51722,
      "considering characteristics specific": 21317,
      "characteristics specific tasks": 16534,
      "specific tasks limits": 111499,
      "effectiveness tasks propose": 34962,
      "tasks propose architecture": 119406,
      "propose architecture based": 92554,
      "architecture based gated": 8625,
      "based gated recurrent": 11731,
      "gated recurrent unit": 46515,
      "recurrent unit supports": 97266,
      "unit supports representation": 127212,
      "supports representation learning": 116040,
      "representation learning phrases": 99305,
      "learning phrases arbitrary": 62891,
      "phrases arbitrary granularity": 87408,
      "arbitrary granularity ii": 8583,
      "granularity ii taskspecific": 48748,
      "phrase alignments sentences": 87332,
      "experimental results te": 40710,
      "results te match": 102260,
      "te match observation": 119687,
      "languages using word": 60957,
      "word embedding language": 133013,
      "based word embedding": 12164,
      "called word embedding": 15068,
      "use parallel corpora": 128190,
      "recent work word": 96596,
      "syntactic structure important": 116483,
      "news articles web": 79307,
      "exercise item candidates": 39987,
      "order automatically identify": 82282,
      "language learning exercises": 59242,
      "achieved average precision": 2596,
      "languages like chinese": 60692,
      "previous work mainly": 90535,
      "work mainly focused": 134625,
      "paper propose neural": 84271,
      "process experimental results": 91480,
      "experimental results proposed": 40670,
      "significantly outperforms stateoftheart": 109031,
      "outperforms stateoftheart approaches": 82998,
      "datasets experiment results": 27466,
      "experiment results chinese": 40492,
      "chinese zero pronoun": 16844,
      "languages provide analysis": 60824,
      "machine translation order": 66135,
      "phenomena neural machine": 87237,
      "models use large": 74278,
      "memory usage paper": 68394,
      "traditional machine translation": 122833,
      "model experimental results": 71128,
      "experimental results largescale": 40631,
      "method achieves better": 68593,
      "achieves better translation": 2752,
      "better translation performance": 13751,
      "vocabulary neural machine": 131887,
      "time unlike previous": 122130,
      "unlike previous work": 127444,
      "approach able jointly": 7291,
      "jointly model relevance": 57364,
      "taskspecific word representation": 119647,
      "variety downstream tasks": 130974,
      "propose novel framework": 92920,
      "patterns training data": 85788,
      "training data test": 123549,
      "data test data": 26553,
      "feedforward neural network": 43851,
      "memory natural language": 68352,
      "dependency trees improve": 29244,
      "languages large amounts": 60675,
      "annotated corpora available": 6144,
      "address issue using": 3700,
      "using parallel corpora": 130001,
      "parallel corpora consisting": 84597,
      "resourcepoor resourcerich language": 100917,
      "resourcerich language pairs": 100925,
      "large amounts annotated": 61007,
      "amounts annotated data": 5329,
      "used improve performance": 128583,
      "translation mt evaluation": 124980,
      "manual automatic evaluation": 66982,
      "automatic evaluation methods": 10534,
      "precision recall fmeasure": 88801,
      "features semantic features": 43716,
      "language models deep": 59533,
      "models deep learning": 73022,
      "introduce evaluation methods": 56422,
      "quality estimation qe": 94650,
      "automatic evaluation measures": 10532,
      "machine translation investigate": 66015,
      "translation investigate use": 124878,
      "machine translation nmt": 66086,
      "language model score": 59395,
      "neural network text": 78406,
      "network text classification": 77454,
      "classification multitask learning": 17295,
      "multitask learning neural": 75857,
      "learning neural network": 62842,
      "network based methods": 77162,
      "variety natural language": 131000,
      "processing tasks previous": 91825,
      "tasks previous works": 119398,
      "insufficient training data": 55728,
      "training data paper": 123506,
      "multitask learning framework": 75838,
      "learning framework jointly": 62602,
      "framework jointly learn": 45590,
      "jointly learn multiple": 57348,
      "learn multiple related": 62097,
      "multiple related tasks": 75663,
      "based recurrent neural": 11978,
      "network trained jointly": 77460,
      "trained jointly tasks": 123165,
      "text classification tasks": 120773,
      "tasks proposed models": 119418,
      "proposed models improve": 93495,
      "models improve performance": 73367,
      "features cognate identification": 43404,
      "paper explore use": 83926,
      "purpose cognate identification": 94425,
      "string similarity measures": 113600,
      "different language families": 31191,
      "achieve competitive results": 2448,
      "large text corpora": 61291,
      "propose simple approach": 93051,
      "new stateoftheart performances": 79165,
      "neural networks existing": 78477,
      "networks existing methods": 77593,
      "sentence paper propose": 105975,
      "paper propose deep": 84227,
      "datasets demonstrate efficacy": 27405,
      "demonstrate efficacy proposed": 28730,
      "efficacy proposed architecture": 35016,
      "machine translation stateoftheart": 66239,
      "grammatical error correction": 48697,
      "grammar error correction": 48633,
      "error correction gec": 38295,
      "correction gec tasks": 24152,
      "outperforms previously published": 82962,
      "previously published results": 90616,
      "results conll2014 test": 101626,
      "conll2014 test set": 21107,
      "test set large": 120507,
      "set large margin": 107476,
      "trained publicly available": 123253,
      "publicly available data": 94298,
      "novel approach proposed": 80486,
      "automatically construct parallel": 10741,
      "dialogue machine translation": 30705,
      "translation experiments proposed": 124813,
      "experiments proposed method": 41072,
      "proposed method achieve": 93336,
      "language model adaptation": 59289,
      "bleu points improvement": 14308,
      "word classes adjectives": 132953,
      "outperforms stateoftheart models": 83009,
      "translation models neural": 124958,
      "models neural machine": 73627,
      "machine translation discriminative": 65953,
      "source sentence paper": 110812,
      "paper propose variational": 84342,
      "encoderdecoder model trained": 36602,
      "encoderdecoder model generates": 36600,
      "model generates target": 71242,
      "representations source sentences": 99898,
      "model introduces continuous": 71372,
      "introduces continuous latent": 56607,
      "continuous latent variable": 22622,
      "latent variable explicitly": 61627,
      "variable explicitly model": 130839,
      "order perform efficient": 82383,
      "source target sides": 110841,
      "variational lower bound": 130928,
      "english german translation": 37157,
      "translation tasks proposed": 125360,
      "machine translation achieves": 65885,
      "significant improvements vanilla": 108814,
      "machine translation baselines": 65910,
      "embeddings paper propose": 35851,
      "multiple levels granularity": 75605,
      "generate tree structures": 47040,
      "embeddings different levels": 35650,
      "different levels granularity": 31229,
      "embeddings source target": 35947,
      "neural model trained": 78155,
      "bilingual semantic similarity": 14056,
      "experiments nist chineseenglish": 41042,
      "model achieves substantial": 70610,
      "achieves substantial improvement": 2912,
      "bleu points average": 14305,
      "evaluation nlp methods": 39311,
      "given test set": 48153,
      "nlp task paper": 79703,
      "recognizing textual entailment": 97079,
      "performance standard evaluation": 86735,
      "standard evaluation metrics": 112238,
      "words semantically similar": 134190,
      "approach sets new": 7887,
      "temporal information extraction": 120108,
      "vanilla recurrent neural": 130819,
      "word embeddings trained": 133227,
      "distant supervision rules": 32478,
      "relations events time": 98158,
      "events time expressions": 39607,
      "scale training data": 103758,
      "deep reinforcement learning": 28410,
      "learning dialogue generation": 62501,
      "generation recent neural": 47583,
      "recent neural models": 96482,
      "reinforcement learning paper": 97819,
      "reinforcement learning model": 97816,
      "model future reward": 71217,
      "using policy gradient": 130028,
      "policy gradient methods": 87953,
      "proposed algorithm generates": 93175,
      "understanding existing models": 126841,
      "models discourse coherence": 73082,
      "models learn distinguish": 73481,
      "models produce coherent": 73809,
      "work achieves stateoftheart": 134352,
      "stateoftheart performance multiple": 112832,
      "pseudo training data": 94187,
      "zero pronoun resolution": 135476,
      "pronoun resolution existing": 92342,
      "resolution existing approaches": 100762,
      "relying annotated data": 98756,
      "shared task organizers": 108075,
      "lack annotated data": 58679,
      "data major obstacle": 26110,
      "major obstacle progress": 66576,
      "data better performance": 25703,
      "problem paper propose": 91157,
      "paper propose simple": 84321,
      "propose simple novel": 93066,
      "simple novel approach": 109488,
      "novel approach automatically": 80470,
      "automatically generate largescale": 10775,
      "clozestyle reading comprehension": 17903,
      "reading comprehension neural": 95999,
      "results proposed approach": 102062,
      "proposed approach significantly": 93214,
      "approach significantly outperforms": 7907,
      "outperforms stateoftheart systems": 83016,
      "sentiment classification problem": 106710,
      "transfer knowledge learned": 124066,
      "source language lowresource": 110776,
      "arabic sentiment classification": 8553,
      "sentiment classification demonstrate": 106698,
      "systems neural machine": 117019,
      "memory paper propose": 68365,
      "neural machine translator": 78137,
      "representation source sentence": 99414,
      "incorporating external knowledge": 53538,
      "external knowledge neural": 42014,
      "knowledge neural machine": 58082,
      "generic neural machine": 47804,
      "discrete translation lexicons": 32181,
      "lexicons neural machine": 63932,
      "translation neural machine": 125008,
      "understanding meaning sentence": 126886,
      "method alleviate problem": 68629,
      "time neural machine": 122057,
      "machine translation simultaneous": 66221,
      "translation simultaneous translation": 125264,
      "translation introduce novel": 124875,
      "novel decoding algorithm": 80534,
      "existing neural machine": 40228,
      "translation based neural": 124665,
      "based neural machine": 11888,
      "contrary common belief": 22678,
      "results significantly improve": 102183,
      "use semi supervised": 128262,
      "conditional random fields": 20782,
      "automatic manual evaluation": 10585,
      "machine translation focused": 65989,
      "apply neural machine": 7197,
      "machine translation task": 66259,
      "extensive comparison using": 41867,
      "comparison using various": 19586,
      "neural translation systems": 78724,
      "machine translation significantly": 66219,
      "wmt 2016 shared": 132772,
      "shared news translation": 107995,
      "news translation task": 79388,
      "systems language pairs": 116967,
      "language pairs trained": 59782,
      "corpus additional training": 23647,
      "additional training data": 3588,
      "systems human evaluation": 116939,
      "improve neural machine": 52438,
      "machine translation recently": 66192,
      "recently achieved impressive": 96649,
      "neural mt models": 78212,
      "morphological features partofspeech": 74702,
      "features partofspeech tags": 43651,
      "translation systems experiments": 125311,
      "training test sets": 123914,
      "lexicons word vector": 63937,
      "representations proven useful": 99833,
      "reach stateoftheart results": 95904,
      "datasets morphologically rich": 27582,
      "text representation learning": 121246,
      "representation learning propose": 99310,
      "learning propose new": 62924,
      "new active learning": 78764,
      "active learning al": 3110,
      "text classification convolutional": 120715,
      "classification convolutional neural": 17160,
      "convolutional neural networks": 23199,
      "neural networks cnns": 78454,
      "effort neural models": 35179,
      "approach document classification": 7501,
      "empirical results method": 36186,
      "results method outperforms": 101921,
      "method outperforms baseline": 69008,
      "document classification tasks": 32969,
      "best knowledge work": 13356,
      "neural models text": 78203,
      "models text classification": 74183,
      "experiments chineseenglish dataset": 40823,
      "continuous bag words": 22613,
      "model neural network": 71573,
      "proven surprisingly successful": 93734,
      "embeddings word embeddings": 36029,
      "word embeddings directly": 133085,
      "neural network learns": 78320,
      "encoder decoder architecture": 36505,
      "based machine translation": 11823,
      "multiple target languages": 75722,
      "work explore idea": 134511,
      "encoder decoder architectures": 36506,
      "training data available": 123432,
      "data available training": 25685,
      "available training data": 11132,
      "data available case": 25673,
      "report promising results": 99030,
      "introduce new method": 56486,
      "present novel unsupervised": 89626,
      "novel unsupervised approach": 80765,
      "exploiting syntactic information": 41482,
      "existing unsupervised methods": 40329,
      "stateoftheart supervised models": 112982,
      "news article summaries": 79286,
      "stateoftheart systems human": 112990,
      "sentiment analysis approach": 106567,
      "propose sentiment analysis": 93039,
      "newly created corpus": 79268,
      "wikipedia talk page": 132680,
      "best model achieves": 13369,
      "model achieves promising": 70584,
      "detection online discussions": 30011,
      "outperform stateoftheart approaches": 82751,
      "stateoftheart approaches datasets": 112582,
      "model achieves f1": 70570,
      "achieves f1 scores": 2787,
      "linear chain crf": 64331,
      "language model sparse": 59403,
      "features neural network": 43626,
      "neural network joint": 78310,
      "network joint model": 77291,
      "language model unsupervised": 59416,
      "translation tasks neural": 125359,
      "neural sequence sequence": 78673,
      "sequence sequence learning": 107056,
      "paradigm machine translation": 84539,
      "machine translation achieving": 65886,
      "achieving competitive results": 2941,
      "recently published methods": 96745,
      "methods used neural": 69832,
      "multimodal machine translation": 75441,
      "micro f1 score": 70036,
      "neural response generation": 78655,
      "consider incorporating topic": 21206,
      "incorporating topic information": 53569,
      "chatbots end propose": 16670,
      "joint attention mechanism": 57265,
      "topic vectors topic": 122584,
      "topic words message": 122587,
      "obtained pretrained lda": 81395,
      "pretrained lda model": 90119,
      "evaluation metrics human": 39288,
      "response generation models": 101205,
      "simple neural network": 109483,
      "neural network word": 78416,
      "source target word": 110845,
      "given target word": 48147,
      "machine translation quality": 66180,
      "translation quality estimation": 125170,
      "quality estimation machine": 94648,
      "estimation machine translation": 38652,
      "notoriously difficult task": 80419,
      "information extraction propose": 54595,
      "approach biomedical information": 7393,
      "large body work": 61044,
      "texts complex networks": 121479,
      "network model able": 77326,
      "machine learning context": 65791,
      "detection social media": 30057,
      "media text messages": 68169,
      "set rules detect": 107570,
      "learning approaches work": 62367,
      "state art approaches": 112428,
      "rely hand crafted": 98702,
      "hand crafted features": 49316,
      "network based models": 77164,
      "require handcrafted features": 100142,
      "languages paper propose": 60782,
      "propose neural network": 92824,
      "network based model": 77163,
      "word character level": 132943,
      "corpora available language": 23419,
      "leads improved performance": 61937,
      "meaningful feature representations": 67717,
      "experiment language pairs": 40470,
      "dutch word embeddings": 34285,
      "word embeddings linguistic": 133151,
      "word embeddings recently": 133198,
      "variety tasks research": 131022,
      "evaluation dialect identification": 39181,
      "using word vector": 130385,
      "vector representations capture": 131349,
      "capture linguistic properties": 15331,
      "using dimensionality reduction": 129580,
      "semantically similar words": 105379,
      "representation neural machine": 99354,
      "continuous vector representations": 22643,
      "continuous vector representation": 22642,
      "vector representation word": 131345,
      "representation word embedding": 99464,
      "encoder decoder recurrent": 36511,
      "networks neural machine": 77675,
      "machine translation need": 66080,
      "source target words": 110847,
      "words based context": 133835,
      "word embedding vectors": 133041,
      "embedding vectors using": 35528,
      "improves translation quality": 53060,
      "translation quality neural": 125179,
      "quality neural machine": 94729,
      "pos tagging using": 88240,
      "using crosslingual projection": 129520,
      "paper introduce novel": 84000,
      "approach sequence tagging": 7883,
      "bridge gap normative": 14592,
      "uses dependency trees": 129223,
      "improves neural machine": 52998,
      "complexity neural machine": 19929,
      "translation nmt systems": 125059,
      "luong et al": 65726,
      "named entity ne": 76051,
      "sequence sequence model": 107058,
      "translation model used": 124945,
      "bilingual training corpus": 14064,
      "tasks evaluation word": 119107,
      "evaluation word similarity": 39447,
      "partofspeech tagging demonstrate": 85550,
      "recurrent convolutional neural": 97168,
      "achieving new stateoftheart": 2963,
      "new stateoftheart performance": 79162,
      "semantic parsers map": 105138,
      "leverage information contained": 63590,
      "knowledge bases kbs": 57807,
      "recently proposed methods": 96740,
      "models arbitrary language": 72768,
      "knowledge base prior": 57781,
      "significantly improved performance": 108936,
      "performance stateoftheart baselines": 86744,
      "opendomain natural language": 81973,
      "natural language question": 76518,
      "language question answering": 60017,
      "question answering task": 95111,
      "dependency language models": 29153,
      "dependency parsing paper": 29189,
      "parsing paper present": 85182,
      "paper present approach": 84082,
      "present approach improve": 89380,
      "approach improve accuracy": 7619,
      "transitionbased dependency parser": 124503,
      "language models parser": 59616,
      "effectiveness proposed approach": 34926,
      "achieve competitive accuracy": 2445,
      "achieved stateoftheart accuracy": 2704,
      "competitive results english": 19684,
      "crossdocument coreference resolution": 24882,
      "coreference resolution named": 23368,
      "resolution named entities": 100773,
      "report results using": 99039,
      "collection news articles": 18482,
      "word2vec mikolov et": 133678,
      "mikolov et al": 70080,
      "et al 2013a": 38697,
      "despite promising results": 29717,
      "reproduce results paper": 100083,
      "rigorous empirical evaluation": 102851,
      "using models trained": 129891,
      "models trained large": 74210,
      "using pretrained word": 130050,
      "release source code": 98473,
      "models discriminating similar": 73086,
      "discriminating similar languages": 32201,
      "language social media": 60101,
      "social media messages": 110390,
      "step linguistic processing": 113279,
      "dissimilar language pairs": 32398,
      "language pairs propose": 59776,
      "models paper introduce": 73701,
      "model jointly learns": 71384,
      "word vectors experiments": 133636,
      "phrase similarity tasks": 87370,
      "dependency parsing using": 29209,
      "semantics pragmatics paper": 105445,
      "pragmatics paper present": 88763,
      "paper present novel": 84124,
      "contextfree grammars cfgs": 22362,
      "critical task natural": 24837,
      "previous works focused": 90571,
      "collect large corpus": 18383,
      "paper publicly available": 84390,
      "based dynamic programming": 11666,
      "kneserney language model": 57723,
      "experiments english czech": 40932,
      "significant perplexity reductions": 108846,
      "features deep learning": 43440,
      "deep learning methods": 28280,
      "methods automatically detect": 69332,
      "process widely used": 91588,
      "phrase based statistical": 87336,
      "based statistical machine": 12072,
      "methods focus sentence": 69515,
      "comparison existing methods": 19544,
      "primary goal thesis": 90651,
      "grammar induction task": 48641,
      "new word analogy": 79250,
      "word analogy corpus": 132917,
      "words word embedding": 134316,
      "word embedding methods": 133019,
      "nlp natural language": 79651,
      "word embeddings english": 133098,
      "english words phrases": 37351,
      "goal paper explore": 48379,
      "stateoftheart word embedding": 113029,
      "introduce new corpus": 56477,
      "word analogy task": 132918,
      "experiment word2vec glove": 40521,
      "results corpus corpus": 101644,
      "corpus available research": 23675,
      "available research community": 11092,
      "play critical role": 87688,
      "development test sets": 30432,
      "poses major challenge": 88268,
      "languages significant differences": 60873,
      "differences word order": 30981,
      "word order paper": 133379,
      "dependency parse tree": 29157,
      "translation output experiments": 125082,
      "statistically significant improvement": 113184,
      "compared stateoftheart statistical": 19461,
      "high dimensional spaces": 50063,
      "embeddings machine translation": 35797,
      "translation outofvocabulary words": 125080,
      "alleviate problem propose": 5044,
      "problem propose use": 91184,
      "source word model": 110866,
      "target language model": 117640,
      "uses word embeddings": 129300,
      "machine translation obtain": 66131,
      "consistent improvements translation": 21390,
      "improvements translation quality": 52934,
      "englishspanish language pair": 37430,
      "different languages exhibit": 31200,
      "overall best performing": 83218,
      "spoken language understanding": 111996,
      "language understanding paper": 60249,
      "language understanding introduce": 60230,
      "bidirectional long short": 13935,
      "long short term": 65093,
      "short term memory": 108248,
      "term memory networks": 120235,
      "deep learning sequence": 28314,
      "sequence labelling task": 106996,
      "input output sequences": 55383,
      "word attention mechanism": 132931,
      "limitation propose novel": 64166,
      "standard atis dataset": 112193,
      "experiments proposed model": 41079,
      "proposed model robust": 93479,
      "robust speech recognition": 103073,
      "speech recognition errors": 111750,
      "native english speakers": 76221,
      "gap human performance": 46456,
      "representation learning models": 99299,
      "models substantial differences": 74113,
      "causal language model": 15808,
      "language understanding work": 60296,
      "language identification social": 59142,
      "identification social media": 51444,
      "language identification method": 59134,
      "unsupervised approach using": 127604,
      "approach using large": 8007,
      "using large corpus": 129787,
      "using latent dirichlet": 129798,
      "results suggest certain": 102229,
      "sentence embeddings using": 105847,
      "using auxiliary prediction": 129375,
      "auxiliary prediction tasks": 10883,
      "fixed length vectors": 44600,
      "averaging word vectors": 11229,
      "based hidden states": 11756,
      "hidden states recurrent": 49914,
      "subsequent machine learning": 114815,
      "machine learning tasks": 65831,
      "encoded sentence representations": 36483,
      "word content word": 132966,
      "prediction task using": 89137,
      "analysis sheds light": 5844,
      "sentence embedding methods": 105834,
      "representations natural language": 99768,
      "language processing using": 59983,
      "human language processing": 50901,
      "challenges natural language": 16178,
      "process large data": 91520,
      "language processing architectures": 59843,
      "language models large": 59577,
      "translation smt neural": 125272,
      "englishgerman language pair": 37381,
      "better translation quality": 13752,
      "analysis provides useful": 5787,
      "sentiment classification sentiment": 106712,
      "social media increasingly": 110386,
      "traditional sentiment analysis": 122864,
      "sentiment analysis methods": 106609,
      "sentiment analysis social": 106657,
      "analysis social media": 5860,
      "social media content": 110361,
      "present approach leverage": 89381,
      "publicly available research": 94320,
      "available research purposes": 11093,
      "sequence sequence architecture": 107053,
      "response generation neural": 101206,
      "generation neural machine": 47507,
      "generates target sentence": 47183,
      "target sentence word": 117700,
      "sentence word word": 106129,
      "paper proposed novel": 84348,
      "proposed novel approach": 93513,
      "generation short text": 47620,
      "proposed approach outperforms": 93203,
      "approach outperforms stateoftheart": 7777,
      "outperforms stateoftheart approach": 82997,
      "automatic manual evaluations": 10586,
      "improve language models": 52404,
      "language models study": 59664,
      "network language models": 77299,
      "variety neural network": 131008,
      "sentiment analysis automatic": 106571,
      "sentiment analysis language": 106602,
      "sparse vector space": 111241,
      "model paper explore": 71692,
      "paper explore alternative": 83914,
      "word representations word": 133480,
      "word similarity benchmarks": 133557,
      "model competitive stateoftheart": 70864,
      "real world data": 96091,
      "radically benchmark corpora": 95442,
      "nlp models trained": 79645,
      "natural language variation": 76603,
      "context propose new": 22234,
      "task automatically generating": 117915,
      "facts knowledge base": 42916,
      "bleu point improvement": 14300,
      "correlation human judgments": 24233,
      "semantic similarity relatedness": 105289,
      "based human curated": 11759,
      "contextual information available": 22470,
      "vector space representation": 131381,
      "improves correlation human": 52964,
      "method compares favorably": 68713,
      "different word embedding": 31567,
      "machine translation efforts": 65966,
      "different languages key": 31206,
      "algorithm outperforms current": 4777,
      "improvement bleu points": 52688,
      "applications neural language": 6976,
      "neural language modeling": 77938,
      "function used word2vec": 46051,
      "contrastive estimation nce": 22729,
      "word representations unlike": 133477,
      "nce language models": 76703,
      "processing tasks word": 91831,
      "objective function experimental": 81079,
      "function experimental results": 46016,
      "experimental results popular": 40667,
      "reviews sentiment analysis": 102630,
      "language processing work": 59987,
      "text classification methods": 120740,
      "applied new domains": 7101,
      "dependency parsing present": 29194,
      "single deep neural": 109725,
      "reach stateoftheart performance": 95903,
      "languages universal dependencies": 60938,
      "universal dependencies treebank": 127293,
      "word source language": 133578,
      "supervised learning paper": 115773,
      "characterbased encoderdecoder model": 16490,
      "consists recurrent neural": 21491,
      "neural networks encoder": 78473,
      "bidirectional recurrent neural": 13970,
      "vector representation decoder": 131342,
      "representation decoder generates": 99205,
      "attentionbased recurrent neural": 10095,
      "attention mechanism jointly": 9902,
      "experiments different datasets": 40907,
      "different datasets proposed": 31079,
      "proposed encoderdecoder model": 93266,
      "model able achieve": 70509,
      "able achieve significantly": 1598,
      "traditional statistical models": 122872,
      "italian paper present": 57097,
      "data automatically generated": 25664,
      "field computational linguistics": 43936,
      "paper presents overview": 84196,
      "tts automatic speech": 125915,
      "used select appropriate": 128742,
      "domain knowledge text": 33561,
      "experiments english russian": 40936,
      "networks propose novel": 77712,
      "propose novel semantic": 92964,
      "uses word character": 129299,
      "task semantic tagging": 118675,
      "significantly outperforms prior": 109026,
      "english universal dependencies": 37326,
      "deep multitask learning": 28341,
      "multitask learning shared": 75869,
      "models achieved impressive": 72682,
      "data paper propose": 26216,
      "model external memory": 71159,
      "proposed architectures improve": 93228,
      "task amrtotext generation": 117874,
      "results machine translation": 101900,
      "lack parallel corpora": 58732,
      "previous work relied": 90543,
      "advantage neural models": 4053,
      "translation nmt heavily": 125034,
      "nmt heavily relies": 79872,
      "learn semantic representations": 62140,
      "semantic representations input": 105236,
      "representations input sentences": 99696,
      "languages natural word": 60755,
      "natural word delimiters": 76626,
      "word delimiters chinese": 132984,
      "experiment results chineseenglish": 40493,
      "results chineseenglish translation": 101575,
      "chineseenglish translation demonstrate": 16860,
      "distributional semantics present": 32731,
      "distributional semantics model": 32729,
      "vector representations based": 131348,
      "simple vector operations": 109541,
      "recent previous work": 96494,
      "languages lowresource languages": 60708,
      "languages spoken language": 60891,
      "language resources likely": 60054,
      "resources likely annotated": 101002,
      "likely annotated translations": 64135,
      "annotated translations transcriptions": 6252,
      "translated speech data": 124564,
      "speech translation systems": 111837,
      "making use data": 66872,
      "translations present model": 125487,
      "using dynamic time": 129612,
      "dynamic time warping": 34330,
      "extremely lowresource scenario": 42605,
      "opened new research": 81984,
      "utilize network structures": 130519,
      "propose effective approach": 92640,
      "structure experimental results": 113858,
      "experimental results chinese": 40565,
      "evaluation rnn architectures": 39374,
      "sentence classification task": 105786,
      "task recurrent neural": 118616,
      "neural networks achieved": 78436,
      "networks achieved stateoftheart": 77503,
      "achieved stateoftheart results": 2707,
      "present empirical study": 89467,
      "empirical study using": 36206,
      "pooling model achieves": 88045,
      "model achieves best": 70554,
      "achieves best performance": 2738,
      "best performance datasets": 13392,
      "model does outperform": 71020,
      "learning sentence representation": 63002,
      "recently progress learning": 96737,
      "sentence representations used": 106044,
      "models typically treat": 74258,
      "end propose novel": 36825,
      "propose novel attention": 92902,
      "extensive experiments demonstrate": 41905,
      "experiments demonstrate proposed": 40889,
      "demonstrate proposed methods": 28845,
      "proposed methods significantly": 93417,
      "methods significantly improve": 69751,
      "significantly improve stateoftheart": 108932,
      "sentence representation models": 106032,
      "using bidirectional recurrent": 129403,
      "neural networks work": 78608,
      "crosslingual annotation projection": 24928,
      "using recurrent neural": 130089,
      "neural networks rnn": 78560,
      "parallel corpus source": 84623,
      "corpus source target": 24005,
      "source target language": 110835,
      "applicable wide range": 6831,
      "propose method include": 92771,
      "parallel corpora obtained": 84606,
      "strategies neural machine": 113478,
      "sentence recent work": 106017,
      "extend previous work": 41793,
      "englishgerman wmt16 englishromanian": 37395,
      "state art neural": 112458,
      "based natural language": 11879,
      "features word embeddings": 43787,
      "word embeddings evaluated": 133102,
      "results suggest model": 102233,
      "data syntactic structures": 26527,
      "models language change": 73449,
      "translation related languages": 125211,
      "tweets labeled using": 126034,
      "labeled using hashtags": 58478,
      "distant supervision classifiers": 32466,
      "smt neural machine": 110288,
      "united nations parallel": 127227,
      "efficient neural machine": 35098,
      "current neural machine": 25308,
      "resources paper presents": 101026,
      "paper presents approach": 84149,
      "evaluation machine translation": 39259,
      "mt evaluation metric": 74965,
      "evaluation metric meteor": 39281,
      "available github page": 11002,
      "word embeddings word": 133247,
      "word embeddings demonstrated": 133077,
      "benefit nlp tasks": 12985,
      "vector representations current": 131350,
      "propose novel models": 92945,
      "novel models improve": 80650,
      "improve word embeddings": 52585,
      "word embeddings unsupervised": 133232,
      "word embeddings based": 133053,
      "embeddings outperform original": 35846,
      "machine translation memory": 66043,
      "language statistical machine": 60124,
      "systems recent years": 117102,
      "translation lowresource languages": 124919,
      "lowresource languages present": 65525,
      "hindienglish machine translation": 50490,
      "lowresource language pair": 65503,
      "augmenting training data": 10347,
      "sentence previous methods": 106002,
      "previous methods use": 90430,
      "attention mechanism local": 9904,
      "word embeddings use": 133233,
      "performance standard test": 86736,
      "standard test set": 112321,
      "recursive neural networks": 97287,
      "neural networks sequential": 78574,
      "sequential recurrent neural": 107249,
      "sentence pairs existing": 105970,
      "given pair sentences": 48073,
      "tasks semantic similarity": 119485,
      "task automatically identifying": 117916,
      "power recurrent neural": 88651,
      "propose predictive model": 92995,
      "information model achieves": 54769,
      "paper empirically explore": 83890,
      "stacked bidirectional lstm": 112123,
      "models obtain stateoftheart": 73655,
      "used learn better": 128608,
      "auxiliary task multitask": 10890,
      "models significantly better": 74039,
      "models trained text": 74234,
      "lstm networks model": 65651,
      "networks model trained": 77661,
      "neural network paper": 78368,
      "submission shared task": 114736,
      "level quality estimation": 63494,
      "machine translation wmt16": 66300,
      "shared task predict": 108085,
      "paper propose novel": 84287,
      "propose novel approach": 92896,
      "language model rnnlm": 59391,
      "model rnnlm architecture": 71937,
      "different natural language": 31278,
      "vector space language": 131372,
      "approach language independent": 7663,
      "translation nmt new": 125045,
      "nmt new approach": 79932,
      "new approach machine": 78789,
      "great progress recent": 49016,
      "progress recent years": 92180,
      "fluent inadequate translations": 44710,
      "tu et al": 125922,
      "et al 2016b": 38709,
      "et al 2016a": 38708,
      "et al 2017": 38710,
      "conventional statistical machine": 22902,
      "leverage advantages models": 63575,
      "work propose incorporate": 134722,
      "endtoend manner experimental": 36917,
      "manner experimental results": 66947,
      "experimental results chineseenglish": 40566,
      "translation proposed approach": 125155,
      "proposed approach achieves": 93183,
      "achieves significant consistent": 2853,
      "significant consistent improvements": 108743,
      "consistent improvements stateoftheart": 21388,
      "translation nmt significantly": 125054,
      "nmt significantly improved": 79962,
      "improved translation quality": 52651,
      "machine translation shown": 66215,
      "input target language": 55450,
      "target language neural": 117642,
      "language neural machine": 59713,
      "machine translation pbmt": 66154,
      "german translation task": 47921,
      "translation task using": 125346,
      "mt bleu points": 74962,
      "demonstration neural machine": 28988,
      "large number training": 61191,
      "work present approach": 134675,
      "32 language pairs": 930,
      "approach yields similar": 8031,
      "yields similar results": 135436,
      "simple effective method": 109411,
      "method makes use": 68953,
      "information target language": 55029,
      "target language source": 117652,
      "language source language": 60105,
      "annotation projection method": 6359,
      "used previous work": 128697,
      "previous work setting": 90547,
      "previous work results": 90545,
      "universal dependencies corpora": 127284,
      "byte pair encoding": 14998,
      "using byte pair": 129422,
      "currently best performing": 25401,
      "character sequences basic": 16466,
      "languages writing systems": 60984,
      "characters input output": 16610,
      "reduces computational complexity": 97381,
      "improves translation performance": 53058,
      "approaches typically based": 8386,
      "neural networks rnns": 78561,
      "suffer problem generating": 115239,
      "problem generating short": 91067,
      "utterances paper propose": 130660,
      "systems open domain": 117032,
      "encoderdecoder neural networks": 36614,
      "encoderdecoder neural models": 36612,
      "monotone sequencetosequence tasks": 74641,
      "research fields including": 100514,
      "speech recognition investigate": 111757,
      "learning distributed representations": 62512,
      "develop simple effective": 30232,
      "simple effective model": 109413,
      "vector space propose": 131380,
      "analyze strengths weaknesses": 6009,
      "methods public datasets": 69702,
      "movies tv shows": 74917,
      "experimental results indicate": 40623,
      "results indicate proposed": 101858,
      "outperform current stateoftheart": 82705,
      "current stateoftheart methods": 25358,
      "sequencetosequence neural network": 107189,
      "paper demonstrates neural": 83805,
      "neural sequencetosequence models": 78678,
      "state art close": 112432,
      "close state art": 17834,
      "results existing datasets": 101774,
      "work focus task": 134533,
      "focus task extracting": 44830,
      "news articles labels": 79297,
      "articles labels derived": 9195,
      "previous machine reading": 90422,
      "introduce novel neural": 56517,
      "novel neural architecture": 80667,
      "models trained endtoend": 74201,
      "resulting relative improvement": 101467,
      "remains open paper": 98813,
      "social media experiments": 110379,
      "processing nlp tasks": 91755,
      "using term frequencyinverse": 130281,
      "term frequencyinverse document": 120220,
      "frequencyinverse document frequency": 45856,
      "document frequency tfidf": 33013,
      "syntactic semantic relationships": 116475,
      "classification paper proposes": 17326,
      "paper proposes novel": 84363,
      "distributed vector representation": 32631,
      "model rnnlm long": 71938,
      "rnnlm long shortterm": 102955,
      "rnn language model": 102919,
      "train translation systems": 123048,
      "framework investigate different": 45586,
      "drawn lot attention": 34135,
      "weak supervision propose": 132184,
      "using long short": 129832,
      "term memory lstm": 120229,
      "memory lstm network": 68337,
      "learning unified framework": 63134,
      "unified framework experimental": 127109,
      "framework experimental results": 45534,
      "proposed method significantly": 93398,
      "method significantly outperform": 69140,
      "significantly outperform baseline": 108980,
      "outperform baseline methods": 82690,
      "rarely stated explicitly": 95764,
      "neural sequencetosequence model": 78677,
      "vectorspace word representations": 131502,
      "prior works use": 90768,
      "proposed method improves": 93373,
      "word approach outperforms": 132926,
      "approach outperforms prior": 7775,
      "outperforms prior works": 82967,
      "empirical study vietnamese": 36207,
      "vietnamese dependency parsing": 131675,
      "parsing paper presents": 85184,
      "paper presents empirical": 84165,
      "experimental results neural": 40660,
      "results neural networkbased": 101986,
      "significantly better traditional": 108893,
      "labeled attachment score": 58425,
      "unlabeled attachment score": 127370,
      "cogalexv shared task": 18255,
      "shared task showing": 108096,
      "multiple semantic relations": 75682,
      "simple neural networks": 109484,
      "despite using fewer": 29741,
      "transfer learning settings": 124133,
      "model simultaneously learns": 72051,
      "language models propose": 59636,
      "class language models": 17044,
      "language models treat": 59679,
      "natural language descriptions": 76281,
      "novel neural network": 80670,
      "network architecture task": 77146,
      "reduces error rate": 97384,
      "gap model trained": 46467,
      "model trained data": 72185,
      "data neural machine": 26164,
      "translation nmt achieved": 125021,
      "nmt achieved remarkable": 79826,
      "achieved remarkable progress": 2685,
      "remarkable progress past": 98840,
      "progress past years": 92172,
      "problem propose novel": 91178,
      "input source sentence": 55440,
      "experiments proposed framework": 41071,
      "proposed framework significantly": 93298,
      "framework significantly improves": 45684,
      "significantly improves adequacy": 108941,
      "improves adequacy nmt": 52947,
      "make publicly available": 66714,
      "rouge evaluation metric": 103293,
      "popular research topic": 88126,
      "experimental results showed": 40698,
      "showed promising performance": 108384,
      "types texts different": 126369,
      "set named entities": 107503,
      "named entities usually": 76040,
      "detection natural language": 30000,
      "training data approach": 123428,
      "use target language": 128313,
      "algorithm makes use": 4769,
      "new recurrent neural": 79099,
      "features conditional random": 43419,
      "conditional random field": 20775,
      "random field crf": 95478,
      "recurrent neural models": 97192,
      "neural models achieved": 78162,
      "features prove important": 43677,
      "work introduce new": 134577,
      "introduce new recurrent": 56498,
      "neural crf model": 77882,
      "model different neural": 70995,
      "structured prediction tasks": 114030,
      "tasks experiments model": 119123,
      "experiments model outperforms": 41015,
      "model outperforms stateoftheart": 71668,
      "outperforms stateoftheart methods": 83006,
      "chinese word segmentation": 16835,
      "word segmentation pos": 133503,
      "segmentation pos tagging": 104619,
      "neural network named": 78360,
      "network named entity": 77366,
      "entity recognition chinese": 37998,
      "recognition chinese social": 96831,
      "chinese social media": 16821,
      "social media focus": 110383,
      "focus named entity": 44793,
      "recognition ner chinese": 96923,
      "ner chinese social": 77028,
      "propose semisupervised learning": 93034,
      "learning model based": 62758,
      "deep learning model": 28283,
      "improvement previous stateoftheart": 52748,
      "previous stateoftheart result": 90482,
      "hybrid neural network": 51189,
      "brings big challenge": 14644,
      "using prior knowledge": 130057,
      "gated recurrent units": 46519,
      "recurrent units gru": 97273,
      "processed convolutional neural": 91595,
      "neural network generate": 78302,
      "model extends existing": 71149,
      "results extensive experiments": 101788,
      "extensive experiments public": 41925,
      "experiments public data": 41090,
      "public data sets": 94245,
      "using pointer network": 130021,
      "works mainly focused": 134962,
      "models incorporate contextual": 73385,
      "models paper propose": 73706,
      "paper propose endtoend": 84232,
      "propose endtoend neural": 92652,
      "pointer network ptrnet": 87827,
      "alleviate error propagation": 5035,
      "error propagation problem": 38319,
      "utilize contextual information": 130500,
      "information experimental results": 54550,
      "experimental results effectiveness": 40609,
      "results effectiveness proposed": 101746,
      "test collection produced": 120429,
      "word error rates": 133274,
      "quality assessment new": 94597,
      "embeddings based neural": 35566,
      "neural network approaches": 78243,
      "task machine learning": 118366,
      "data affects performance": 25586,
      "affects performance machine": 4248,
      "rapid development deep": 95714,
      "development deep learning": 30381,
      "language processing widely": 59984,
      "processing widely used": 91857,
      "widely used feature": 132570,
      "word order information": 133376,
      "features partofspeech tagging": 43650,
      "systems present new": 117071,
      "present new model": 89588,
      "word embeddings make": 133156,
      "word embedding models": 133024,
      "model construction training": 70904,
      "evaluate word embeddings": 38945,
      "use word embeddings": 128362,
      "word embedding second": 133032,
      "word representation introduce": 133437,
      "introduce joint training": 56441,
      "document representation models": 33065,
      "representation models including": 99341,
      "models present new": 73773,
      "new model recurrent": 79034,
      "widely used various": 132595,
      "order address issues": 82272,
      "address issues present": 3705,
      "stateoftheart methods open": 112712,
      "minimum spanning tree": 70219,
      "require large amounts": 100155,
      "paper aims provide": 83728,
      "aims provide comprehensive": 4565,
      "offered tei guidelines": 81576,
      "representation lexical information": 99317,
      "rule based statistical": 103351,
      "employ different methods": 36259,
      "detect different types": 29801,
      "pos tagged corpus": 88218,
      "english indian language": 37171,
      "narrow search space": 76193,
      "word embeddings natural": 133166,
      "embeddings natural language": 35823,
      "capture lexical semantics": 15328,
      "utilize distributed representations": 130504,
      "dictionary definitions words": 30878,
      "model architectures based": 70685,
      "multiple data sets": 75532,
      "error analysis suggests": 38287,
      "words multiple meanings": 134067,
      "word representations estimated": 133454,
      "corpus paper propose": 23924,
      "propose new neural": 92873,
      "new neural network": 79056,
      "proposed neural network": 93508,
      "neural network trained": 78408,
      "speech recognition engine": 111747,
      "widely used speech": 132588,
      "used speech recognition": 128775,
      "opensource natural language": 82023,
      "integrated crisis early": 55774,
      "crisis early warning": 24784,
      "complementary entity recognition": 19742,
      "work reviewed product": 134788,
      "address problem complementary": 3729,
      "problem complementary entity": 90972,
      "entity recognition cer": 37997,
      "method solve problem": 69158,
      "solve problem propose": 110607,
      "novel unsupervised method": 80767,
      "proposed method does": 93361,
      "method does require": 68785,
      "labeled data training": 58448,
      "demonstrate proposed approach": 28833,
      "proposed approach effective": 93193,
      "customer service propose": 25447,
      "propose novel probabilistic": 92956,
      "outperforms heuristic baselines": 82910,
      "knowledge machine learning": 58059,
      "sources external information": 110890,
      "linked open data": 64647,
      "used modern nlp": 128640,
      "modern nlp techniques": 74409,
      "techniques word embeddings": 120014,
      "conceptnet knowledge graph": 20598,
      "improving natural language": 53123,
      "combined word embeddings": 18675,
      "resources wordnet dbpedia": 101069,
      "demonstrate stateoftheart results": 28870,
      "driven great progress": 34153,
      "documents address problem": 33178,
      "address problem introducing": 3737,
      "community question answering": 19091,
      "provides useful information": 94094,
      "existing unsupervised method": 40328,
      "classifier second stage": 17579,
      "human annotated data": 50741,
      "data conduct experiments": 25772,
      "proposed method effective": 93364,
      "characterlevel neural machine": 16576,
      "machine translation attractive": 65904,
      "agreement long distances": 4396,
      "largescale data set": 61417,
      "wmt englishgerman translation": 132788,
      "englishgerman translation task": 37388,
      "rules report results": 103435,
      "systems perform better": 117057,
      "bytepair encoding bpe": 15005,
      "sentiment polarity lexicons": 106773,
      "paper propose replace": 84315,
      "machine translation address": 65887,
      "classification task new": 17435,
      "data selection method": 26385,
      "translation quality compared": 125165,
      "compared stateoftheart method": 19458,
      "detection neural networks": 30003,
      "neural networks attention": 78441,
      "proven effective natural": 93723,
      "approaches use external": 8391,
      "use external resources": 128046,
      "set new state": 107506,
      "new state art": 79147,
      "large set linguistic": 61270,
      "set linguistic features": 107481,
      "tense aspect mood": 120181,
      "representations widely used": 99977,
      "widely used text": 132593,
      "multiclass multilabel classification": 75057,
      "multilabel classification tasks": 75153,
      "outperform previous stateoftheart": 82729,
      "liu et al": 64841,
      "achieve significant reduction": 2542,
      "achieves best worlds": 2743,
      "time space complexity": 122105,
      "gained lot attention": 46366,
      "commonly used text": 18973,
      "using continuous vector": 129489,
      "language model used": 59418,
      "model used improve": 72255,
      "multilingual language models": 75269,
      "language models language": 59573,
      "multinomial logistic regression": 75472,
      "logistic regression model": 65037,
      "participation shared task": 85366,
      "shared task detecting": 108041,
      "forum information retrieval": 45306,
      "information retrieval evaluation": 54932,
      "evaluated test set": 39015,
      "2016 shared task": 618,
      "teams participated shared": 119729,
      "participated shared task": 85346,
      "second highest score": 104413,
      "score systems participated": 104131,
      "neural networks representation": 78556,
      "interpretability paper propose": 56228,
      "paper propose general": 84245,
      "propose general methodology": 92698,
      "using reinforcement learning": 130102,
      "multiple nlp tasks": 75630,
      "nlp tasks including": 79726,
      "sentencelevel sentiment analysis": 106174,
      "sentiment analysis document": 106586,
      "neural models deep": 78171,
      "present novel neural": 89617,
      "learning sequence labeling": 63011,
      "introduce neural network": 56468,
      "percentage points f1": 85925,
      "social media mining": 110391,
      "word embeddings capture": 133057,
      "outperforms previous work": 82957,
      "performances comparable supervised": 86873,
      "extensive empirical experiments": 41871,
      "rapid growth social": 95720,
      "growth social media": 49196,
      "social media data": 110366,
      "experimental results demonstrate": 40584,
      "results demonstrate proposed": 101698,
      "demonstrate proposed model": 28847,
      "proposed model obtains": 93454,
      "topic modeling classification": 122545,
      "apply machine learning": 7182,
      "crosslingual dependency parsing": 24938,
      "truly lowresource languages": 125860,
      "graphbased neural network": 48906,
      "languages compared previous": 60455,
      "compared previous state": 19416,
      "previous state art": 90466,
      "art neural machine": 9070,
      "attention researchers field": 9996,
      "modified loss function": 74447,
      "new loss function": 78998,
      "generation domain adaptation": 47377,
      "response generation conversational": 101200,
      "based sequence sequence": 12026,
      "propose twophase approach": 93124,
      "approach achieves good": 7305,
      "achieves good results": 2794,
      "translation language directions": 124888,
      "introduced neural machine": 56588,
      "conduct multifaceted evaluation": 20885,
      "stateoftheart neural machine": 112778,
      "machine translation phrasebased": 66157,
      "effect sentence length": 34612,
      "neural network distinguishing": 78295,
      "achieve high performance": 2476,
      "high performance nlp": 50097,
      "systems notoriously difficult": 117028,
      "methods proven effective": 69698,
      "relations paper present": 98232,
      "lexical syntactic information": 63829,
      "syntactic information successfully": 116422,
      "parsing universal dependencies": 85273,
      "universal dependencies ud": 127295,
      "parser requires training": 84993,
      "conversation response generation": 22969,
      "baseline sequencetosequence models": 12308,
      "user generated comments": 128986,
      "largescale social media": 61510,
      "social media news": 110393,
      "document classification paper": 32965,
      "text documents current": 120884,
      "use deep neural": 127986,
      "rich set features": 102786,
      "gives best results": 48182,
      "2016 evaluation campaign": 608,
      "sequence model neural": 107009,
      "domain adaptation techniques": 33457,
      "language understanding nlu": 60243,
      "understanding nlu tasks": 126911,
      "neural network dnn": 78296,
      "methods consider tasks": 69394,
      "sequence labeling problem": 106977,
      "labeling problem word": 58528,
      "labels paper propose": 58629,
      "paper propose alternative": 84218,
      "propose alternative approach": 92544,
      "propose neural models": 92823,
      "slot filling tasks": 110063,
      "deep memory networks": 28337,
      "endtoend machine learning": 36915,
      "machine learning architecture": 65778,
      "deep memory network": 28336,
      "conventional machine learning": 22882,
      "stateoftheart deep learning": 112631,
      "features sentiment analysis": 43721,
      "sentiment analysis sentiments": 106655,
      "usergenerated short text": 129076,
      "semantic pragmatic levels": 105182,
      "used sentiment analysis": 128750,
      "sentiment analysis sarcasm": 106648,
      "analysis sarcasm detection": 5830,
      "sarcasm detection cognitive": 103641,
      "detection cognitive features": 29900,
      "statistical classification using": 113083,
      "classification using enhanced": 17490,
      "using enhanced feature": 129631,
      "enhanced feature set": 37508,
      "parts speech pos": 85587,
      "speech pos tagging": 111725,
      "using machine learning": 129846,
      "machine learning technique": 65832,
      "deep learning survey": 28318,
      "survey natural language": 116180,
      "knowledge recent advances": 58136,
      "recent advances machine": 96407,
      "advances machine learning": 3995,
      "machine learning ml": 65802,
      "learning ml natural": 62753,
      "ml natural language": 70437,
      "new deep learning": 78877,
      "deep learning networks": 28301,
      "translation nmt models": 125039,
      "prepositional phrase attachment": 89312,
      "aims answer questions": 4502,
      "introduce syntactic information": 56548,
      "distillation neural machine": 32517,
      "translation knowledge distillation": 124884,
      "translation quality ensemble": 125169,
      "training leads better": 123684,
      "leads better translation": 61924,
      "processing deep neural": 91658,
      "field natural language": 43968,
      "processing nlp convolutional": 91730,
      "nlp convolutional neural": 79592,
      "network cnn recurrent": 77197,
      "cnn recurrent neural": 18019,
      "state art nlp": 112461,
      "art nlp tasks": 9073,
      "representative nlp tasks": 100015,
      "neural conversation model": 77874,
      "models capable generating": 72875,
      "novel fully datadriven": 80578,
      "setting approach yields": 107733,
      "yields significant improvements": 135429,
      "significant improvements competitive": 108792,
      "long standing problem": 65133,
      "entity recognition text": 38082,
      "semantic knowledge base": 105091,
      "finegrained entity types": 44351,
      "loc org misc": 64902,
      "make datasets publicly": 66653,
      "datasets publicly available": 27655,
      "namedentity recognition ner": 76138,
      "feature machine translation": 43293,
      "trained neural network": 123219,
      "neural network work": 78418,
      "detection paper proposes": 30018,
      "paper proposes use": 84371,
      "representation words word": 99471,
      "words word embeddings": 134317,
      "combine different methods": 18624,
      "different methods proposed": 31258,
      "overall f1 score": 83231,
      "recent work shows": 96588,
      "maps natural language": 67167,
      "natural language logical": 76370,
      "experiments question answering": 41101,
      "outperforms strong baselines": 83026,
      "linguistic resources paper": 64550,
      "using universal dependencies": 130339,
      "suffers data sparsity": 115253,
      "low similarity scores": 65399,
      "method generate fully": 68855,
      "entity semantic relatedness": 38101,
      "error rate compared": 38328,
      "document classification models": 32963,
      "classification models outperform": 17280,
      "method achieve better": 68584,
      "achieve better accuracy": 2425,
      "using deep learning": 129542,
      "specific natural language": 111469,
      "nlp tasks comparison": 79711,
      "popularity word embedding": 88158,
      "use deep learning": 127982,
      "results proposed approaches": 102066,
      "relatively little attention": 98398,
      "machine translation previous": 66169,
      "translation previous work": 125140,
      "paper propose hybrid": 84251,
      "machine translation approach": 65895,
      "experiments language pairs": 40973,
      "multilingual corpus translations": 75226,
      "languages english german": 60538,
      "english german italian": 37154,
      "german italian dutch": 47900,
      "approach based crosslingual": 7382,
      "semantic analysis based": 104978,
      "discourse representation theory": 32084,
      "supervised learning procedure": 115776,
      "procedure real life": 91398,
      "proposed approach detect": 93191,
      "learning procedure using": 62914,
      "massive text corpora": 67338,
      "important various tasks": 52291,
      "existing methods rely": 40192,
      "text corpora new": 120830,
      "experts paper propose": 41247,
      "achieves better performance": 2746,
      "partofspeech pos tags": 85540,
      "knowledge base wikipedia": 57794,
      "compared stateoftheart methods": 19459,
      "shown significant improvements": 108523,
      "realworld datasets different": 96163,
      "datasets different domains": 27425,
      "different domains languages": 31114,
      "words paper present": 134103,
      "test sets compared": 120527,
      "neural networks propose": 78542,
      "propose deep learning": 92616,
      "sequence labeling approach": 106967,
      "model recurrent neural": 71872,
      "memory lstm cells": 68327,
      "novel attention mechanism": 80501,
      "rnn compare model": 102912,
      "information extraction scientific": 54599,
      "extraction scientific literature": 42475,
      "learning algorithms able": 62323,
      "maximum entropy model": 67534,
      "generation text generation": 47679,
      "present case study": 89401,
      "dialogue generation model": 30677,
      "possible responses given": 88429,
      "models experiments model": 73205,
      "model generates diverse": 71238,
      "train recurrent neural": 123005,
      "sequencetosequence model attention": 107166,
      "achieves bleu score": 2754,
      "dialogue generation models": 30678,
      "models trained different": 74199,
      "different degrees data": 31088,
      "use reinforcement learning": 128237,
      "lack direct parallel": 58701,
      "direct parallel corpus": 31798,
      "relatively simple neural": 98413,
      "neural network recurrent": 78380,
      "features recurrent neural": 43692,
      "natural language inference": 76325,
      "work investigate using": 134593,
      "training data improve": 123476,
      "data improve generalization": 26021,
      "learning models using": 62797,
      "training deep neural": 123576,
      "neural networks dnns": 78469,
      "provided human annotators": 93970,
      "language inference nli": 59177,
      "inference nli task": 54181,
      "training set size": 123852,
      "present results empirical": 89668,
      "text classification based": 120712,
      "based distributed word": 11651,
      "distributed word representation": 32636,
      "graphbased dependency parsing": 48895,
      "dependency parsing dependency": 29176,
      "parsing dependency parsing": 85098,
      "dependency parsing important": 29181,
      "parsing important nlp": 85127,
      "important nlp task": 52198,
      "semantic similarity measures": 105284,
      "propose new hybrid": 92855,
      "usual vector space": 130401,
      "automatic text generation": 10694,
      "word segmentation word": 133508,
      "segmentation word segmentation": 104654,
      "far paper present": 43095,
      "achieve high accuracy": 2473,
      "external linguistic knowledge": 42031,
      "external knowledge used": 42021,
      "model text comprehension": 72164,
      "text comprehension tasks": 120809,
      "achieve new stateoftheart": 2503,
      "new stateoftheart results": 79167,
      "analysis learned representations": 5676,
      "paper explores problem": 83933,
      "vectors language models": 131448,
      "similarity propose new": 109288,
      "completely unsupervised experiments": 19785,
      "experiments publicly available": 41097,
      "publicly available dataset": 94300,
      "compared strong supervised": 19470,
      "strong supervised baselines": 113720,
      "discovery massive text": 32143,
      "text mining nlp": 121124,
      "research previous studies": 100585,
      "large scale study": 61262,
      "study propose novel": 114487,
      "demonstrate proposed framework": 28838,
      "sequence labelling tasks": 106997,
      "recognition asr systems": 96816,
      "proposes novel technique": 93614,
      "improved performance tasks": 52627,
      "improving neural machine": 53127,
      "machine translation conditional": 65928,
      "generative adversarial nets": 47710,
      "paper proposes approach": 84351,
      "generative adversarial net": 47709,
      "results proposed model": 102079,
      "proposed model consistently": 93440,
      "model consistently outperforms": 70893,
      "englishgerman chineseenglish translation": 37373,
      "chineseenglish translation tasks": 16863,
      "dialogue systems recently": 30786,
      "supervised learning problem": 115774,
      "question answering tasks": 95112,
      "reinforcement learning method": 97814,
      "policy gradient algorithm": 87951,
      "problem generating natural": 91065,
      "crosslingual semantic textual": 25005,
      "semantic textual similarity": 105329,
      "textual similarity sts": 121743,
      "english spanish arabic": 37282,
      "understanding flow information": 126847,
      "method msc using": 68974,
      "informative sentences maintaining": 55154,
      "sentences maintaining grammaticality": 106392,
      "propose hierarchical recurrent": 92712,
      "hierarchical recurrent neural": 49995,
      "results neural networks": 101987,
      "achieve better performance": 2431,
      "like social media": 64094,
      "social media twitter": 110429,
      "media twitter facebook": 68177,
      "document classification problems": 32967,
      "document classification approaches": 32962,
      "approaches study aims": 8358,
      "preprocessing feature extraction": 89324,
      "english wikipedia articles": 37340,
      "text simplification using": 121308,
      "evaluation metric machine": 39279,
      "metric machine translation": 69885,
      "topics natural language": 122641,
      "processing nlp literature": 91741,
      "automatic mt evaluation": 10612,
      "language pairs english": 59747,
      "english target language": 37302,
      "additional linguistic features": 3543,
      "features achieve good": 43349,
      "evaluation methods investigate": 39275,
      "external linguistic resources": 42032,
      "vector machine svm": 131316,
      "word representations paper": 133465,
      "paper propose extension": 84238,
      "different word representations": 31573,
      "significant number new": 108829,
      "language understanding present": 60252,
      "neural network classifiers": 78277,
      "color reference games": 18532,
      "novel crosslingual transfer": 80519,
      "crosslingual transfer method": 25034,
      "neural encoderdecoder model": 77906,
      "state art monolingual": 112451,
      "use labeled data": 128109,
      "data highresource language": 26005,
      "performance lowresource language": 86507,
      "language pairs different": 59745,
      "sentiment analysis important": 106597,
      "analysis important task": 5650,
      "important task scientific": 52270,
      "task scientific paper": 118665,
      "scientific paper analysis": 104014,
      "requires large annotated": 100287,
      "large annotated corpus": 61030,
      "classifying positive negative": 17672,
      "formed feature space": 45226,
      "feature space examined": 43316,
      "sentence mapped features": 105940,
      "mapped features input": 67124,
      "features input classifiers": 43570,
      "supervised classification using": 115738,
      "classification using 10crossvalidation": 17483,
      "using 10crossvalidation scheme": 129308,
      "10crossvalidation scheme evaluation": 203,
      "scheme evaluation conducted": 103922,
      "word embeddings effective": 133092,
      "performed better overall": 86914,
      "better overall classification": 13643,
      "using word embeddings": 130380,
      "word embeddings important": 133128,
      "important problems machine": 52216,
      "problems machine translation": 91340,
      "different surface forms": 31468,
      "propose use word": 93146,
      "word embeddings perform": 133181,
      "datasets experimental results": 27468,
      "results proposed methods": 102076,
      "proposed methods outperform": 93414,
      "methods outperform previous": 69656,
      "methods combining lexical": 69379,
      "lexical syntactic features": 63828,
      "applications information extraction": 6941,
      "information extraction question": 54597,
      "extraction question answering": 42458,
      "large training corpus": 61298,
      "manually annotated data": 67027,
      "compare performance domainspecific": 19274,
      "directed acyclic graph": 31813,
      "set features based": 107444,
      "tai et al": 117496,
      "attentional encoderdecoder model": 10046,
      "report consistent improvements": 98985,
      "like sentiment analysis": 64091,
      "sentiment analysis semantic": 106651,
      "learning models better": 62775,
      "better traditional baselines": 13746,
      "representations work better": 99996,
      "study supervised unsupervised": 114528,
      "model low resource": 71480,
      "low resource settings": 65395,
      "estimate semantic similarity": 38636,
      "morphologically complex language": 74760,
      "input word embedding": 55473,
      "embedding models using": 35459,
      "word embeddings explore": 133111,
      "embeddings capture semantic": 35585,
      "novel approach modeling": 80484,
      "social media using": 110432,
      "proposed model outperforms": 93458,
      "social media websites": 110434,
      "media websites electronic": 68183,
      "websites electronic newspapers": 132301,
      "electronic newspapers internet": 35257,
      "newspapers internet forums": 79410,
      "internet forums allow": 56191,
      "forums allow visitors": 45312,
      "allow visitors leave": 5100,
      "visitors leave comments": 131779,
      "leave comments read": 63260,
      "comments read interact": 18822,
      "read interact exchange": 95933,
      "interact exchange free": 55932,
      "exchange free participants": 39942,
      "free participants malicious": 45762,
      "participants malicious intentions": 85322,
      "paper present comprehensive": 84086,
      "model jointly predicts": 71388,
      "present new annotated": 89579,
      "new annotated dataset": 78774,
      "annotated dataset containing": 6174,
      "dataset containing excerpts": 26826,
      "trolls interactions users": 125839,
      "word embeddings multilingual": 133165,
      "semeval 2017 task": 105501,
      "task multilingual crosslingual": 118416,
      "multilingual word embeddings": 75403,
      "word embeddings combination": 133067,
      "expression generation using": 41738,
      "work present simple": 134688,
      "larger training data": 61385,
      "training data sets": 123535,
      "evaluation shows method": 39397,
      "recursive neural network": 97286,
      "need feature engineering": 76810,
      "discourse coherence evaluation": 32043,
      "proposed model significantly": 93480,
      "significantly outperforms existing": 109012,
      "outperforms existing strong": 82903,
      "characterbased word embeddings": 16504,
      "paper study impact": 84443,
      "distributions training testing": 32774,
      "evaluate robustness different": 38917,
      "robustness different models": 103093,
      "models convolutional neural": 72989,
      "neural networks recurrent": 78551,
      "networks recurrent neural": 77728,
      "characters byte pair": 16598,
      "different nlp tasks": 31301,
      "nlp tasks morphological": 79741,
      "tagging machine translation": 117405,
      "achieved promising results": 2672,
      "despite recent success": 29723,
      "handle larger vocabulary": 49389,
      "larger vocabulary training": 61389,
      "vocabulary training complexity": 131911,
      "training complexity decoding": 123389,
      "complexity decoding complexity": 19907,
      "number target words": 80977,
      "using statistical approach": 130221,
      "applied translation task": 7135,
      "nmt model achieves": 79895,
      "baseline nmt model": 12276,
      "using neural machine": 129953,
      "translate source sentences": 124545,
      "abstract meaning representation": 1769,
      "meaning representation parsing": 67675,
      "representation parsing abstract": 99370,
      "parsing abstract meaning": 85058,
      "meaning representation amr": 67669,
      "languages propose method": 60820,
      "propose method based": 92766,
      "method based annotation": 68661,
      "based annotation projection": 11508,
      "language parallel corpus": 59798,
      "target language using": 117660,
      "english source language": 37280,
      "spanish german chinese": 111161,
      "does require access": 33390,
      "annotations target languages": 6470,
      "graph convolutional encoders": 48780,
      "translation present simple": 125135,
      "present simple effective": 89701,
      "simple effective approach": 109404,
      "incorporating syntactic structure": 53564,
      "syntactic structure neural": 116485,
      "attentionbased encoderdecoder models": 10071,
      "neural networks developed": 78464,
      "trees source sentences": 125715,
      "words hidden states": 133981,
      "hidden states encoder": 49912,
      "neural networks evaluate": 78476,
      "translation experiments different": 124812,
      "experiments different types": 40909,
      "recent years seen": 96633,
      "language inference problem": 59189,
      "crosslingual word embeddings": 25045,
      "machine translation best": 65912,
      "units natural language": 127260,
      "textual similarity tasks": 121744,
      "network models including": 77347,
      "adjoining grammars tags": 3869,
      "adversarial multitask learning": 4146,
      "learning text classification": 63103,
      "text classification neural": 120749,
      "classification neural network": 17307,
      "network models shown": 77355,
      "models shown promising": 74030,
      "tasks paper propose": 119363,
      "paper propose adversarial": 84215,
      "conduct extensive experiments": 20870,
      "different text classification": 31488,
      "classification tasks demonstrates": 17444,
      "learned proposed model": 62247,
      "reinforcement learning external": 97810,
      "online discussion forum": 81768,
      "natural language state": 76535,
      "external knowledge source": 42019,
      "reported results task": 99074,
      "leveraging multisource nmt": 63693,
      "translation extensive experiments": 124820,
      "chinesetoenglish translation task": 16873,
      "translation task model": 125336,
      "bleu points best": 14307,
      "multitask learning semantic": 75865,
      "deep neural architecture": 28348,
      "able significantly improve": 1687,
      "significantly improve state": 108930,
      "improve state art": 52548,
      "state art semantic": 112472,
      "art semantic dependency": 9090,
      "using handengineered features": 129726,
      "approaches improve performance": 8187,
      "achieving new state": 2961,
      "state art code": 112433,
      "art code opensource": 9041,
      "code opensource available": 18126,
      "model propose new": 71811,
      "stateoftheart results english": 112900,
      "characterlevel language models": 16571,
      "sequences word tokens": 107147,
      "previously generated words": 90600,
      "typologically diverse languages": 126498,
      "languages demonstrate effectiveness": 60490,
      "demonstrate effectiveness model": 28715,
      "dynamic knowledge graph": 34313,
      "knowledge graph embeddings": 57964,
      "graph embeddings study": 48799,
      "goal openended dialogue": 48372,
      "poses new challenges": 88270,
      "new challenges existing": 78824,
      "existing dialogue systems": 40113,
      "structured knowledge unstructured": 114009,
      "propose neural model": 92822,
      "neural model dynamic": 78147,
      "automatic human evaluations": 10557,
      "human evaluations model": 50849,
      "model effective achieving": 71042,
      "word sense induction": 133530,
      "despite simplicity approach": 29731,
      "excellent results outperforming": 39921,
      "competitive stateoftheart methods": 19694,
      "gold standard datasets": 48441,
      "language processing knowledge": 59877,
      "used information retrieval": 128591,
      "descriptions paper propose": 29490,
      "paper propose approach": 84221,
      "evaluating machine translation": 39072,
      "small set sentences": 110200,
      "skipgram negative sampling": 109990,
      "similarity word pairs": 109335,
      "finally discuss potential": 44172,
      "directions future work": 31842,
      "negative sampling sgns": 76950,
      "sampling sgns word": 103608,
      "word embedding model": 133022,
      "networks word embeddings": 77817,
      "word embeddings detecting": 133081,
      "network measurements applied": 77321,
      "binary classification task": 14120,
      "bag words bow": 11362,
      "using complex networks": 129461,
      "networks support vector": 77788,
      "vector machine superior": 131315,
      "using linguistic features": 129817,
      "taken results indicate": 117515,
      "duluth semeval2017 task": 34260,
      "paper describes duluth": 83818,
      "describes duluth systems": 29400,
      "duluth systems participated": 34264,
      "participated semeval2017 task": 85343,
      "sense disambiguation measures": 105662,
      "text feature extraction": 120950,
      "feature extraction techniques": 43281,
      "using neural network": 129956,
      "stateoftheart neural network": 112784,
      "language processing language": 59878,
      "generation text classification": 47678,
      "bidirectional language models": 13933,
      "language models pretrained": 59626,
      "models pretrained word": 73788,
      "standard component neural": 112214,
      "representations produce context": 99826,
      "relatively little labeled": 98400,
      "little labeled data": 64818,
      "data paper demonstrate": 26208,
      "evaluate model standard": 38868,
      "standard datasets named": 112225,
      "datasets named entity": 27588,
      "additional labeled data": 3534,
      "latent variable models": 61631,
      "models shown facilitate": 74027,
      "response generation opendomain": 101208,
      "inject linguistic knowledge": 55259,
      "recent advances gpu": 96403,
      "advances gpu hardware": 3991,
      "neural networks achieve": 78435,
      "achieve significant gains": 2535,
      "previous best models": 90394,
      "english penn treebank": 37238,
      "tagging lowresource languages": 117401,
      "lowresource languages using": 65530,
      "languages using bilingual": 60952,
      "using bilingual dictionary": 129411,
      "data readily available": 26319,
      "readily available languages": 95967,
      "framework takes advantage": 45711,
      "bilingual dictionary propose": 14034,
      "propose novel neural": 92950,
      "network model joint": 77329,
      "word embeddings substantial": 133215,
      "lowresource neural machine": 65544,
      "lowresource language pairs": 65504,
      "propose novel data": 92908,
      "novel data augmentation": 80525,
      "data augmentation approach": 25629,
      "generating new sentence": 47237,
      "lowresource settings method": 65564,
      "method improves translation": 68892,
      "bleu points baseline": 14306,
      "efficient natural language": 35096,
      "natural language response": 76527,
      "method natural language": 68982,
      "feedforward neural networks": 43852,
      "neural networks using": 78601,
      "effectiveness feature set": 34893,
      "performance improvements various": 86453,
      "order better understand": 82287,
      "task convolutional neural": 118024,
      "classification paper presents": 17324,
      "uses convolutional neural": 129212,
      "conducted experiments benchmark": 20923,
      "languages proposed approach": 60822,
      "comparable performance stateofart": 19158,
      "learning universal sentence": 63138,
      "universal sentence representations": 127334,
      "language inference data": 59169,
      "modern nlp systems": 74408,
      "rely word embeddings": 98753,
      "trained unsupervised manner": 123322,
      "unsupervised manner large": 127662,
      "trained using supervised": 123331,
      "stanford natural language": 112371,
      "language inference datasets": 59171,
      "wide range transfer": 132515,
      "transfer tasks like": 124193,
      "transfer learning nlp": 124116,
      "learning nlp tasks": 62850,
      "work present method": 134680,
      "present method automatically": 89547,
      "method automatically generate": 68656,
      "approach sequence sequence": 7881,
      "maps input sequence": 67163,
      "sequence variable length": 107097,
      "neural networks introduce": 78494,
      "based entirely convolutional": 11686,
      "entirely convolutional neural": 37731,
      "accuracy deep lstm": 2132,
      "deep lstm setup": 28332,
      "englishgerman wmt14 englishfrench": 37392,
      "wmt14 englishfrench translation": 132812,
      "order magnitude faster": 82358,
      "support development evaluation": 115963,
      "short message service": 108219,
      "message service sms": 68499,
      "use machine learning": 128133,
      "result compared english": 101365,
      "strong baseline future": 113629,
      "compositional neural networks": 20122,
      "neural networks proven": 78545,
      "networks proven effective": 77716,
      "proven effective learning": 93721,
      "shared compositional function": 107968,
      "address issue introducing": 3689,
      "experimental results typical": 40715,
      "effectiveness proposed models": 34937,
      "dependency parsing transitionbased": 29205,
      "error propagation paper": 38317,
      "leverage lexical information": 63599,
      "achieve significantly better": 2544,
      "joint modeling approach": 57295,
      "latent variables experimental": 61637,
      "variables experimental results": 130856,
      "corpora joint model": 23508,
      "model outperform stateoftheart": 71626,
      "better predictive performance": 13676,
      "pretrained word vectors": 90227,
      "word vectors paper": 133643,
      "machine learning used": 65836,
      "learning used produce": 63148,
      "paper presents simple": 84203,
      "internal representations learned": 56171,
      "proposed model performs": 93471,
      "model performs comparably": 71738,
      "performs comparably stateoftheart": 86992,
      "tensor product representation": 120192,
      "et al 2003": 38684,
      "methods language pairs": 69574,
      "variety text mining": 131025,
      "paper proposes simple": 84368,
      "simple neural model": 109482,
      "fed softmax layer": 43810,
      "ensemble based model": 37579,
      "event trigger identification": 39551,
      "trigger identification using": 125785,
      "complex handcrafted features": 19814,
      "window paper propose": 132711,
      "goals artificial intelligence": 48419,
      "natural language existing": 76295,
      "existing work natural": 40338,
      "work natural language": 134646,
      "natural language learning": 76367,
      "training data training": 123555,
      "data training data": 26572,
      "generalization behavior different": 46768,
      "learning process humans": 62917,
      "validate effectiveness proposed": 130718,
      "data paper discusses": 26210,
      "model convolutional neural": 70922,
      "neural networks dependency": 78463,
      "rich languages present": 102761,
      "compose word representations": 20063,
      "outperforms previous best": 82934,
      "word embeddings help": 133122,
      "data analysis methods": 25597,
      "information predicateargument structure": 54855,
      "zero anaphora resolution": 135472,
      "structure analysis pasa": 113808,
      "anaphora resolution zar": 6072,
      "relationships predicates arguments": 98329,
      "predicates arguments semantic": 88862,
      "use syntactic information": 128309,
      "modern natural language": 74400,
      "systems paper present": 117045,
      "training neural machine": 123734,
      "neural sequencetosequence architectures": 78676,
      "wmt multimodal machine": 132802,
      "language understanding lu": 60234,
      "address issue paper": 3692,
      "transfer learning approach": 124077,
      "representation transfer learning": 99447,
      "transfer learning approaches": 124081,
      "effectiveness proposed method": 34930,
      "communications social media": 19048,
      "social media colloquial": 110358,
      "large amounts monolingual": 61017,
      "amounts monolingual data": 5352,
      "translation statistical machine": 125293,
      "neural networks present": 78537,
      "neural networks cnn": 78453,
      "partofspeech tagging morphological": 85553,
      "tagging morphological tagging": 117411,
      "networks rnns proved": 77747,
      "effective nlp tasks": 34726,
      "nlp tasks despite": 79713,
      "models proved effective": 73831,
      "work propose solution": 134744,
      "rnn lstm gru": 102925,
      "tasks spoken language": 119519,
      "language understanding slu": 60268,
      "semantic similarity semantic": 105290,
      "semantic similarity measure": 105283,
      "word2vec word embedding": 133689,
      "semantic similarity methods": 105285,
      "improve performance word": 52493,
      "performance word embedding": 86858,
      "network architectures measure": 77149,
      "conduct comprehensive experiments": 20831,
      "results verify effectiveness": 102318,
      "verify effectiveness approach": 131581,
      "words context syntactic": 133881,
      "context word sense": 22328,
      "embedding models based": 35453,
      "standard neural network": 112281,
      "large unannotated text": 61311,
      "partofspeech taggers dependency": 85545,
      "taggers dependency parsers": 117365,
      "dependency parsers trained": 29169,
      "model proven effective": 71820,
      "learning sentence representations": 63003,
      "representations capturing sentence": 99535,
      "layer encoder decoder": 61712,
      "encoder decoder help": 36508,
      "help model generalize": 49752,
      "word embedding initialization": 133012,
      "learning better sentence": 62408,
      "empirically proposed model": 36241,
      "language processing techniques": 59975,
      "techniques used develop": 120008,
      "propose novel fully": 92923,
      "human evaluation neural": 50824,
      "machine translation pure": 66179,
      "finegrained manual evaluation": 44366,
      "annotators interannotator agreement": 6494,
      "best performing neural": 13405,
      "alzheimers disease ad": 5260,
      "features derived word": 43446,
      "text natural language": 121140,
      "using dependency structures": 129560,
      "documents different domains": 33218,
      "machine translation training": 66274,
      "nmt models usually": 79925,
      "based sentence length": 12016,
      "training widely used": 123951,
      "different datasets results": 31080,
      "corpora using different": 23614,
      "account results indicate": 2037,
      "error reduction general": 38342,
      "statistical language model": 113099,
      "overcome data sparsity": 83279,
      "natural language different": 76287,
      "set used build": 107627,
      "parser dyer et": 84945,
      "dyer et al": 34291,
      "vinyals et al": 131730,
      "automatic synonym discovery": 10683,
      "text corpora news": 120831,
      "existing methods require": 40193,
      "require training data": 100203,
      "paper study problem": 84446,
      "knowledge base entities": 57770,
      "task propose novel": 118575,
      "novel framework called": 80574,
      "entities experimental results": 37780,
      "results prove effectiveness": 102084,
      "effectiveness proposed framework": 34929,
      "translation nmt recently": 125052,
      "field machine translation": 43965,
      "translation nmt suffers": 125058,
      "translation address problem": 124627,
      "al 2017 proposed": 4622,
      "bahdanau et al": 11388,
      "chineseenglish translation task": 16862,
      "forward translation backtranslation": 45330,
      "datadriven natural language": 26672,
      "language generation systems": 59098,
      "dataset poses new": 27098,
      "play crucial role": 87690,
      "achieves competitive performance": 2770,
      "training test time": 123915,
      "unsupervised representation learning": 127704,
      "representation learning model": 99298,
      "evaluating natural language": 39081,
      "automated metrics bleu": 10457,
      "widely used machine": 132572,
      "dialogue response generation": 30734,
      "generation previous work": 47559,
      "metrics correlate strongly": 69944,
      "correlate strongly human": 24203,
      "correlate human judgment": 24195,
      "conduct empirical study": 20839,
      "currently available corpora": 25397,
      "character embedding models": 16427,
      "use additional phonetic": 127883,
      "approach superior stateoftheart": 7948,
      "achieved excellent results": 2620,
      "tasks paper present": 119362,
      "using deep neural": 129546,
      "neural networks context": 78459,
      "achieve similar results": 2548,
      "similar results compared": 109139,
      "crosslingual sentiment analysis": 25012,
      "translation current approaches": 124727,
      "data using bilingual": 26609,
      "bilingual vector space": 14068,
      "use single linear": 128280,
      "data low cost": 26098,
      "event extraction systems": 39516,
      "data sets different": 26422,
      "machine learning classifiers": 65788,
      "acquisition paper investigates": 3049,
      "language learning using": 59246,
      "using computational models": 129465,
      "neural language generation": 77935,
      "language generation work": 59103,
      "neural natural language": 78224,
      "content generated text": 21885,
      "aspects generated text": 9385,
      "recurrent neural architecture": 97187,
      "learning multitask learning": 62819,
      "paper propose multitask": 84270,
      "propose multitask learning": 92808,
      "multitask learning architecture": 75825,
      "types recurrent neural": 126346,
      "extensive experiments benchmark": 41895,
      "experiments benchmark datasets": 40794,
      "datasets text classification": 27758,
      "text classification model": 120742,
      "nonprojective dependency parsing": 80279,
      "parsing bidirectional lstm": 85081,
      "mitigate error propagation": 70368,
      "conll 2017 ud": 21080,
      "2017 ud shared": 643,
      "ud shared task": 126533,
      "using ensemble methods": 129635,
      "ensemble methods using": 37601,
      "qualitative data analysis": 94559,
      "analysis paper presents": 5739,
      "large document collections": 61085,
      "information retrieval systems": 54948,
      "stateoftheart natural language": 112768,
      "techniques large document": 119916,
      "approach consists main": 7452,
      "learning present new": 62903,
      "positive negative training": 88335,
      "negative training examples": 76958,
      "supervised relation extraction": 115829,
      "explicit negative examples": 41339,
      "significantly improve performance": 108928,
      "improve performance compared": 52461,
      "performance compared stateoftheart": 86237,
      "datasets different languages": 27430,
      "speech recognition natural": 111762,
      "recognition natural language": 96918,
      "language understanding language": 60232,
      "evaluate method using": 38857,
      "user logs commercial": 129011,
      "experimental results demonstrated": 40603,
      "translation task paper": 125337,
      "task paper describes": 118500,
      "target monolingual corpora": 117677,
      "document retrieval task": 33070,
      "establish strong baseline": 38595,
      "syntactic semantic structures": 116477,
      "translation paper proposes": 125095,
      "attention mechanism used": 9916,
      "outofvocabulary oov problem": 82674,
      "results reveal proposed": 102140,
      "text datasets visualizing": 120857,
      "datasets visualizing relevant": 27795,
      "visualizing relevant words": 131843,
      "advanced machine learning": 3946,
      "paper propose extract": 84239,
      "texts summarize contents": 121626,
      "summarize contents documents": 115600,
      "contents documents belonging": 21988,
      "discovering trending topics": 32130,
      "new york times": 79258,
      "york times article": 135454,
      "times article snippets": 122167,
      "simple language model": 109452,
      "learning language models": 62673,
      "language models training": 59674,
      "models trained simple": 74228,
      "simple negative sampling": 109480,
      "learn word embeddings": 62192,
      "discriminative training approaches": 32219,
      "machine translation framework": 65990,
      "largescale chineseenglish translation": 61405,
      "lead significant improvements": 61868,
      "significant improvements translation": 108812,
      "discover groups words": 32116,
      "based svm classifier": 12098,
      "supervised learning models": 115772,
      "model automatic interpretation": 70711,
      "rogets thesaurus implementation": 103150,
      "method consistently outperforms": 68731,
      "better comparable results": 13541,
      "results reported literature": 102129,
      "deep learning predicting": 28306,
      "new method based": 79008,
      "method based bidirectional": 68663,
      "based bidirectional long": 11550,
      "lstm neural network": 65655,
      "evaluate proposed architecture": 38898,
      "news articles news": 79301,
      "social media contain": 110360,
      "datasets approach improves": 27315,
      "approach improves performance": 7628,
      "word embeddings improve": 133129,
      "embeddings improve performance": 35748,
      "improve performance neural": 52474,
      "performance neural model": 86568,
      "results variety tasks": 102309,
      "tasks word similarity": 119608,
      "similarity word analogy": 109333,
      "word embeddings downstream": 133091,
      "downstream nlp tasks": 34015,
      "nlp tasks limited": 79736,
      "outofvocabulary oov words": 82675,
      "unlike prior work": 127449,
      "original word embedding": 82564,
      "word embedding corpus": 133008,
      "intrinsic extrinsic evaluations": 56361,
      "naive bayes logistic": 76017,
      "bayes logistic regression": 12570,
      "results better ones": 101553,
      "classical machine learning": 17090,
      "twitter social media": 126125,
      "social media users": 110431,
      "requires training data": 100327,
      "tasks demonstrate approach": 119035,
      "novel neural model": 80669,
      "using small sets": 130190,
      "sets training data": 107720,
      "results benchmark datasets": 101541,
      "outperforms stateoftheart unsupervised": 83021,
      "morphologically rich language": 74768,
      "rich language features": 102756,
      "words morphological rules": 134063,
      "words paper introduces": 134102,
      "recognition paper describes": 96953,
      "problem named entity": 91136,
      "best knowledge dataset": 13346,
      "embeddings word sense": 36033,
      "word embeddings model": 133162,
      "update neural networks": 127798,
      "word embeddings learning": 133148,
      "use word sense": 128364,
      "proposed method compare": 93351,
      "method compare performance": 68709,
      "compare performance word": 19280,
      "performance word embeddings": 86859,
      "word similarity tasks": 133571,
      "tasks text classification": 119556,
      "model provides better": 71826,
      "text classification learning": 120733,
      "work introduce temporal": 134581,
      "text classification novel": 120752,
      "novel approach based": 80471,
      "word embeddings paper": 133176,
      "approach text classification": 7972,
      "model widely used": 72317,
      "used computer vision": 128444,
      "word vectors order": 133642,
      "better performance using": 13664,
      "text mining tasks": 121125,
      "tasks text categorization": 119555,
      "model yields better": 72337,
      "yields better performance": 135401,
      "apply neural networks": 7199,
      "neural networks increasingly": 78492,
      "parsing semantic role": 85226,
      "semantic role labeling": 105246,
      "network able learn": 77120,
      "models sentence representation": 74002,
      "features features extracted": 43514,
      "sentiment analysis paper": 106623,
      "analysis paper describes": 5736,
      "analysis task build": 5885,
      "workshop building linguistically": 135003,
      "building linguistically generalizable": 14859,
      "linguistically generalizable nlp": 64586,
      "generalizable nlp systems": 46758,
      "opinion mining sentiment": 82092,
      "social media research": 110418,
      "problem propose paper": 91180,
      "data collection method": 25742,
      "weakly supervised approach": 132195,
      "temporal relations events": 120123,
      "existing temporal relation": 40313,
      "trained supervised manner": 123299,
      "weakly supervised learning": 132196,
      "supervised learning approach": 115764,
      "rich commonsense knowledge": 102729,
      "domain specific knowledge": 33650,
      "specific knowledge addition": 111455,
      "weakly supervised trained": 132199,
      "achieves comparable performance": 2759,
      "comparable performance stateoftheart": 19159,
      "performance stateoftheart supervised": 86748,
      "stateoftheart supervised systems": 112983,
      "machine translation examine": 65977,
      "machine learning approach": 65772,
      "learned word embeddings": 62287,
      "numerous natural language": 81024,
      "nlp tasks recent": 79759,
      "tasks recent years": 119442,
      "main contribution paper": 66410,
      "contribution paper develop": 22774,
      "et al 2013": 38696,
      "new vector space": 79239,
      "demonstrate effectiveness approach": 28706,
      "measures meaning similarity": 67881,
      "meaning similarity sentences": 67693,
      "generation question answering": 47580,
      "question answering qa": 95076,
      "short answer grading": 108201,
      "mt quality estimation": 74986,
      "limitations existing models": 64175,
      "set carefully selected": 107388,
      "shared task data": 108038,
      "produce high quality": 91899,
      "use recurrent neural": 128231,
      "neural networks lstm": 78510,
      "paper presents method": 84180,
      "large number latent": 61180,
      "longterm label dependencies": 65207,
      "neural mt systems": 78214,
      "paper describes university": 83857,
      "university edinburghs submissions": 127351,
      "task submitted systems": 118752,
      "czech german polish": 25491,
      "translation systems trained": 125316,
      "monolingual training data": 74626,
      "dependency grammar induction": 29142,
      "grammar induction neural": 48639,
      "training data study": 123544,
      "competitive current stateoftheart": 19652,
      "network model based": 77327,
      "model based novel": 70739,
      "models sutskever et": 74143,
      "sutskever et al": 116227,
      "evaluate model automatic": 38863,
      "exploiting linguistic resources": 41470,
      "resources neural machine": 101019,
      "using multitask learning": 129925,
      "used statistical machine": 128784,
      "model jointly training": 71389,
      "training natural language": 123726,
      "information improve performance": 54670,
      "additional linguistic resources": 3545,
      "15 bleu points": 342,
      "bleu points lowresource": 14309,
      "multitask learning scheme": 75864,
      "fundamental component nlp": 46090,
      "component nlp tasks": 19996,
      "nlp tasks named": 79742,
      "recognition machine translation": 96906,
      "machine translation popular": 66158,
      "using convolutional neural": 129497,
      "task propose new": 118574,
      "evaluate model large": 38865,
      "outperforms existing models": 82894,
      "outperforms prior work": 82966,
      "important role natural": 52243,
      "role natural language": 103206,
      "et al 2002": 38683,
      "et al 2005": 38688,
      "work propose method": 134725,
      "model proposed model": 71816,
      "chinese english translation": 16764,
      "proposed model achieves": 93425,
      "model achieves significant": 70589,
      "significant improvements baseline": 108784,
      "various test sets": 131219,
      "nmt achieved notable": 79823,
      "achieved notable success": 2660,
      "notable success recent": 80378,
      "memoryaugmented nmt mnmt": 68401,
      "oov words based": 81870,
      "experiments chineseenglish translation": 40827,
      "introduce dataset created": 56407,
      "model aspect extraction": 70690,
      "aspect extraction sentiment": 9328,
      "extraction sentiment classification": 42484,
      "sentiment classification model": 106701,
      "model outperforms baseline": 71636,
      "results experiments domain": 101782,
      "finegrained opinion mining": 44371,
      "deep learning based": 28253,
      "learning based natural": 62393,
      "processing deep learning": 91657,
      "learn hierarchical representations": 62061,
      "recently variety model": 96775,
      "variety model designs": 130995,
      "model designs methods": 70975,
      "designs methods blossomed": 29656,
      "methods blossomed context": 69354,
      "context natural language": 22192,
      "processing nlp paper": 91746,
      "paper review significant": 84410,
      "numerous nlp tasks": 81027,
      "nlp tasks provide": 79755,
      "compare contrast various": 19232,
      "past present future": 85649,
      "deep learning nlp": 28304,
      "dialog systems attracting": 30595,
      "attracting increasing attention": 10166,
      "attention academia industry": 9789,
      "datasets paper propose": 27612,
      "facilitate research direction": 42790,
      "source language input": 110774,
      "information input text": 54692,
      "predicted using text": 88971,
      "work use stateoftheart": 134871,
      "neural network methods": 78327,
      "methods tackle problem": 69790,
      "translation smt framework": 125270,
      "languages chinese japanese": 60444,
      "model sentiment classification": 71980,
      "radicals chinese characters": 95445,
      "respectively results suggest": 101163,
      "word embeddings present": 133185,
      "effective approach learning": 34631,
      "learning word sense": 63168,
      "word sense embeddings": 133529,
      "existing word embeddings": 40334,
      "word embeddings clustering": 133066,
      "stateoftheart unsupervised wsd": 113022,
      "identification discourse relations": 51382,
      "discourse relations challenging": 32077,
      "relations challenging task": 98120,
      "challenging task natural": 16324,
      "effectively paper propose": 34836,
      "corpora text simplification": 23601,
      "signal discourse relation": 108686,
      "evaluating word embeddings": 39111,
      "sentence boundary detection": 105781,
      "speech transcripts paper": 111830,
      "discourse analysis involves": 32040,
      "capture semantic syntactic": 15385,
      "semantic syntactic morphological": 105317,
      "high accuracy speech": 50034,
      "speech recognition requires": 111771,
      "data domain adaptation": 25865,
      "performed high accuracy": 86924,
      "labeled data target": 58446,
      "data target domain": 26540,
      "domain work propose": 33699,
      "work propose approach": 134713,
      "data consisting pairs": 25780,
      "teacherstudent ts learning": 119707,
      "model evaluate proposed": 71109,
      "significant improvements accuracy": 108782,
      "reductions word error": 97474,
      "paper presents models": 84182,
      "word embedding features": 133010,
      "embedding features model": 35408,
      "model achieves state": 70593,
      "achieves state art": 2874,
      "variety nlp applications": 131010,
      "major arabic dialects": 66535,
      "using limited training": 129812,
      "limited training data": 64292,
      "modern standard arabic": 74416,
      "task wmt 2017": 118847,
      "human automatic evaluations": 50761,
      "endtoend trainable neural": 36981,
      "trainable neural network": 123056,
      "tracking taskoriented dialog": 122764,
      "present novel endtoend": 89606,
      "taskoriented dialog systems": 118885,
      "issue api calls": 56990,
      "knowledge base kb": 57773,
      "responses successfully complete": 101298,
      "successfully complete taskoriented": 115183,
      "proposed model produces": 93474,
      "restaurant search domain": 101326,
      "second dialog state": 104402,
      "dialog state tracking": 30588,
      "state tracking challenge": 112521,
      "tracking challenge dstc2": 122748,
      "experiment results proposed": 40503,
      "given dialog history": 48016,
      "network models using": 77358,
      "question answering question": 95088,
      "answering question answering": 6680,
      "answering qa systems": 6674,
      "ways natural language": 132168,
      "trained endtoend using": 123133,
      "using questionanswer pairs": 130076,
      "questionanswer pairs supervision": 95248,
      "experimental results datasets": 40577,
      "results datasets framework": 101664,
      "consistently improves performance": 21419,
      "improves performance achieving": 53007,
      "performance achieving competitive": 86118,
      "competitive results despite": 19682,
      "results despite use": 101717,
      "use simple qa": 128278,
      "training seq2seq models": 123841,
      "models language models": 73452,
      "sequencetosequence seq2seq models": 107207,
      "generating natural language": 47234,
      "language sentences machine": 60080,
      "sentences machine translation": 106389,
      "machine translation image": 66000,
      "translation image captioning": 124855,
      "speech recognition performance": 111768,
      "leveraging unlabeled data": 63712,
      "leverages pretrained language": 63654,
      "pretrained language model": 90047,
      "language model training": 59412,
      "effectiveness speech recognition": 34952,
      "speech recognition task": 111782,
      "new domain using": 78889,
      "existing nmt systems": 40240,
      "translating ambiguous words": 124586,
      "language pairs demonstrate": 59744,
      "demonstrate models improve": 28802,
      "improve performance nmt": 52477,
      "performance nmt systems": 86578,
      "contrast previous work": 22705,
      "model trained using": 72217,
      "automatic training data": 10701,
      "baseline logistic regression": 12242,
      "neural network baseline": 78264,
      "average f1 scores": 11190,
      "network model outperforms": 77335,
      "previous work suggests": 90555,
      "architecture named entity": 8702,
      "entity recognition disambiguation": 38011,
      "crucial natural language": 25152,
      "machine translation standard": 66238,
      "paper propose different": 84228,
      "datasets russian language": 27692,
      "perform extensive error": 86000,
      "extensive error analysis": 41875,
      "logical forms easily": 65006,
      "textual entailment datasets": 121694,
      "results comparable stateoftheart": 101595,
      "finegrained domain adaptation": 44346,
      "nlp models typically": 79646,
      "different market sectors": 31246,
      "techniques semisupervised learning": 119980,
      "learning domain adaptation": 62518,
      "domain adaptation limited": 33436,
      "semantic relatedness semantic": 105197,
      "systems semantic relatedness": 117131,
      "shared task russian": 108091,
      "largescale crowdsourcing study": 61415,
      "application machine translation": 6862,
      "stateoftheart performance analysis": 112807,
      "source target tokens": 110843,
      "algorithms natural language": 4870,
      "distributional representation words": 32716,
      "nlp tasks models": 79740,
      "models learn features": 73483,
      "learn features automatically": 62049,
      "explicit feature engineering": 41322,
      "neural models natural": 78183,
      "sequence tagging tasks": 107084,
      "perform comprehensive analysis": 85972,
      "important contextual words": 52138,
      "work explore using": 134518,
      "scale free text": 103722,
      "vectors word vectors": 131490,
      "word vectors provide": 133645,
      "large number features": 61177,
      "quantitative qualitative analysis": 94877,
      "shows approach achieves": 108550,
      "approach achieves comparable": 7303,
      "achieves comparable accuracy": 2757,
      "intrinsic evaluation word": 56357,
      "evaluation word embeddings": 39443,
      "word embeddings introduce": 133138,
      "intrinsic evaluation metric": 56354,
      "evaluation metric word": 39282,
      "metric word embeddings": 69912,
      "different vector representations": 31553,
      "machine translation demonstrate": 65947,
      "word embeddings extended": 133113,
      "experimental results sentence": 40694,
      "perform comparably better": 85962,
      "better stateoftheart approaches": 13727,
      "benchmark datasets sentence": 12803,
      "including sequence tagging": 53373,
      "addresses problem identifying": 3803,
      "language models proposed": 59638,
      "models proposed method": 73824,
      "input output layers": 55382,
      "kobayashi et al": 58339,
      "gu et al": 49210,
      "addition construct new": 3414,
      "ability capture word": 1493,
      "using novel dataset": 129982,
      "model outperformed baseline": 71629,
      "outperformed baseline model": 82778,
      "knowledge bases neural": 57808,
      "machine translation work": 66303,
      "work presented paper": 134692,
      "machine translation translate": 66279,
      "translation quality domainspecific": 125168,
      "machine translation methods": 66045,
      "domain adaptation translation": 33459,
      "furthermore perform experiments": 46200,
      "model training data": 72219,
      "language models trained": 59669,
      "intensive care units": 55888,
      "endtoend neural network": 36944,
      "topic model simultaneously": 122541,
      "experiments mimiciii dataset": 41010,
      "models significantly outperform": 74046,
      "response generation propose": 101211,
      "generation propose simple": 47573,
      "decoding method based": 28106,
      "human evaluation results": 50830,
      "datadriven dialogue systems": 26660,
      "corpora social media": 23590,
      "range nlp tasks": 95592,
      "random fields crfs": 95486,
      "capture longrange dependencies": 15339,
      "substantial improvements strong": 114865,
      "neural mt engines": 78211,
      "systems submitted wmt": 117166,
      "systems built using": 116766,
      "sequencetosequence models lstm": 107180,
      "wide variety problems": 132527,
      "classification ranking tasks": 17364,
      "sentence document level": 105826,
      "task empirical results": 118136,
      "results number tasks": 102000,
      "performed russian language": 86932,
      "tackle challenge propose": 117286,
      "sequence prediction problem": 107032,
      "problem experimental results": 91043,
      "significantly improves accuracy": 108940,
      "use text classification": 128327,
      "linguistically motivated features": 64593,
      "decoder neural machine": 28035,
      "achieved remarkable performance": 2682,
      "contextual information timestep": 22474,
      "based sequence model": 12025,
      "positive negative reviews": 88333,
      "available natural language": 11052,
      "natural language generator": 76322,
      "twitter customer service": 126102,
      "dialogue acts given": 30636,
      "customer service dialogue": 25446,
      "neural networks applied": 78438,
      "encoderdecoder neural machine": 36610,
      "bayesian nonparametric model": 12586,
      "used create bilingual": 128463,
      "presents approach task": 89819,
      "multilayer recurrent neural": 75184,
      "neural network approach": 78242,
      "approach substantially outperforms": 7940,
      "substantially outperforms previous": 114908,
      "previous work terms": 90557,
      "work terms bleu": 134849,
      "respectively bleu score": 101128,
      "paper makes use": 84043,
      "british national corpus": 14661,
      "word vectors evaluated": 133635,
      "bag words approach": 11361,
      "methods based distributional": 69334,
      "based distributional semantics": 11655,
      "vietnamese word segmentation": 131702,
      "rules stored exception": 103443,
      "stored exception structure": 113385,
      "exception structure new": 39925,
      "structure new rules": 113919,
      "new rules added": 79115,
      "rules added correct": 103379,
      "rules experimental results": 103399,
      "experimental results benchmark": 40561,
      "approach outperforms previous": 7773,
      "outperforms previous stateoftheart": 82947,
      "previous stateoftheart approaches": 90470,
      "presents empirical study": 89845,
      "translation models work": 124967,
      "machine translation method": 66044,
      "current stateoftheart method": 25357,
      "slightly lower accuracy": 110026,
      "room future improvement": 103260,
      "collecting training data": 18453,
      "benchmark datasets work": 12810,
      "work propose evaluate": 134719,
      "dataset constructed using": 26822,
      "learning domainspecific word": 62525,
      "domainspecific word embeddings": 33923,
      "space widely used": 111083,
      "variety nlp tasks": 131011,
      "entity recognition syntactic": 38074,
      "recognition syntactic parsing": 97017,
      "parsing sentiment analysis": 85231,
      "embedding methods word2vec": 35445,
      "methods word2vec glove": 69855,
      "large text corpus": 61293,
      "methods fail produce": 69507,
      "knowledge domain vocabulary": 57883,
      "specifically propose general": 111581,
      "results demonstrated effectiveness": 101709,
      "method learning domainspecific": 68937,
      "language models scale": 59654,
      "noisecontrastive estimation nce": 80080,
      "performance neural machine": 86565,
      "learning rate schedule": 62937,
      "models outperforms stateoftheart": 73691,
      "training data machine": 123489,
      "trained clean data": 123091,
      "new corpus called": 78846,
      "training data finally": 123464,
      "reinforcement learning rl": 97823,
      "language generation dialogue": 59070,
      "rich linguistic knowledge": 102768,
      "mitigating impact speech": 70386,
      "impact speech recognition": 51892,
      "using sequencetosequence model": 130161,
      "task domain adaptation": 118116,
      "domain adaptation problem": 33447,
      "text different domains": 120874,
      "hidden states similar": 49915,
      "errors experimental results": 38369,
      "cornell movie dialog": 23390,
      "movie dialog dataset": 74895,
      "machine learning based": 65780,
      "learning based method": 62389,
      "approach used paper": 7998,
      "machine svm model": 65873,
      "sentiment tendency review": 106805,
      "word frequency word": 133295,
      "multisource neural machine": 75787,
      "performance language pairs": 86483,
      "train machine translation": 122953,
      "systems paper propose": 117048,
      "model showed improvement": 72006,
      "machine translation performance": 66156,
      "sequencetosequence model generates": 107170,
      "given context propose": 48001,
      "propose dual encoder": 92635,
      "language widely studied": 60341,
      "languages domains genres": 60516,
      "pose significant challenges": 88257,
      "statistical analysis methods": 113078,
      "crossdomain sentiment classification": 24901,
      "sentiment classification word": 106720,
      "word similarity prediction": 133564,
      "attentive neural architecture": 10121,
      "neural architecture joint": 77834,
      "processing nlp techniques": 91760,
      "simple effective neural": 109414,
      "joint model performs": 57292,
      "format work propose": 45212,
      "work propose new": 134730,
      "new joint model": 78967,
      "extensive comparative study": 41865,
      "proposed joint model": 93317,
      "encourage model focus": 36759,
      "attentive neural architectures": 10122,
      "attention neural networks": 9957,
      "tasks language understanding": 119242,
      "language understanding tasks": 60280,
      "intent classification slot": 55897,
      "classification slot filling": 17409,
      "classification namedentity recognition": 17298,
      "memory lstm networks": 68338,
      "classification proposed method": 17358,
      "different sources data": 31440,
      "formal lexical semantics": 45168,
      "language understanding dialogue": 60215,
      "dialogue policy learning": 30723,
      "learning language understanding": 62674,
      "understanding lu dialogue": 126881,
      "experiments benchmark dataset": 40793,
      "benchmark dataset proposed": 12758,
      "models word embeddings": 74349,
      "word embeddings neural": 133169,
      "word vectors vector": 133651,
      "textual similarity natural": 121740,
      "similarity natural language": 109274,
      "language inference tasks": 59201,
      "small sample sizes": 110193,
      "paper aims address": 83726,
      "machine translation explore": 65987,
      "improvements bleu score": 52820,
      "natural language natural": 76374,
      "reasoning language vision": 96262,
      "receiving increasing attention": 96383,
      "existing data sets": 40102,
      "natural language visual": 76604,
      "language visual reasoning": 60335,
      "representation word taking": 99467,
      "word taking account": 133601,
      "local context information": 64907,
      "sentence representation learning": 106030,
      "training data extract": 123463,
      "question answer sentences": 95017,
      "new model selection": 79035,
      "finally propose use": 44224,
      "automatic evaluation method": 10533,
      "evaluation shows model": 39398,
      "shows model achieves": 108599,
      "technical report describes": 119752,
      "16th early 20th": 396,
      "early 20th century": 34385,
      "translation underresourced languages": 125399,
      "underresourced languages paper": 126730,
      "translation computerassisted translation": 124703,
      "input sentence sequence": 55424,
      "distributional semantics despite": 32724,
      "results natural language": 101975,
      "language machine translation": 59265,
      "large number languages": 61179,
      "provide training development": 93944,
      "training development test": 123583,
      "development test splits": 30433,
      "semantic sentiment analysis": 105271,
      "various fields including": 131100,
      "marketing political science": 67245,
      "answering questions requires": 6683,
      "people express social": 85876,
      "named entity tagger": 76127,
      "word embeddings open": 133173,
      "open information extraction": 81905,
      "question answering built": 95027,
      "demonstrate utility corpus": 28904,
      "similarity task showing": 109317,
      "model trained corpus": 72184,
      "better results models": 13704,
      "results models trained": 101960,
      "model outperforms state": 71665,
      "outperforms state art": 82992,
      "state art models": 112450,
      "paper propose task": 84332,
      "language model capture": 59305,
      "pretraining natural language": 90296,
      "language inference neural": 59176,
      "using attentionbased sequencetosequence": 129357,
      "pretrained glove word2vec": 90040,
      "framework machine translation": 45606,
      "distributed vector representations": 32632,
      "vector representations fed": 131354,
      "experiment benchmark datasets": 40453,
      "wmt metrics shared": 132799,
      "metrics shared task": 70008,
      "obtain best results": 81262,
      "perform series experiments": 86067,
      "series experiments analyze": 107274,
      "convolutional recurrent neural": 23213,
      "correlates human judgments": 24218,
      "par state art": 84524,
      "finegrained entity typing": 44352,
      "text document corpus": 120882,
      "freely available research": 45787,
      "approaches learning word": 8213,
      "semantics natural language": 105439,
      "natural language paper": 76379,
      "models efficiently learn": 73123,
      "word vectors investigate": 133637,
      "resulting word embeddings": 101480,
      "word embeddings outperform": 133174,
      "challenges sentiment analysis": 16212,
      "social media decade": 110372,
      "respect specific topic": 101104,
      "sentiment emotion analysis": 106736,
      "election social media": 35238,
      "use social media": 128284,
      "tweets natural language": 126040,
      "textual entailment task": 121699,
      "present new dataset": 89584,
      "strong neural baselines": 113694,
      "textual visual features": 121754,
      "features recent years": 43688,
      "optical character recognition": 82157,
      "character recognition ocr": 16459,
      "new stateoftheart task": 79178,
      "task deep learning": 118050,
      "monolingual word embeddings": 74631,
      "surge social media": 116088,
      "translation paper propose": 125092,
      "propose new deep": 92845,
      "word embeddings different": 133082,
      "shared embedding space": 107977,
      "embedding space finally": 35497,
      "outperforms stateoftheart baseline": 82999,
      "terms classification accuracy": 120292,
      "neural network translation": 78411,
      "network translation models": 77464,
      "systems manual evaluation": 116997,
      "supervised learning using": 115784,
      "languages training data": 60927,
      "training data supervised": 123546,
      "given small set": 48128,
      "training data robust": 123530,
      "baseline large margin": 12239,
      "multiturn dialogue dataset": 75916,
      "evaluate existing approaches": 38835,
      "translation parallel data": 125099,
      "methods learning crosslingual": 69588,
      "learning crosslingual word": 62476,
      "dictionaries parallel corpora": 30864,
      "corpora recent studies": 23569,
      "need parallel data": 76845,
      "showed encouraging results": 108376,
      "monolingual word embedding": 74629,
      "word embedding spaces": 133034,
      "information model outperforms": 54770,
      "model outperforms existing": 71645,
      "crosslingual tasks language": 25021,
      "tasks language pairs": 119241,
      "language pairs experiments": 59759,
      "experiments demonstrate method": 40885,
      "distant language pairs": 32452,
      "limited parallel data": 64262,
      "method fully unsupervised": 68849,
      "unsupervised machine translation": 127660,
      "machine translation code": 65923,
      "dictionaries publicly available": 30868,
      "paper present work": 84146,
      "experiments pos tagging": 41061,
      "pos tagging dependency": 88228,
      "vietnamese language processing": 131679,
      "plays central role": 87721,
      "downstream tasks paper": 34047,
      "mapping event mentions": 67130,
      "task neural machine": 118453,
      "translation nmt struggles": 125057,
      "parallel corpus used": 84627,
      "corpus used train": 24057,
      "evaluation results best": 39366,
      "word vectors lexical": 133640,
      "word vector space": 133630,
      "effectiveness robustness proposed": 34945,
      "sentiment analysis spanish": 106661,
      "analysis spanish tweets": 5865,
      "svm convolutional neural": 116238,
      "classifier best performance": 17526,
      "word embeddings particularly": 133179,
      "customer feedback analysis": 25439,
      "paper describes systems": 83851,
      "2017 shared task": 640,
      "shared task customer": 108035,
      "task customer feedback": 118042,
      "cnn bidirectional lstm": 17993,
      "baseline model best": 12256,
      "model best performing": 70771,
      "best performing model": 13403,
      "embedding text classification": 35516,
      "previous works treat": 90579,
      "information makes difficult": 54748,
      "models jointly learn": 73434,
      "text classification semantic": 120763,
      "text classification models": 120743,
      "neural conversation models": 77875,
      "data model training": 26136,
      "model training paper": 72221,
      "proposing multitask learning": 93625,
      "multitask learning approach": 75823,
      "learning approach training": 62352,
      "approach leads significant": 7671,
      "leads significant improvements": 61955,
      "improvements baseline model": 52810,
      "large quantities data": 61229,
      "text coherence analysis": 120795,
      "network paper propose": 77382,
      "propose novel deep": 92911,
      "trained endtoend fashion": 123130,
      "proposed model evaluated": 93445,
      "results demonstrate effectiveness": 101673,
      "significant improvement stateoftheart": 108776,
      "learning semantic similarity": 62996,
      "task semantic similarity": 118674,
      "embeddings social media": 35944,
      "social media platforms": 110403,
      "sources information discuss": 110898,
      "provide new insights": 93881,
      "paper propose transformation": 84336,
      "word embeddings extracted": 133114,
      "outperforms original version": 82930,
      "achieves best accuracy": 2737,
      "best accuracy stanford": 13299,
      "accuracy stanford dependencies": 2286,
      "conversion penn treebank": 23088,
      "short text classification": 108256,
      "model languages present": 71412,
      "multilingual text classification": 75383,
      "does require parallel": 33396,
      "require parallel data": 100185,
      "vector machine classifier": 131312,
      "word embeddings character": 133062,
      "analysis languages english": 5670,
      "english french japanese": 37144,
      "french japanese spanish": 45818,
      "slot filling paper": 110051,
      "importance coreference resolution": 52052,
      "coreference resolution systems": 23372,
      "provide experimental results": 93820,
      "performance slot filling": 86722,
      "induction task learning": 54037,
      "data work investigate": 26631,
      "sequence learning neural": 107001,
      "learning neural machine": 62840,
      "language generation problem": 59092,
      "recent work investigated": 96574,
      "work investigated use": 134595,
      "propose tackle problem": 93105,
      "representations sequence sequence": 99883,
      "latent variables capture": 61636,
      "sequence sequence models": 107061,
      "shown useful various": 108537,
      "semantic syntactic information": 105315,
      "syntactic information words": 116424,
      "word vectors trained": 133648,
      "people paper propose": 85883,
      "skipgram model using": 109987,
      "social network data": 110439,
      "word vectors obtained": 133641,
      "sentence completion task": 105793,
      "word vectors shown": 133647,
      "freely available dataset": 45781,
      "incorporates number novel": 53510,
      "baseline machine translation": 12246,
      "systems machine translation": 116989,
      "machine translation lowresource": 66034,
      "phrasebased statistical mt": 87397,
      "swiss german dialects": 116275,
      "advances deep learning": 3981,
      "deep learning dl": 28268,
      "paper proposes new": 84360,
      "image captioning architecture": 51763,
      "including long shortterm": 53314,
      "grammatical structure sentence": 48723,
      "knowledge base triples": 57793,
      "context address problem": 22003,
      "use neural networks": 128169,
      "train evaluate models": 122935,
      "promising results improving": 92295,
      "traditional statistical machine": 122869,
      "search space standard": 104350,
      "translation quality different": 125167,
      "complex natural language": 19841,
      "natural language phenomena": 76383,
      "paper introduces new": 84009,
      "information extraction tasks": 54606,
      "support wide range": 116019,
      "event extraction tasks": 39518,
      "corpora neural machine": 23538,
      "nmt new paradigm": 79935,
      "parallel data scarce": 84650,
      "domains language pairs": 33799,
      "language pairs paper": 59771,
      "different generation strategies": 31163,
      "word embeddings deep": 133074,
      "processing nlp models": 91744,
      "nlp models require": 79642,
      "word embeddings resulting": 133201,
      "maximize compression rate": 67510,
      "tasks proposed method": 119415,
      "proposed method improve": 93371,
      "improve model performance": 52417,
      "multitask multilingual modelling": 75882,
      "learning multiple tasks": 62817,
      "especially lowresource languages": 38473,
      "lowresource languages benefit": 65509,
      "created new word": 24679,
      "original english word2vec": 82517,
      "english word2vec word": 37345,
      "word2vec word analogy": 133687,
      "specific linguistic aspects": 111461,
      "corpora basic evaluation": 23424,
      "basic evaluation word": 12521,
      "evaluation word similarities": 39446,
      "word representation models": 133440,
      "tokens training data": 122345,
      "able create meaningful": 1627,
      "quality resulting word": 94779,
      "response generation strategies": 101212,
      "simple question answering": 109497,
      "model semantic composition": 71964,
      "treestructured neural network": 125729,
      "yield significant improvements": 135352,
      "significant improvements existing": 108794,
      "model stanford nli": 72082,
      "stanford sentiment treebank": 112382,
      "previous best known": 90392,
      "dirichlet allocation lda": 31923,
      "vietnamese paper presents": 131688,
      "vietnamese partofspeech pos": 131690,
      "nguyen et al": 79473,
      "learning recent work": 62944,
      "models structured prediction": 74107,
      "beam search optimization": 12608,
      "larger wmt14 englishfrench": 61393,
      "englishfrench translation task": 37368,
      "empirical results demonstrate": 36183,
      "nlpcc 2017 shared": 79795,
      "model unsupervised morphological": 72250,
      "present language independent": 89526,
      "language independent unsupervised": 59165,
      "model handles problem": 71274,
      "handles problem data": 49411,
      "problem data sparsity": 90994,
      "yields improved word": 135417,
      "improved word embeddings": 52657,
      "word similarity task": 133570,
      "large training set": 61303,
      "training set standard": 123853,
      "standard test sets": 112322,
      "features addition features": 43356,
      "detailed qualitative analysis": 29784,
      "qualitative analysis demonstrates": 94552,
      "questionanswer pairs cover": 95247,
      "code publicly available": 18141,
      "achieved recent success": 2677,
      "natural language narratives": 76373,
      "information extracted text": 54560,
      "results provide insight": 102089,
      "comparable translation performance": 19190,
      "translation challenging problem": 124692,
      "introduce new evaluation": 56482,
      "new evaluation method": 78907,
      "online forums users": 81775,
      "use topic models": 128333,
      "important step natural": 52257,
      "neural conversational models": 77878,
      "using large set": 129791,
      "response selection retrievalbased": 101241,
      "produce distributed representations": 91884,
      "ubuntu dialogue corpus": 126521,
      "demonstrate significant performance": 28865,
      "significant performance gains": 108838,
      "gains proposed method": 46406,
      "retrievalbased question answering": 102448,
      "study transfer learning": 114536,
      "knowledge learned resourcerich": 58048,
      "transfer learning methods": 124105,
      "source target domains": 110833,
      "propose efficient effective": 92644,
      "model extensive experiments": 71154,
      "paraphrase identification natural": 84819,
      "identification natural language": 51407,
      "promising performance compared": 92286,
      "models transfer learning": 74239,
      "transfer learning method": 124100,
      "transfer learning model": 124106,
      "bring significant improvements": 14636,
      "representations capture underlying": 99532,
      "propose novel variant": 92976,
      "representations word embeddings": 99982,
      "word embeddings beginning": 133054,
      "large scale human": 61247,
      "scale human evaluation": 103724,
      "human evaluation report": 50829,
      "word embeddings diverse": 133090,
      "dialogue systems use": 30792,
      "potential ethical issues": 88553,
      "ethical issues arise": 38749,
      "tasks word embeddings": 119605,
      "performance nlp tasks": 86575,
      "nlp tasks recurrent": 79760,
      "recurrent convolutional architectures": 97165,
      "widely used natural": 132576,
      "based deep learning": 11629,
      "various text classification": 131221,
      "dependencybased word embeddings": 29259,
      "neural distributional models": 77899,
      "syntactic context words": 116387,
      "superior performance tasks": 115692,
      "word representation model": 133439,
      "word embeddings glove": 133120,
      "different embedding models": 31122,
      "nmt models generally": 79914,
      "translate sentences isolation": 124541,
      "documentlevel information work": 33145,
      "information work propose": 55107,
      "work propose augment": 134715,
      "language understanding models": 60238,
      "understanding models introduce": 126891,
      "present neural architecture": 89572,
      "real user data": 96086,
      "commercial personal assistant": 18838,
      "significant performance improvement": 108843,
      "performance improvement stateoftheart": 86444,
      "slot tagging models": 110068,
      "using contextual information": 129483,
      "sentiment analysis develop": 106583,
      "develop machine learning": 30210,
      "machine learning classification": 65786,
      "supervised machine learning": 115788,
      "machine learning algorithm": 65766,
      "maximizing lower bound": 67524,
      "monte carlo sampling": 74653,
      "experimental results automatic": 40558,
      "automatic metrics human": 10603,
      "significantly outperform stateoftheart": 108987,
      "outperform stateoftheart methods": 82758,
      "terms response quality": 120377,
      "learning slot filling": 63032,
      "slot filling spoken": 110057,
      "filling spoken language": 44056,
      "representations slot filling": 99896,
      "filling task spoken": 44059,
      "task spoken language": 118736,
      "work propose adversarial": 134710,
      "propose adversarial training": 92535,
      "adversarial training method": 4176,
      "learning common features": 62444,
      "representations shared multiple": 99889,
      "models trained individual": 74206,
      "slot filling f1": 110044,
      "slot filling performance": 110053,
      "finite dimensional vector": 44534,
      "dimensional vector spaces": 31756,
      "deep natural language": 28343,
      "recent work shown": 96585,
      "content words annotated": 21977,
      "product paper address": 92041,
      "closely related problems": 17863,
      "design neural network": 29551,
      "neural network called": 78271,
      "semisupervised sequence labeling": 105622,
      "extensive experiments model": 41915,
      "outperforms stateoftheart baselines": 83001,
      "sentence model based": 105951,
      "model based encoderdecoder": 70729,
      "based encoderdecoder architecture": 11675,
      "sequence learning problem": 107002,
      "make following contributions": 66675,
      "create parallel corpora": 24635,
      "language model achieved": 59287,
      "model achieved stateoftheart": 70550,
      "model able correctly": 70511,
      "domain neural machine": 33595,
      "hard negative mining": 49471,
      "achieves better results": 2751,
      "better results terms": 13708,
      "terms average precision": 120280,
      "compared models trained": 19389,
      "models trained proposed": 74224,
      "trained proposed approach": 123248,
      "hierarchical text generation": 50015,
      "reinforcement learning experiments": 97809,
      "language processing recently": 59939,
      "machine learning neural": 65818,
      "learning neural networks": 62844,
      "neural networks methods": 78513,
      "train neural networks": 122988,
      "specific aspects product": 111408,
      "attentive memory networks": 10118,
      "leveraging external knowledge": 63678,
      "external knowledge bases": 42009,
      "information conveyed user": 54456,
      "machine reading task": 65868,
      "information retrieval community": 54931,
      "attentive memory network": 10117,
      "datasets commonly used": 27358,
      "used evaluating machine": 128526,
      "evaluating machine reading": 39071,
      "achieves performance comparable": 2831,
      "performance comparable stateoftheart": 86224,
      "comparable stateoftheart models": 19180,
      "amazon alexa prize": 5267,
      "forums social networks": 45320,
      "morphological syntactic properties": 74737,
      "neural headline generation": 77927,
      "language generation tasks": 59101,
      "generation tasks model": 47672,
      "distributions source target": 32770,
      "model outperforms current": 71643,
      "outperforms current stateoftheart": 82877,
      "headline generation task": 49597,
      "lexical semantic resources": 63816,
      "knowledgebased word sense": 58259,
      "data available social": 25681,
      "available social media": 11105,
      "areas natural language": 8895,
      "language processing sentiment": 59943,
      "processing sentiment analysis": 91793,
      "different domains different": 31111,
      "domains different languages": 33763,
      "different languages paper": 31209,
      "controlled experiments experiments": 22836,
      "good interannotator agreement": 48483,
      "best knowledge study": 13353,
      "stateoftheart speech recognition": 112967,
      "speech recognition systems": 111781,
      "recognition systems rely": 97020,
      "systems rely heavily": 117107,
      "model pronunciation lexicon": 71807,
      "language model build": 59303,
      "expert domain knowledge": 41219,
      "available linguistic resources": 11033,
      "lstmbased model trained": 65682,
      "error rate comparable": 38327,
      "research propose novel": 100594,
      "capability deep learning": 15180,
      "showing competitive results": 108401,
      "text summarization approaches": 121345,
      "novel multitask learning": 80660,
      "extensive set experiments": 41956,
      "english poetry corpus": 37245,
      "subcorpus gutenberg english": 114658,
      "gutenberg english poetry": 49288,
      "semantic analysis significant": 104982,
      "research digital humanities": 100471,
      "empirical results obtained": 36191,
      "important concepts understanding": 52129,
      "individual group different": 53911,
      "validate novel approach": 130728,
      "novel approach detecting": 80475,
      "neural mt nmt": 78213,
      "different text types": 31492,
      "assess translation quality": 9494,
      "100 million words": 167,
      "11 relative improvement": 227,
      "speakers target language": 111328,
      "recurrent neural machine": 97189,
      "latent random variables": 61596,
      "experiments chineseenglish englishgerman": 40824,
      "chineseenglish englishgerman translation": 16850,
      "englishgerman translation tasks": 37389,
      "translation tasks demonstrate": 125350,
      "tasks demonstrate proposed": 119040,
      "significant improvements conventional": 108793,
      "sentiment classification existing": 106699,
      "learning models achieved": 62772,
      "models achieved great": 72679,
      "achieved great success": 2632,
      "sentiment classification typically": 106718,
      "consider word interaction": 21239,
      "compared traditional word": 19482,
      "traditional word embeddings": 122885,
      "learning experimental results": 62564,
      "experimental results comparable": 40572,
      "performance stateoftheart methods": 86745,
      "efficient text classification": 35120,
      "text classification using": 120783,
      "classification using treestructured": 17496,
      "principal component analysis": 90671,
      "dimension reduction technique": 31747,
      "dimension reduction methods": 31746,
      "component analysis pca": 19967,
      "results support vector": 102241,
      "better performance stateoftheart": 13662,
      "stateoftheart recurrent neural": 112882,
      "information extraction knowledge": 54573,
      "extraction knowledge base": 42363,
      "order promote research": 82389,
      "promote research area": 92320,
      "annotated different types": 6185,
      "text taken wikipedia": 121358,
      "tasks neural machine": 119339,
      "nmt models provide": 79920,
      "vector representations learned": 131355,
      "morphological syntactic tasks": 74738,
      "syntactic tasks paper": 116492,
      "tasks paper investigate": 119361,
      "representations learned different": 99726,
      "parallel data use": 84659,
      "models extract features": 73221,
      "analysis yields interesting": 5943,
      "instance higher layers": 55602,
      "lower layers tend": 65438,
      "better partofspeech tagging": 13649,
      "pos tags word": 88246,
      "using feedforward neural": 129674,
      "train sequencetosequence neural": 123018,
      "information encoderdecoder framework": 54527,
      "important preprocessing step": 52210,
      "nlp tasks questionanswering": 79758,
      "specific word senses": 111518,
      "text data available": 120850,
      "human judgment experiment": 50881,
      "compare performance approaches": 19270,
      "learning goaloriented dialogue": 62619,
      "goaloriented dialogue systems": 48412,
      "dialogue systems colloquially": 30770,
      "systems colloquially known": 116788,
      "colloquially known goal": 18526,
      "known goal oriented": 58299,
      "goal oriented chatbots": 48374,
      "oriented chatbots help": 82488,
      "chatbots help users": 16673,
      "help users achieve": 49786,
      "users achieve predefined": 129091,
      "achieve predefined goal": 2512,
      "predefined goal book": 88825,
      "goal book movie": 48334,
      "book movie ticket": 14409,
      "movie ticket closed": 74908,
      "ticket closed domain": 121958,
      "closed domain step": 17844,
      "domain step understand": 33659,
      "step understand users": 113308,
      "understand users goal": 126787,
      "users goal using": 129125,
      "goal using natural": 48396,
      "language understanding techniques": 60287,
      "understanding techniques goal": 126994,
      "techniques goal known": 119896,
      "goal known bot": 48361,
      "known bot manage": 58290,
      "bot manage dialogue": 14482,
      "manage dialogue achieve": 66897,
      "dialogue achieve goal": 30626,
      "achieve goal conducted": 2464,
      "goal conducted respect": 48339,
      "conducted respect learnt": 20941,
      "respect learnt policy": 101087,
      "learnt policy success": 63251,
      "policy success dialogue": 87966,
      "success dialogue depends": 115074,
      "dialogue depends quality": 30661,
      "depends quality policy": 29290,
      "quality policy turn": 94754,
      "policy turn reliant": 87971,
      "turn reliant availability": 125977,
      "reliant availability highquality": 98635,
      "availability highquality training": 10910,
      "highquality training data": 50411,
      "training data policy": 123513,
      "data policy learning": 26244,
      "policy learning method": 87959,
      "learning method instance": 62722,
      "method instance deep": 68913,
      "instance deep reinforcement": 55596,
      "reinforcement learning domain": 97803,
      "learning domain specificity": 62520,
      "domain specificity available": 33654,
      "specificity available data": 111615,
      "available data typically": 10972,
      "data typically low": 26588,
      "typically low allow": 126442,
      "low allow training": 65348,
      "allow training good": 5094,
      "training good dialogue": 123643,
      "good dialogue policies": 48474,
      "introduce transfer learning": 56557,
      "learning method mitigate": 62726,
      "method mitigate effects": 68964,
      "mitigate effects low": 70364,
      "effects low indomain": 34991,
      "low indomain data": 65367,
      "indomain data availability": 53961,
      "data availability transfer": 25667,
      "availability transfer learning": 10929,
      "transfer learning based": 124082,
      "learning based approach": 62384,
      "based approach improves": 11512,
      "approach improves bots": 7624,
      "improves bots success": 52956,
      "bots success rate": 14490,
      "success rate 20": 115120,
      "rate 20 relative": 95774,
      "20 relative terms": 523,
      "relative terms distant": 98380,
      "terms distant domains": 120309,
      "distant domains double": 32447,
      "domains double close": 33769,
      "double close domains": 33962,
      "close domains compared": 17812,
      "domains compared model": 33746,
      "compared model transfer": 19385,
      "model transfer learning": 72225,
      "transfer learning transfer": 124140,
      "learning transfer learning": 63120,
      "transfer learning chatbots": 124085,
      "learning chatbots learn": 62427,
      "chatbots learn policy": 16677,
      "learn policy 10": 62113,
      "policy 10 times": 87945,
      "10 times faster": 150,
      "times faster finally": 122174,
      "faster finally transfer": 43175,
      "finally transfer learning": 44242,
      "learning approach complementary": 62343,
      "approach complementary additional": 7438,
      "complementary additional processing": 19734,
      "additional processing warmstarting": 3560,
      "processing warmstarting joint": 91853,
      "warmstarting joint application": 132027,
      "joint application gives": 57259,
      "application gives best": 6853,
      "gives best outcomes": 48180,
      "word embeddings words": 133255,
      "word meanings lexical": 133362,
      "computational model represents": 20408,
      "natural language despite": 76284,
      "address issue introduce": 3688,
      "properties word embeddings": 92493,
      "word cooccurrence statistics": 132974,
      "good results paper": 48500,
      "networks proposed model": 77714,
      "layers encoder decoder": 61774,
      "arabic chinese english": 8496,
      "properties proposed model": 92475,
      "endtoend deep neural": 36885,
      "pretrained word embedding": 90213,
      "training set address": 123846,
      "evaluate effectiveness proposed": 38828,
      "understanding recurrent neural": 126941,
      "remains open challenge": 98812,
      "language models extracting": 59554,
      "text documents using": 120889,
      "work present novel": 134685,
      "text categorization problem": 120700,
      "language present method": 59821,
      "ontology commonly used": 81839,
      "outperforms existing methods": 82892,
      "proposed framework consists": 93286,
      "vectors convolutional neural": 131423,
      "work tackle problem": 134840,
      "combine convolutional neural": 18618,
      "vectors word embedding": 131487,
      "word embedding representations": 133031,
      "using trained models": 130302,
      "use convolutional neural": 127969,
      "improvements language pairs": 52865,
      "stateoftheart nmt systems": 112794,
      "data set automatically": 26400,
      "set automatically extracted": 107369,
      "release data set": 98447,
      "data set use": 26416,
      "paper proposes method": 84357,
      "knearest neighbors knn": 57718,
      "neighbors knn model": 76998,
      "multilayer perceptron mlp": 75181,
      "different evaluation metrics": 31133,
      "evaluation metrics used": 39296,
      "deep contextualized word": 28218,
      "contextualized word representations": 22580,
      "contextualized word representation": 22579,
      "deep bidirectional language": 28203,
      "bidirectional language model": 13931,
      "language model bilm": 59301,
      "representations easily added": 99612,
      "easily added existing": 34434,
      "added existing models": 3359,
      "existing models significantly": 40208,
      "models significantly improve": 74041,
      "state art challenging": 112431,
      "including question answering": 53356,
      "question answering textual": 95117,
      "answering textual entailment": 6715,
      "entailment sentiment analysis": 37676,
      "response selection multiturn": 101238,
      "long input sequences": 65078,
      "train models using": 122965,
      "improvements compared previous": 52829,
      "information extraction oie": 54584,
      "downstream tasks including": 34039,
      "relation extraction question": 98021,
      "perform significantly worse": 86070,
      "text encyclopedic text": 120909,
      "similarity words sentences": 109343,
      "words sentences using": 134199,
      "calculating semantic similarity": 15033,
      "problem area natural": 90941,
      "semantic similarity words": 105297,
      "models word similarity": 74352,
      "deep transfer learning": 28427,
      "models typically trained": 74257,
      "training work propose": 123955,
      "inductive transfer learning": 54052,
      "different learning tasks": 31225,
      "processing nlp domain": 91732,
      "experimental evaluations proposed": 40539,
      "evaluations proposed approach": 39481,
      "proposed approach reduces": 93209,
      "artificial intelligence ai": 9247,
      "extractive document summarization": 42552,
      "document level data": 33033,
      "bridge gap present": 14593,
      "release annotated corpus": 98433,
      "recent advances artificial": 96394,
      "advances artificial intelligence": 3976,
      "requires understanding data": 100330,
      "paper propose novelty": 84311,
      "propose novelty metric": 92980,
      "based text content": 12115,
      "leads accurate results": 61918,
      "previous work used": 90562,
      "used models large": 128638,
      "model superior performance": 72115,
      "synthetic natural data": 116633,
      "compared standard approach": 19448,
      "generation natural language": 47498,
      "experts nonexperts results": 41245,
      "representations semantic parsing": 99872,
      "improve natural language": 52431,
      "annotated parallel corpus": 6220,
      "computing precision recall": 20525,
      "abstract meaning representations": 1775,
      "improving sentiment analysis": 53164,
      "sentiment analysis arabic": 106569,
      "makes sentiment analysis": 66806,
      "analysis arabic challenging": 5501,
      "recent years deep": 96604,
      "years deep neural": 135251,
      "results sentiment classification": 102158,
      "sentiment classification natural": 106703,
      "classification natural language": 17300,
      "processing applications word": 91621,
      "word embedding word": 133043,
      "neural networks different": 78466,
      "accuracy sentiment classification": 2273,
      "publicly available arabic": 94292,
      "underlying natural language": 126691,
      "supervised deep learning": 115749,
      "paper propose hierarchical": 84250,
      "pair text snippets": 83460,
      "microsoft research paraphrase": 70061,
      "dataset extensive experiments": 26920,
      "extensive experiments proposed": 41919,
      "improve performance existing": 52467,
      "learning models based": 62774,
      "semeval2018 task 11": 105532,
      "machine comprehension paper": 65753,
      "comprehension paper describes": 20208,
      "task 11 machine": 117812,
      "11 machine comprehension": 221,
      "machine comprehension using": 65754,
      "comprehension using commonsense": 20241,
      "using commonsense knowledge": 129455,
      "commonsense knowledge use": 19003,
      "passage question answers": 85609,
      "incorporate commonsense knowledge": 53454,
      "official test data": 81606,
      "data code publicly": 25736,
      "language processing area": 59844,
      "given set words": 48123,
      "syntactic semantic analysis": 116464,
      "semantic features text": 105053,
      "results deep learning": 101669,
      "suggestion model based": 115363,
      "sentence completion challenge": 105792,
      "chinese paper propose": 16800,
      "endtoend neural model": 36942,
      "makes difficult train": 66772,
      "difficult train model": 31669,
      "training data model": 123494,
      "model copying mechanism": 70928,
      "work introduce task": 134580,
      "standard named entity": 112273,
      "distantly supervised dataset": 32491,
      "neural models task": 78199,
      "large improvements strong": 61111,
      "improvements strong baseline": 52921,
      "neural question answering": 78641,
      "entirely attention mechanism": 37728,
      "attention mechanism transformer": 9915,
      "vaswani et al": 131288,
      "stateoftheart results machine": 112907,
      "does explicitly model": 33348,
      "present alternative approach": 89364,
      "englishtogerman englishtofrench translation": 37447,
      "translation tasks approach": 125348,
      "approach yields improvements": 8028,
      "yields improvements 13": 135421,
      "improvements 13 bleu": 52790,
      "improvement translation quality": 52780,
      "endtoend goaloriented dialog": 36905,
      "advancements deep learning": 3970,
      "goaloriented dialog systems": 48410,
      "systems achieve good": 116712,
      "work address limitations": 134355,
      "model word order": 72321,
      "neural network able": 78234,
      "bengalienglish codemixed data": 13042,
      "codemixed data using": 18171,
      "text remains challenging": 121241,
      "remains challenging task": 98792,
      "approach language identification": 7662,
      "language identification word": 59149,
      "word level low": 133343,
      "deep lstm models": 28331,
      "ensemble models using": 37613,
      "bridging anaphora resolution": 14604,
      "current models word": 25303,
      "semantic similarity information": 105281,
      "hou et al": 50677,
      "models work propose": 74361,
      "propose new language": 92856,
      "new language modeling": 78975,
      "language modeling paradigm": 59452,
      "language models models": 59603,
      "approach allows seamlessly": 7350,
      "lexical items language": 63774,
      "model experiments multiple": 71133,
      "language modeling tasks": 59472,
      "language models utilize": 59686,
      "binary multiclass classification": 14133,
      "language inference sentence": 59194,
      "evaluating sentence representations": 39099,
      "question naturally arises": 95188,
      "human parity translation": 50922,
      "translations widely used": 125507,
      "translation task chinese": 125330,
      "stateoftheart translation quality": 113018,
      "shared task word": 108107,
      "task word sense": 118852,
      "describes results shared": 29429,
      "results shared task": 102168,
      "sense induction wsi": 105685,
      "languages rich morphology": 60855,
      "new lexical resource": 78994,
      "representations boost performance": 99525,
      "semantic natural language": 105119,
      "platform paper describes": 87655,
      "methods natural language": 69632,
      "rajpurkar et al": 95467,
      "aspectlevel sentiment classification": 9368,
      "classification sentiment analysis": 17393,
      "research multilingual crosslingual": 100559,
      "multilingual crosslingual sentiment": 75229,
      "require large number": 100164,
      "large number resources": 61187,
      "aspectlevel sentiment analysis": 9367,
      "researchers working languages": 100719,
      "embeddings preserving semantic": 35875,
      "preserving semantic information": 89939,
      "ones experimental results": 81685,
      "experimental results semantic": 40692,
      "similarity text classification": 109321,
      "text classification sentiment": 120764,
      "sentiment analysis tasks": 106667,
      "applications machine learning": 6962,
      "text processing tasks": 121213,
      "different natural languages": 31281,
      "languages consider problem": 60465,
      "novel approach called": 80472,
      "retrieved knowledge base": 102462,
      "network architecture learns": 77141,
      "stateoftheart approaches based": 112581,
      "wordlevel language modeling": 133741,
      "language modeling problem": 59457,
      "parallel sentence pairs": 84679,
      "deep neural model": 28351,
      "trained parallel corpus": 123234,
      "based surface features": 12095,
      "matter neural machine": 67488,
      "using dependency context": 129555,
      "bilingual word embeddings": 14070,
      "using parser trained": 130008,
      "paper studies problem": 84440,
      "problem short text": 91228,
      "experimental results framework": 40618,
      "framework significantly outperforms": 45686,
      "significantly outperforms baselines": 109005,
      "outperforms baselines substantial": 82851,
      "learning general purpose": 62610,
      "distributed sentence representations": 32627,
      "success natural language": 115102,
      "trained large amounts": 123174,
      "large amounts text": 61021,
      "text unsupervised manner": 121390,
      "range nlp problems": 95591,
      "success learning representations": 115091,
      "remains open problem": 98814,
      "problem recent work": 91198,
      "unsupervised supervised learning": 127729,
      "objectives learn general": 81139,
      "representations work present": 99997,
      "effective multitask learning": 34716,
      "single model train": 109764,
      "model train model": 72179,
      "multiple training objectives": 75734,
      "sentences extensive experiments": 106311,
      "improvements previous methods": 52899,
      "transfer learning lowresource": 124099,
      "learning lowresource settings": 62698,
      "lowresource settings using": 65565,
      "codemixed social media": 18182,
      "text social media": 121313,
      "media platforms twitter": 68149,
      "platforms twitter facebook": 87674,
      "south asian languages": 110964,
      "languages codemixed data": 60450,
      "noisy social media": 80121,
      "using distributed representations": 129597,
      "words experiments reveal": 133950,
      "representations resourcepoor languages": 99858,
      "resourcepoor languages using": 100915,
      "languages using siamese": 60954,
      "learning approaches sentiment": 62363,
      "approaches sentiment analysis": 8335,
      "novel method called": 80630,
      "siamese network architecture": 108652,
      "network architecture sentiment": 77144,
      "architecture sentiment analysis": 8743,
      "resourcepoor languages jointly": 100912,
      "languages jointly training": 60657,
      "jointly training resourcerich": 57397,
      "training resourcerich languages": 123816,
      "resourcerich languages using": 100934,
      "using siamese network": 130169,
      "model consists twin": 70896,
      "consists twin bidirectional": 21501,
      "twin bidirectional long": 126089,
      "neural networks bilstm": 78448,
      "networks bilstm rnn": 77526,
      "bilstm rnn shared": 14104,
      "rnn shared parameters": 102933,
      "shared parameters joined": 108000,
      "parameters joined contrastive": 84758,
      "joined contrastive loss": 57251,
      "contrastive loss function": 22734,
      "loss function based": 65267,
      "function based similarity": 46000,
      "based similarity metric": 12045,
      "similarity metric model": 109268,
      "metric model learns": 69890,
      "representations resourcepoor resourcerich": 99861,
      "resourcerich language common": 100924,
      "common sentiment space": 18926,
      "space using similarity": 111076,
      "using similarity metric": 130174,
      "similarity metric based": 109267,
      "model projects sentences": 71803,
      "projects sentences similar": 92244,
      "closer sentences different": 17878,
      "farther experiments largescale": 43112,
      "experiments largescale datasets": 40984,
      "datasets resourcerich languages": 27680,
      "resourcerich languages english": 100927,
      "english spanish resourcepoor": 37285,
      "spanish resourcepoor languages": 111168,
      "resourcepoor languages hindi": 100909,
      "languages hindi telugu": 60621,
      "hindi telugu reveal": 50474,
      "sentiment analysis approaches": 106568,
      "approaches based distributional": 8076,
      "distributional semantics semantic": 32733,
      "semantics semantic rules": 105461,
      "semantic rules lexicon": 105263,
      "rules lexicon lists": 103417,
      "lexicon lists deep": 63894,
      "lists deep neural": 64716,
      "neural network representations": 78384,
      "asr natural language": 9439,
      "task experiment different": 118171,
      "sequencetosequence neural models": 107188,
      "dataset approach produces": 26742,
      "availability large amounts": 10916,
      "sequencetosequence models generate": 107179,
      "leveraging transfer learning": 63710,
      "different languages english": 31198,
      "fewshot text classification": 43918,
      "word embeddings human": 133125,
      "classifier accurately predict": 17518,
      "corpora documents classified": 23462,
      "fewshot learning approach": 43905,
      "uses pretrained word": 129266,
      "word embeddings tested": 133224,
      "results code available": 101584,
      "resource indian languages": 100850,
      "annotated gold standard": 6200,
      "gold standard corpus": 48438,
      "disambiguation wsd tasks": 32001,
      "yield significant performance": 135353,
      "universal decompositional semantics": 127281,
      "input sequences considering": 55433,
      "model reinforcement learning": 71884,
      "develop evaluation metrics": 30197,
      "performance proposed models": 86635,
      "propose simple fast": 93061,
      "similar input sentence": 109101,
      "target sentences aligned": 117703,
      "existing nmt model": 40239,
      "nmt model used": 79903,
      "model used translate": 72258,
      "nmt translation results": 80000,
      "domain translation tasks": 33684,
      "sentiment analysis todays": 106673,
      "machine learning natural": 65815,
      "trained tested using": 123310,
      "data driven approaches": 25871,
      "statistical machine learning": 113105,
      "learning approach used": 62354,
      "classification deep learning": 17176,
      "deep learning algorithms": 28246,
      "parallel corpus using": 84628,
      "machine translation significant": 66218,
      "parallel corpus filtering": 84617,
      "models trained using": 74235,
      "source word embeddings": 110865,
      "pretrained source embeddings": 90190,
      "received attention lately": 96358,
      "given semantic relation": 48114,
      "paper describes process": 83841,
      "describes process building": 29427,
      "textual information paper": 121714,
      "information paper addresses": 54822,
      "scenario small number": 103832,
      "user embeddings user": 128978,
      "entity recognition using": 38084,
      "word representations recent": 133470,
      "representations recent work": 99841,
      "field crf model": 43944,
      "skipgram cbow glove": 109982,
      "syntactic semantic processing": 116472,
      "different syntactic semantic": 31470,
      "signals recorded reading": 108705,
      "neural reinforcement learning": 78646,
      "address challenges propose": 3664,
      "learning method learn": 62725,
      "neural network representation": 78383,
      "representation sentences model": 99406,
      "results approach outperforms": 101515,
      "approach outperforms baseline": 7765,
      "results semeval2018 task": 102149,
      "et al 2018": 38716,
      "improved f1 score": 52605,
      "paper present study": 84138,
      "texts russian national": 121598,
      "russian national corpus": 103496,
      "serve starting point": 107301,
      "data present new": 26266,
      "learning semantic parsers": 62993,
      "semantic parsers multiple": 105139,
      "important error analysis": 52157,
      "information extraction applications": 54565,
      "persons killed police": 87155,
      "semisupervised learning framework": 105606,
      "large unlabeled data": 61314,
      "highquality unlabeled samples": 50416,
      "demonstrate proposed method": 28840,
      "proposed method obtain": 93380,
      "domain present novel": 33614,
      "present novel methodology": 89614,
      "supervised learning data": 115767,
      "texts paper aim": 121568,
      "large scale datasets": 61246,
      "data text pairs": 26560,
      "results demonstrate models": 101694,
      "interpreting neural models": 56289,
      "neural models based": 78166,
      "classification task importantly": 17432,
      "interpret neural models": 56213,
      "shed light limitations": 108154,
      "pos tagging parsing": 88237,
      "parser achieves stateoftheart": 84916,
      "stateoftheart performance downstream": 112821,
      "report results preliminary": 99037,
      "results preliminary experiments": 102048,
      "monolingual multilingual models": 74600,
      "ntuaslp semeval2018 task": 80820,
      "paper present deeplearning": 84095,
      "competed semeval2018 task": 19608,
      "semeval2018 task irony": 105538,
      "task irony detection": 118316,
      "irony detection english": 56939,
      "detection english tweets": 29936,
      "models based recurrent": 72813,
      "augmented selfattention mechanism": 10336,
      "selfattention mechanism order": 104886,
      "word2vec word embeddings": 133690,
      "word embeddings pretrained": 133187,
      "collection 550 million": 18456,
      "550 million english": 1117,
      "million english tweets": 70098,
      "handcrafted features lexicons": 49346,
      "data furthermore provide": 25965,
      "proposed models ranked": 93498,
      "open domain dialogue": 81892,
      "combining machine learning": 18733,
      "rule based approaches": 103348,
      "present experimental results": 89481,
      "tasks relation extraction": 119451,
      "structure paper present": 113928,
      "paper present methods": 84111,
      "dependencies graph representation": 29087,
      "case study swedish": 15624,
      "recent years neural": 96623,
      "neural extractive summarization": 77914,
      "neural coherence model": 77864,
      "using unlabeled data": 130341,
      "unlabeled data empirical": 127382,
      "data empirical results": 25887,
      "empirical results proposed": 36192,
      "learning method train": 62730,
      "simultaneously experimental results": 109676,
      "outperforms existing baselines": 82889,
      "cnndaily mail dataset": 18034,
      "present simple method": 89706,
      "experiments english japanese": 40935,
      "describing clinical trials": 29449,
      "method learns representations": 68940,
      "stance detection using": 112171,
      "memory networks present": 68359,
      "fake news challenge": 43013,
      "news challenge dataset": 79315,
      "problem propose method": 91176,
      "existing paraphrase databases": 40251,
      "novel approach learn": 80481,
      "resulting sentence embeddings": 101471,
      "question answering cqa": 95031,
      "language inference task": 59199,
      "task extensive experiments": 118192,
      "performance neural models": 86569,
      "words input sentence": 134005,
      "models make predictions": 73548,
      "neural models trained": 78204,
      "trained maximum likelihood": 123198,
      "reduction accuracy loss": 97439,
      "representation language model": 99274,
      "pretrained language models": 90061,
      "language models lms": 59592,
      "brought significant improvements": 14717,
      "significant improvements various": 108815,
      "capture linguistic information": 15330,
      "different layers model": 31223,
      "better robustness experiments": 13713,
      "robustness experiments benchmark": 103099,
      "benchmark datasets demonstrate": 12772,
      "datasets demonstrate effectiveness": 27398,
      "demonstrate effectiveness method": 28712,
      "transfer transfer learning": 124200,
      "transfer learning sentence": 124130,
      "learning sentence embeddings": 63001,
      "sentence embeddings semantic": 105844,
      "embeddings semantic similarity": 35926,
      "transfer downstream tasks": 124049,
      "transfer learning techniques": 124137,
      "received relatively little": 96373,
      "little attention paper": 64796,
      "attention paper propose": 9965,
      "propose transfer learning": 93118,
      "semantic similarity tasks": 105292,
      "greatly affects performance": 49045,
      "unsupervised natural language": 127678,
      "generation denoising autoencoders": 47361,
      "generating text structured": 47276,
      "text structured data": 121333,
      "tasks question answering": 119427,
      "build natural language": 14790,
      "use denoising autoencoder": 127992,
      "denoising autoencoder reconstruct": 28997,
      "introduce noise training": 56507,
      "given structured data": 48140,
      "leads stateoftheart performance": 61965,
      "stateoftheart performance model": 112829,
      "shed new light": 108157,
      "word embeddings nlp": 133170,
      "different sources news": 31443,
      "websites social media": 132306,
      "semantic parsing introduce": 105149,
      "endtoend model novel": 36929,
      "dataset experiments model": 26915,
      "model outperforms strong": 71679,
      "modelling present novel": 72617,
      "largescale domain classification": 61430,
      "intelligent personal digital": 55853,
      "personal digital assistants": 87117,
      "digital assistants ipdas": 31717,
      "application spoken language": 6886,
      "language understanding capabilities": 60210,
      "thousands overlapping domains": 121927,
      "language understanding task": 60279,
      "paper propose set": 84320,
      "additional contextual information": 3502,
      "effectiveness approach extensive": 34868,
      "approach extensive experiments": 7561,
      "data augmentation methods": 25638,
      "different types noise": 31538,
      "proposed data augmentation": 93248,
      "sentence classification tasks": 105787,
      "standard rnn language": 112296,
      "language model generating": 59329,
      "language model known": 59337,
      "stateoftheart results multiple": 112911,
      "results multiple datasets": 101971,
      "endtoend taskoriented dialog": 36977,
      "dialog systems endtoend": 30598,
      "usually suffer challenge": 130457,
      "suffer challenge incorporating": 115211,
      "knowledge bases paper": 57809,
      "propose novel simple": 92966,
      "neural generative model": 77921,
      "multihop attention mechanism": 75137,
      "attention mechanism helps": 9899,
      "stateoftheart performance different": 112819,
      "tweets universal dependencies": 126067,
      "kong et al": 58343,
      "efficiency propose new": 35040,
      "argument reasoning comprehension": 8951,
      "reasoning comprehension task": 96238,
      "comprehension task requires": 20236,
      "task requires significant": 118646,
      "given small size": 48130,
      "small size dataset": 110207,
      "dataset best model": 26768,
      "best model uses": 13374,
      "model achieves mean": 70575,
      "test set accuracy": 120494,
      "relatively low performance": 98405,
      "machine translation graph": 65998,
      "graph convolutional networks": 48786,
      "structure source sentences": 113960,
      "representations neural machine": 99775,
      "machine translation use": 66290,
      "convolutional networks gcns": 23172,
      "language pair using": 59734,
      "contain rich information": 21757,
      "present machine learning": 89539,
      "syntactic features using": 116406,
      "using support vector": 130246,
      "vector machines svms": 131327,
      "corpus translation studies": 24049,
      "lexical information perform": 63771,
      "distinguish translated original": 32577,
      "conference machine translation": 20977,
      "machine translation wmt": 66299,
      "methods applied successfully": 69313,
      "recent work questioned": 96583,
      "scheduled multitask learning": 103885,
      "translation neural encoderdecoder": 125006,
      "learning linguistic knowledge": 62694,
      "large parallel corpus": 61201,
      "wmt14 english german": 132810,
      "language inference propose": 59190,
      "train natural language": 122975,
      "extracting entities relations": 42210,
      "entities relations text": 37861,
      "text important task": 121044,
      "modeling local context": 72465,
      "statistics large corpus": 113203,
      "facts external knowledge": 42913,
      "applied different domains": 7061,
      "using corpuslevel statistics": 129509,
      "based local context": 11809,
      "corpora different domains": 23457,
      "different domains demonstrate": 31108,
      "domains demonstrate effectiveness": 33760,
      "stateoftheart open systems": 112796,
      "propose approach based": 92549,
      "approach based combination": 7380,
      "nonprojective transitionbased parsing": 80282,
      "unsupervised word sense": 127757,
      "sense disambiguation underresourced": 105670,
      "disambiguation underresourced languages": 31992,
      "languages paper present": 60778,
      "sense target word": 105702,
      "word embedding based": 133007,
      "semeval 2018 task": 105503,
      "achieving f1 score": 2945,
      "native language identification": 76225,
      "language identification dataset": 59129,
      "language identification nli": 59135,
      "english spanish german": 37284,
      "second language acquisition": 104415,
      "dataset present results": 27102,
      "present results obtained": 89674,
      "machine translation traditional": 66272,
      "nist chinesetoenglish wmt": 79502,
      "chinesetoenglish wmt englishtogerman": 16876,
      "semantic parsing performance": 105157,
      "multitask learning significantly": 75870,
      "indomain outofdomain settings": 53975,
      "syntax neural machine": 116548,
      "word frequency analysis": 133292,
      "free text data": 45767,
      "machine learning model": 65808,
      "different classification models": 31040,
      "classification models using": 17283,
      "models using ensemble": 74302,
      "embeddings generated different": 35732,
      "targetoriented sentiment classification": 117788,
      "sentiment classification aims": 106693,
      "perform classification task": 85960,
      "classification task propose": 17436,
      "propose new model": 92870,
      "model overcome issues": 71687,
      "extract salient features": 42111,
      "representations words sentence": 99993,
      "experiments model achieves": 41012,
      "model achieves new": 70576,
      "achieves new stateoftheart": 2821,
      "stateoftheart performance benchmarks": 112811,
      "new natural language": 79046,
      "natural language domains": 76291,
      "learning methods applied": 62736,
      "propose novel approaches": 92900,
      "advances state art": 4021,
      "computational linguistics artificial": 20393,
      "linguistics artificial intelligence": 64605,
      "artificial intelligence past": 9255,
      "models vector representations": 74328,
      "evaluation downstream tasks": 39196,
      "downstream tasks like": 34043,
      "tasks like word": 119264,
      "like word sense": 64111,
      "models demonstrate effectiveness": 73034,
      "spectrum paper propose": 111634,
      "probabilistic soft logic": 90829,
      "model outperforms stateofart": 71667,
      "recent work neural": 96580,
      "work neural networkbased": 134652,
      "existing methods mainly": 40189,
      "methods mainly focus": 69609,
      "transfer different languages": 124046,
      "address issue propose": 3697,
      "evaluate proposed model": 38901,
      "high correlation human": 50051,
      "automatic article commenting": 10496,
      "improve user engagement": 52582,
      "chatbots paper proposes": 16681,
      "proposes new task": 93608,
      "new task automatic": 79197,
      "text generation paper": 120995,
      "rewards reinforcement learning": 102686,
      "generate long coherent": 46968,
      "state art task": 112479,
      "hierarchical convolutional neural": 49946,
      "propose novel task": 92968,
      "build new dataset": 14793,
      "neural network achieve": 78235,
      "model outperforms baselines": 71638,
      "baselines large margin": 12421,
      "dataset code available": 26788,
      "estimation qe machine": 38663,
      "qe machine translation": 94531,
      "natural language models": 76372,
      "language models understand": 59681,
      "models require large": 73933,
      "spoken dialog systems": 111976,
      "dialog systems enable": 30597,
      "machine translation human": 65999,
      "utterances high accuracy": 130645,
      "unsupervised supervised techniques": 127730,
      "parts speech sentiment": 85588,
      "sentiment association scores": 106686,
      "context paper investigate": 22209,
      "paper investigate role": 84023,
      "datasets penn treebank": 27617,
      "penn treebank wikitext2": 85864,
      "sensitive order words": 105743,
      "grave et al": 48991,
      "stateofthe art results": 112563,
      "art results task": 9088,
      "machine translation sentence": 66207,
      "present new corpus": 89581,
      "reinforcement learning approach": 97796,
      "content main challenge": 21903,
      "lack parallel data": 58733,
      "data solve problem": 26467,
      "results approach significantly": 101518,
      "performance bleu score": 86187,
      "task requires effective": 118642,
      "dialogue systems paper": 30780,
      "systems paper presents": 117046,
      "collected social media": 18436,
      "various unsupervised supervised": 131234,
      "models response selection": 73950,
      "experimental results confirm": 40573,
      "results confirm proposed": 101621,
      "attentive neural network": 10123,
      "neural network sequence": 78396,
      "improvement bleu score": 52689,
      "models promising text": 73814,
      "promising text generation": 92311,
      "text generation tasks": 121012,
      "great social scientists": 49026,
      "human behavior computer": 50764,
      "behavior computer scientists": 12651,
      "build automatic methods": 14742,
      "automatic methods infer": 10594,
      "methods infer social": 69551,
      "infer social contexts": 54105,
      "social context interaction": 110330,
      "rnn language models": 102920,
      "guide decoding process": 49235,
      "human evaluation demonstrates": 50810,
      "semantic relation classification": 105203,
      "classification semantic relations": 17389,
      "semantic relations terms": 105215,
      "relation classification task": 97957,
      "different existing approaches": 31137,
      "existing approaches use": 40064,
      "learning models built": 62776,
      "distributional word vector": 32747,
      "features proposed model": 43675,
      "proposed model uses": 93486,
      "model uses combination": 72264,
      "commonsense knowledge base": 18991,
      "semantic relatedness using": 105200,
      "using machine translation": 129850,
      "translation paper provides": 125096,
      "paper provides comparative": 84382,
      "provides comparative analysis": 94011,
      "semantic models dsms": 105115,
      "use machine translation": 128135,
      "experimental results significant": 40702,
      "spearman correlation using": 111345,
      "using stateoftheart machine": 130217,
      "stateoftheart machine translation": 112695,
      "machine translation approaches": 65896,
      "introduced machine translation": 56581,
      "machine translation languages": 66024,
      "recent advances word": 96425,
      "word representations greatly": 133455,
      "syntactically similar source": 116518,
      "stateoftheart single model": 112954,
      "decomposable attention model": 28151,
      "challenge dataset models": 16017,
      "comprehension present new": 20216,
      "introduce new neural": 56490,
      "new neural models": 79055,
      "exploit alternative mechanisms": 41405,
      "lstm input encoding": 65629,
      "models improve accuracy": 73366,
      "dataset models available": 27038,
      "spanbased constituency parsing": 111143,
      "achieved competitive accuracies": 2609,
      "et al 2017a": 38715,
      "current state art": 25341,
      "state art accuracy": 112427,
      "endtoend discourse parsing": 36894,
      "sentences penn treebank": 106443,
      "contextualized word vectors": 22584,
      "word vectors pretrained": 133644,
      "form transfer learning": 45138,
      "lack training data": 58763,
      "obtained stateofthe art": 81416,
      "art machine translation": 9056,
      "data lowresourced language": 26102,
      "build nmt systems": 14796,
      "based universal sentence": 12145,
      "sentence representations capture": 106034,
      "capture wide range": 15421,
      "based character word": 11568,
      "character word ngrams": 16479,
      "sentence representations using": 106046,
      "improve automatic evaluation": 52340,
      "machine translation experimental": 65983,
      "translation experimental results": 124810,
      "dataset proposed method": 27119,
      "method achieves stateoftheart": 68608,
      "modeling multiple word": 72482,
      "models capture different": 72878,
      "capture different aspects": 15290,
      "inspired propose model": 55578,
      "word embeddings evaluating": 133103,
      "novel sentence embedding": 80720,
      "methods textual similarity": 69802,
      "use handcrafted features": 128081,
      "model substantially improves": 72102,
      "improves strong baselines": 53051,
      "strong baselines data": 113645,
      "data models publicly": 26139,
      "models publicly available": 73843,
      "rise neural networks": 102867,
      "neural networks particularly": 78534,
      "partofspeech tagging accuracy": 85548,
      "paper investigate models": 84018,
      "models use recurrent": 74279,
      "partofspeech morphological tagging": 85530,
      "important challenging task": 52117,
      "challenging task especially": 16318,
      "especially low resource": 38469,
      "low resource scenarios": 65393,
      "simple powerful technique": 109494,
      "high low resource": 50087,
      "neural networks sentence": 78570,
      "sentence pair modeling": 105966,
      "critical nlp tasks": 24821,
      "nlp tasks paraphrase": 79748,
      "tasks paraphrase identification": 119368,
      "paraphrase identification semantic": 84824,
      "identification semantic textual": 51437,
      "stateoftheart neural models": 112782,
      "neural models tasks": 78200,
      "rely pretrained word": 98731,
      "tasks paper study": 119365,
      "representations sentence pair": 99878,
      "single sentence input": 109794,
      "language modeling machine": 59441,
      "modeling machine translation": 72471,
      "social media datasets": 110371,
      "generators natural language": 47788,
      "challenging task paper": 16329,
      "learning phrase embeddings": 62889,
      "phrase embeddings using": 87347,
      "rotten tomatoes dataset": 103287,
      "online reviews important": 81801,
      "analysis lowresource languages": 5685,
      "machine translation bilingual": 65914,
      "large amounts parallel": 61019,
      "amounts parallel data": 5354,
      "information source target": 54994,
      "word embeddings language": 133141,
      "machine translation analysis": 65893,
      "embedding space provides": 35499,
      "resourcepoor target language": 100920,
      "annotated data language": 6160,
      "bootstrapping natural language": 14461,
      "approaches improve results": 8188,
      "paper describes participation": 83838,
      "weighted average word": 132338,
      "average word embeddings": 11218,
      "datasets best result": 27335,
      "based lexical features": 11802,
      "model control degree": 70912,
      "proposed model task": 93485,
      "model furthermore investigate": 71212,
      "crosslingual document classification": 24941,
      "covers languages english": 24557,
      "german french spanish": 47894,
      "provide strong baselines": 93928,
      "word sentence embeddings": 133538,
      "paper describes submissions": 83849,
      "adam mickiewicz university": 3168,
      "machine translation improve": 66003,
      "translation improve performance": 124858,
      "translations paper propose": 125484,
      "alleviate problem using": 5046,
      "strong attentionbased nmt": 113623,
      "increase vocabulary size": 53623,
      "words training data": 134277,
      "data set furthermore": 26408,
      "word order source": 133381,
      "order source target": 82410,
      "method achieves comparable": 68595,
      "translation quality stateoftheart": 125187,
      "training sequencetosequence models": 123844,
      "machine translation speech": 66236,
      "translation speech recognition": 125283,
      "speech recognition tasks": 111783,
      "models built using": 72870,
      "solve wide range": 110628,
      "wide range tasks": 132514,
      "tasks including neural": 119184,
      "including neural machine": 53337,
      "machine translation automatic": 65905,
      "translation automatic speech": 124659,
      "speech recognition speech": 111775,
      "recognition speech synthesis": 97008,
      "various neural network": 131151,
      "representations latent space": 99720,
      "term memory network": 120234,
      "network gated recurrent": 77263,
      "different embedding methods": 31121,
      "embedding methods including": 35442,
      "largescale knowledge graphs": 61452,
      "knowledge graphs provide": 57991,
      "short textual description": 108265,
      "paper introduce dynamic": 83991,
      "discern relevant information": 32012,
      "models paper proposes": 73707,
      "apply proposed method": 7208,
      "machine translation grammatical": 65995,
      "translation grammatical error": 124844,
      "english grammatical error": 37163,
      "trained different initialization": 123117,
      "dependency parsing neural": 29188,
      "model achieves improvements": 70574,
      "structured prediction models": 114028,
      "parser abstract meaning": 84908,
      "achieve stateoftheart accuracy": 2557,
      "outperform strong baselines": 82764,
      "information extraction particular": 54590,
      "named entity classification": 76042,
      "datasets conll2003 ontonotes": 27374,
      "obtained results indicate": 81405,
      "detecting different types": 29831,
      "competitive stateofthe art": 19692,
      "stateofthe art methods": 112561,
      "despite long history": 29708,
      "recognition ner task": 96943,
      "language processing community": 59853,
      "high number oov": 50090,
      "number oov words": 80930,
      "characterbased neural models": 16498,
      "neural sequence labeling": 78670,
      "sequence labeling models": 106974,
      "models perform better": 73721,
      "15 f1 points": 345,
      "2019 shared task": 709,
      "shared task crosslingual": 108034,
      "semantic similarity datasets": 105280,
      "learning approach based": 62342,
      "approach jointly learns": 7655,
      "approach achieves stateoftheart": 7322,
      "stateoftheart performance relation": 112841,
      "representations used predict": 99962,
      "anderson et al": 6096,
      "shared semantic space": 108009,
      "used variety linguistic": 128840,
      "corpora paper propose": 23554,
      "raw corpus experiments": 95851,
      "outperforms existing stateoftheart": 82899,
      "existing stateoftheart models": 40293,
      "multiple languages english": 75592,
      "open domain suggestion": 81900,
      "domain suggestion mining": 33665,
      "task suggestion mining": 118764,
      "annotation study based": 6379,
      "based observations propose": 11910,
      "natural language corpora": 76277,
      "translation present challenge": 125131,
      "present challenge set": 89403,
      "data publicly available": 26306,
      "domain adversarial training": 33466,
      "recognition paper propose": 96958,
      "paper propose domain": 84229,
      "labeled source domain": 58470,
      "source domain data": 110748,
      "unlabeled target domain": 127410,
      "target domain data": 117607,
      "encourage model learn": 36760,
      "character error rate": 16432,
      "compared baseline trained": 19332,
      "document describes findings": 32987,
      "held concert annual": 49682,
      "concert annual conference": 20696,
      "association computational linguistics": 9620,
      "summarize research trends": 115610,
      "research trends papers": 100654,
      "trends papers presented": 125752,
      "papers presented proceedings": 84509,
      "participants tasked creating": 85328,
      "systems accurate efficient": 116710,
      "multilingual neural machine": 75307,
      "machine translation taskspecific": 66264,
      "attention models simple": 9935,
      "models simple effective": 74051,
      "simple effective technique": 109416,
      "quality possible translation": 94758,
      "zeroshot translation directions": 135535,
      "translation directions model": 124756,
      "representation learning text": 99312,
      "resource poor languages": 100870,
      "promising results text": 92303,
      "results text classification": 102270,
      "enhance text classification": 37496,
      "term memory bilstm": 120228,
      "memory bilstm network": 68294,
      "model learns representation": 71450,
      "model classification tasks": 70829,
      "classification tasks sentiment": 17457,
      "tasks sentiment analysis": 119490,
      "sentiment analysis emoji": 106588,
      "experiments proposed models": 41086,
      "achieve new state": 2501,
      "summarization dialogue systems": 115496,
      "systems create new": 116823,
      "baseline recurrent neural": 12293,
      "additional trainable parameters": 3586,
      "number baselines including": 80847,
      "language model sequencetosequence": 59399,
      "taskoriented dialogue dialogue": 118891,
      "dialogue dialogue state": 30666,
      "models taskoriented dialogue": 74162,
      "dialogue history response": 30688,
      "models dialogue state": 73065,
      "multidomain taskoriented dialogue": 75098,
      "taskoriented dialogue dataset": 118890,
      "models automatic human": 72790,
      "automatic human evaluation": 10551,
      "model neural machine": 71568,
      "usually trained large": 130462,
      "bilingual sentence pairs": 14059,
      "translate sentence time": 124536,
      "order handle issue": 82334,
      "handle issue propose": 49384,
      "way proposed model": 132125,
      "proposed model capture": 93438,
      "nist chineseenglish translation": 79497,
      "translation tasks experiments": 125354,
      "tasks experiments demonstrate": 119122,
      "achieves substantial improvements": 2913,
      "substantial improvements baseline": 114860,
      "fundamental building block": 46086,
      "systems propose novel": 117083,
      "evaluation metrics datasets": 39286,
      "2018 shared task": 677,
      "shared task second": 108092,
      "required human effort": 100217,
      "received significant attention": 96376,
      "word segmentation using": 133507,
      "using conditional random": 129471,
      "language processing algorithms": 59834,
      "languages word segmentation": 60972,
      "arabic script cursive": 8548,
      "uses conditional random": 129208,
      "using manually annotated": 129858,
      "corpus publicly available": 23952,
      "semantic parsing recent": 105161,
      "recent research efforts": 96507,
      "experiments proposed approach": 41068,
      "proposed approach able": 93180,
      "userproduct information sentiment": 129085,
      "current models deal": 25302,
      "paper present hybrid": 84106,
      "short long range": 108217,
      "long range dependency": 65083,
      "encoded word vectors": 36488,
      "significantly better compared": 108885,
      "models training data": 74237,
      "detection englishhindi codemixed": 29938,
      "englishhindi codemixed social": 37402,
      "media content corpus": 68093,
      "content corpus baseline": 21865,
      "user generated data": 128988,
      "computational linguistics past": 20398,
      "semantic understanding text": 105343,
      "posting social media": 88499,
      "social media called": 110355,
      "media called codemixing": 68087,
      "codemixed tweets paper": 18191,
      "paper analyze task": 83733,
      "englishhindi codemixed tweets": 37405,
      "corpus provide baseline": 23948,
      "toolkit paper describes": 122417,
      "crf inference layer": 24749,
      "active area research": 3104,
      "linear mixed effects": 64353,
      "mixed effects models": 70396,
      "using corpus able": 129507,
      "evaluation sentence embeddings": 39384,
      "techniques past years": 119947,
      "universal sentence encoders": 127333,
      "wide variety downstream": 132520,
      "downstream tasks work": 34061,
      "perform comprehensive evaluation": 85973,
      "tasks simple approach": 119503,
      "simple approach using": 109364,
      "language model deep": 59311,
      "contextdependent word embeddings": 22357,
      "word embeddings proved": 133194,
      "better results tasks": 13707,
      "representations words distributed": 99992,
      "capture syntactic semantic": 15406,
      "semantic information words": 105081,
      "model consider word": 70886,
      "original test set": 82553,
      "method convolutional neural": 68746,
      "language understanding slot": 60266,
      "understanding slot filling": 126957,
      "slot filling important": 110046,
      "word sentence paper": 133541,
      "combines convolutional neural": 18681,
      "translation propose simple": 125153,
      "propose simple effective": 93054,
      "simple effective way": 109418,
      "domain mismatch training": 33584,
      "test effectiveness proposed": 120449,
      "proposed method results": 93397,
      "fundamental task natural": 46119,
      "development information extraction": 30392,
      "building knowledge graph": 14851,
      "knowledge graph natural": 57971,
      "graph natural language": 48836,
      "rich source knowledge": 102789,
      "used semantic interpretation": 128744,
      "major challenge stateoftheart": 66545,
      "entities multiple sentences": 37824,
      "automatically extract features": 10768,
      "shown great results": 108478,
      "using densely connected": 129553,
      "tagging namedentity recognition": 117423,
      "obtain stateoftheart performance": 81328,
      "tagging pos tagging": 117438,
      "intelligent dialogue systems": 55849,
      "dialogue systems work": 30797,
      "final matching score": 44108,
      "aggregation experimental results": 4359,
      "results model outperforms": 101941,
      "incremental learning framework": 53732,
      "high cost data": 50054,
      "cost data annotation": 24349,
      "framework allows rapid": 45421,
      "maintaining high accuracy": 66518,
      "framework publicly available": 45665,
      "publicly available datasets": 94301,
      "work study problem": 134826,
      "soft attention mechanisms": 110487,
      "quantitative qualitative evaluation": 94878,
      "evaluation analysis model": 39119,
      "contextual language model": 22478,
      "high accuracy variety": 50037,
      "goaloriented conversational agents": 48408,
      "conversational agents natural": 22989,
      "recognizing named entities": 97075,
      "speedup orders magnitude": 111887,
      "qualitative quantitative evaluations": 94574,
      "humanities social sciences": 51043,
      "traditional acoustic features": 122790,
      "models outperform previously": 73680,
      "outperform previously reported": 82731,
      "systems previous work": 117074,
      "paper introduce neural": 83997,
      "joint learning model": 57287,
      "approach significantly improves": 7901,
      "context spoken language": 22279,
      "taskoriented dialogue systems": 118894,
      "dialogue systems spoken": 30790,
      "parsing natural language": 85170,
      "natural language user": 76597,
      "language user utterances": 60307,
      "user utterances semantic": 129054,
      "utterances semantic frames": 130667,
      "dialogue state tracker": 30748,
      "state tracker dst": 112519,
      "track dialogue state": 122728,
      "network rnn based": 77414,
      "experiments demonstrate effectiveness": 40882,
      "shortterm memory blstm": 108321,
      "sequence alignment algorithm": 106916,
      "neural networks trained": 78592,
      "machine translation trained": 66273,
      "standard national varieties": 112275,
      "brazilian european portuguese": 14571,
      "machine translation report": 66199,
      "09 bleu points": 99,
      "human evaluation experiment": 50811,
      "dialogue systems based": 30766,
      "work small set": 134809,
      "understanding nlu module": 126909,
      "dependency parsing syntactic": 29202,
      "parsing syntactic dependency": 85252,
      "parser dozat manning": 84943,
      "dozat manning 2017": 34069,
      "perform comparative analysis": 85964,
      "major contribution work": 66550,
      "rulebased machine translation": 103367,
      "machine translation rulebased": 66205,
      "machine translation data": 65938,
      "success nlp tasks": 115110,
      "inflectional morphological systems": 54294,
      "arabic sentiment analysis": 8552,
      "wide range application": 132497,
      "range application areas": 95540,
      "lstm networks proven": 65653,
      "nlp tasks sentiment": 79767,
      "twitter challenging task": 126097,
      "arabic language sentiment": 8520,
      "sentiment classification tasks": 106715,
      "classification tasks arabic": 17439,
      "sentiment analysis different": 106584,
      "analysis different datasets": 5577,
      "different sentiment classification": 31416,
      "recent works shown": 96599,
      "task sentiment analysis": 118692,
      "training deep learning": 123575,
      "applications neural machine": 6977,
      "language modelling task": 59492,
      "using linear transformations": 129814,
      "analogy task languages": 5425,
      "new intrinsic evaluation": 78963,
      "method crosslingual semantic": 68753,
      "languages different language": 60502,
      "english german spanish": 37156,
      "monolingual semantic spaces": 74611,
      "shared space using": 108014,
      "bilingual semantic space": 14057,
      "learning deep learning": 62487,
      "deep learning techniques": 28320,
      "successful natural language": 115163,
      "systems require large": 117113,
      "require large annotated": 100161,
      "large annotated data": 61031,
      "paper explores use": 83935,
      "domain invariant features": 33552,
      "new target domains": 79194,
      "target domains languages": 117612,
      "monolingual word vectors": 74632,
      "core nlp tasks": 23330,
      "targeted sentiment analysis": 117770,
      "interaction prior work": 55963,
      "prior work nlp": 90757,
      "direction power pairs": 31825,
      "power pairs participants": 88647,
      "pairs participants email": 83605,
      "improvement stateoftheart methods": 52767,
      "amounts unlabeled data": 5365,
      "corpora achieve robustness": 23398,
      "understanding nlu models": 126908,
      "large scale nlu": 61257,
      "results approach achieves": 101514,
      "training data furthermore": 123467,
      "mutual information input": 75977,
      "nmt neural machine": 79928,
      "neural machine translationnmt": 78135,
      "evaluation metrics bleu": 39285,
      "overall translation quality": 83264,
      "issues paper propose": 57065,
      "scores human evaluations": 104178,
      "automatic metric bleu": 10598,
      "expensive human annotation": 40423,
      "model pretrained large": 71777,
      "large parallel corpora": 61200,
      "latent representation source": 61599,
      "based prior knowledge": 11953,
      "evaluation experimental results": 39212,
      "performance quality estimation": 86645,
      "areas sentiment analysis": 8902,
      "neural networks gained": 78485,
      "propose novel architecture": 92901,
      "popular benchmark datasets": 88081,
      "benchmark datasets compared": 12770,
      "datasets compared previous": 27363,
      "existing pretrained word": 40258,
      "word vectors used": 133649,
      "used text classification": 128809,
      "conditional variational autoencoder": 20793,
      "explore novel approach": 41571,
      "empirically compare proposed": 36220,
      "model recent approaches": 71861,
      "achieves significant improvement": 2856,
      "deep generative model": 28235,
      "source target sentences": 110839,
      "neural networks perform": 78535,
      "amortised variational inference": 5325,
      "model machine translation": 71486,
      "synthetic data experiments": 116617,
      "standard neural machine": 112278,
      "speech translation corpus": 111832,
      "corpus paper presents": 23923,
      "paper presents extension": 84171,
      "lowresource parallel corpus": 65553,
      "endangered language griko": 36840,
      "collected cleaned processed": 18407,
      "cleaned processed illustrate": 17709,
      "processed illustrate use": 91599,
      "illustrate use zeroresource": 51747,
      "unsupervised word discovery": 127752,
      "discovery dataset available": 32137,
      "dataset available online": 26755,
      "computational language documentation": 20384,
      "language documentation experiments": 58979,
      "named entities text": 76037,
      "use computational techniques": 127951,
      "computational techniques identify": 20438,
      "nlp applications paper": 79571,
      "based semantic similarity": 12011,
      "corpus proposed method": 23944,
      "proposed method relies": 93396,
      "entity recognition techniques": 38081,
      "recent word embeddings": 96565,
      "word embeddings models": 133163,
      "models semantic similarity": 73991,
      "deep learning natural": 28298,
      "deep learning architectures": 28252,
      "applications computational linguistics": 6906,
      "recommendations future research": 97093,
      "loss neural machine": 65285,
      "distances word embedding": 32443,
      "word embedding space": 133033,
      "loss function encourages": 65271,
      "data sets proposed": 26431,
      "sets proposed method": 107699,
      "interactive neural machine": 56025,
      "active learning techniques": 3117,
      "machine translation main": 66040,
      "adapting neural machine": 3317,
      "model propose novel": 71812,
      "propose novel methods": 92942,
      "mechanism neural machine": 68018,
      "approaches large margin": 8208,
      "semeval 2016 task": 105499,
      "department computer science": 29038,
      "human language technology": 50902,
      "use distributed representations": 128009,
      "multilingual semantic network": 75359,
      "experimental results outperform": 40664,
      "gender bias neural": 46590,
      "processing nlp systems": 91753,
      "biases training data": 13887,
      "quantify gender bias": 94850,
      "neural coreference resolution": 77880,
      "rnnbased language models": 102945,
      "gender bias preserving": 46591,
      "word embedding debiasing": 133009,
      "pretrained embeddings methods": 90028,
      "parallel corpus mining": 84621,
      "sentence embeddings paper": 105842,
      "embeddings paper presents": 35850,
      "novel training method": 80754,
      "data able train": 25552,
      "neural speech translation": 78686,
      "deep learning systems": 28319,
      "source sentences target": 110824,
      "generate complete sentence": 46913,
      "taskspecific training data": 119643,
      "original training data": 82557,
      "training data achieve": 123420,
      "indian languages hindi": 53822,
      "native speakers languages": 76235,
      "following annotation guidelines": 44960,
      "word embedding used": 133040,
      "building nlp systems": 14874,
      "nlp systems including": 79698,
      "conversational agents work": 22990,
      "reduce training time": 97364,
      "time number parameters": 122061,
      "endtoend speech recognition": 36967,
      "speech recognition existing": 111752,
      "systems typically built": 117211,
      "recent progress endtoend": 96496,
      "progress endtoend speech": 92148,
      "single model jointly": 109761,
      "improve recognition accuracy": 52517,
      "accuracy work present": 2320,
      "manner evaluate proposed": 66945,
      "switchboard conversational speech": 116281,
      "conversational speech corpus": 23032,
      "temporal relation classification": 120120,
      "classification models work": 17284,
      "learned word representations": 62288,
      "loss component evaluate": 65260,
      "evaluate approach task": 38794,
      "temporal relation extraction": 120121,
      "relation extraction particular": 98008,
      "relation extraction clinical": 97982,
      "better taskspecific embeddings": 13740,
      "adversarial domain adaptation": 4112,
      "generation dialogue systems": 47369,
      "learning source domain": 63040,
      "sufficient annotated data": 115273,
      "data source domain": 26472,
      "domain limited labeled": 33573,
      "limited labeled data": 64245,
      "model trained source": 72209,
      "trained source domain": 123289,
      "knowledge source domain": 58179,
      "multichannel convolutional neural": 75043,
      "representations word character": 99979,
      "paper presents systematic": 84208,
      "representation experimental results": 99233,
      "generally improves performance": 46862,
      "model achieves competitive": 70565,
      "competitive performance stateoftheart": 19667,
      "analysis pos tagging": 5756,
      "biomedical event extraction": 14183,
      "event extraction biomedical": 39513,
      "information extraction task": 54605,
      "processing tasks partofspeech": 91824,
      "neural models data": 78170,
      "models data available": 73011,
      "neural models outperform": 78187,
      "detailed empirical study": 29770,
      "models pos tagging": 73755,
      "large training dataset": 61301,
      "largescale parallel corpus": 61492,
      "study machine translation": 114433,
      "languages method combines": 60725,
      "best knowledge large": 13350,
      "strong baseline task": 113638,
      "trained language modeling": 123169,
      "sentences binary classification": 106232,
      "sentence encoding models": 105855,
      "language modeling task": 59471,
      "response selection task": 101242,
      "multiple languages key": 75593,
      "lowresource language data": 65501,
      "experiments dataset demonstrate": 40866,
      "dataset demonstrate effectiveness": 26852,
      "demonstrate effectiveness methods": 28714,
      "nmt systems suffer": 79980,
      "fundamental shortcoming sequence": 46115,
      "shortcoming sequence generation": 108279,
      "sequence generation tasks": 106952,
      "generation address issue": 47296,
      "issue propose novel": 57025,
      "propose novel model": 92943,
      "joint training strategy": 57329,
      "construct new dataset": 21643,
      "propose novel endtoend": 92918,
      "dataset experimental results": 26910,
      "experimental results method": 40637,
      "results method significantly": 101926,
      "significantly outperform baselines": 108982,
      "gated graph neural": 46507,
      "graph neural networks": 48846,
      "neural networks knowledge": 78498,
      "knowledge base question": 57784,
      "base question answering": 11476,
      "question answering based": 95024,
      "based semantic parsing": 12010,
      "semantic parsing paper": 105156,
      "address problem learning": 3739,
      "work largely focused": 134604,
      "encode graph structure": 36432,
      "outperform baseline models": 82692,
      "error analysis confirms": 38281,
      "manually annotated dataset": 67028,
      "correlation human judgements": 24231,
      "issues limit performance": 57058,
      "semimarkov conditional random": 105581,
      "demonstrate effectiveness multilingual": 28717,
      "stateoftheart baselines various": 112597,
      "benchmark dataset code": 12753,
      "dataset publicly available": 27134,
      "internal structure words": 56177,
      "stateoftheart results recently": 112918,
      "terms efficiency accuracy": 120317,
      "deep recurrent neural": 28404,
      "language generation models": 59076,
      "sentences experiments translation": 106306,
      "performance generally improved": 86416,
      "experiment results method": 40495,
      "mean average precision": 67596,
      "adversarial neural networks": 4156,
      "data target language": 26542,
      "language adversarial training": 58819,
      "different types adversarial": 31524,
      "tasks dependency parsing": 119048,
      "persian sentiment analysis": 87085,
      "rise social media": 102871,
      "opinions products services": 82128,
      "sentiment analysis automatically": 106572,
      "sentiment positive negative": 106780,
      "deep learning recently": 28307,
      "powerful machine learning": 88669,
      "sentiment analysis limited": 106603,
      "analysis limited work": 5681,
      "learning models deep": 62778,
      "deep convolutional neural": 28224,
      "movie reviews dataset": 74903,
      "proposed deep learning": 93256,
      "shallow multilayer perceptron": 107925,
      "based machine learning": 11822,
      "performance deep learning": 86277,
      "wellstudied problem natural": 132437,
      "statistical language processing": 113101,
      "language processing features": 59864,
      "nlp tasks text": 79773,
      "question answering named": 95062,
      "answering named entity": 6649,
      "field image processing": 43955,
      "processing speech recognition": 91799,
      "tasks semantic role": 119482,
      "role labeling dependency": 103183,
      "labeling dependency parsing": 58496,
      "dependency parsing compare": 29174,
      "model captures information": 70811,
      "new words emerge": 79254,
      "minimum semantic units": 70215,
      "semantic units human": 105346,
      "units human languages": 127246,
      "novel soft loss": 80733,
      "soft loss function": 110498,
      "sememe knowledge base": 105488,
      "knowledge base hownet": 57772,
      "test set outperforms": 120511,
      "subset test set": 114838,
      "important research topic": 52231,
      "research topic natural": 100643,
      "topic natural language": 122555,
      "performance various natural": 86844,
      "sentence embedding models": 105836,
      "sentence embedding method": 105833,
      "close embedding space": 17816,
      "paraphrase identification datasets": 84817,
      "learning graph embeddings": 62623,
      "approach learning graph": 7678,
      "structural measures pairwise": 113781,
      "measures pairwise node": 67885,
      "pairwise node similarities": 83679,
      "model learns representations": 71451,
      "userdefined graph distance": 129065,
      "shortest path distance": 108304,
      "path distance distance": 85670,
      "distance distance measures": 32414,
      "distance measures information": 32425,
      "measures information graph": 67873,
      "information graph structure": 54647,
      "graph structure account": 48870,
      "evaluation proposed model": 39341,
      "model semantic similarity": 71967,
      "semantic similarity word": 105294,
      "tasks using various": 119590,
      "similarity measures approach": 109262,
      "approach yields competitive": 8026,
      "yields competitive results": 135406,
      "model computationally efficient": 70876,
      "translation systems translate": 125317,
      "especially lowresource scenarios": 38474,
      "word embeddings aims": 133047,
      "dimensional vector space": 31755,
      "vector space order": 131378,
      "experimental results shown": 40701,
      "presented results based": 89796,
      "13 bleu points": 283,
      "high quality corpus": 50113,
      "corpus target language": 24033,
      "learning rl finetune": 62975,
      "rewards evaluate approach": 102684,
      "evaluate approach chinese": 38791,
      "demonstrates effectiveness proposed": 28943,
      "translation proposed method": 125156,
      "filling f1 score": 44041,
      "neural models ability": 78159,
      "models ability capture": 72638,
      "interaction lexical syntactic": 55951,
      "current stateoftheart neural": 25362,
      "prediction neural machine": 89088,
      "grammatical errors make": 48701,
      "real grammatical errors": 96066,
      "nlp applications information": 79568,
      "popular word embedding": 88146,
      "extensive experiments conducted": 41900,
      "shows competitive results": 108567,
      "word embedding stability": 133035,
      "paper provide dataset": 84375,
      "thorough manual analysis": 121892,
      "developed annotation scheme": 30252,
      "annotated dataset provided": 6175,
      "news websites world": 79399,
      "previous works focus": 90570,
      "reducing gender bias": 97418,
      "bias abusive language": 13791,
      "abusive language detection": 1845,
      "detection abusive language": 29864,
      "language detection models": 58953,
      "different pretrained word": 31351,
      "effectively reduce gender": 34843,
      "reduce gender bias": 97330,
      "unsupervised way previous": 127748,
      "way previous work": 132121,
      "novel method automatically": 80629,
      "data augmentation textbased": 25654,
      "simple data augmentation": 109394,
      "data augmentation strategy": 25649,
      "words source sentence": 134230,
      "source sentence target": 110815,
      "sentence target sentence": 106101,
      "experiments translation datasets": 41186,
      "datasets different scales": 27431,
      "comparable performances strong": 19165,
      "sennrich et al": 105645,
      "nonnative english speakers": 80242,
      "arabic dialect identification": 8504,
      "test set recently": 120516,
      "obtained stateoftheart results": 81418,
      "stateoftheart results various": 112929,
      "results various text": 102314,
      "tasks arabic dialect": 118936,
      "dialect identification native": 30521,
      "identification native language": 51404,
      "language identification paper": 59136,
      "test set report": 120517,
      "deep probabilistic logic": 28395,
      "wide range nlp": 132510,
      "learning paper propose": 62870,
      "learning deep neural": 62488,
      "demonstrate promise approach": 28827,
      "semantic relations using": 105216,
      "human language paper": 50900,
      "task predicting semantic": 118544,
      "yielding new stateoftheart": 135381,
      "dialogue generation generating": 30674,
      "representations experimental results": 99645,
      "results automatic human": 101531,
      "human evaluations demonstrate": 50847,
      "capable generating responses": 15201,
      "compared baseline models": 19331,
      "models code available": 72916,
      "tasks like speech": 119263,
      "texts paper present": 121569,
      "machine learning task": 65830,
      "svm random forest": 116246,
      "training neural language": 123732,
      "generated text compared": 47138,
      "task language modeling": 118334,
      "high dimensional space": 50062,
      "results wordlevel language": 102344,
      "reducing number parameters": 97431,
      "stateoftheart language model": 112683,
      "number model parameters": 80913,
      "text representation paper": 121249,
      "representation paper present": 99366,
      "paper present method": 84110,
      "representation input sentence": 99266,
      "proposed method capable": 93350,
      "languages previous work": 60812,
      "previous work suggested": 90554,
      "neural dependency parsers": 77888,
      "lead better performance": 61844,
      "parameter sharing strategies": 84723,
      "pairs related languages": 83625,
      "obtains significant improvements": 81476,
      "pos tags dependency": 88242,
      "dependency parsing provide": 29196,
      "provide comprehensive analysis": 93786,
      "parser previous studies": 84988,
      "previous studies shown": 90499,
      "especially morphologically rich": 38480,
      "using noisy channel": 129974,
      "noisy channel model": 80092,
      "model deep neural": 70958,
      "deep neural language": 28350,
      "paper presents model": 84181,
      "model model uses": 71523,
      "language model scores": 59396,
      "using lstm language": 129842,
      "model improves stateoftheart": 71319,
      "neural networks recent": 78548,
      "recent years natural": 96620,
      "years natural language": 135275,
      "representations various tasks": 99971,
      "proposes simple effective": 93618,
      "neural network acnn": 78237,
      "model uses convolutional": 72265,
      "previous best result": 90397,
      "task mapping natural": 118377,
      "natural language commands": 76275,
      "visualisation diagnostic classifiers": 131819,
      "recently enjoyed surge": 96688,
      "word similarity dataset": 133559,
      "release dataset annotation": 98449,
      "universal dependency parsing": 127298,
      "semantic annotation scheme": 104987,
      "multitask learning code": 75828,
      "weighted finite state": 132349,
      "work recurrent neural": 134764,
      "neural models use": 78205,
      "model performs better": 71736,
      "language modeling text": 59474,
      "modeling text classification": 72563,
      "results validate hypothesis": 102307,
      "significant gains stateoftheart": 108759,
      "error reduction benchmark": 38340,
      "reduction benchmark datasets": 97441,
      "multilingual corpus wikipedia": 75227,
      "corpora models trained": 23532,
      "encode different aspects": 36427,
      "translation systems wmt18": 125322,
      "wmt18 news translation": 132845,
      "narayan et al": 76162,
      "models social media": 74064,
      "social media language": 110388,
      "evaluate approach different": 38792,
      "using linguistically motivated": 129820,
      "parallel corpora bilingual": 84596,
      "demonstrate effectiveness approaches": 28709,
      "entity recognition languages": 38021,
      "low resource languages": 65389,
      "experiments machine translation": 40994,
      "machine translation exploiting": 65986,
      "expressed natural language": 41719,
      "existing datasets introduce": 40105,
      "built natural language": 14930,
      "natural language entailment": 76293,
      "based approach allows": 11511,
      "model effectively encodes": 71046,
      "text generation neural": 120994,
      "neural text generation": 78708,
      "lead substantial improvements": 61874,
      "improvements strong baselines": 52922,
      "baselines machine translation": 12427,
      "neural models endtoend": 78174,
      "strategy reinforcement learning": 113538,
      "supervised learning reinforcement": 115778,
      "learning reinforcement learning": 62952,
      "learning domainspecific knowledge": 62524,
      "approach recently proposed": 7843,
      "human evaluation shows": 50834,
      "systems achieve higher": 116715,
      "task success rate": 118757,
      "machine translation propose": 66175,
      "source sentence word": 110819,
      "modern neural models": 74403,
      "machine translation improves": 66006,
      "approach sentiment analysis": 7879,
      "using unsupervised clustering": 130345,
      "respect previous work": 101096,
      "research distant supervision": 100477,
      "neural partofspeech tagger": 78630,
      "lowresource languages model": 65520,
      "resulting new state": 101458,
      "state art access": 112426,
      "transfer natural language": 124162,
      "models namedentity recognition": 73604,
      "lexical items languages": 63775,
      "demonstrate methods achieve": 28785,
      "language modeling approach": 59424,
      "address issues paper": 3703,
      "comparison state art": 19576,
      "answering qa datasets": 6668,
      "used sequencetosequence models": 128755,
      "models attention mechanism": 72783,
      "attention mechanism learn": 9903,
      "hard monotonic attention": 49469,
      "tasks image captioning": 119165,
      "xu et al": 135226,
      "pronouns longstanding challenge": 92358,
      "statistical neural mt": 113144,
      "test suite enables": 120541,
      "enables finegrained analysis": 36387,
      "sheds light difficulties": 108162,
      "shows promising results": 108613,
      "considerable room improvement": 21259,
      "incorporating syntactic information": 53563,
      "improve strong baselines": 52553,
      "coreference resolution achieving": 23361,
      "competitive performance tasks": 19671,
      "language processing greatly": 59867,
      "introduction attention mechanism": 56655,
      "sentiment classification datasets": 106696,
      "classification datasets demonstrate": 17172,
      "embeddings neural machine": 35829,
      "target word embeddings": 117747,
      "word embeddings target": 133219,
      "faster training better": 43195,
      "leverage prior knowledge": 63617,
      "shows effectiveness method": 108573,
      "performance stateoftheart systems": 86749,
      "parsing propose novel": 85202,
      "model semantic parsing": 71965,
      "parsing converts natural": 85089,
      "converts natural language": 23109,
      "natural language utterance": 76601,
      "machine interpretable meaning": 65762,
      "interpretable meaning representations": 56244,
      "previous stateoftheart models": 90478,
      "inference extensive experiments": 54143,
      "extensive experiments standard": 41931,
      "approach able achieve": 7289,
      "able achieve stateoftheart": 1603,
      "wordlevel quality estimation": 133748,
      "information target word": 55030,
      "words partofspeech tags": 134110,
      "partofspeech tags languages": 85567,
      "feedforward recurrent neural": 43854,
      "global context sentence": 48231,
      "dual conditional crossentropy": 34229,
      "conditional crossentropy filtering": 20754,
      "parallel corpora work": 84612,
      "parallel data sentence": 84651,
      "data sentence pair": 26395,
      "translation models trained": 124965,
      "models trained clean": 74194,
      "parallel data achieve": 84631,
      "higher bleu scores": 50168,
      "trained parallel data": 123235,
      "data evaluate method": 25907,
      "shared task parallel": 108081,
      "task parallel corpus": 118513,
      "multilingual information extraction": 75260,
      "information extraction pipeline": 54592,
      "extraction pipeline automatically": 42434,
      "large collections unstructured": 61056,
      "unstructured textual data": 127589,
      "conversational agents chatbots": 22988,
      "inspired recent advances": 55581,
      "recent advances neural": 96412,
      "advances neural machine": 4004,
      "applied various tasks": 7142,
      "new research directions": 79107,
      "information retrieval techniques": 54950,
      "seq2seq model outperforms": 106901,
      "text foreign language": 120960,
      "translation propose novel": 125152,
      "translation nmt different": 125031,
      "results demonstrate model": 101690,
      "demonstrate model significantly": 28798,
      "model significantly outperform": 72025,
      "recent works neural": 96598,
      "explore neural architectures": 41567,
      "exploit source targetside": 41442,
      "bleu manual evaluation": 14291,
      "machine translation noisy": 66128,
      "systems publicly available": 117090,
      "publicly available parallel": 94316,
      "available parallel corpora": 11072,
      "dataset machine translation": 27008,
      "consisting noisy comments": 21458,
      "noisy comments reddit": 80096,
      "professionally sourced translations": 92088,
      "sentences language pair": 106369,
      "data augmentation neural": 25639,
      "improve generalization ability": 52390,
      "generalization ability models": 46763,
      "investigate data augmentation": 56738,
      "data augmentation techniques": 25653,
      "empirical results approach": 36177,
      "network machine reading": 77309,
      "machine reading comprehension": 65850,
      "reading comprehension paper": 96000,
      "comprehension paper introduce": 20209,
      "reading comprehension model": 95990,
      "extends existing approaches": 41826,
      "existing approaches perspectives": 40055,
      "interaction question passage": 55965,
      "combining contributions enhance": 18715,
      "information extracting ability": 54562,
      "finally explore applications": 44186,
      "responses neural conversation": 101286,
      "conversation models tend": 22963,
      "tend generate safe": 120151,
      "address challenge propose": 3656,
      "challenge propose simple": 16089,
      "arora et al": 9020,
      "al 2016 evaluate": 4612,
      "evaluate approach variety": 38796,
      "using automatic metrics": 129369,
      "metrics human judgments": 69970,
      "method outperforms stateoftheart": 69030,
      "translation nmt lowresource": 125037,
      "input meaning representation": 55363,
      "representation model trained": 99331,
      "model word segmentation": 72323,
      "word segmentation morphological": 133496,
      "segmentation morphological tagging": 104604,
      "structured prediction framework": 114026,
      "morphological tagging tasks": 74747,
      "model adopt approaches": 70639,
      "mcdonald et al": 67582,
      "training data use": 123561,
      "plays important role": 87731,
      "understanding slu essential": 126960,
      "slu essential component": 110114,
      "essential component conversational": 38547,
      "component conversational systems": 19973,
      "provide informative cues": 93854,
      "informative cues better": 55135,
      "content history utterances": 21888,
      "recent utterances important": 96561,
      "utterances important recent": 130649,
      "important recent ones": 52224,
      "model automatically learn": 70714,
      "experiments benchmark dialogue": 40801,
      "benchmark dialogue state": 12815,
      "dialogue state tracking": 30750,
      "tracking challenge dstc4": 122750,
      "challenge dstc4 dataset": 16029,
      "learning present novel": 62904,
      "supervised reinforcement learning": 115825,
      "test time experiments": 120545,
      "image captioning datasets": 51765,
      "datasets method achieves": 27565,
      "leads significant improvement": 61954,
      "current dialogue systems": 25277,
      "zhang et al": 135538,
      "introduce new dataset": 56480,
      "achieving stateoftheart results": 2987,
      "conll 2018 shared": 21084,
      "shared task universal": 108101,
      "dependency parsing pipeline": 29193,
      "language closely related": 58881,
      "data augmentation spoken": 25645,
      "augmentation spoken language": 10304,
      "text generative models": 121023,
      "variational autoencoder vae": 130914,
      "sentences paper propose": 106428,
      "propose novel generative": 92924,
      "helps alleviate data": 49810,
      "data scarcity issue": 26373,
      "task partofspeech tagging": 118523,
      "single language english": 109748,
      "tasks languages work": 119244,
      "languages work examine": 60975,
      "require manually annotated": 100178,
      "text present new": 121194,
      "grounded word embedding": 49117,
      "model word embedding": 72319,
      "methods learning word": 69590,
      "word embeddings word2vec": 133254,
      "word cooccurrence information": 132972,
      "grounded word embeddings": 49118,
      "vectors word embeddings": 131488,
      "neural networks robust": 78569,
      "annotation artifacts neural": 6279,
      "neural generation methods": 77919,
      "sentence planning surface": 105986,
      "planning surface realization": 87645,
      "endtoend learning framework": 36912,
      "work shown neural": 134802,
      "content single sentence": 21948,
      "single sentence multiple": 109795,
      "seen training systematically": 104549,
      "large training corpora": 61297,
      "neural models learn": 78180,
      "parseme shared task": 84903,
      "shared task automatic": 108026,
      "task automatic identification": 117913,
      "multiword expression mwe": 75944,
      "employ neural architecture": 36273,
      "participated open track": 85339,
      "open track parseme": 81943,
      "track parseme shared": 122733,
      "grammar induction systems": 48640,
      "al 2018 modern": 4634,
      "johnson et al": 57246,
      "results english chinese": 101754,
      "english chinese german": 37087,
      "models usually trained": 74312,
      "using teacher forcing": 130271,
      "suffers exposure bias": 115257,
      "problem exposure bias": 91048,
      "results nist chinesetoenglish": 101992,
      "nist chinesetoenglish translation": 79500,
      "chinesetoenglish translation tasks": 16874,
      "translation tasks method": 125355,
      "improvement 15 bleu": 52669,
      "neural networks paper": 78529,
      "networks paper describes": 77688,
      "lstmbased neural network": 65686,
      "lemmas partofspeech tags": 63333,
      "extractive reading comprehension": 42559,
      "machine translation despite": 65948,
      "translation despite recent": 124746,
      "reading comprehension rc": 96008,
      "lack largescale datasets": 58730,
      "rc training data": 95877,
      "training data given": 123472,
      "translation nmt model": 125038,
      "data nonenglish languages": 26180,
      "method experimental results": 68824,
      "results datasets method": 101665,
      "achieved recurrent neural": 2679,
      "standard future work": 112244,
      "semantic relations vector": 105217,
      "relations vector space": 98279,
      "vector space capturing": 131370,
      "space capturing semantic": 110978,
      "capturing semantic relations": 15487,
      "words vector space": 134305,
      "pairs paper propose": 83600,
      "cooccurrences word pairs": 23244,
      "models experimental results": 73198,
      "comparable stateoftheart model": 19179,
      "recognizing lexical semantic": 97071,
      "semantic relations recognizing": 105214,
      "important task applications": 52263,
      "solve problem proposed": 110610,
      "learned unsupervised manner": 62276,
      "pairs experimental results": 83540,
      "previous neural approaches": 90436,
      "based hidden markov": 11753,
      "markov models hmms": 67267,
      "based expectation maximization": 11697,
      "translation work investigates": 125433,
      "multihead attention models": 75127,
      "models based transformer": 72817,
      "based transformer architecture": 12128,
      "transformer architecture demonstrate": 124292,
      "using proposed approach": 130060,
      "proposed approach achieve": 93182,
      "sentence vector representations": 106124,
      "current neural sentence": 25315,
      "models present method": 73772,
      "address challenge developing": 3651,
      "datasets used experiments": 27777,
      "available paper propose": 11070,
      "novel model called": 80645,
      "finegrained semantic meanings": 44378,
      "help model learn": 49753,
      "model learn semantic": 71434,
      "experimental results dataset": 40575,
      "results dataset proposed": 101657,
      "dataset proposed model": 27122,
      "model outperforms competitive": 71640,
      "outperforms competitive baselines": 82868,
      "years deep learning": 135248,
      "learning methods achieved": 62735,
      "methods achieved great": 69286,
      "premise hypothesis paper": 89286,
      "hypothesis paper propose": 51270,
      "word embeddings experimental": 133107,
      "embeddings experimental results": 35709,
      "embeddings significantly improve": 35938,
      "textual entailment models": 121695,
      "data set consists": 26404,
      "information available online": 54389,
      "querybased extractive summarisation": 94999,
      "using data provided": 129532,
      "data training neural": 26574,
      "sentences training data": 106527,
      "training data automatically": 123431,
      "training data allow": 123424,
      "effect training different": 34617,
      "training data second": 123532,
      "data second propose": 26379,
      "second propose method": 104447,
      "traditional generative models": 122822,
      "generative models generate": 47743,
      "recently researchers attempted": 96754,
      "performance paper propose": 86597,
      "propose new framework": 92851,
      "novel generative model": 80583,
      "generation experimental results": 47399,
      "experimental results approaches": 40555,
      "patient health records": 85701,
      "electronic health records": 35252,
      "health records ehrs": 49628,
      "named entity extraction": 76045,
      "training neural networks": 123742,
      "task particularly challenging": 118521,
      "manually engineered features": 67059,
      "provide strong inductive": 93930,
      "network make use": 77312,
      "representations used improve": 99961,
      "evaluation gold standard": 39228,
      "attention recent years": 9986,
      "word order using": 133385,
      "transfer multitask learning": 124160,
      "indepth error analysis": 53798,
      "neural classification model": 77860,
      "accuracy neural classifier": 2215,
      "opendomain conversational agents": 81961,
      "variational autoencoder model": 130913,
      "neural models terms": 78201,
      "models terms bleu": 74177,
      "task paper presents": 118507,
      "paper presents nicts": 84187,
      "presents nicts participation": 89877,
      "nicts participation wmt18": 79479,
      "participation wmt18 shared": 85370,
      "translation nmt using": 125063,
      "wmt18 shared news": 132848,
      "score sentence pairs": 104121,
      "10 million words": 140,
      "systems empirical results": 116863,
      "nmt systems trained": 79981,
      "data achieve promising": 25561,
      "achieve promising performance": 2516,
      "paper propose latent": 84256,
      "information source sentence": 54992,
      "topic distribution words": 122510,
      "contribute better understanding": 22747,
      "entity type classification": 38116,
      "specifically propose hierarchical": 111582,
      "results model achieves": 101934,
      "information extraction existing": 54570,
      "number training examples": 80986,
      "languages large number": 60676,
      "present comprehensive evaluation": 89413,
      "building conversation systems": 14830,
      "sequencetosequence generation task": 107161,
      "task given sequence": 118245,
      "topic facilitate development": 122517,
      "create new dataset": 24632,
      "new dataset containing": 78865,
      "establish baseline results": 38579,
      "baseline results dataset": 12301,
      "using different models": 129575,
      "information background knowledge": 54392,
      "background knowledge required": 11309,
      "models case studies": 72890,
      "large amounts unlabeled": 61024,
      "amounts unlabeled text": 5366,
      "learning algorithm improves": 62319,
      "labeled unlabeled data": 58476,
      "unlabeled data labeled": 127387,
      "standard supervised learning": 112314,
      "supervised learning used": 115783,
      "word embeddings shown": 133208,
      "introduce convolutional neural": 56398,
      "neural network structure": 78399,
      "parallel corpora model": 84605,
      "utility natural language": 130486,
      "capturing longrange dependencies": 15479,
      "longrange dependencies paper": 65180,
      "framework jointly extract": 45589,
      "event triggers arguments": 39554,
      "graph convolution networks": 48778,
      "experiment results demonstrate": 40494,
      "proposed framework achieves": 93283,
      "framework achieves competitive": 45409,
      "achieves competitive results": 2775,
      "competitive results compared": 19681,
      "results compared stateoftheart": 101606,
      "reduce vocabulary size": 97367,
      "machine translation dataset": 65940,
      "hate speech detection": 49533,
      "detection twitter present": 30089,
      "online hate speech": 81777,
      "occurrence hate speech": 81511,
      "hate speech commonly": 49532,
      "outperform state art": 82748,
      "using significantly fewer": 130171,
      "significantly fewer parameters": 108914,
      "compared previous methods": 19414,
      "machine learning major": 65797,
      "scale data available": 103710,
      "spanning multiple domains": 111190,
      "order magnitude larger": 82361,
      "magnitude larger previous": 66386,
      "data collection procedure": 25744,
      "data sets baseline": 26418,
      "novel natural language": 80662,
      "language generation task": 59099,
      "generated recurrent neural": 47112,
      "dataset experiment results": 26908,
      "method outperforms baselines": 69010,
      "opportunities future research": 82143,
      "propose simple robust": 93068,
      "sentence representations inspired": 106038,
      "word surrounding context": 133598,
      "sentence model semantic": 105952,
      "nlp tasks model": 79739,
      "tasks model shows": 119302,
      "model shows superior": 72015,
      "shows superior performance": 108636,
      "superior performance compared": 115686,
      "large amounts labelled": 61015,
      "amounts labelled data": 5347,
      "languages sentence discourse": 60861,
      "global phenomenon multilingual": 48261,
      "phenomenon multilingual communities": 87254,
      "codeswitching automatic speech": 18227,
      "present efforts building": 89464,
      "language identification language": 59133,
      "language understanding spoken": 60275,
      "understanding spoken language": 126972,
      "understanding slu systems": 126965,
      "systems widely used": 117237,
      "acoustic model language": 3015,
      "model language model": 71408,
      "error rate cer": 38326,
      "significantly outperforms traditional": 109041,
      "e2e nlg challenge": 34362,
      "dialogue systems recent": 30785,
      "shared task aims": 108024,
      "winograd schema challenge": 132728,
      "schema challenge wsc": 103895,
      "reasoning task requires": 96315,
      "choice plausible alternatives": 16897,
      "neural transitionbased model": 78719,
      "sentence length based": 105922,
      "recognition work propose": 97047,
      "present theoretical analysis": 89742,
      "stateoftheart performance benchmark": 112809,
      "language understanding current": 60213,
      "models limited ability": 73508,
      "network capable representing": 77181,
      "applied proposed model": 7111,
      "text sequence labeling": 121287,
      "english chinese corpora": 37084,
      "word embeddings universal": 133231,
      "universal dependencies english": 127285,
      "patterns natural language": 85762,
      "natural language description": 76280,
      "datatotext natural language": 27816,
      "parallel data relies": 84649,
      "offtheshelf nlp tools": 81627,
      "language understanding traditional": 60290,
      "task machine comprehension": 118365,
      "outperforms baseline methods": 82844,
      "exact match accuracy": 39719,
      "accuracy f1 score": 2161,
      "task multitask learning": 118426,
      "paper present multitask": 84114,
      "dataset model outperforms": 27032,
      "outperforms prior neural": 82964,
      "10 percentage points": 145,
      "sequential neural networks": 107242,
      "processing nlp research": 91752,
      "processing morphologically rich": 91717,
      "dependencies improve performance": 29089,
      "performance neural network": 86570,
      "models empirical results": 73138,
      "information retrieval natural": 54941,
      "retrieval natural language": 102414,
      "language generation paper": 59089,
      "generation paper propose": 47530,
      "user generated text": 128989,
      "method outperforms previous": 69017,
      "pronouns frequently omitted": 92353,
      "frequently omitted prodrop": 45875,
      "omitted prodrop languages": 81656,
      "prodrop languages chinese": 91867,
      "languages chinese generally": 60441,
      "chinese generally leading": 16769,
      "generally leading significant": 46867,
      "leading significant challenges": 61905,
      "significant challenges respect": 108735,
      "challenges respect production": 16206,
      "respect production complete": 101098,
      "production complete translations": 92052,
      "wang et al": 132008,
      "al 2018 proposed": 4637,
      "novel reconstructionbased approach": 80700,
      "reconstructionbased approach alleviating": 97126,
      "dropped pronoun dp": 34182,
      "dp translation problems": 34075,
      "problems neural machine": 91349,
      "encoder decoder representations": 36512,
      "significantly improves translation": 108958,
      "machine translation universal": 66282,
      "propose new architecture": 92835,
      "new architecture based": 78802,
      "introducing interlingual loss": 56637,
      "interlingual loss additional": 56125,
      "loss additional training": 65253,
      "additional training objective": 3590,
      "training objective adding": 123754,
      "objective adding forcing": 81062,
      "adding forcing interlingual": 3379,
      "forcing interlingual loss": 45020,
      "interlingual loss able": 56122,
      "loss able train": 65247,
      "able train multiple": 1697,
      "train multiple encoders": 122970,
      "multiple encoders decoders": 75550,
      "encoders decoders language": 36640,
      "decoders language sharing": 28066,
      "language sharing common": 60089,
      "similar input sentences": 109102,
      "sentence different languages": 105823,
      "architecture capable learning": 8634,
      "learning universal language": 63136,
      "universal morphology universal": 127315,
      "universal morphology unimorph": 127314,
      "universal dependencies v2": 127296,
      "present critical evaluation": 89425,
      "learning text representations": 63104,
      "representation learning methods": 99297,
      "large performance gains": 61207,
      "performance gains various": 86405,
      "manual feature engineering": 66999,
      "classification problem image": 17345,
      "problem image classification": 91077,
      "image classification problem": 51776,
      "features extracted automatically": 43506,
      "vector representations experimental": 131352,
      "social media corpus": 110365,
      "super characters method": 115669,
      "different languages including": 31202,
      "chinese japanese korean": 16777,
      "previous work neural": 90537,
      "classification unsupervised topic": 17479,
      "evaluation text simplification": 39424,
      "machine translation large": 66025,
      "parallel corpora automatically": 84594,
      "single translation model": 109806,
      "new annotated corpus": 78773,
      "annotated corpus based": 6147,
      "corpus nus sms": 23908,
      "nus sms corpus": 81034,
      "paper focuses study": 83954,
      "previous work propose": 90541,
      "work propose use": 134748,
      "propose use novel": 93142,
      "empirical results model": 36187,
      "results model able": 101933,
      "significantly better results": 108889,
      "wide range applications": 132499,
      "paper proposes framework": 84355,
      "entity recognition work": 38085,
      "used train evaluate": 128819,
      "entity recognition models": 38023,
      "glove word embeddings": 48305,
      "measuring semantic similarity": 67930,
      "nlp applications despite": 79567,
      "existing approaches leverage": 40053,
      "approaches leverage machine": 8215,
      "leverage machine translation": 63601,
      "machine translation techniques": 66265,
      "applications spoken language": 7020,
      "target translation language": 117735,
      "demonstrate superiority method": 28881,
      "task significant improvement": 118714,
      "word representations learned": 133458,
      "word embeddings experiments": 133109,
      "adversarial squad datasets": 4166,
      "bilingual contextual word": 14024,
      "contextual word similarity": 22523,
      "word similarity bcws": 133555,
      "model recently proposed": 71864,
      "extensive empirical analysis": 41869,
      "standard datasets demonstrate": 112222,
      "corpus language modeling": 23854,
      "scale language modeling": 103730,
      "language modeling lm": 59440,
      "plays key role": 87737,
      "applications speech recognition": 7018,
      "speech recognition machine": 111759,
      "publicly available large": 94309,
      "available large datasets": 11028,
      "training time compared": 123923,
      "information neural machine": 54800,
      "translation previous studies": 125139,
      "information improve translation": 54671,
      "pays special attention": 85819,
      "model trained dataset": 72186,
      "annotation experimental results": 6309,
      "experimental results various": 40719,
      "language pairs neural": 59768,
      "local context words": 64911,
      "consider global context": 21202,
      "model achieved good": 70547,
      "task oriented dialogs": 118486,
      "task oriented dialog": 118484,
      "oriented dialog systems": 82492,
      "incorporate external knowledge": 53471,
      "current stateoftheart models": 25360,
      "present design implementation": 89442,
      "word order change": 133375,
      "large manually annotated": 61137,
      "manually annotated datasets": 67029,
      "created new dataset": 24678,
      "previously released datasets": 90618,
      "prior work particular": 90759,
      "multimodal dialogue generation": 75426,
      "dialogue generation work": 30685,
      "based recently released": 11976,
      "strong baselines terms": 113656,
      "textbased similarity metrics": 121429,
      "language models performing": 59619,
      "dialogue systems conversational": 30773,
      "systems conversational agents": 116817,
      "demonstrate deep learning": 28698,
      "generating relevant responses": 47254,
      "real world users": 96095,
      "multitask deep learning": 75815,
      "learning approaches evaluate": 62358,
      "approaches evaluate performance": 8148,
      "words sequence labeling": 134205,
      "neural models recently": 78195,
      "proven useful nlp": 93736,
      "nlp tasks gap": 79723,
      "models learning representations": 73493,
      "learning representations words": 62962,
      "word sentence representations": 133542,
      "words composing characters": 133872,
      "supervised training corpora": 115847,
      "model sequence labeling": 71986,
      "including named entity": 53330,
      "entity recognition partofspeech": 38057,
      "recognition partofspeech tagging": 96961,
      "indepth analysis shows": 53795,
      "obtains new stateoftheart": 81469,
      "external knowledge resources": 42018,
      "resources machine translation": 101009,
      "approach based deep": 7383,
      "neural networks state": 78583,
      "networks state art": 77779,
      "pairs english french": 83527,
      "english chinese english": 37086,
      "neural models achieve": 78161,
      "different neural machine": 31292,
      "models lowresource language": 73533,
      "model local dependencies": 71470,
      "valuable source information": 130771,
      "source information improve": 110766,
      "evaluation different tasks": 39185,
      "textual social media": 121746,
      "consistently outperforms stateoftheart": 21434,
      "stateoftheart sequence tagging": 112951,
      "sequence tagging model": 107079,
      "appropriate word segmentation": 8445,
      "selfattention network san": 104894,
      "recently attracted increasing": 96662,
      "multiheaded attention mechanism": 75134,
      "information different representation": 54490,
      "different representation subspaces": 31381,
      "representation subspaces different": 99426,
      "subspaces different positions": 114847,
      "al 2017 work": 4623,
      "propose novel convolutional": 92906,
      "multiple attention heads": 75498,
      "experimental results wmt14": 40724,
      "results wmt14 englishtogerman": 102339,
      "wmt14 englishtogerman translation": 132824,
      "translation task demonstrate": 125331,
      "task demonstrate proposed": 118062,
      "approach outperforms strong": 7779,
      "outperforms strong transformer": 83029,
      "strong transformer baseline": 113725,
      "comparing previous work": 19514,
      "previous work model": 90536,
      "transfer learning multilingual": 124109,
      "conversational ai systems": 22992,
      "user intents corresponding": 129002,
      "learning models task": 62794,
      "make use existing": 66743,
      "models lowresource languages": 73534,
      "lowresource languages development": 65512,
      "data paper present": 26213,
      "new data set": 78855,
      "translating training data": 124611,
      "method using multilingual": 69211,
      "machine translation encoder": 65968,
      "contextual word representations": 22522,
      "examples target language": 39886,
      "training data lowresource": 123487,
      "multilingual contextual word": 75221,
      "better results using": 13709,
      "static embeddings compare": 113064,
      "amounts target language": 5358,
      "target language data": 117635,
      "automatically identify relevant": 10785,
      "identify relevant information": 51546,
      "account global context": 2026,
      "publicly available state": 94324,
      "available state art": 11112,
      "state art natural": 112454,
      "art natural language": 9066,
      "linear programming model": 64362,
      "web search results": 132255,
      "perceived quality generated": 85914,
      "parallel corpora using": 84610,
      "using crosslingual word": 129521,
      "alternative approach based": 5223,
      "parallel corpora experiments": 84600,
      "generation propose new": 47571,
      "new task named": 79202,
      "dialogue summarization tasks": 30761,
      "summarization tasks meeting": 115579,
      "results using new": 102304,
      "word embeddings propose": 133190,
      "personality traits text": 87134,
      "text data different": 120851,
      "data different languages": 25843,
      "similar semantic meaning": 109142,
      "target language english": 117636,
      "language english words": 59006,
      "vector space using": 131388,
      "languages english lowresource": 60544,
      "english lowresource languages": 37202,
      "achieve average fscore": 2417,
      "relatively good performance": 98391,
      "better classification results": 13537,
      "natural language based": 76271,
      "based local global": 11810,
      "solve task propose": 110624,
      "contrast existing methods": 22693,
      "experimental results existing": 40616,
      "longform text generation": 65171,
      "challenging task previous": 16331,
      "task previous works": 118560,
      "new neural language": 79052,
      "language models recent": 59645,
      "models recent years": 73880,
      "recent years witnessed": 96643,
      "various language modeling": 131114,
      "language modeling techniques": 59473,
      "dialogue conducted based": 30647,
      "hu et al": 50708,
      "al 2014 work": 4609,
      "neural architectures using": 77841,
      "image text representations": 51799,
      "guo et al": 49283,
      "et al 2019": 38728,
      "al 2019 dataset": 4641,
      "metrics human evaluations": 69968,
      "neural word sense": 78731,
      "sense annotated corpora": 105652,
      "taking advantage knowledge": 117542,
      "order reduce number": 82397,
      "reduce number different": 97344,
      "number different sense": 80871,
      "different sense tags": 31410,
      "disambiguate words lexical": 31943,
      "words lexical database": 134034,
      "leads state art": 61962,
      "wsd evaluation tasks": 135171,
      "evaluation tasks improving": 39417,
      "training data addition": 123422,
      "results significantly outperform": 102185,
      "significantly outperform state": 108985,
      "wide variety topics": 132530,
      "variety topics including": 131027,
      "used train systems": 128821,
      "real conversations collected": 96060,
      "2017 alexa prize": 627,
      "user utterances responses": 129053,
      "consistently outperforms strong": 21435,
      "make data collection": 66647,
      "short social media": 108237,
      "tokens extensive experiments": 122309,
      "extensive experiments datasets": 41903,
      "models achieve better": 72659,
      "map natural language": 67111,
      "natural language instructions": 76360,
      "training time search": 123927,
      "using beam search": 129385,
      "known training time": 58329,
      "dramatically improves performance": 34096,
      "compared standard beam": 19449,
      "standard beam search": 112201,
      "crosslingual transfer natural": 25035,
      "work languages english": 134601,
      "paper present general": 84103,
      "form weak supervision": 45147,
      "existing neural architectures": 40226,
      "collect release large": 18391,
      "release large dataset": 98455,
      "evaluated automatic metrics": 38953,
      "important research direction": 52229,
      "language model fusion": 59327,
      "using external language": 129661,
      "external language model": 42024,
      "context target language": 22292,
      "simple transfer learning": 109535,
      "filling paper describes": 44050,
      "propose use neural": 93141,
      "networks combined traditional": 77539,
      "traditional evaluation methods": 122816,
      "effective feature representation": 34675,
      "deep learningbased nlp": 28329,
      "sentence understanding tasks": 106115,
      "feature decay algorithms": 43259,
      "data selection techniques": 26388,
      "increase performance model": 53610,
      "use training data": 128338,
      "training data possible": 123516,
      "contain parallel sentences": 21752,
      "data selection approach": 26384,
      "decay algorithms fda": 27912,
      "test set source": 120519,
      "obtain statistically significant": 81331,
      "bleu points strong": 14313,
      "perform large vocabulary": 86022,
      "explore potential use": 41576,
      "speech commands dataset": 111658,
      "commands dataset shows": 18794,
      "new model implicit": 79030,
      "shows good performance": 108582,
      "make good use": 66679,
      "good use relevant": 48509,
      "use relevant knowledge": 128243,
      "able generate responses": 1648,
      "longterm dependencies work": 65201,
      "information training data": 55053,
      "training data propose": 123517,
      "data propose novel": 26291,
      "propose novel reinforcement": 92959,
      "experimental results natural": 40658,
      "better performance existing": 13657,
      "performance existing methods": 86351,
      "experiment proposed model": 40487,
      "proposed model outperformed": 93456,
      "lowresource natural language": 65541,
      "recent deep learning": 96447,
      "training procedure experiments": 123789,
      "experiments showed proposed": 41135,
      "outperform previous models": 82726,
      "training data scarce": 123531,
      "generative adversarial learning": 47708,
      "paper propose multimodal": 84268,
      "generative adversarial network": 47711,
      "generation reinforcement learning": 47591,
      "reinforcement learning process": 97820,
      "dialog systems consider": 30596,
      "model captures user": 70812,
      "qualitative performance improvements": 94569,
      "improvements stateoftheart methods": 52916,
      "fake news detection": 43015,
      "sources including social": 110895,
      "including social media": 53376,
      "social media feeds": 110380,
      "media feeds news": 68114,
      "feeds news blogs": 43858,
      "news blogs online": 79311,
      "blogs online newspapers": 14375,
      "various deep learning": 131076,
      "detecting fake news": 29835,
      "develop models based": 30215,
      "cnn bidirectional long": 17992,
      "outperforms current state": 82875,
      "model opinion target": 71613,
      "opinion target extraction": 82104,
      "targetbased sentiment analysis": 117758,
      "sentiment analysis endtoend": 106592,
      "novel unified model": 80762,
      "stacked recurrent neural": 112130,
      "produce final output": 91890,
      "mechanism models relation": 68015,
      "benchmark datasets framework": 12785,
      "datasets framework achieves": 27490,
      "useful downstream applications": 128881,
      "domains labeled data": 33797,
      "labeled data available": 58431,
      "unsupervised domain adaptation": 127627,
      "previous work bridging": 90522,
      "work bridging anaphora": 134402,
      "anaphora resolution poesio": 6069,
      "resolution poesio et": 100777,
      "poesio et al": 87783,
      "et al 2004": 38685,
      "al 2004 hou": 4594,
      "2004 hou et": 547,
      "et al 2013b": 38698,
      "new word embeddings": 79252,
      "approach achieves competitive": 7304,
      "networks model problem": 77660,
      "results bridging anaphora": 101559,
      "relies recurrent neural": 98657,
      "neural networks order": 78527,
      "significantly outperforming baseline": 108999,
      "outperforming baseline models": 82799,
      "baseline models improvement": 12262,
      "running time information": 103472,
      "closely related words": 17866,
      "words new terms": 134078,
      "word level phrase": 133344,
      "level phrase level": 63489,
      "task paper present": 118506,
      "english low resource": 37200,
      "low resource mt": 65392,
      "challenge neural machine": 16065,
      "translation models usually": 124966,
      "synthetic data improve": 116619,
      "data improve translation": 26027,
      "model built using": 70796,
      "test set model": 120508,
      "hierarchical multitask approach": 49982,
      "used various natural": 128842,
      "nlp downstream applications": 79609,
      "work introduce hierarchical": 134575,
      "model trained multitask": 72202,
      "trained multitask learning": 123210,
      "multitask learning setup": 75868,
      "tasks model trained": 119303,
      "entity recognition entity": 38013,
      "entity mention detection": 37970,
      "lower layers model": 65437,
      "problem twophase task": 91274,
      "twophase task detecting": 126155,
      "pairs followed organizing": 83549,
      "followed organizing pairs": 44950,
      "suffer error propagation": 115219,
      "issue propose new": 57024,
      "new framework named": 78926,
      "demonstrate effectiveness efficiency": 28710,
      "communityspecific sentiment lexicons": 19107,
      "existing knowledge bases": 40148,
      "common sense knowledge": 18922,
      "entities paper propose": 37834,
      "neural multitask learning": 78217,
      "tasks multitask learning": 119318,
      "artificial intelligence research": 9256,
      "building dialogue systems": 14837,
      "dialogue systems shed": 30788,
      "systems shed light": 117140,
      "improve quality generated": 52508,
      "quality generated responses": 94674,
      "approach adopted paper": 7340,
      "dataset demonstrate model": 26856,
      "competitive baselines terms": 19643,
      "wide variety applications": 132519,
      "linguistic features based": 64481,
      "average word vectors": 11219,
      "information logistic regression": 54736,
      "diversitypromoting objective function": 32903,
      "objective function neural": 81083,
      "neural dialogue generation": 77893,
      "dialogue systems widely": 30796,
      "likelihood estimation mle": 64119,
      "context paper propose": 22210,
      "function encourages model": 46013,
      "encourages model generate": 36771,
      "neural networks sequence": 78572,
      "neural networks gnns": 78490,
      "attracted attention paper": 10143,
      "results nlp tasks": 101995,
      "providing better interpretability": 94105,
      "lexical resources improve": 63801,
      "tagging natural language": 117425,
      "deep learning revolution": 28312,
      "working lowresource languages": 134900,
      "additional linguistic information": 3544,
      "recent neural approaches": 96476,
      "combining best worlds": 18709,
      "translation mt tasks": 124987,
      "lowresource language lrl": 65502,
      "language pairs parallel": 59773,
      "pairs parallel corpora": 83603,
      "parallel corpora exist": 84599,
      "seq2seq models perform": 106903,
      "limited parallel corpora": 64261,
      "work character level": 134411,
      "character level nmt": 16447,
      "improves bleu score": 52954,
      "hierarchical neural network": 49988,
      "learning recent years": 62945,
      "recent years sequencetosequence": 96636,
      "networks attention mechanism": 77513,
      "sentences paper present": 106427,
      "paper present hierarchical": 84105,
      "hierarchical deep neural": 49952,
      "sequencetosequence neural networks": 107192,
      "based sequencetosequence neural": 12032,
      "achieve superior results": 2582,
      "retrievalbased dialogue systems": 102442,
      "article describes new": 9122,
      "exploit hierarchical information": 41418,
      "verb argument structure": 131517,
      "artificial neural networks": 9263,
      "sentences labeled grammatical": 106366,
      "verbal argument structure": 131531,
      "using sentence embedding": 130147,
      "information present word": 54861,
      "present word embeddings": 89766,
      "reference resolution dialogue": 97529,
      "slotfilling paradigm user": 110078,
      "paradigm user refer": 84552,
      "user refer slots": 129034,
      "refer slots context": 97500,
      "slots context conversation": 110085,
      "context conversation goal": 22040,
      "conversation goal contextual": 22953,
      "goal contextual understanding": 48343,
      "contextual understanding resolve": 22508,
      "understanding resolve referring": 126944,
      "resolve referring expressions": 100806,
      "referring expressions appropriate": 97576,
      "expressions appropriate slots": 41748,
      "appropriate slots context": 8439,
      "annotated data target": 6169,
      "need annotated data": 76780,
      "little annotated data": 64789,
      "target language paper": 117645,
      "language paper explore": 59789,
      "paper explore different": 83915,
      "models improved performance": 73369,
      "performance recently proposed": 86660,
      "improving robustness neural": 53159,
      "training data access": 123419,
      "paper explore problem": 83924,
      "neural dialog models": 77891,
      "sequential inference models": 107228,
      "response selection paper": 101239,
      "paper presents endtoend": 84168,
      "dialogue technology challenges": 30805,
      "technology challenges dstc7": 120044,
      "enhanced sequential inference": 37520,
      "sequential inference model": 107227,
      "task proposed model": 118581,
      "original esim model": 82521,
      "address challenge outofvocabulary": 3652,
      "challenge outofvocabulary oov": 16074,
      "oov words second": 81872,
      "words second attentive": 134182,
      "second attentive hierarchical": 104390,
      "attentive hierarchical recurrent": 10112,
      "hierarchical recurrent encoder": 49992,
      "recurrent encoder ahre": 97173,
      "released evaluation results": 98482,
      "paper presents description": 84162,
      "evaluation metrics english": 39287,
      "editing natural language": 34548,
      "image editing program": 51782,
      "program using natural": 92118,
      "users natural language": 129146,
      "nlp pipelines paper": 79664,
      "information sources like": 54996,
      "neural ner models": 78228,
      "understanding nlu systems": 126910,
      "systems taskoriented dialog": 117188,
      "dialog systems slot": 30602,
      "systems slot filling": 117147,
      "handcrafted features endtoend": 49341,
      "recent state art": 96516,
      "using word character": 130378,
      "character level embeddings": 16441,
      "learning unsupervised word": 63145,
      "unsupervised word embeddings": 127754,
      "word embeddings popular": 133183,
      "lead suboptimal performance": 61871,
      "performance propose novel": 86627,
      "propose novel learning": 92935,
      "general nlp tasks": 46688,
      "tasks improve performance": 119170,
      "word embeddings better": 133056,
      "conditional text generation": 20790,
      "context neural machine": 22197,
      "parikh et al": 84863,
      "interactive speech recognition": 56031,
      "data new application": 26169,
      "existing training data": 40322,
      "applications paper propose": 6987,
      "improves word error": 53063,
      "error rate 15": 38324,
      "data available application": 25671,
      "random fields crf": 95484,
      "fields crf model": 43993,
      "model partofspeech tagging": 71717,
      "used sequence sequence": 128753,
      "sequence sequence tasks": 107067,
      "quality generated sequences": 94675,
      "sequences paper propose": 107131,
      "target sequence proposed": 117706,
      "experiments neural machine": 41036,
      "tasks text summarization": 119558,
      "text summarization task": 121352,
      "task method outperforms": 118388,
      "method outperforms conventional": 69013,
      "bleu scores wmt14": 14336,
      "wmt14 englishgerman wmt17": 132818,
      "englishgerman wmt17 chineseenglish": 37398,
      "wmt17 chineseenglish translation": 132839,
      "advances natural language": 3999,
      "language processing help": 59868,
      "language documentation process": 58980,
      "based recent advances": 11973,
      "massively multilingual neural": 67343,
      "scenario approach achieves": 103820,
      "achieves high accuracy": 2798,
      "multitask learning improve": 75842,
      "language understanding recently": 60262,
      "language understanding building": 60209,
      "neural networkbased natural": 78428,
      "networkbased natural language": 77490,
      "collect training data": 18394,
      "training data generation": 123471,
      "real natural language": 96072,
      "address challenge using": 3660,
      "language understanding evaluate": 60219,
      "attentionbased encoderdecoder model": 10070,
      "various types speech": 131232,
      "paper propose supervised": 84328,
      "attention mechanism utilizes": 9917,
      "supervised attention attention": 115729,
      "sentiment analysis neural": 106616,
      "analysis neural network": 5718,
      "sentiment analysis seen": 106650,
      "selfattention networks sans": 104896,
      "basic building block": 12514,
      "networks shown effective": 77766,
      "sentiment analysis demonstrate": 106582,
      "french corpus encyclopedic": 45803,
      "sequence labeling model": 106973,
      "model jointly optimizes": 71386,
      "representations recurrent neural": 99844,
      "networks rnns learn": 77745,
      "learn continuous vector": 62010,
      "words test hypothesis": 134263,
      "learning based nlp": 62397,
      "based cosine similarity": 11617,
      "cosine similarity measure": 24341,
      "results obtained proposed": 102008,
      "outperform results obtained": 82739,
      "obtained support vector": 81420,
      "nlp models despite": 79640,
      "processing nlp interpretability": 91737,
      "nlp interpretability remains": 79625,
      "interpretability remains challenge": 56233,
      "endtoend neural models": 36943,
      "nlp tasks propose": 79753,
      "correlation analysis based": 24222,
      "analysis unsupervised method": 5924,
      "present comprehensive analysis": 89412,
      "dalvi et al": 25524,
      "model test time": 72159,
      "propose novel methodology": 92941,
      "dialog technology challenges": 30611,
      "technology challenges dstc6": 120043,
      "approaches based neural": 8080,
      "neural networks identifying": 78491,
      "method achieved best": 68588,
      "achieved best performance": 2598,
      "equal error rate": 38224,
      "dependency parsing propose": 29195,
      "multitask learning model": 75850,
      "learning model joint": 62761,
      "word segmentation partofspeech": 133498,
      "segmentation partofspeech pos": 104611,
      "bist graphbased dependency": 14233,
      "graphbased dependency parser": 48891,
      "dependency parser kiperwasser": 29161,
      "parser kiperwasser goldberg": 84968,
      "kiperwasser goldberg 2016": 57701,
      "vietnamese benchmark datasets": 131670,
      "benchmark datasets experimental": 12782,
      "experimental results joint": 40626,
      "constituency parsing selfattention": 21530,
      "large model sizes": 61152,
      "100 million parameters": 166,
      "separate models language": 106857,
      "multilingual pretraining finetuning": 75343,
      "small number parameters": 110177,
      "32 relative error": 933,
      "languages finally demonstrate": 60587,
      "stateoftheart results 11": 112891,
      "languages including english": 60635,
      "literature review machine": 64772,
      "review machine translation": 102555,
      "past couple years": 85637,
      "machine translation poses": 66159,
      "numerous methods proposed": 81021,
      "statistical spoken dialogue": 113168,
      "dialogue systems usually": 30795,
      "systems usually rely": 117227,
      "propose novel dialogue": 92913,
      "relations multiple entities": 98213,
      "insights strengths weaknesses": 55547,
      "track dialog technology": 122726,
      "stateoftheart models use": 112753,
      "neural networks explicitly": 78480,
      "model interactions different": 71358,
      "multiturn response selection": 75924,
      "outperforms previous models": 82939,
      "previous models including": 90433,
      "models including stateoftheart": 73383,
      "models achieves new": 72702,
      "language model trained": 59409,
      "model trained text": 72214,
      "trained text corpus": 123312,
      "induce distributed representations": 54002,
      "multilingual model used": 75282,
      "learn distributed representations": 62029,
      "languages similar languages": 60875,
      "semantic syntactic similarity": 105320,
      "paper propose sentence": 84318,
      "based semantic parser": 12009,
      "semantic parsing dataset": 105146,
      "experimental results outperforms": 40665,
      "pretrained sentence encoders": 90185,
      "corpus linguistic acceptability": 23868,
      "linguistic acceptability cola": 64419,
      "bert devlin et": 13096,
      "devlin et al": 30469,
      "al 2018 gpt": 4633,
      "gpt radford et": 48555,
      "radford et al": 95434,
      "complex syntactic structures": 19885,
      "dialog technology challenge": 30610,
      "challenge paper introduces": 16078,
      "technology challenges dstc": 120042,
      "building dialog systems": 14835,
      "modeling approaches applied": 72377,
      "endtoend dialog systems": 36888,
      "participants achieved impressive": 85312,
      "using stateoftheart endtoend": 130216,
      "intelligent natural language": 55851,
      "stateoftheart methods achieve": 112704,
      "improvement majorityclass baseline": 52722,
      "establish new baseline": 38585,
      "model achieves success": 70613,
      "improvements machine learning": 52872,
      "require vast amounts": 100206,
      "understanding passenger intents": 126917,
      "amie automatedvehicle multimodal": 5320,
      "automatedvehicle multimodal incabin": 10484,
      "multimodal incabin experience": 75432,
      "scenarios experimental results": 103847,
      "experimental results achieved": 40549,
      "using distributional semantics": 129601,
      "presents new challenges": 89874,
      "significant research topic": 108859,
      "research topic field": 100642,
      "topic field natural": 122520,
      "nlp tasks sentence": 79765,
      "tasks sentence classification": 119488,
      "classification document summarization": 17191,
      "supervised unsupervised learning": 115851,
      "sentence neural networkbased": 105957,
      "embedding models generally": 35455,
      "yielded stateoftheart performance": 135373,
      "stateoftheart performance models": 112830,
      "data study propose": 26514,
      "proposed method practical": 93395,
      "training examples improving": 123623,
      "learning optimal transport": 62862,
      "models commonly trained": 72930,
      "fail capture longrange": 42953,
      "semantic structure present": 105310,
      "based optimal transport": 11915,
      "distribution extensive experiments": 32655,
      "wide variety nlp": 132526,
      "tasks including machine": 119179,
      "machine translation abstractive": 65881,
      "abstractive text summarization": 1824,
      "summarization image captioning": 115517,
      "image captioning crosslingual": 51764,
      "crosslingual language model": 24965,
      "language model pretraining": 59378,
      "model pretraining recent": 71782,
      "recent studies demonstrated": 96522,
      "crosslingual language models": 24967,
      "parallel data new": 84641,
      "language model objective": 59359,
      "results crosslingual classification": 101649,
      "supervised machine translation": 115795,
      "state art absolute": 112425,
      "improving previous state": 53147,
      "obtain new state": 81303,
      "outperforming previous best": 82817,
      "code pretrained models": 18133,
      "pretrained models publicly": 90148,
      "semisupervised text classification": 105629,
      "embeddings propose novel": 35888,
      "text classification method": 120739,
      "word embeddings outperforms": 133175,
      "word embeddings empirically": 133095,
      "text classification datasets": 120717,
      "automated essay scoring": 10438,
      "appropriate given context": 8421,
      "meaning polysemous word": 67661,
      "word given context": 133299,
      "given context using": 48002,
      "endtoend natural language": 36935,
      "paper provides comprehensive": 84384,
      "analysis shared task": 5842,
      "avenues future research": 11168,
      "future research based": 46294,
      "shared task aimed": 108023,
      "novel automatic human": 80504,
      "vanilla seq2seq models": 130822,
      "learning approach neural": 62348,
      "dialogue systems chatbots": 30769,
      "improvements current stateoftheart": 52838,
      "models like memory": 73506,
      "conversational intelligence challenge": 23010,
      "46 absolute improvement": 1041,
      "carry sentiment analysis": 15559,
      "present work track": 89770,
      "models widely used": 74340,
      "follows twostage pipeline": 44994,
      "according official evaluation": 1992,
      "official evaluation results": 81598,
      "generative models language": 47744,
      "experiments largescale machine": 40985,
      "translation language modeling": 124889,
      "wmt14 englishgerman test": 132815,
      "englishgerman test set": 37386,
      "word embeddings training": 133229,
      "shared task introduce": 108055,
      "raw text input": 95859,
      "achieved competitive performance": 2610,
      "3rd official evaluation": 997,
      "dialogue systems attracted": 30763,
      "attracted increasing attention": 10151,
      "li et al": 63954,
      "work propose endtoend": 134718,
      "medical knowledge graph": 68210,
      "topic transition dialogue": 122580,
      "language understanding natural": 60240,
      "hierarchical hidden markov": 49968,
      "supervised classification task": 115737,
      "propose framework based": 92687,
      "experimental results positive": 40668,
      "different input representations": 31184,
      "relatively small amounts": 98415,
      "small amounts training": 110132,
      "amounts training data": 5363,
      "easy data augmentation": 34476,
      "synonym replacement random": 116349,
      "extension transformer architecture": 41841,
      "architecture makes use": 8692,
      "trained bilingual data": 123082,
      "available data sets": 10971,
      "present novel semantic": 89622,
      "novel semantic framework": 80715,
      "semantic framework modeling": 105061,
      "use framework construct": 128059,
      "covering entirety universal": 24540,
      "entirety universal dependencies": 37738,
      "dependencies english web": 29080,
      "english web treebank": 37336,
      "web treebank use": 132267,
      "treebank use dataset": 125650,
      "data code available": 25733,
      "crosslingual transfer nlp": 25037,
      "transfer nlp models": 124167,
      "lowresource target language": 65569,
      "distant languages propose": 32456,
      "strong baselines including": 113650,
      "best individual model": 13341,
      "rely parallel corpora": 98725,
      "processing large amounts": 91692,
      "reading comprehension crc": 95978,
      "dataset propose novel": 27114,
      "language model bert": 59298,
      "performance end task": 86338,
      "approach highly effective": 7606,
      "competitive performance supervised": 19669,
      "lowresource machine translation": 65533,
      "majority language pairs": 66604,
      "little parallel data": 64823,
      "language pairs lack": 59763,
      "freely publicly available": 45791,
      "publicly available benchmarks": 94293,
      "outofdomain parallel data": 82657,
      "data freely available": 25960,
      "report baseline performance": 98979,
      "supervised weakly supervised": 115856,
      "code reproduce experiments": 18150,
      "learned vector representations": 62281,
      "lexical semantics tasks": 63818,
      "downstream tasks performance": 34048,
      "word entity embeddings": 133264,
      "word embeddings variety": 133237,
      "sentiment analysis problems": 106633,
      "art performance benchmark": 9077,
      "sentence level applying": 105926,
      "unstructured natural language": 127576,
      "model machine reading": 71485,
      "reading comprehension answer": 95975,
      "comprehension answer reranking": 20165,
      "recent advances deep": 96397,
      "advances deep neural": 3984,
      "neural networks language": 78499,
      "networks language modeling": 77638,
      "language modeling language": 59437,
      "deep neural models": 28352,
      "dialog systems models": 30599,
      "models able generate": 72643,
      "lot training data": 65332,
      "data build reliable": 25713,
      "traditional approaches based": 122794,
      "approaches based information": 8079,
      "different models using": 31266,
      "train model using": 122959,
      "framework improve performance": 45567,
      "word representations contextual": 133450,
      "translation question answering": 125191,
      "known word embeddings": 58332,
      "dimensionality reduction technique": 31760,
      "language translation systems": 60188,
      "widely used languages": 132571,
      "based sentiment analysis": 12021,
      "model sentiment analysis": 71979,
      "sentiment analysis performance": 106627,
      "network rnn model": 77420,
      "neural network used": 78414,
      "biomedical natural language": 14198,
      "language processing despite": 59860,
      "despite recent advances": 29719,
      "recent advances natural": 96409,
      "language processing statistical": 59946,
      "biomedical clinical text": 14167,
      "publicly available models": 94311,
      "models paper describes": 73697,
      "neural named entity": 78219,
      "recognition ner important": 96931,
      "important task nlp": 52268,
      "explore transfer learning": 41600,
      "baseline neural models": 12273,
      "test f1 score": 120456,
      "text classification knowledge": 120729,
      "classification knowledge powered": 17243,
      "knowledge powered attention": 58112,
      "contextual information poses": 22473,
      "poses great challenge": 88266,
      "knowledge external knowledge": 57925,
      "semantic representation short": 105232,
      "purpose measuring importance": 94433,
      "measuring importance knowledge": 67914,
      "importance knowledge introduce": 52063,
      "knowledge introduce attention": 58024,
      "introduce attention mechanisms": 56383,
      "attention mechanisms propose": 9927,
      "information unlike traditional": 55070,
      "unlike traditional approaches": 127453,
      "experiments public datasets": 41092,
      "public datasets different": 94251,
      "different tasks experimental": 31479,
      "tasks experimental results": 119119,
      "studies model outperforms": 114255,
      "multilingual sentence embedding": 75361,
      "additive margin softmax": 3643,
      "multilingual sentence embeddings": 75362,
      "nmt models achieve": 79906,
      "similar performance models": 109119,
      "performance models trained": 86541,
      "language pairs lastly": 59765,
      "results compared current": 101600,
      "compared current stateoftheart": 19355,
      "new stateoftheart level": 79160,
      "performance large scale": 86489,
      "diverse set languages": 32846,
      "recent advances computational": 96396,
      "representation paper propose": 99367,
      "representation text documents": 99437,
      "embeddings word embedding": 36028,
      "classification tasks compare": 17442,
      "range recent stateoftheart": 95595,
      "recent stateoftheart methods": 96519,
      "approach code available": 7423,
      "neural models cnn": 78168,
      "tasks neural models": 119341,
      "models performed better": 73739,
      "ensemble model combining": 37606,
      "entity discovery linking": 37925,
      "discovery linking edl": 32141,
      "text speech text": 121325,
      "major problem online": 66578,
      "certain linguistic features": 15948,
      "faithfulness original text": 43010,
      "translation systems typically": 125318,
      "target language cases": 117634,
      "translation quality measured": 125178,
      "units phrases sentences": 127265,
      "phrases sentences documents": 87454,
      "new framework models": 78925,
      "models different levels": 73071,
      "framework text classification": 45714,
      "task evaluation results": 118160,
      "statistical methods paper": 113128,
      "methods paper introduce": 69665,
      "text propose novel": 121219,
      "language model using": 59420,
      "contextual word embeddings": 22519,
      "neural networks keyword": 78497,
      "reinforcement learning algorithm": 97794,
      "shows proposed model": 108619,
      "proposed model generate": 93449,
      "multiple languages single": 75597,
      "offline training online": 81616,
      "individual models language": 53920,
      "models language pair": 73454,
      "limitations paper propose": 64184,
      "individual models trained": 53922,
      "multilingual model trained": 75281,
      "knowledge distillation experiments": 57866,
      "translation datasets demonstrate": 124736,
      "handle multiple languages": 49394,
      "learn general representation": 62053,
      "queries paper propose": 94933,
      "way model able": 132109,
      "results datasets proposed": 101666,
      "datasets proposed model": 27641,
      "existing stateoftheart approaches": 40291,
      "graph attention networks": 48761,
      "models proposed text": 73827,
      "text matching problem": 121104,
      "nlp tasks existing": 79717,
      "short texts paraphrase": 108263,
      "paraphrase identification question": 84822,
      "identification question answering": 51428,
      "structural information long": 113773,
      "long text documents": 65138,
      "text documents paper": 120887,
      "graph attention network": 48760,
      "graph convolutional network": 48781,
      "convolutional network gcn": 23164,
      "attention mechanisms applied": 9920,
      "results datasets demonstrate": 101659,
      "outperforms stateoftheart deep": 83003,
      "deep matching models": 28334,
      "static word embeddings": 113068,
      "sequence tagging models": 107080,
      "models constituent parsing": 72971,
      "high error rates": 50071,
      "large label sets": 61122,
      "multitask learning jointly": 75844,
      "jointly learn predict": 57351,
      "parsers english chinese": 85021,
      "english chinese penn": 37090,
      "chinese penn treebanks": 16802,
      "reinforcement learning based": 97800,
      "optimization neural machine": 82199,
      "training data neural": 123500,
      "reinforcement learning learn": 97813,
      "using transfer learning": 130310,
      "parallel corpus machine": 84619,
      "languages english propose": 60546,
      "lexicosyntactic features languages": 63941,
      "demonstrate method outperforms": 28781,
      "relation extraction datasets": 97985,
      "datasets digital humanities": 27434,
      "digital humanities domain": 31724,
      "highquality datasets digital": 50378,
      "evaluation language models": 39249,
      "language models specifically": 59660,
      "fantasy novel book": 43064,
      "novel book series": 80508,
      "word2vec glove fasttext": 133676,
      "suitability word embedding": 115392,
      "relation extraction tasks": 98030,
      "easily extended new": 34450,
      "new datasets tasks": 78872,
      "evaluate approach using": 38795,
      "shows significant improvement": 108625,
      "significant improvement translation": 108779,
      "multilingual word sense": 75404,
      "humanhuman dialogue dataset": 51032,
      "f1 score heldout": 42682,
      "score heldout test": 104077,
      "heldout test set": 49691,
      "best fit specific": 13336,
      "embeddings used downstream": 36004,
      "embedding models task": 35457,
      "task achieve new": 117832,
      "stateoftheart results task": 112921,
      "task benchmark dataset": 117928,
      "nlp tasks source": 79769,
      "tasks source code": 119508,
      "source code data": 110716,
      "code data available": 18077,
      "acoustic patterns using": 3022,
      "neural sequence models": 78672,
      "level individual posts": 63459,
      "nature social media": 76669,
      "method dependency parsing": 68765,
      "takes account problem": 117524,
      "languages model relies": 60730,
      "multiple source languages": 75696,
      "crosslingual word representations": 25046,
      "training target language": 123903,
      "language improves accuracy": 59157,
      "languages achieve average": 60383,
      "syntactic semantic representations": 116476,
      "entity recognition task": 38077,
      "recognition task named": 97023,
      "ner important step": 77049,
      "finally introduce novel": 44201,
      "introduce novel metric": 56515,
      "model shown improve": 72011,
      "comparison different approaches": 19537,
      "model provides significant": 71827,
      "provides significant performance": 94075,
      "variety language understanding": 130985,
      "language understanding problems": 60256,
      "text experiments demonstrate": 120937,
      "introduced bert model": 56573,
      "present detailed analysis": 89444,
      "factors contribute effective": 42885,
      "contribute effective pretraining": 22751,
      "describes machine translation": 29409,
      "generation tasks machine": 47669,
      "open domain question": 81896,
      "domain question answering": 33622,
      "question answering current": 95032,
      "answering current approaches": 6616,
      "approaches natural language": 8247,
      "work explores feasibility": 134522,
      "opendomain question answering": 81977,
      "answering qa based": 6666,
      "multitask learning experiments": 75835,
      "using conversational context": 129495,
      "parsing pointer networks": 85195,
      "sentences left right": 106378,
      "length input sentence": 63361,
      "ma et al": 65735,
      "achieving best accuracy": 2929,
      "best accuracy date": 13296,
      "accuracy date english": 2127,
      "date english ptb": 27826,
      "english ptb dataset": 37254,
      "majority languages tested": 66606,
      "emerging research topic": 36073,
      "select appropriate response": 104691,
      "endtoend dialogue model": 36891,
      "technology challenge dstc7": 120040,
      "characteristics natural language": 16531,
      "large set candidates": 61269,
      "recent success transformer": 96543,
      "transformer model paper": 124343,
      "model paper proposes": 71698,
      "attention mechanism based": 9886,
      "based multihead attention": 11862,
      "model based transformer": 70748,
      "capability proposed model": 15186,
      "downstream task performance": 34031,
      "linguistic information encoded": 64489,
      "probing task performance": 90910,
      "performance diverse set": 86308,
      "postagging dependency parsing": 88460,
      "dependency parsing semantic": 29197,
      "role labeling named": 103187,
      "labeling named entity": 58515,
      "entity recognition natural": 38028,
      "high positive correlation": 50102,
      "correlation downstream tasks": 24228,
      "downstream tasks especially": 34036,
      "lowresource languages neural": 65521,
      "systems achieve high": 116714,
      "case lowresource languages": 15595,
      "languages present simple": 60806,
      "using dependency trees": 129561,
      "universal dependencies project": 127289,
      "partofspeech tagging task": 85562,
      "promising results using": 92305,
      "using handcrafted features": 129724,
      "diverse informative responses": 32817,
      "outofvocabulary words demonstrate": 82683,
      "characterlevel recurrent neural": 16581,
      "virtual keyboard smartphones": 131746,
      "popular social networking": 88130,
      "cross entropy loss": 24861,
      "new candidate terms": 78819,
      "english french german": 37142,
      "product description generation": 92033,
      "methods mainly use": 69611,
      "paper explore new": 83920,
      "power neural networks": 88645,
      "transformer sequence modeling": 124378,
      "experiments realworld datasets": 41107,
      "realworld datasets demonstrate": 96159,
      "datasets demonstrate proposed": 27412,
      "discontinuous constituency parsing": 32028,
      "introduce novel transition": 56520,
      "stack data structure": 112118,
      "parser obtains stateoftheart": 84976,
      "obtains stateoftheart results": 81479,
      "results english german": 101756,
      "english german discontinuous": 37149,
      "described natural language": 29378,
      "commonsense knowledge present": 19000,
      "present novel method": 89613,
      "assess models performance": 9483,
      "establishes new stateoftheart": 38610,
      "new stateoftheart model": 79161,
      "methods perform task": 69672,
      "word experimental results": 133279,
      "sequence modeling toolkit": 107016,
      "researchers developers train": 100686,
      "developers train custom": 30331,
      "train custom models": 122916,
      "translation summarization language": 125302,
      "summarization language modeling": 115524,
      "modeling text generation": 72564,
      "semantics sentence representations": 105465,
      "propose generative model": 92704,
      "determining natural language": 30151,
      "models capture semantic": 72882,
      "text generation using": 121021,
      "original input text": 82528,
      "approaches widely used": 8403,
      "grounded language learning": 49108,
      "methods improve performance": 69541,
      "improve performance strong": 52488,
      "structured meaning representations": 114015,
      "background information use": 11303,
      "new stateoftheart existing": 79158,
      "increase f1 score": 53599,
      "compared existing datasets": 19367,
      "datasets code data": 27349,
      "highquality embeddings rare": 50382,
      "embeddings rare words": 35898,
      "rare words paper": 95750,
      "compared previous work": 19420,
      "previous work attentive": 90520,
      "speech recognition research": 111772,
      "previous studies mainly": 90498,
      "process paper propose": 91545,
      "extensive experiments largescale": 41913,
      "algorithm outperforms stateoftheart": 4778,
      "effectively boost performance": 34791,
      "learning spoken language": 63047,
      "language understanding typically": 60291,
      "understanding slu models": 126963,
      "new language present": 78976,
      "different languages evaluate": 31199,
      "languages evaluate approach": 60557,
      "models outperform stateoftheart": 73685,
      "japanese predicate argument": 57196,
      "predicate argument structure": 88848,
      "argument structure analysis": 8961,
      "build single model": 14809,
      "single model solving": 109763,
      "address problem present": 3740,
      "present multitask learning": 89565,
      "multitask learning method": 75848,
      "performance tasks compared": 86788,
      "models achieved stateoftheart": 72693,
      "overall f1 scores": 83232,
      "employ neural networks": 36274,
      "nary relation extraction": 76202,
      "relations expressed single": 98168,
      "expressed single sentences": 41722,
      "crosssentence relation extraction": 25079,
      "relation extraction combines": 97983,
      "text spans document": 121319,
      "noisy labels distant": 80110,
      "labels distant supervision": 58592,
      "distant supervision experiments": 32470,
      "relation extraction methods": 97998,
      "elmo embeddings peters": 35314,
      "embeddings peters et": 35861,
      "peters et al": 87209,
      "nlp community recent": 79588,
      "performance downstream nlp": 86319,
      "word embeddings like": 133150,
      "et al proposed": 38742,
      "evaluate different methods": 38821,
      "different methods combine": 31257,
      "language model order": 59361,
      "best possible performance": 13410,
      "suggestion mining paper": 115361,
      "paper present neural": 84115,
      "present neural model": 89574,
      "neural model architecture": 78145,
      "submitted semeval2019 task": 114771,
      "suggestion mining online": 115358,
      "mining online reviews": 70250,
      "online reviews forums": 81800,
      "crossdomain suggestion mining": 24905,
      "achieve better results": 2433,
      "performed experiments different": 86921,
      "function best model": 46004,
      "best model achieved": 13368,
      "using sequencetosequence models": 130162,
      "sequencetosequence models present": 107181,
      "models present novel": 73774,
      "text knowledge graph": 121072,
      "comparable stateoftheart systems": 19183,
      "given sentence target": 48117,
      "contextually relevant information": 22589,
      "dataset created automatically": 26837,
      "results large room": 101881,
      "large room improvement": 61240,
      "promising directions future": 92273,
      "directions future research": 31841,
      "improve model robustness": 52418,
      "introduce simple effective": 56537,
      "corpus short texts": 23993,
      "texts using neural": 121646,
      "present novel data": 89602,
      "datasets grammatical error": 27499,
      "supervised learning approaches": 115765,
      "achieve high precision": 2478,
      "information extraction using": 54610,
      "introduce general framework": 56429,
      "iteratively refine span": 57145,
      "refine span representations": 97588,
      "extraction tasks multiple": 42510,
      "tasks multiple datasets": 119315,
      "f1 score improvement": 42686,
      "stringtotree neural machine": 113609,
      "networks paper propose": 77693,
      "paper propose structured": 84326,
      "evaluate performance proposed": 38887,
      "performance proposed model": 86634,
      "dataset model achieves": 27028,
      "unsupervised abstractive sentence": 127595,
      "sentence compression neural": 105799,
      "sequencetosequence models currently": 107178,
      "processing tasks require": 91828,
      "discrete latent variables": 32168,
      "apply proposed model": 7209,
      "input pretrained language": 55399,
      "reinforcement learning proposed": 97822,
      "model does require": 71021,
      "achieving promising results": 2971,
      "results unsupervised sentence": 102295,
      "unsupervised sentence compression": 127718,
      "representations neural language": 99773,
      "models embedding matrix": 73131,
      "language models nlm": 59608,
      "strong generalization capability": 113679,
      "representation rare words": 99393,
      "representations rare words": 99838,
      "performance proposed method": 86631,
      "semantically syntactically similar": 105382,
      "words evaluate proposed": 133943,
      "english speech recognition": 37292,
      "reduces word error": 97404,
      "improves recognition accuracy": 53037,
      "word similarity datasets": 133560,
      "similarity words used": 109344,
      "speech transcription errors": 111828,
      "transcription errors paper": 123998,
      "conversational speech phenomena": 23033,
      "text generation model": 120992,
      "text generation techniques": 121018,
      "given input text": 48045,
      "text training data": 121372,
      "decoder evaluate proposed": 28008,
      "model abstractive text": 70523,
      "model achieves strong": 70608,
      "achieves strong performance": 2910,
      "training data general": 123468,
      "translation training data": 125390,
      "models languages using": 73458,
      "minimum description length": 70206,
      "description length paradigm": 29467,
      "models produce better": 73808,
      "using parallel text": 130005,
      "sentence embeddings observe": 105840,
      "sets parallel data": 107694,
      "news articles using": 79304,
      "hard paper propose": 49476,
      "wikidata knowledge base": 132631,
      "semantic search engine": 105268,
      "complex word identification": 19895,
      "word identification cwi": 133311,
      "task identifying words": 118270,
      "cwi shared task": 25474,
      "shared task released": 108088,
      "task released data": 118629,
      "used neural networks": 128658,
      "neural networks multitask": 78518,
      "networks multitask learning": 77667,
      "multitask learning paper": 75859,
      "learning paper present": 62868,
      "learning models achieve": 62769,
      "strong baselines future": 113647,
      "using word movers": 130382,
      "word movers distance": 133368,
      "different languages different": 31196,
      "semantic document representations": 105036,
      "surface realization shared": 116073,
      "realization shared task": 96116,
      "shared task 2018": 108017,
      "present language model": 89528,
      "model lm based": 71468,
      "availability large datasets": 10919,
      "model trained entire": 72192,
      "entire training set": 37722,
      "generating pun sentence": 47249,
      "pun sentence given": 94390,
      "sentence given pair": 105892,
      "lack large corpus": 58726,
      "content paper propose": 21918,
      "paper propose unsupervised": 84339,
      "propose unsupervised approach": 93134,
      "crossdomain knowledge graphs": 24892,
      "knowledge graphs wikidata": 57994,
      "entities knowledge graphs": 37808,
      "copy mechanism generate": 23289,
      "textual descriptions entities": 121691,
      "evaluation demonstrate method": 39175,
      "demonstrate method significantly": 28782,
      "spoken question answering": 112017,
      "question answering adversarial": 95019,
      "question answering sqa": 95104,
      "recognition asr errors": 96810,
      "domain adaptation task": 33455,
      "model learn domaininvariant": 71429,
      "learn domaininvariant features": 62032,
      "proposed model results": 93478,
      "better previous best": 13682,
      "previous best model": 90393,
      "provide strong empirical": 93929,
      "strong empirical evidence": 113670,
      "state art sota": 112475,
      "paper introduce iterative": 83993,
      "train semantic parsers": 123012,
      "contextdependent semantic parsing": 22355,
      "semantic parsing task": 105171,
      "significantly improve accuracy": 108926,
      "accuracy semantic parsing": 2271,
      "dataset training data": 27243,
      "data augmentation method": 25637,
      "allowing model learn": 5115,
      "extensive experiments method": 41914,
      "experiments method achieves": 40998,
      "method achieves consistent": 68598,
      "achieves consistent improvements": 2779,
      "improvements existing approaches": 52853,
      "novel language representation": 80612,
      "language representation model": 60039,
      "methods achieving new": 69291,
      "stateoftheart results chinese": 112896,
      "chinese natural language": 16793,
      "processing tasks including": 91816,
      "including natural language": 53333,
      "language inference semantic": 59193,
      "entity recognition sentiment": 38069,
      "recognition sentiment analysis": 96995,
      "sentiment analysis question": 106640,
      "analysis question answering": 5793,
      "question answering demonstrate": 95040,
      "data work propose": 26632,
      "approach demonstrates strong": 7479,
      "data language modeling": 26071,
      "prior knowledge just": 90708,
      "compare different types": 19240,
      "nlp models simple": 79644,
      "compositional data augmentation": 20110,
      "propose simple data": 93052,
      "synthetic training examples": 116652,
      "past future neural": 85644,
      "future neural machine": 46284,
      "shown neural machine": 108497,
      "nmt models benefit": 79909,
      "benefit explicitly modeling": 12968,
      "translated untranslated contents": 124577,
      "approach achieves substantial": 7326,
      "languages best knowledge": 60428,
      "parallel bible corpus": 84589,
      "spanish french italian": 111159,
      "test set achieving": 120496,
      "results model effectively": 101939,
      "model effectively predict": 71047,
      "better automatic evaluation": 13514,
      "opendomain dialogue systems": 81969,
      "dialogue systems automatic": 30765,
      "systems automatic evaluation": 116746,
      "referencebased metrics bleu": 97549,
      "share common words": 107950,
      "human judgments paper": 50884,
      "using contextualized word": 129486,
      "contextualized word embeddings": 22575,
      "better evaluation metrics": 13578,
      "articles using neural": 9225,
      "using different languages": 129572,
      "using ngram language": 129968,
      "language model method": 59347,
      "method achieves high": 68601,
      "achieves high performance": 2799,
      "methods best accuracy": 69345,
      "task information extraction": 118297,
      "extraction scientific publications": 42476,
      "neural networks leveraging": 78504,
      "compared stateoftheart baselines": 19455,
      "evaluation neural machine": 39308,
      "deep learning approach": 28247,
      "learning based model": 62391,
      "linguistic features domain": 64482,
      "features domain specific": 43464,
      "domain specific information": 33649,
      "model evaluated using": 71111,
      "significant improvement compared": 108768,
      "specific linguistic features": 111462,
      "sentence encoders language": 105852,
      "encoders language modeling": 36653,
      "pretraining language modeling": 90270,
      "best knowledge effort": 13347,
      "corpus open information": 23917,
      "extraction oie systems": 42417,
      "extract relations arguments": 42106,
      "arguments natural language": 8988,
      "tasks knowledge base": 119235,
      "event schema induction": 39539,
      "paper release analyze": 84398,
      "text english wikipedia": 120912,
      "information confidence scores": 54438,
      "task binary classification": 117939,
      "limited availability resources": 64211,
      "language processing technologies": 59976,
      "sentiment analysis systems": 106665,
      "obtained best results": 81351,
      "comprehensive biomedical knowledge": 20250,
      "showed best performance": 108370,
      "semeval 2019 task": 105505,
      "wmt biomedical translation": 132780,
      "biomedical translation shared": 14212,
      "translation shared task": 125248,
      "shared task paper": 108077,
      "paper describes machine": 83825,
      "translation systems developed": 125310,
      "shared task systems": 108097,
      "translation using moses": 125419,
      "participated translation directions": 85353,
      "create training data": 24648,
      "shared task evaluation": 108046,
      "distributional semantics provides": 32732,
      "method deep neural": 68760,
      "language generation neural": 59080,
      "present paper aim": 89636,
      "information retrieval perspective": 54945,
      "interannotator agreement study": 56047,
      "low agreement scores": 65346,
      "unified language model": 127114,
      "understanding generation paper": 126853,
      "generation paper presents": 47529,
      "language model unilm": 59415,
      "understanding generation tasks": 126854,
      "employing shared transformer": 36321,
      "shared transformer network": 108114,
      "bert glue benchmark": 13135,
      "squad 20 coqa": 112074,
      "stateoftheart results natural": 112912,
      "generative question answering": 47756,
      "absolute improvement squad": 1738,
      "pretrained models available": 90134,
      "generation tasks including": 47668,
      "tasks including language": 119178,
      "existing methods limited": 40188,
      "capture structure output": 15403,
      "model natural language": 71558,
      "previous work focuses": 90530,
      "results demonstrate framework": 101682,
      "competitive baselines achieves": 19640,
      "stateoftheart performance unsupervised": 112852,
      "performance unsupervised approach": 86825,
      "using twodimensional word": 130330,
      "twodimensional word embedding": 126138,
      "models recent work": 73877,
      "recent work super": 96591,
      "work super characters": 134835,
      "stateoftheart results text": 112924,
      "open domain dialogues": 81894,
      "dialogues experimental results": 30832,
      "experimental results public": 40685,
      "dataset shows proposed": 27201,
      "interactive demo ready": 56016,
      "demo ready workshop": 28641,
      "generation nlg critical": 47513,
      "utterances natural language": 130658,
      "tasks graphemetophoneme conversion": 119153,
      "ask following question": 9296,
      "english sentiment analysis": 37275,
      "sentiment analysis sentiment": 106653,
      "analysis sentiment analysis": 5838,
      "sentiment analysis opinion": 106619,
      "analysis opinion mining": 5730,
      "machine learning method": 65798,
      "waikato environment knowledge": 131985,
      "environment knowledge analysis": 38187,
      "knowledge analysis weka": 57751,
      "comments experimental results": 18814,
      "propose novel techniques": 92970,
      "neural language representation": 77957,
      "language representation models": 60042,
      "representation models bert": 99334,
      "pretrained largescale corpora": 90116,
      "consistently improve performance": 21416,
      "improve performance various": 52492,
      "existing pretrained language": 40256,
      "rarely consider incorporating": 95755,
      "incorporating knowledge graphs": 53544,
      "knowledge graphs kgs": 57986,
      "graphs kgs provide": 48963,
      "external knowledge paper": 42015,
      "stateoftheart model bert": 112720,
      "common nlp tasks": 18901,
      "source code paper": 110721,
      "code paper obtained": 18128,
      "language paper introduces": 59790,
      "masked language models": 67296,
      "pretrained bidirectional language": 89998,
      "language models stateoftheart": 59663,
      "stateoftheart performance wide": 112855,
      "reading comprehension natural": 95996,
      "comprehension natural language": 20204,
      "language inference sentiment": 59195,
      "inference sentiment analysis": 54217,
      "monolingual multilingual language": 74599,
      "tasks different languages": 119062,
      "learning multilingual model": 62813,
      "tasks reading comprehension": 119435,
      "reading comprehension paraphrase": 96002,
      "paraphrase detection sentiment": 84812,
      "detection sentiment analysis": 30048,
      "models russian language": 73972,
      "using topic models": 130298,
      "different previous work": 31356,
      "address data sparsity": 3672,
      "data sparsity issue": 26484,
      "propose jointly model": 92734,
      "extensive experimental results": 41884,
      "results largescale datasets": 101886,
      "stateoftheart models based": 112726,
      "highquality annotated data": 50369,
      "trained outofdomain data": 123229,
      "extraction task using": 42507,
      "near humanlevel performance": 76711,
      "accuracy stateoftheart models": 2288,
      "stateoftheart models struggle": 112748,
      "adversarial filtering af": 4124,
      "based speech act": 12064,
      "novel annotated corpus": 80462,
      "approach problem jointly": 7819,
      "training neural network": 123739,
      "previous work proposed": 90542,
      "competitive previous work": 19675,
      "attention mechanisms neural": 9926,
      "translation image video": 124856,
      "revised human agent": 102646,
      "reduce human effort": 97332,
      "implemented following clientserver": 51963,
      "following clientserver architecture": 44966,
      "developed website communicates": 30321,
      "website communicates neural": 132293,
      "task propose supervised": 118577,
      "perform extensive comparison": 85999,
      "representations benchmark datasets": 99517,
      "models outperform previous": 73679,
      "logical forms graph": 65007,
      "structured information entities": 114003,
      "semantic parsing tasks": 105174,
      "tasks present approach": 119392,
      "graph neural network": 48841,
      "neural network gnn": 78303,
      "generate logical forms": 46966,
      "outperforms existing approaches": 82887,
      "text passages written": 121173,
      "entities provide valuable": 37851,
      "short text passages": 108260,
      "better understand aspects": 13755,
      "using semantic similarity": 130141,
      "news social media": 79367,
      "temporal relations existing": 120124,
      "distant supervision method": 32472,
      "results provide insights": 102090,
      "random walk model": 95519,
      "better fit data": 13591,
      "fit data better": 44578,
      "translation nmt shown": 125053,
      "present largescale study": 89534,
      "present methods evaluating": 89554,
      "analysis effects different": 5588,
      "structured prediction problem": 114029,
      "accordingly propose novel": 2017,
      "propose novel hierarchical": 92927,
      "results benchmark dataset": 101540,
      "benchmark dataset demonstrate": 12755,
      "user input produce": 128997,
      "leads suboptimal performance": 61970,
      "translation systems built": 125309,
      "paper conduct systematic": 83779,
      "augmentation neural machine": 10293,
      "translation data augmentation": 124733,
      "accuracy deep learning": 2131,
      "computer vision tasks": 20501,
      "study natural language": 114451,
      "limited paper present": 64259,
      "machine translation different": 65952,
      "multiple semantically similar": 75684,
      "methods experimental results": 69488,
      "machine translation datasets": 65941,
      "superiority method strong": 115703,
      "method strong baselines": 69165,
      "driving force recent": 34163,
      "multiple attention mechanisms": 75499,
      "different parts input": 31322,
      "multilabel text classification": 75159,
      "release new dataset": 98462,
      "substantially larger previous": 114902,
      "employ labelwise attention": 36269,
      "labelwise attention networks": 58661,
      "best overall performance": 13388,
      "used automatically build": 128403,
      "obtains best performance": 81455,
      "networks rnns widely": 77749,
      "rnns widely used": 102985,
      "sentiment analysis news": 106618,
      "benchmark data sets": 12751,
      "sets experimental results": 107670,
      "compared previous models": 19415,
      "model achieves better": 70558,
      "open research problem": 81929,
      "incremental dialog state": 53725,
      "dialog state tracker": 30587,
      "challenge dstc2 dataset": 16027,
      "model achieve better": 70534,
      "example english word": 39781,
      "paper present algorithm": 84077,
      "solve problems propose": 110613,
      "propose new dataset": 92842,
      "experiments method improves": 41003,
      "stateoftheart performance datasets": 112818,
      "available web service": 11143,
      "sense induction word": 105682,
      "induction word sense": 54041,
      "induction wsi task": 54045,
      "language models elmo": 59543,
      "methods propose method": 69692,
      "achieves human level": 2806,
      "human level performance": 50907,
      "answer open question": 6550,
      "factual accuracy generated": 42928,
      "accuracy generated text": 2173,
      "bilingual evaluation understudy": 14038,
      "release new largescale": 98463,
      "new largescale dataset": 78988,
      "human evaluation study": 50838,
      "network sequence modeling": 77428,
      "great success various": 49032,
      "input forget gates": 55341,
      "models extensive experiments": 73215,
      "experiments nlp tasks": 41044,
      "paper presents computational": 84160,
      "quantitative qualitative results": 94881,
      "qualitative results demonstrate": 94576,
      "method substantially outperforms": 69169,
      "substantially outperforms existing": 114907,
      "different writing systems": 31581,
      "languages language families": 60671,
      "language model called": 59304,
      "documents consist long": 33208,
      "masked language modeling": 67292,
      "language modeling mlm": 59443,
      "sentence prediction tasks": 105998,
      "additionally present new": 3620,
      "long text generation": 65139,
      "coherence generated text": 18302,
      "speech pos tags": 111726,
      "detailed description dataset": 29765,
      "range word embedding": 95614,
      "language processing artificial": 59846,
      "processing artificial intelligence": 91627,
      "generation recent years": 47586,
      "approaches statistical approaches": 8352,
      "provide comprehensive review": 93789,
      "domain dialogue systems": 33505,
      "important application natural": 52099,
      "learning semantic parser": 62992,
      "semantic parsing aims": 105142,
      "transform natural language": 124254,
      "language nl utterances": 59720,
      "meaning representations mrs": 67681,
      "prior work paper": 90758,
      "tasks joint learning": 119230,
      "joint learning framework": 57285,
      "learning framework demonstrate": 62597,
      "framework demonstrate effectiveness": 45486,
      "universal conceptual cognitive": 127277,
      "conceptual cognitive annotation": 20658,
      "entity event mentions": 37933,
      "pilot annotation study": 87507,
      "novel neural approaches": 80666,
      "sentiment analysis dataset": 106580,
      "automatically constructed datasets": 10743,
      "propose new metric": 92868,
      "existing text generation": 40315,
      "text generation metrics": 120991,
      "information extraction based": 54567,
      "wiseman et al": 132741,
      "reinforcement learning propose": 97821,
      "achieve better bleu": 2426,
      "better bleu scores": 13522,
      "using domain specific": 129606,
      "analysis tasks using": 5889,
      "method transfer learning": 69198,
      "lowresource named entity": 65538,
      "entity recognition recent": 38063,
      "recognition recent years": 96983,
      "recognition ner especially": 96927,
      "resources lowresource languages": 101007,
      "leading poor performance": 61902,
      "improve performance ner": 52473,
      "using attention based": 129352,
      "significant performance improvements": 108844,
      "come play increasingly": 18771,
      "propose neural framework": 92818,
      "results proposed framework": 102067,
      "methods previous work": 69684,
      "directed acyclic graphs": 31814,
      "model using combination": 72276,
      "achieves new state": 2819,
      "cloze test dataset": 17896,
      "performance stateoftheart models": 86746,
      "translation work present": 125434,
      "present novel approaches": 89599,
      "improve translation performance": 52571,
      "representations individual words": 99690,
      "wmt14 englishtogerman englishtofrench": 132822,
      "model consistently improves": 70891,
      "improves performance strong": 53019,
      "performance strong transformer": 86761,
      "model vaswani et": 72301,
      "encoder neural machine": 36547,
      "stateoftheart nmt model": 112793,
      "nmt model transformer": 79902,
      "translation performance experiment": 125108,
      "performance experiment results": 86354,
      "propose unsupervised method": 93135,
      "large amounts web": 61027,
      "amounts web data": 5368,
      "increasingly popular recent": 53707,
      "new training data": 79226,
      "quantitative qualitative experiments": 94880,
      "language inference paraphrase": 59186,
      "inference paraphrase identification": 54188,
      "learning word embedding": 63165,
      "separate word embedding": 106866,
      "standard word embedding": 112338,
      "cooccurrence statistics used": 23238,
      "external knowledge base": 42007,
      "knowledge base available": 57763,
      "standard word embeddings": 112339,
      "existing methods focus": 40183,
      "applied text classification": 7127,
      "success deep learning": 115068,
      "improve text classification": 52563,
      "text classification specifically": 120769,
      "models designed capture": 73050,
      "capture context information": 15280,
      "method benchmark datasets": 68674,
      "text classification experimental": 120719,
      "classification experimental results": 17203,
      "prove method achieves": 93701,
      "achieves stateoftheart performances": 2895,
      "training neural response": 123744,
      "dialogue systems despite": 30775,
      "inspired recent success": 55582,
      "recent success pretraining": 96540,
      "propose effective method": 92641,
      "tasks propose novel": 119410,
      "small indomain dataset": 110160,
      "systems natural language": 117011,
      "speech recognition hypotheses": 111753,
      "models suffer slow": 74128,
      "speech translation task": 111838,
      "outperforms examined baselines": 82885,
      "method sentiment analysis": 69126,
      "parser domain adaptation": 84941,
      "domain adaptation propose": 33448,
      "domain adaptation method": 33439,
      "existing benchmark datasets": 40081,
      "benchmark datasets biomedical": 12769,
      "applied proposed method": 7110,
      "shows significant performance": 108628,
      "generate abstractive summaries": 46896,
      "propose endtoend trainable": 92655,
      "novel hierarchical attention": 80591,
      "hierarchical attention mechanism": 49931,
      "improves previous stateoftheart": 53030,
      "previous stateoftheart results": 90483,
      "machine learning paper": 65822,
      "learning paper presents": 62869,
      "supervised learning experiments": 115768,
      "machine translation chinese": 65922,
      "language work present": 60348,
      "novel endtoend neural": 80558,
      "structured attention mechanism": 113993,
      "attention mechanism models": 9906,
      "largescale multilabel text": 61471,
      "text classification lmtc": 120735,
      "state art methods": 112448,
      "elmo embeddings improve": 35312,
      "obtaining best results": 81435,
      "accuracy nlp tasks": 2218,
      "models costly train": 72995,
      "environmentally carbon footprint": 38199,
      "based findings propose": 11720,
      "deep transition architecture": 28436,
      "architecture sequence labeling": 8745,
      "labeling current stateoftheart": 58491,
      "current stateoftheart systems": 25366,
      "consecutive hidden states": 21164,
      "modeling global information": 72438,
      "models paper try": 73711,
      "paper try address": 84482,
      "try address issues": 125893,
      "address issues propose": 3706,
      "entire sentence experiments": 37712,
      "given training data": 48164,
      "task outperforms best": 118493,
      "outperforms best reported": 82860,
      "best reported results": 13427,
      "establish new stateoftheart": 38589,
      "present work progress": 89769,
      "change time use": 16372,
      "google books corpus": 48518,
      "previous studies typically": 90503,
      "effective approach improve": 34630,
      "intrinsic extrinsic evaluation": 56360,
      "presented trained using": 89800,
      "intrinsic evaluation embeddings": 56353,
      "extrinsic evaluation tasks": 42617,
      "morphological tagging text": 74748,
      "tagging text classification": 117458,
      "neural network using": 78415,
      "text classification propose": 120757,
      "corpus news articles": 23902,
      "underlying language understanding": 126683,
      "stateoftheart nlp techniques": 112791,
      "novel task predicting": 80746,
      "performance range nlp": 86649,
      "nlp tasks attention": 79707,
      "wide range natural": 132507,
      "rare unseen words": 95746,
      "learning representations rare": 62961,
      "training embedding model": 123603,
      "form knowledge graph": 45098,
      "contrast existing approaches": 22692,
      "existing approaches typically": 40063,
      "typically make use": 126446,
      "use vector space": 128356,
      "model learn better": 71427,
      "learn better representations": 61995,
      "task method improves": 118387,
      "nested entity mentions": 77105,
      "objective function train": 81084,
      "lowresource languages lrls": 65519,
      "machine translation terms": 66266,
      "translation terms adequacy": 125373,
      "terms adequacy fluency": 120267,
      "data augmentation utilizing": 25658,
      "effective way alleviate": 34778,
      "problems paper propose": 91355,
      "framework data augmentation": 45480,
      "machine translation uses": 66292,
      "targetside monolingual data": 117802,
      "related highresource language": 97866,
      "highresource language hrl": 50423,
      "experiments lowresource datasets": 40991,
      "settings data augmentation": 107794,
      "approach recurrent neural": 7845,
      "languages present new": 60803,
      "present new datasets": 89585,
      "models trained languages": 74209,
      "applications transfer learning": 7030,
      "named entities using": 76039,
      "multitask neural networks": 75885,
      "networks social media": 77776,
      "named entities document": 76029,
      "key task nlp": 57607,
      "task nlp applications": 118461,
      "stateoftheart approaches task": 112584,
      "task reach high": 118603,
      "reach high performance": 95892,
      "social media domains": 110375,
      "processing social media": 91796,
      "word embeddings partofspeech": 133180,
      "model uses multitask": 72268,
      "prediction systems outperform": 89130,
      "workshop noisy usergenerated": 135012,
      "noisy usergenerated text": 80132,
      "question answering study": 95107,
      "multimodal approaches visual": 75416,
      "question answering vqa": 95124,
      "continual learning methods": 22599,
      "matching model response": 67413,
      "model response selection": 71920,
      "noisy training data": 80129,
      "training set model": 123850,
      "teacher student models": 119699,
      "results public data": 102094,
      "data sets indicate": 26427,
      "existing matching models": 40172,
      "paper address task": 83711,
      "address task extracting": 3772,
      "left right context": 63300,
      "important role identifying": 52241,
      "methods prior work": 69687,
      "language models perform": 59617,
      "models perform equally": 73725,
      "recent developments natural": 96455,
      "developments natural language": 30444,
      "multilingual sentence representations": 75363,
      "french german spanish": 45811,
      "german spanish russian": 47914,
      "sentences parallel corpus": 106434,
      "trained natural language": 123212,
      "language models lexical": 59585,
      "contextual information words": 22475,
      "attention simultaneous machine": 10012,
      "simultaneous machine translation": 109656,
      "translation simultaneous machine": 125262,
      "translate source sentence": 124544,
      "model attends source": 70700,
      "sentence soft attention": 106082,
      "recently proposed waitk": 96743,
      "cambridge university engineering": 15080,
      "elastic weight consolidation": 35232,
      "weight consolidation ewc": 132324,
      "test sets using": 120535,
      "combination checkpoint averaging": 18553,
      "speech translation recent": 111835,
      "translation recent advances": 125199,
      "output automatic speech": 83052,
      "information lattice structure": 54719,
      "spanishenglish speech translation": 111180,
      "outperforms transformer baseline": 83039,
      "pairs propose novel": 83612,
      "model proposed paper": 71817,
      "outperforms existing research": 82896,
      "sentiment classification neural": 106706,
      "sentiment analysis current": 106578,
      "models sentiment analysis": 74005,
      "sentiment analysis make": 106607,
      "make use word": 66748,
      "language modeling objective": 59449,
      "problem paper proposes": 91160,
      "method significantly improve": 69135,
      "models english chinese": 73154,
      "chinese event detection": 16766,
      "language models greatly": 59564,
      "highquality natural language": 50399,
      "able generate diverse": 1645,
      "generate diverse set": 46932,
      "diverse set candidate": 32844,
      "recent work focused": 96572,
      "work perform extensive": 134669,
      "mention boundary detection": 68417,
      "does scale large": 33401,
      "scale large corpora": 103733,
      "manually annotated experts": 67031,
      "novel neural approach": 80665,
      "absolute improvement stateoftheart": 1739,
      "improvement stateoftheart results": 52768,
      "study problem generating": 114483,
      "compared previous works": 19421,
      "single sentence paragraph": 109796,
      "parts extensive experiments": 85576,
      "generate fluent text": 46945,
      "language models capable": 59524,
      "seen training time": 104550,
      "knowledge graph language": 57970,
      "facts knowledge graph": 42917,
      "wikidata knowledge graph": 132632,
      "achieves significantly better": 2864,
      "performance strong baseline": 86757,
      "strong baseline language": 113631,
      "baseline language model": 12236,
      "language models ability": 59494,
      "large language models": 61128,
      "language models generating": 59558,
      "text recurrent neural": 121232,
      "linguistic knowledge encoded": 64502,
      "multiturn conversations hierarchical": 75913,
      "responses given previous": 101272,
      "given previous utterances": 48088,
      "conversation takes place": 22977,
      "results model significantly": 101949,
      "significantly improves quality": 108954,
      "compared baseline methods": 19329,
      "instances relation extraction": 55634,
      "transfer learning helps": 124096,
      "2019 evaluation campaign": 697,
      "incorporate information negation": 53476,
      "neural sentiment classifiers": 78667,
      "results test set": 102265,
      "improving model performance": 53116,
      "language pairs achieve": 59738,
      "tasks vietnamese language": 119598,
      "paper reviews stateoftheart": 84412,
      "word segmentation existing": 133494,
      "implementing machine learning": 51983,
      "learning techniques improve": 63093,
      "word segmentation systems": 133506,
      "text classification feature": 120723,
      "methods proposed recently": 69695,
      "forcing model focus": 45024,
      "helps model performance": 49826,
      "machine translation domain": 65959,
      "report paper describes": 99022,
      "machine translation developed": 65950,
      "oregon state university": 82455,
      "machine translation robustness": 66204,
      "large parallel data": 61202,
      "data social media": 26464,
      "backtranslated monolingual data": 11325,
      "monolingual data using": 74583,
      "using model trained": 129889,
      "generative model discriminative": 47736,
      "models consistently outperform": 72967,
      "embedding propose novel": 35476,
      "given dialogue context": 48018,
      "knowledge experimental results": 57913,
      "results showed proposed": 102173,
      "showed proposed method": 108388,
      "proposed method improved": 93372,
      "nmt models learn": 79918,
      "study sentiment analysis": 114513,
      "sentiment analysis benefits": 106574,
      "train test machine": 123039,
      "test machine learning": 120468,
      "sentiment analysis multiple": 106612,
      "unlabeled monolingual data": 127402,
      "leads better results": 61923,
      "information conversational text": 54454,
      "labeled data limited": 58436,
      "paper propose weaklysupervised": 84345,
      "information extraction framework": 54571,
      "relation extraction task": 98029,
      "model longrange context": 71473,
      "following recent work": 44984,
      "language using neural": 60314,
      "technical note describes": 119750,
      "translation neural networks": 125015,
      "neural networks stateoftheart": 78585,
      "mt remains unclear": 74989,
      "paper systematically compare": 84461,
      "statistical mt models": 113136,
      "model size data": 72055,
      "gain significant improvements": 46351,
      "sentence representations learned": 106039,
      "representations learned deep": 99725,
      "multiple neural network": 75626,
      "data best knowledge": 25699,
      "attention neural machine": 9955,
      "machine translation sequencetosequence": 66212,
      "use attention mechanisms": 127901,
      "presents empirical evaluation": 89844,
      "word discovery unsegmented": 132995,
      "specifically address problem": 111523,
      "models outperform baselines": 73675,
      "evaluating language model": 39065,
      "language model finetuning": 59325,
      "languages suffer lack": 60905,
      "apply stateoftheart techniques": 7219,
      "introduce new benchmark": 56476,
      "benchmark language modeling": 12834,
      "systems paper introduce": 117044,
      "learning adversarial training": 62314,
      "adversarial training demonstrate": 4173,
      "words appear training": 133819,
      "specifically propose novel": 111584,
      "context information word": 22146,
      "modernizing historical documents": 74435,
      "generate new version": 46976,
      "written modern version": 135136,
      "modern version documents": 74426,
      "conducted user study": 20956,
      "sequences recurrent neural": 107135,
      "intermediate layer representations": 56134,
      "sentence token level": 106107,
      "present use cases": 89758,
      "word embeddings second": 133203,
      "publications using deep": 94285,
      "domain expert annotations": 33526,
      "introduce new model": 56488,
      "learning techniques using": 63098,
      "compared baseline model": 19330,
      "set best practices": 107379,
      "mining nlp techniques": 70247,
      "goal paper presents": 48380,
      "deep learning approaches": 28249,
      "learning approaches models": 62359,
      "language processing automatic": 59848,
      "important nlp applications": 52197,
      "neural network proposed": 78378,
      "word level sentence": 133347,
      "level sentence level": 63502,
      "layer improve performance": 61724,
      "measures accuracy precision": 67850,
      "accuracy precision recall": 2241,
      "precision recall f1score": 88800,
      "graph based neural": 48765,
      "using syntactic semantic": 130260,
      "task syntactic semantic": 118770,
      "important context words": 52135,
      "work introduce novel": 134578,
      "introduce novel graphbased": 56513,
      "novel graphbased neural": 80587,
      "experiments demonstrate advantage": 40876,
      "advantage proposed model": 4057,
      "wmt19 robustness task": 132860,
      "submission wmt19 robustness": 114742,
      "task mainly focuses": 118371,
      "translating noisy text": 124601,
      "experimental results revealed": 40691,
      "translation improves translation": 124863,
      "improves translation accuracy": 53057,
      "different perspectives using": 31332,
      "present results systematic": 89679,
      "universal sentence encoder": 127332,
      "models respectively based": 73947,
      "semantic space using": 105305,
      "retrieval question answering": 102425,
      "transfer learning tasks": 124135,
      "neural constituency parsers": 77871,
      "stateoftheart results benchmark": 112893,
      "results benchmark treebanks": 101545,
      "substantially improves performance": 114899,
      "rich input representations": 102749,
      "bias natural language": 13822,
      "popular natural language": 88107,
      "inference nli datasets": 54177,
      "spurious correlations data": 112064,
      "adversarial learning used": 4140,
      "models learn representations": 73487,
      "empirical study language": 36201,
      "language models chinese": 59528,
      "important role alleviating": 52237,
      "models extensively used": 73217,
      "approach tackle problem": 7953,
      "use recently proposed": 128229,
      "proposed neural models": 93507,
      "paper conduct extensive": 83778,
      "conduct extensive empirical": 20868,
      "extensive empirical study": 41873,
      "statistical neural language": 113143,
      "language models experimental": 59549,
      "experimental results different": 40605,
      "approach bring significant": 7400,
      "bring significant improvement": 14635,
      "detection russian study": 30041,
      "training datasets using": 123571,
      "present approach based": 89373,
      "bleu scores language": 14334,
      "documentlevel neural machine": 33154,
      "translation paper describes": 125087,
      "paper describes microsoft": 83831,
      "submissions wmt19 news": 114760,
      "wmt19 news translation": 132857,
      "news translation shared": 79386,
      "shared task englishgerman": 108043,
      "machine translation deep": 65946,
      "deep transformer models": 28431,
      "synthetic parallel data": 116638,
      "source language monolingual": 110777,
      "preliminary human evaluation": 89274,
      "score higher human": 104080,
      "naver labs europes": 76679,
      "labs europes systems": 58674,
      "describes systems submitted": 29437,
      "task aims improve": 117869,
      "noise social media": 80067,
      "extracted social media": 42183,
      "according automatic metrics": 1969,
      "automatic metrics bleu": 10602,
      "human evaluation proposed": 50826,
      "language pairs according": 59737,
      "domain adaptation language": 33435,
      "large textual corpora": 61295,
      "crosslingual natural language": 24983,
      "corpus speech recognition": 24013,
      "fields natural language": 44003,
      "relation detection knowledge": 97968,
      "detection knowledge base": 29979,
      "question answering relation": 95096,
      "answering relation detection": 6691,
      "step natural language": 113282,
      "question answering previous": 95073,
      "seen training data": 104546,
      "training data unseen": 123559,
      "seen unseen relations": 104553,
      "method greatly improve": 68867,
      "greatly improve performance": 49054,
      "stateoftheart code data": 112611,
      "models learn language": 73485,
      "model hierarchical structure": 71282,
      "capture similar information": 15396,
      "typologically different languages": 126494,
      "information learned network": 54724,
      "paper explore task": 83925,
      "explore task predicting": 41594,
      "task predicting human": 118543,
      "train neural network": 122985,
      "evaluation natural language": 39305,
      "training data multitask": 123497,
      "data multitask learning": 26148,
      "knowledge plays critical": 58104,
      "existing dialogue models": 40112,
      "propose novel transformerbased": 92972,
      "responses generated model": 101270,
      "significantly outperform competitive": 108983,
      "outperform competitive baselines": 82698,
      "dataset manually annotated": 27013,
      "manually annotated sentence": 67033,
      "classification models trained": 17282,
      "models trained dataset": 74198,
      "neural generative models": 77922,
      "shared task present": 108086,
      "wmt 2019 shared": 132776,
      "language pairs englishgerman": 59752,
      "learning approaches using": 62366,
      "propose new ensemble": 92847,
      "propose simple technique": 93069,
      "submitted systems achieve": 114775,
      "achieve best results": 2422,
      "framework distributional semantics": 45501,
      "word embeddings recent": 133196,
      "shown correlate poorly": 108458,
      "correlate poorly human": 24198,
      "hierarchical attention model": 49932,
      "answering set questions": 6699,
      "propose new hierarchical": 92852,
      "new hierarchical attention": 78943,
      "manually annotated corpora": 67025,
      "corpora lowresource languages": 23522,
      "processing nlp task": 91754,
      "finetune pretrained language": 44414,
      "finetuning pretrained language": 44494,
      "compare proposed method": 19285,
      "method crosslingual transfer": 68754,
      "partofspeech tagging pos": 85560,
      "word representations present": 133467,
      "models commonly used": 72931,
      "supervised unsupervised neural": 115852,
      "supervised unsupervised approaches": 115850,
      "language models supervised": 59666,
      "task data set": 118045,
      "systematic comparison neural": 116666,
      "compare performance current": 19271,
      "performance current stateoftheart": 86268,
      "extensive feature engineering": 41938,
      "problem training neural": 91268,
      "distributed narrow cone": 32616,
      "power word embeddings": 88659,
      "word embeddings analyze": 133050,
      "experiments language modeling": 40970,
      "better performance baseline": 13652,
      "neural network input": 78308,
      "conversational artificial intelligence": 22995,
      "conversational dataset experiments": 23001,
      "understanding recently pretrained": 126938,
      "recently pretrained models": 96734,
      "results various language": 102313,
      "crucial role natural": 25163,
      "language processing current": 59857,
      "source codes pretrained": 110733,
      "codes pretrained models": 18211,
      "pretrained models released": 90151,
      "cuni systems unsupervised": 25240,
      "task wmt 2019": 118848,
      "news shared task": 79363,
      "et al 2018b": 38727,
      "trained monolingual data": 123206,
      "monolingual data followed": 74572,
      "parallel data synthetic": 84654,
      "wmt19 shared task": 132862,
      "word embeddings case": 133060,
      "data past events": 26227,
      "models trained english": 74202,
      "data machine translation": 26106,
      "bert bidirectional encoder": 13079,
      "bidirectional encoder representations": 13914,
      "encoder representations transformers": 36568,
      "shared task dataset": 108039,
      "task dataset metric": 118047,
      "metric achieves stateoftheart": 69869,
      "translation systems based": 125308,
      "systems based transformer": 116752,
      "based transformer model": 12130,
      "aspectbased opinion mining": 9354,
      "partofspeech pos tagger": 85532,
      "advantages existing methods": 4080,
      "validate effectiveness method": 130716,
      "conversational machine comprehension": 23013,
      "machine comprehension conversational": 65747,
      "comprehension conversational machine": 20170,
      "machine comprehension mc": 65750,
      "proven significantly challenging": 93731,
      "words paper propose": 134105,
      "model effectively capture": 71045,
      "shows competitive performance": 108566,
      "competitive performance compared": 19659,
      "performance compared existing": 86231,
      "compared existing stateoftheart": 19369,
      "existing stateoftheart methods": 40292,
      "neural networks effective": 78471,
      "data natural language": 26154,
      "paper present dataset": 84091,
      "similar language translation": 109105,
      "area research years": 8882,
      "paper study performance": 84445,
      "performance varies depending": 86837,
      "language pair statistical": 59733,
      "language pair proposed": 59732,
      "word representations sentence": 133472,
      "extracting valuable information": 42245,
      "learning model task": 62766,
      "scarcity labeled data": 103807,
      "semisupervised learning technique": 105612,
      "utilize unlabeled data": 130535,
      "unlabeled data improve": 127385,
      "dataset model outperformed": 27031,
      "relative error reduction": 98352,
      "amr abstract meaning": 5382,
      "representing meaning natural": 100058,
      "make better use": 66628,
      "better use information": 13761,
      "techniques natural language": 119936,
      "nlp deep learning": 79600,
      "volume data order": 131938,
      "models requires large": 73939,
      "language paper presents": 59793,
      "techniques deep learning": 119863,
      "outperform conventional methods": 82701,
      "methods applied task": 69314,
      "error analysis reveals": 38285,
      "neural dependency parser": 77887,
      "nlpcc 2019 shared": 79798,
      "task semisupervised domain": 118684,
      "semisupervised domain adaptation": 105601,
      "dependency parsing based": 29172,
      "adapt different domains": 3174,
      "demonstrate model performs": 28796,
      "model performs competitively": 71740,
      "dependency treebanks evaluate": 29241,
      "tasks pos tagging": 119385,
      "improve performance different": 52463,
      "entity recognition present": 38060,
      "recognition present corpus": 96968,
      "news articles manually": 79299,
      "baseline experiments corpus": 12220,
      "outofdomain test sets": 82663,
      "especially recurrent neural": 38502,
      "network rnn models": 77421,
      "establishing strong baselines": 38623,
      "strong baselines new": 113652,
      "bert paper presents": 13198,
      "stateoftheart models tasks": 112750,
      "semantic parsing using": 105176,
      "datasets english chinese": 27449,
      "establish strong baselines": 38596,
      "reading comprehension machine": 95987,
      "comprehension machine reading": 20192,
      "words work propose": 134324,
      "attention mechanism better": 9887,
      "linguistically motivated word": 64594,
      "verify effectiveness proposed": 131583,
      "improvement strong baselines": 52772,
      "used data augmentation": 128470,
      "data augmentation technique": 25652,
      "target monolingual data": 117678,
      "backtranslated training data": 11327,
      "training data work": 123566,
      "naturally occurring text": 76647,
      "empirical evidence support": 36164,
      "capture human preferences": 15312,
      "collect largescale dataset": 18386,
      "heavily rely humanannotated": 49657,
      "rely humanannotated data": 98711,
      "data expensive obtain": 25919,
      "tackle training data": 117313,
      "training data bottleneck": 123439,
      "training data experimental": 123460,
      "data experimental results": 25922,
      "experimental results commonly": 40570,
      "results commonly used": 101589,
      "test set approach": 120497,
      "performance compared previous": 86234,
      "systems trained using": 117204,
      "training data method": 123493,
      "significantly outperforms previous": 109021,
      "outperforms previous methods": 82936,
      "methods reducing error": 69714,
      "noisy channel modeling": 80093,
      "neural noisy channel": 78617,
      "source target sentence": 110838,
      "models trained billions": 74193,
      "wmt17 germanenglish translation": 132841,
      "consistently outperform strong": 21426,
      "outperform strong alternatives": 82762,
      "relatively little research": 98402,
      "abductive natural language": 1471,
      "generation present study": 47552,
      "introduce challenge dataset": 56390,
      "multiplechoice question answering": 75756,
      "conditional generation task": 20756,
      "analysis leads new": 5674,
      "deep pretrained language": 28392,
      "language named entity": 59700,
      "different languages like": 31207,
      "languages like english": 60693,
      "language paper propose": 59794,
      "feature based svm": 43255,
      "data transfer learning": 26580,
      "dialogue systems realworld": 30784,
      "limited number examples": 64257,
      "stateoftheart dialogue generation": 112640,
      "using annotated data": 129336,
      "annotated data leveraging": 6162,
      "stanford multidomain dialogue": 112369,
      "dialogue dataset consisting": 30656,
      "scheduling weather information": 103891,
      "best model terms": 13373,
      "model terms bleu": 72155,
      "terms bleu entity": 120284,
      "bleu entity f1": 14281,
      "entity f1 scores": 37939,
      "word subword units": 133595,
      "using language model": 129781,
      "language model rescoring": 59389,
      "machine translation track": 66271,
      "wmt19 news shared": 132855,
      "machine translation pbsmt": 66155,
      "model pretrained language": 71774,
      "nmt models using": 79924,
      "parallel data propose": 84647,
      "conneau et al": 21114,
      "language model select": 59397,
      "applied improve quality": 7079,
      "translation lowresource language": 124918,
      "pairs previous works": 83610,
      "previous works usually": 90582,
      "setting paper propose": 107768,
      "experiments ted talks": 41173,
      "ted talks multilingual": 120053,
      "attention networks document": 9949,
      "neural networks recently": 78549,
      "achieved stateoftheart performance": 2705,
      "word cooccurrence networks": 132973,
      "standard text classification": 112324,
      "impact different components": 51868,
      "performance code publicly": 86212,
      "applications existing methods": 6927,
      "results experimental results": 101779,
      "finegrained sentiment analysis": 44382,
      "stateoftheart relation extraction": 112887,
      "relation extraction model": 97999,
      "conditional language generation": 20759,
      "language generation large": 59074,
      "large pretrained language": 61216,
      "work explore methods": 134512,
      "adapting pretrained language": 3323,
      "pretrained transformer models": 90198,
      "graphbased dependency parsers": 48894,
      "neural networks continuous": 78460,
      "word embeddings allow": 133049,
      "information global sentence": 54643,
      "evaluating contextualized embeddings": 39045,
      "tagging lemmatization dependency": 117398,
      "lemmatization dependency parsing": 63338,
      "present extensive evaluation": 89489,
      "udpipe 20 bestperforming": 126539,
      "20 bestperforming systems": 490,
      "bestperforming systems conll": 13479,
      "systems conll 2018": 116803,
      "shared task overall": 108076,
      "contextualized word embedding": 22573,
      "characterlevel word embeddings": 16589,
      "report stateoftheart results": 99049,
      "tasks compared results": 118994,
      "language processing computational": 59855,
      "monolingual data demonstrated": 74567,
      "data demonstrated helpful": 25824,
      "demonstrated helpful improving": 28917,
      "helpful improving translation": 49798,
      "improving translation quality": 53180,
      "information word embedding": 55096,
      "playing important role": 87718,
      "leverage monolingual data": 63605,
      "representations nmt model": 99784,
      "translation quality experimental": 125171,
      "quality experimental results": 94661,
      "chineseenglish germanenglish machine": 16854,
      "germanenglish machine translation": 47935,
      "tasks proposed model": 119417,
      "strong transformer baselines": 113726,
      "effectiveness approach lowresource": 34870,
      "proven effective method": 93722,
      "currently publicly available": 25415,
      "datasets models trained": 27580,
      "articles publicly available": 9210,
      "experiments demonstrate deep": 40881,
      "transfer learning different": 124090,
      "pretraining text generation": 90342,
      "sequencetosequence seq2seq pretraining": 107208,
      "conduct experiments text": 20858,
      "experiments text generation": 41180,
      "generation tasks abstractive": 47666,
      "tasks abstractive summarization": 118913,
      "performance strong baselines": 86759,
      "machine translation sequence": 66209,
      "sequence prediction tasks": 107033,
      "crossentropy loss leads": 24914,
      "wmt 2014 englishgerman": 132770,
      "englishgerman englishfrench translation": 37377,
      "comparison strong baselines": 19580,
      "dialogue act labels": 30631,
      "input dialogue act": 55320,
      "novel approach dialogue": 80476,
      "prediction auxiliary task": 89034,
      "inflected forms word": 54282,
      "edit distance experiments": 34533,
      "sequence labeling neural": 106975,
      "pos tagging ner": 88235,
      "model improves overall": 71317,
      "used various tasks": 128845,
      "work conduct indepth": 134430,
      "survey recent literature": 116187,
      "translation present experiments": 125133,
      "conduct detailed analysis": 20835,
      "working natural language": 134904,
      "speech recognition especially": 111751,
      "especially neural networks": 38488,
      "named entity recognizers": 76126,
      "stateoftheart models named": 112739,
      "models named entity": 73600,
      "large amounts labeled": 61013,
      "amounts labeled data": 5343,
      "involving crosslingual transfer": 56913,
      "effective training data": 34763,
      "second language learners": 104419,
      "attracted attention recently": 10144,
      "datasets multitask learning": 27586,
      "multitask learning applied": 75822,
      "proposed method performs": 93392,
      "method performs better": 69048,
      "performs better stateoftheart": 86987,
      "better stateoftheart baselines": 13728,
      "finetuning neural machine": 44483,
      "test set particular": 120512,
      "using transductive data": 130307,
      "transductive data selection": 124028,
      "test set provided": 120515,
      "small subset data": 110214,
      "sentiment classification text": 106716,
      "user product information": 129022,
      "improve performance sentiment": 52484,
      "performance sentiment classification": 86703,
      "sentiment classification models": 106702,
      "unlike previous models": 127442,
      "method achieves significant": 68606,
      "significant improvements standard": 108804,
      "approach attention mechanism": 7369,
      "transfer tasks model": 124194,
      "patent claim generation": 85663,
      "augmented inventing inventors": 10327,
      "inventors conceive better": 56693,
      "conceive better inventions": 20550,
      "finetuning pretrained model": 44497,
      "stateoftheart pretrained models": 112871,
      "set new stateoftheart": 107508,
      "tasks transfer learning": 119573,
      "model named entity": 71554,
      "entity recognition propose": 38061,
      "entity recognition based": 37996,
      "combines word embeddings": 18704,
      "previous work focused": 90529,
      "word character embeddings": 132942,
      "information named entity": 54785,
      "information different granularity": 54486,
      "fundamental problem natural": 46106,
      "pretraining multitask learning": 90294,
      "multitask learning learn": 75847,
      "modelagnostic metalearning algorithm": 72347,
      "metalearning algorithm maml": 68545,
      "proposed models outperform": 93497,
      "models outperform strong": 73686,
      "tasks efficiently effectively": 119084,
      "semantic parsing train": 105175,
      "multilingual semantic graphbank": 75358,
      "features significantly boost": 43729,
      "structure directly encoding": 113837,
      "transformer machine translation": 124338,
      "source target contexts": 110830,
      "introduce gate mechanism": 56427,
      "using pointwise mutual": 130025,
      "information extensive experiments": 54555,
      "extensive experiments translation": 41934,
      "bleu score strong": 14332,
      "consistent improvements baseline": 21383,
      "representations nlp tasks": 99782,
      "nlp tasks using": 79776,
      "embeddings make use": 35800,
      "syntactic features obtained": 116404,
      "features obtained automatically": 43637,
      "transfer linguistic tasks": 124147,
      "human response patterns": 50956,
      "provide valuable information": 93953,
      "quantitative qualitative analyses": 94876,
      "parameters training set": 84792,
      "sample training data": 103569,
      "data outperforms baseline": 26200,
      "human conceptual knowledge": 50785,
      "models word meaning": 74350,
      "datasets limited size": 27550,
      "space models word": 111027,
      "compare model previous": 19259,
      "model previous approach": 71784,
      "approach performs better": 7797,
      "tasks finally discuss": 119131,
      "method used develop": 69207,
      "differences language use": 30961,
      "changes language use": 16386,
      "potential implications understanding": 88562,
      "achieved new stateoftheart": 2658,
      "paper explore ability": 83913,
      "ability nmt encoders": 1554,
      "encoder hidden states": 36531,
      "outperform word embeddings": 82773,
      "provide relevant information": 93907,
      "decoder hidden states": 28020,
      "syntax neural language": 116546,
      "language models recurrent": 59651,
      "models recurrent neural": 73893,
      "models perform poorly": 73727,
      "zeroshot text classification": 135530,
      "chang et al": 16355,
      "et al 2008": 38691,
      "task specific training": 118730,
      "specific training data": 111508,
      "jointly sequence labeling": 57388,
      "proposed model effective": 93443,
      "empirical results benchmark": 36178,
      "datasets demonstrate approach": 27395,
      "demonstrate approach achieve": 28666,
      "machine translation pretraining": 66168,
      "propose novel crosslingual": 92907,
      "propose new pretraining": 92879,
      "crosslingual masked language": 24974,
      "masked language model": 67288,
      "language model cmlm": 59306,
      "input text stream": 55453,
      "candidates time step": 15149,
      "time step experiments": 122113,
      "models encoder decoder": 73148,
      "present publicly available": 89655,
      "publicly available annotated": 94290,
      "available annotated datasets": 10942,
      "supervised semisupervised settings": 115832,
      "demonstrating effectiveness proposed": 28972,
      "provide qualitative analysis": 93901,
      "language understanding pretrained": 60253,
      "understanding pretrained language": 126924,
      "language models achieved": 59496,
      "achieved great successes": 2636,
      "pretraining language models": 90271,
      "tasks including named": 119181,
      "language inference xnli": 59203,
      "chain conditional random": 15972,
      "aims ground entity": 4536,
      "dataset analysis shows": 26733,
      "representational similarity analysis": 99479,
      "similarity analysis rsa": 109196,
      "require large training": 100167,
      "pretrained language encoders": 90046,
      "help better understand": 49699,
      "model based graph": 70732,
      "significantly outperforms current": 109010,
      "translation recently proposed": 125207,
      "languages english single": 60547,
      "english single model": 37278,
      "performance low resource": 86504,
      "massively multilingual nmt": 67346,
      "multilingual nmt model": 75316,
      "classification sequence labeling": 17398,
      "compare strong baseline": 19299,
      "multilingual bert mbert": 75203,
      "discriminative neural model": 32213,
      "model crosslingual word": 70943,
      "alignment introduce novel": 4965,
      "transformerbased machine translation": 124420,
      "translation model experiments": 124940,
      "orthogonal transformations embedding": 82589,
      "transformations embedding space": 124279,
      "property word embeddings": 92508,
      "word analogies solved": 132915,
      "trained word embeddings": 123338,
      "contextaware neural machine": 22343,
      "recent years studies": 96638,
      "use documentlevel context": 128013,
      "current sentence experiments": 25334,
      "subword language models": 115024,
      "models fast accurate": 73240,
      "25 times faster": 820,
      "propose new evaluation": 92848,
      "work propose neural": 134728,
      "neural components designed": 77867,
      "model using generative": 72278,
      "using generative adversarial": 129702,
      "adversarial neural network": 4155,
      "neural network learn": 78319,
      "fast named entity": 43146,
      "language models knowledge": 59572,
      "models knowledge bases": 73440,
      "improvements downstream nlp": 52844,
      "present training data": 89745,
      "structured knowledge bases": 114008,
      "wide range stateoftheart": 132512,
      "stateoftheart pretrained language": 112869,
      "language models finetuning": 59556,
      "question answering supervised": 95109,
      "opendomain qa systems": 81975,
      "language corpus contains": 58923,
      "corpus contains 26000": 23723,
      "approach taskoriented dialogue": 7965,
      "currently available different": 25398,
      "available different languages": 10983,
      "machine translation modern": 66059,
      "nmt systems produce": 79978,
      "sequencetosequence model maps": 107171,
      "large improvements translation": 61112,
      "translation contextual phenomena": 124710,
      "bleu score conduct": 14325,
      "conduct human evaluation": 20879,
      "human evaluation strong": 50837,
      "using available data": 129377,
      "widely used research": 132585,
      "compare results models": 19291,
      "results highlight importance": 101826,
      "simultaneous translation simultaneous": 109667,
      "bleu scores similar": 14335,
      "modeling hierarchical structure": 72441,
      "ordered neurons lstm": 82434,
      "translation task proposed": 125342,
      "task proposed approach": 118579,
      "language processing based": 59849,
      "language work propose": 60349,
      "current input previous": 25288,
      "datasets establish new": 27454,
      "establish new state": 38586,
      "state art datasets": 112438,
      "close large gap": 17824,
      "simple effective solution": 109415,
      "named entity types": 76130,
      "jointly learning align": 57354,
      "learning align translate": 62329,
      "models state art": 74095,
      "state art machine": 112444,
      "closely related task": 17865,
      "statistical word alignment": 113177,
      "model produce accurate": 71797,
      "translation model training": 124944,
      "based word alignment": 12163,
      "gives better results": 48186,
      "social media newspaper": 110394,
      "tasks requiring significant": 119467,
      "models use single": 74282,
      "negative polarity item": 76942,
      "latest developments natural": 61653,
      "use pretrained transformer": 128209,
      "pretrained transformer networks": 90200,
      "transformer model focus": 124341,
      "stateoftheart results models": 112910,
      "question answering kbqa": 95057,
      "question logical form": 95176,
      "logical forms model": 65008,
      "huge search space": 50727,
      "quality training data": 94814,
      "semantic parsing model": 105151,
      "work propose effective": 134717,
      "method weakly supervised": 69225,
      "weakly supervised training": 132200,
      "news articles event": 79295,
      "extraction news articles": 42414,
      "english news articles": 37222,
      "bilingual lexicon induction": 14045,
      "stateoftheart models task": 112749,
      "demonstrate performance stateoftheart": 28816,
      "systematic comparison methods": 116665,
      "lowresource dependency parsing": 65494,
      "require lots training": 100172,
      "lots training data": 65338,
      "human performance demonstrating": 50929,
      "efficient sentence embedding": 35113,
      "better overall performance": 13644,
      "downstream classification tasks": 33995,
      "selfattention neural machine": 104898,
      "translation models rely": 124964,
      "attention mechanism work": 9918,
      "investigate different approaches": 56743,
      "incorporate syntactic knowledge": 53490,
      "especially long sentences": 38467,
      "compression text compression": 20328,
      "propose fully unsupervised": 92691,
      "sentence compression model": 105798,
      "competitive stateoftheart supervised": 19695,
      "supervised models trained": 115809,
      "large corpus american": 61062,
      "social science studies": 110453,
      "essential natural language": 38559,
      "understanding tasks natural": 126986,
      "language inference machine": 59173,
      "method consistently improves": 68729,
      "contextualized embeddings capture": 22547,
      "context recently proposed": 22244,
      "recognition ner tasks": 96944,
      "czech named entity": 25495,
      "20 report stateoftheart": 527,
      "contextualized span representations": 22571,
      "extraction tasks named": 42511,
      "entity recognition relation": 38065,
      "recognition relation extraction": 96986,
      "relation extraction event": 97992,
      "extraction event extraction": 42329,
      "event extraction framework": 39514,
      "framework achieves stateoftheart": 45410,
      "representations contextualized embeddings": 99570,
      "like bert perform": 64023,
      "recognition ner models": 96934,
      "training data challenging": 123442,
      "data challenging task": 25726,
      "challenging task recent": 16332,
      "task recent work": 118610,
      "work shown promising": 134803,
      "promising results crosslingual": 92292,
      "results crosslingual transfer": 101651,
      "crosslingual transfer highresource": 25029,
      "transfer highresource languages": 124059,
      "highresource languages lowresource": 50427,
      "propose simple efficient": 93059,
      "promising direction improve": 92271,
      "results shed light": 102170,
      "shed light future": 108151,
      "light future research": 63990,
      "future research improving": 46303,
      "given short text": 48126,
      "using dataset tweets": 129535,
      "user study shows": 129046,
      "demonstrate statistically significant": 28872,
      "sequential sentence classification": 107253,
      "task requires understanding": 118648,
      "work pretrained language": 134701,
      "language models bert": 59514,
      "models bert devlin": 72833,
      "stateoftheart results datasets": 112898,
      "embeddings trained different": 35984,
      "training data recent": 123522,
      "recent success transfer": 96541,
      "success transfer learning": 115136,
      "language modeling training": 59476,
      "representations trained different": 99940,
      "improve language modeling": 52403,
      "modeling language models": 72457,
      "models lms predominantly": 73520,
      "types pretrained embeddings": 126340,
      "beneficial neural language": 12952,
      "language modeling knowledge": 59436,
      "representations contextual word": 99567,
      "multiple knowledge bases": 75586,
      "contrast previous approaches": 22703,
      "large raw text": 61235,
      "probing task downstream": 90909,
      "sequence labeling domain": 106968,
      "domain adaptation methods": 33440,
      "lead negative transfer": 61862,
      "paper propose finegrained": 84240,
      "learning target domain": 63084,
      "strong baselines stateoftheart": 113654,
      "stateoftheart sequence labeling": 112950,
      "sentences order predict": 106420,
      "model evaluate method": 71105,
      "using information extraction": 129755,
      "deep representation learning": 28415,
      "work demonstrates potential": 134465,
      "nlp tasks lack": 79729,
      "language model neural": 59355,
      "approach language model": 7664,
      "language model prediction": 59370,
      "translation model jointly": 124941,
      "translation model compared": 124939,
      "englishjapanese machine translation": 37415,
      "language model experimental": 59320,
      "experimental results obtained": 40663,
      "dynamic fusion mechanism": 34308,
      "language model dynamic": 59314,
      "model dynamic fusion": 71034,
      "evaluation text generation": 39423,
      "translation text summarization": 125378,
      "image captioning video": 51773,
      "metrics bleu rouge": 69935,
      "wide variety text": 132529,
      "variety text generation": 131024,
      "ngram based metrics": 79427,
      "general world knowledge": 46733,
      "nlp models better": 79639,
      "models better understand": 72847,
      "evaluate dataset using": 38816,
      "large datasets using": 61075,
      "make effective use": 66664,
      "machine translation semantic": 66206,
      "trained using maximum": 123328,
      "using maximum likelihood": 129866,
      "recent work demonstrated": 96570,
      "nmt systems based": 79973,
      "recent work semantic": 96584,
      "response generation model": 101204,
      "attention mechanism address": 9880,
      "datasets effectiveness model": 27441,
      "outperforming baseline systems": 82800,
      "synthetic data consisting": 116616,
      "outperforms baselines based": 82849,
      "based prior work": 11954,
      "intent detection slot": 55903,
      "detection slot filling": 30055,
      "models fail fully": 73232,
      "performance address issue": 86128,
      "issue paper propose": 57016,
      "scarcity labeled training": 103808,
      "models trained language": 74208,
      "data available source": 25683,
      "source language english": 110773,
      "supervised learning unsupervised": 115782,
      "pretraining unsupervised data": 90350,
      "baselines achieve new": 12346,
      "task oriented dialogue": 118487,
      "oriented dialogue systems": 82495,
      "state tracking dst": 112527,
      "predicting user intent": 89021,
      "stateoftheart dst models": 112648,
      "systems new domains": 117022,
      "address limitations propose": 3716,
      "based reinforcement learning": 11988,
      "models new domains": 73639,
      "method improve models": 68884,
      "recent work language": 96575,
      "training transformer models": 123936,
      "transformer based models": 124305,
      "advance state art": 3932,
      "transformer language model": 124331,
      "model achieve sota": 70539,
      "achieve sota results": 2551,
      "bert model achieves": 13167,
      "achieves sota results": 2872,
      "results race dataset": 102108,
      "graphical neural networks": 48932,
      "neural networks text": 78589,
      "networks text classification": 77795,
      "natural language hard": 76323,
      "neural networks gnn": 78489,
      "model graph structure": 71266,
      "models graph convolutional": 73314,
      "convolutional networks gcn": 23171,
      "local global information": 64923,
      "strong baselines significantly": 113653,
      "novel approach incorporate": 80480,
      "syntax natural language": 116543,
      "inference nli models": 54180,
      "models method uses": 73564,
      "embedding language models": 35421,
      "shown effective improving": 108463,
      "linguistic unit word": 64576,
      "syntactic dependency parsing": 116394,
      "implicit discourse relation": 52006,
      "discourse relation recognition": 32075,
      "description natural language": 29470,
      "work address problem": 134356,
      "variational autoencoder cvae": 130911,
      "present empirical results": 89466,
      "architecture proposed model": 8734,
      "proposed model effectively": 93444,
      "result proposed model": 101397,
      "proposed model yields": 93488,
      "translation quality demonstrating": 125166,
      "languages crosslingual transfer": 60476,
      "word embeddings contextual": 133068,
      "facilitate crosslingual transfer": 42762,
      "wide range target": 132513,
      "target languages demonstrate": 117666,
      "demonstrate adversarial training": 28660,
      "significantly improves overall": 108951,
      "paper explores task": 83934,
      "shown great promise": 108477,
      "perform largescale study": 86025,
      "metrics bleu meteor": 69933,
      "relevant information increases": 98556,
      "explore new approach": 41569,
      "generation aims generate": 47302,
      "text generation models": 120993,
      "disambiguation using deep": 31994,
      "learning approach automatically": 62341,
      "text named entity": 121137,
      "task biomedical domain": 117942,
      "domain named entity": 33589,
      "named entity mentions": 76050,
      "multiple entity types": 75556,
      "source error address": 110756,
      "address issue generating": 3687,
      "task classification problem": 117964,
      "developed deep learningbased": 30267,
      "fully connected layer": 45952,
      "generation nlg tasks": 47518,
      "tasks multiple languages": 119316,
      "monolingual crosslingual settings": 74564,
      "objective encourages model": 81074,
      "different languages shared": 31213,
      "zeroshot crosslingual transfer": 135494,
      "use monolingual data": 128152,
      "monolingual data finetune": 74570,
      "finetune pretrained model": 44416,
      "sequencetosequence model trained": 107175,
      "model trained single": 72208,
      "trained single language": 123279,
      "results question generation": 102105,
      "generation abstractive summarization": 47289,
      "abstractive summarization model": 1814,
      "summarization model outperforms": 115537,
      "methods zeroshot crosslingual": 69861,
      "performance lowresource languages": 86509,
      "word concept embeddings": 132964,
      "embeddings deep learning": 35632,
      "learn word concept": 62189,
      "relations paper propose": 98234,
      "use downstream applications": 128019,
      "significantly outperforms strong": 109038,
      "low resource egyptian": 65385,
      "egyptian arabic dialect": 35211,
      "unlabeled audio data": 127372,
      "audio data collected": 10227,
      "features deep neural": 43442,
      "neural network acoustic": 78238,
      "network acoustic model": 77126,
      "network cnn followed": 77194,
      "neural network tdnn": 78405,
      "sequence discriminative training": 106934,
      "language model decoding": 59310,
      "nbest list rescoring": 76693,
      "achieve good results": 2470,
      "results improved using": 101843,
      "lowest word error": 65466,
      "importance subword information": 52084,
      "lowresource languages recent": 65526,
      "languages recent work": 60834,
      "subword information word": 115022,
      "ability neural models": 1551,
      "truly lowresource scenarios": 125861,
      "morphological tagging named": 74744,
      "conduct systematic study": 20899,
      "data required train": 26345,
      "train word embeddings": 123051,
      "subwordinformed word representation": 115047,
      "pretrained multilingual models": 90166,
      "recently pretrained language": 96731,
      "models achieved remarkable": 72687,
      "achieved remarkable success": 2690,
      "broad range natural": 14677,
      "deep language model": 28241,
      "monolingual language models": 74594,
      "size language model": 109929,
      "model relatively small": 71893,
      "pretrained multilingual model": 90165,
      "tasks including partofspeech": 119187,
      "including partofspeech tagging": 53347,
      "experimental results using": 40717,
      "semantic parsing neural": 105155,
      "meaning representations natural": 67682,
      "natural language utterances": 76602,
      "neural semantic parsers": 78663,
      "semantic parsing datasets": 105147,
      "architecture achieves stateoftheart": 8607,
      "domain mismatch problem": 33583,
      "different things different": 31496,
      "context machine translation": 22172,
      "training machine translation": 123697,
      "systems low resource": 116984,
      "low resource language": 65386,
      "resource language pairs": 100856,
      "networks shown perform": 77767,
      "recognition ner speech": 96939,
      "ner speech pos": 77081,
      "pos tagging chunking": 88227,
      "modern neural network": 74404,
      "field crf output": 43945,
      "crf output layer": 24756,
      "competitive state art": 19690,
      "state art systems": 112478,
      "using finite state": 129681,
      "finite state transducers": 44542,
      "effective way improve": 34779,
      "language generation ability": 59067,
      "given dialog context": 48015,
      "smooth inverse frequency": 110265,
      "sequencetosequence model copy": 107168,
      "generating synthetic data": 47272,
      "decompositional semantics uds": 28161,
      "single semantic graph": 109791,
      "linguistic features extracted": 64483,
      "vulnerabilities natural language": 131970,
      "models paper investigate": 73702,
      "generic language characteristics": 47801,
      "levels lexical syntactic": 63547,
      "propose neural machine": 92819,
      "reinforcement learning technique": 97829,
      "sentiment classification twitter": 106717,
      "nmt models approach": 79907,
      "present new neural": 89589,
      "new neural architecture": 79051,
      "hierarchical multitask architecture": 49983,
      "approach obtains promising": 7747,
      "obtains promising results": 81473,
      "corresponding semantic roles": 24302,
      "significant improvements achieved": 108783,
      "achieved various natural": 2722,
      "processing tasks using": 91830,
      "models automatic speech": 72792,
      "recognition asr task": 96817,
      "training data report": 123525,
      "dataset source code": 27212,
      "source code available": 110715,
      "training models requires": 123715,
      "labeled data learning": 58435,
      "maximize performance test": 67513,
      "performance test set": 86796,
      "substantially improve performance": 114893,
      "improving data efficiency": 53083,
      "data efficiency learning": 25881,
      "forms morphologically rich": 45252,
      "approach models different": 7715,
      "20 relative error": 521,
      "text level graph": 121088,
      "text classification recently": 120761,
      "previous methods based": 90427,
      "problems propose new": 91360,
      "features significantly reduce": 43730,
      "sequence tagging paper": 107081,
      "perform semantic parsing": 86060,
      "cross lingual transfer": 24865,
      "using adversarial learning": 129325,
      "adversarial learning paper": 4139,
      "parsing model based": 85159,
      "adversarial learning domain": 4132,
      "learning domain generalization": 62519,
      "underlying question study": 126694,
      "question study adversarial": 95222,
      "study adversarial learning": 114308,
      "data different sources": 25846,
      "train models higher": 122961,
      "models higher level": 73328,
      "higher level abstraction": 50187,
      "level abstraction order": 63413,
      "abstraction order increase": 1792,
      "order increase robustness": 82345,
      "increase robustness lexical": 53614,
      "robustness lexical stylistic": 103106,
      "strategy evaluated french": 113510,
      "evaluated french corpus": 38977,
      "adversarial learning increases": 4135,
      "learning increases models": 62647,
      "increases models generalization": 53647,
      "models generalization capabilities": 73273,
      "automatic speech transcription": 10671,
      "sensitive noise input": 105741,
      "data language pairs": 26073,
      "language pairs data": 59743,
      "propose new data": 92840,
      "new data augmentation": 78853,
      "vector representations sentences": 131358,
      "datasets natural language": 27591,
      "language inference using": 59202,
      "named entities involved": 76030,
      "languages like arabic": 60690,
      "language generation natural": 59077,
      "endtoend e2e systems": 36897,
      "trained deep learning": 123112,
      "timeconsuming task paper": 122149,
      "task paper propose": 118510,
      "paper propose semisupervised": 84317,
      "annotated data available": 6157,
      "benchmark datasets limited": 12789,
      "limited annotated data": 64205,
      "method achieve competitive": 68585,
      "language processing recent": 59937,
      "progress natural language": 92163,
      "stateofthe art transformer": 112565,
      "augment training data": 10269,
      "training data good": 123473,
      "effective improving performance": 34692,
      "little impact performance": 64811,
      "recent studies revealed": 96524,
      "exploit annotation artifacts": 41407,
      "biases current datasets": 13867,
      "systems address issue": 116723,
      "zeroshot dependency parsing": 135496,
      "monolingual corpora languages": 74559,
      "languages using single": 60956,
      "classification methods dataset": 17269,
      "text classifiers using": 120791,
      "proposed approach yields": 93221,
      "review existing methods": 102549,
      "shows method achieves": 108596,
      "existing methods best": 40177,
      "fewshot relation classification": 43911,
      "relation classification present": 97952,
      "relation classification models": 97950,
      "new test set": 79217,
      "dataset baselines released": 26761,
      "transformer network vaswani": 124356,
      "network vaswani et": 77471,
      "number parameters model": 80934,
      "taskoriented conversational agents": 118882,
      "creating challenge datasets": 24695,
      "data taking advantage": 26538,
      "control generation process": 22806,
      "transfer learning natural": 124113,
      "processing paper present": 91766,
      "use semisupervised training": 128264,
      "language machine learning": 59264,
      "naive bayes decision": 76015,
      "stateoftheart performance natural": 112833,
      "processing tasks text": 91829,
      "classification machine translation": 17260,
      "better performance target": 13663,
      "understanding recent advances": 126935,
      "domain knowledge paper": 33560,
      "knowledge paper present": 58095,
      "obtaining new stateoftheart": 81445,
      "evaluation germanenglish machine": 39224,
      "test suite present": 120542,
      "107 phenomena organized": 196,
      "phenomena organized 14": 87240,
      "organized 14 categories": 82476,
      "numerous downstream applications": 81018,
      "question answering web": 95126,
      "based distributional similarity": 11658,
      "feature selection method": 43307,
      "previous stateoftheart methods": 90472,
      "identify word boundaries": 51577,
      "word segmentation algorithms": 133490,
      "data paper explore": 26211,
      "translation models perform": 124962,
      "models perform unsupervised": 73730,
      "introducing new loss": 56647,
      "loss function jointly": 65274,
      "work liu et": 134618,
      "et al 2019a": 38735,
      "model improve training": 71311,
      "model word representation": 72322,
      "quality word vectors": 94831,
      "neural model biomedical": 78146,
      "structured prediction model": 114027,
      "model achieves performance": 70580,
      "development set model": 30422,
      "representations previous work": 99821,
      "multilingual language model": 75268,
      "sequence labeling methods": 106972,
      "deep contextualized representations": 28217,
      "approach achieves new": 7309,
      "using social media": 130194,
      "usergenerated data tweets": 129073,
      "framework named entity": 45621,
      "paper propose unified": 84337,
      "propose unified framework": 93129,
      "labeling problem propose": 58526,
      "formulate machine reading": 45274,
      "reading comprehension mrc": 95992,
      "comprehension mrc task": 20202,
      "question person mentioned": 95199,
      "person mentioned text": 87099,
      "additionally query encodes": 3628,
      "query encodes informative": 94957,
      "encodes informative prior": 36671,
      "informative prior knowledge": 55147,
      "prior knowledge strategy": 90712,
      "knowledge strategy facilitates": 58188,
      "strategy facilitates process": 113515,
      "facilitates process entity": 42804,
      "process entity extraction": 91474,
      "entity extraction leading": 37935,
      "extraction leading better": 42371,
      "leading better performances": 61886,
      "conll 2003 english": 21074,
      "natural language tools": 76550,
      "annotated sentence level": 6229,
      "evaluating sentence embeddings": 39098,
      "sentence embeddings provide": 105843,
      "provide comprehensive evaluation": 93787,
      "contextual embeddings multilingual": 22459,
      "effective transfer learning": 34765,
      "model uses features": 72267,
      "doing multitask learning": 33420,
      "downstream applications machine": 33990,
      "achieves reasonable accuracy": 2842,
      "lot room improvement": 65329,
      "room improvement especially": 103263,
      "code data publicly": 18086,
      "parallel corpora training": 84609,
      "training neural models": 123738,
      "large training data": 61299,
      "dialogue act types": 30633,
      "neural networks predict": 78536,
      "easyfirst dependency parsing": 34489,
      "word embeddings characterlevel": 133063,
      "embeddings characterlevel word": 35592,
      "uses deep neural": 129220,
      "stateoftheart results paper": 112917,
      "paper proposed new": 84347,
      "proposed new method": 93510,
      "word embeddings method": 133161,
      "embeddings method achieves": 35806,
      "dialogue generation tasks": 30684,
      "systems existing methods": 116885,
      "existing methods tend": 40196,
      "use metalearning framework": 128142,
      "target task finetuning": 117723,
      "module private module": 74505,
      "datasets method outperforms": 27566,
      "outperforms baselines terms": 82853,
      "domain adaptation realworld": 33449,
      "robustness perform extensive": 103124,
      "perform extensive evaluation": 86002,
      "evaluation automatic human": 39126,
      "automatic human shows": 10561,
      "shows significant improvements": 108626,
      "language models introduce": 59570,
      "pretrained neural language": 90172,
      "original training set": 82558,
      "effective domain adaptation": 34666,
      "generate output sequence": 46983,
      "image captioning machine": 51768,
      "captioning machine translation": 15262,
      "like bleu meteor": 64027,
      "approach uses deep": 8003,
      "uses deep learning": 129219,
      "better correlation human": 13554,
      "pretraining text encoders": 90341,
      "important semantic information": 52250,
      "character sequence words": 16464,
      "learning models semantics": 62791,
      "using naturalistic corpora": 129943,
      "models fail generalize": 73233,
      "models solve task": 74067,
      "machine translation community": 65926,
      "sentence source language": 106084,
      "sentence target language": 106100,
      "systems trained different": 117203,
      "trained different data": 123116,
      "dialogue systems user": 30793,
      "prior work focused": 90748,
      "conversation paper investigate": 22966,
      "systems experiments demonstrate": 116892,
      "training single model": 123867,
      "text multiple languages": 121135,
      "present detailed study": 89449,
      "effective crosslingual transfer": 34655,
      "text comes different": 120802,
      "models different languages": 73070,
      "noncontextual word embeddings": 80182,
      "machine translation generally": 65991,
      "translation models generate": 124955,
      "generate target words": 47029,
      "translations experimental results": 125465,
      "experimental results nist": 40661,
      "improvements baseline systems": 52811,
      "compared stateoftheart transformer": 19464,
      "stateoftheart transformer model": 113012,
      "model fewer parameters": 71178,
      "nonautoregressive machine translation": 80156,
      "machine translation nonautoregressive": 66129,
      "machine translation nat": 66077,
      "speed compared autoregressive": 111864,
      "compared autoregressive models": 19326,
      "existing nat models": 40220,
      "large gains accuracy": 61097,
      "training knowledge distillation": 123670,
      "best translation quality": 13466,
      "improving grammatical error": 53101,
      "data synthesis method": 26529,
      "method generate diverse": 68854,
      "error correction based": 38294,
      "translation models different": 124954,
      "english second language": 37269,
      "second language learner": 104418,
      "fluent grammatically correct": 44708,
      "pseudo parallel data": 94184,
      "translation task iwslt": 125333,
      "paper present submission": 84140,
      "language models used": 59683,
      "model trained monolingual": 72201,
      "model trained large": 72197,
      "training language model": 123673,
      "parallel data improve": 84636,
      "based dense representations": 11635,
      "facilitate nlp research": 42784,
      "key idea proposed": 57577,
      "perform downstream tasks": 85985,
      "downstream tasks text": 34057,
      "text classification natural": 120747,
      "classification tasks natural": 17449,
      "robustness neural machine": 103113,
      "machine translation translating": 66280,
      "key challenge machine": 57544,
      "statistical smt neural": 113165,
      "translation nmt paper": 125049,
      "smt nmt models": 110292,
      "test sets unknown": 120534,
      "adaptation neural machine": 3239,
      "success neural machine": 115106,
      "simultaneous neural machine": 109660,
      "machine translation snmt": 66232,
      "corpora language pairs": 23512,
      "efficacy proposed framework": 35017,
      "evaluation paper present": 39318,
      "selection training data": 104843,
      "models deep neural": 73024,
      "progress pretrained language": 92174,
      "language models led": 59583,
      "systems able generate": 116708,
      "topk nucleus sampling": 122671,
      "simple effective decoding": 109408,
      "construct new benchmark": 21642,
      "new benchmark dataset": 78812,
      "benchmark dataset task": 12761,
      "demonstrates model outperforms": 28949,
      "baseline methods large": 12250,
      "methods large margin": 69577,
      "large margin addition": 61142,
      "paper propose multiperspective": 84269,
      "pretrained bert model": 89991,
      "vectors wordpiece vectors": 131492,
      "unsupervised question answering": 127699,
      "supervised relation classification": 115827,
      "relation classification rc": 97953,
      "entity linking el": 37956,
      "original bert model": 82506,
      "intent classification task": 55900,
      "methods word embedding": 69851,
      "canonical correlation analysis": 15153,
      "paper present thorough": 84144,
      "thorough error analysis": 121883,
      "error analysis results": 38284,
      "present effective approach": 89460,
      "demonstrate strong performance": 28875,
      "perform extensive experiments": 86004,
      "syntax semantic role": 116557,
      "aspect based sentiment": 9317,
      "achieving performance comparable": 2968,
      "benefits multitask learning": 13015,
      "proposed method makes": 93379,
      "language multilingual nmt": 59697,
      "nmt machine translation": 79889,
      "original source text": 82548,
      "training data nmt": 123502,
      "data nmt model": 26174,
      "nmt model using": 79904,
      "human evaluation scores": 50832,
      "language model pretrained": 59375,
      "language models ubiquitous": 59680,
      "ubiquitous natural language": 126511,
      "trained english data": 123135,
      "data multiple languages": 26145,
      "models languages english": 73457,
      "languages english limited": 60543,
      "transformerbased language models": 124415,
      "language models languages": 59576,
      "partofspeech tagging dependency": 85551,
      "dependency parsing named": 29184,
      "parsing named entity": 85167,
      "improves state art": 53047,
      "downstream tasks semantic": 34055,
      "parallel sentences web": 84686,
      "20 language pairs": 509,
      "nmt systems language": 79976,
      "english german russian": 37155,
      "achieve excellent results": 2457,
      "results distant languages": 101734,
      "distant languages pairs": 32454,
      "languages pairs like": 60775,
      "asian translation wat": 9292,
      "words textual context": 134267,
      "transformerbased sequencetosequence model": 124439,
      "widely utilized various": 132598,
      "paper propose contextual": 84226,
      "representations neural networks": 99778,
      "neural networks proposed": 78544,
      "classification reading comprehension": 17368,
      "model significant improvements": 72018,
      "significant improvements traditional": 108811,
      "cnn rnn models": 18024,
      "various stateoftheart systems": 131205,
      "recognition ner existing": 96928,
      "structural semantic information": 113794,
      "improve models generalization": 52421,
      "generalization ability different": 46762,
      "different languages introduce": 31204,
      "knowledge learned pretraining": 58047,
      "pretrained models downstream": 90139,
      "models downstream tasks": 73110,
      "focuses applying endtoend": 44888,
      "applying endtoend dialog": 7241,
      "endtoend dialog technologies": 36889,
      "audio visual sceneaware": 10244,
      "visual sceneaware dialog": 131809,
      "paper describes task": 83852,
      "describes task definition": 29439,
      "task definition provided": 118054,
      "definition provided datasets": 28527,
      "evaluation setup track": 39390,
      "setup track summarize": 107860,
      "track summarize results": 122737,
      "summarize results submitted": 115614,
      "results submitted systems": 102221,
      "submitted systems highlight": 114776,
      "systems highlight overall": 116932,
      "highlight overall trends": 50268,
      "overall trends stateoftheart": 83266,
      "trends stateoftheart technologies": 125756,
      "stateoftheart technologies tasks": 113005,
      "generating coherent long": 47203,
      "distractors reading comprehension": 32605,
      "reading comprehension reading": 96009,
      "comprehension reading comprehension": 20225,
      "sequencetosequence seq2seq model": 107206,
      "guide decoder generate": 49232,
      "decoder generate coherent": 28013,
      "metrics achieving stateoftheart": 69919,
      "achieving stateoftheart performance": 2986,
      "stateoftheart performance human": 112824,
      "performance human evaluation": 86433,
      "human evaluation indicates": 50819,
      "compared distractors generated": 19360,
      "distractors generated baselines": 32603,
      "difficulties natural language": 31684,
      "present work annotation": 89768,
      "highly accurate performance": 50289,
      "nlp tools available": 79786,
      "news articles usually": 79306,
      "high performance terms": 50098,
      "understanding human behaviors": 126858,
      "useful range domains": 128920,
      "machine learningbased approaches": 65839,
      "especially spoken language": 38514,
      "spoken language paper": 111989,
      "neural network conduct": 78292,
      "network conduct extensive": 77209,
      "experiments verify effectiveness": 41201,
      "proposed framework outperforms": 93291,
      "framework outperforms stateoftheart": 45643,
      "stateoftheart methods large": 112710,
      "shorttext conversation neural": 108354,
      "researchers propose use": 100706,
      "introduce discrete latent": 56413,
      "discrete latent variable": 32167,
      "results indicate model": 101857,
      "generation models automatic": 47483,
      "features text classification": 43758,
      "vector machine algorithm": 131311,
      "text results suggest": 121265,
      "results suggest using": 102235,
      "recent neural language": 96477,
      "language models able": 59495,
      "model accurately efficiently": 70532,
      "data augmentation based": 25631,
      "based word embeddings": 12165,
      "small sample size": 110192,
      "features downstream applications": 43468,
      "downstream applications summarization": 33992,
      "bert model datasets": 13169,
      "enabling neural machine": 36415,
      "solve challenges propose": 110589,
      "according word alignments": 2010,
      "models stateoftheart transformer": 74097,
      "improvements bleu points": 52819,
      "robust word embeddings": 103084,
      "work propose simple": 134741,
      "propose simple extension": 93060,
      "embeddings outperform baseline": 35845,
      "wide range evaluation": 132504,
      "good performance standard": 48491,
      "language model performs": 59367,
      "responses experimental results": 101266,
      "demonstrate superiority proposed": 28883,
      "superiority proposed model": 115710,
      "stateoftheart translation performance": 113017,
      "alleviate problems propose": 5048,
      "new decoding algorithm": 78874,
      "sacrificing translation quality": 103530,
      "translation quality addition": 125163,
      "using connectionist temporal": 129475,
      "connectionist temporal classification": 21140,
      "performance work propose": 86865,
      "manner proposed method": 66959,
      "temporal classification ctc": 120097,
      "english japanese investigate": 37179,
      "sentencelevel parallel corpora": 106167,
      "bilingual lexicon extraction": 14044,
      "extraction model based": 42391,
      "translation quality language": 125176,
      "quality language pairs": 94700,
      "supported language pairs": 116024,
      "source language introduce": 110775,
      "different languages using": 31216,
      "mining social media": 70260,
      "social media sites": 110419,
      "using neural sequence": 129960,
      "modeling language modeling": 72456,
      "representation learning natural": 99300,
      "datasets different sizes": 27432,
      "lstm models perform": 65645,
      "transformer models perform": 124351,
      "analysis paper propose": 5740,
      "neural network performs": 78372,
      "sentiment analysis emotion": 106589,
      "analysis emotion analysis": 5592,
      "proposed approach based": 93185,
      "approach based bidirectional": 7379,
      "source external knowledge": 110759,
      "external knowledge improve": 42012,
      "benchmark dataset semeval": 12759,
      "compare stateoftheart systems": 19297,
      "performance sentiment analysis": 86702,
      "semantic syntactic properties": 105318,
      "neural network nn": 78365,
      "representations learned large": 99727,
      "learned large unlabeled": 62227,
      "plays essential role": 87729,
      "developing language independent": 30344,
      "word embeddings corpus": 133072,
      "word embeddings stateoftheart": 133212,
      "bag words cbow": 11363,
      "evaluation results demonstrate": 39367,
      "results demonstrate high": 101685,
      "demonstrate high quality": 28754,
      "high quality generated": 50117,
      "studies text classification": 114292,
      "focused english language": 44854,
      "text work propose": 121414,
      "text external knowledge": 120943,
      "dataset outperforms previous": 27080,
      "half million sentences": 49298,
      "experimental evaluation using": 40534,
      "significant improvements task": 108810,
      "lowresource languages lack": 65514,
      "challenge set evaluating": 16099,
      "formal informal language": 45163,
      "sentence embedding space": 105837,
      "text generation amr": 120978,
      "sequencetosequence models successfully": 107183,
      "knowledge pretrained model": 58118,
      "great success natural": 49029,
      "pretraining model bert": 90287,
      "large scale unlabeled": 61265,
      "monolingual data finetuning": 74571,
      "model labeled data": 71401,
      "knowledge paper propose": 58097,
      "proposed approach includes": 93199,
      "pretrained models improve": 90140,
      "nmt experimental results": 79863,
      "experimental results wmt": 40723,
      "wmt english german": 132786,
      "english german german": 37152,
      "german german english": 47896,
      "translation tasks model": 125356,
      "tasks model outperforms": 119300,
      "machine translation review": 66203,
      "translation mt automatic": 124978,
      "translation written text": 125437,
      "single neural network": 109774,
      "recent trends field": 96555,
      "effective data augmentation": 34657,
      "data augmentation approaches": 25630,
      "training taskoriented dialogue": 123906,
      "training data expensive": 123459,
      "data augmentation word": 25659,
      "word sentence level": 133539,
      "obtain significant improvement": 81321,
      "significant improvement strong": 108777,
      "improvement strong baseline": 52771,
      "approaches achieves stateoftheart": 8044,
      "model learn features": 71431,
      "languages evaluation resources": 60560,
      "learn multilingual representations": 62095,
      "compare word embeddings": 19318,
      "word embeddings obtained": 133172,
      "publicly available corpora": 94296,
      "collect high quality": 18379,
      "different architectures learn": 31005,
      "finegrained emotion classification": 44348,
      "sentiment analysis sa": 106646,
      "text deep learning": 120863,
      "graph convolution network": 48777,
      "anger disgust fear": 6107,
      "disgust fear surprise": 32363,
      "annotate new dataset": 6132,
      "new dataset chinese": 78861,
      "sequencelevel knowledge distillation": 107108,
      "train smaller models": 123025,
      "case models trained": 15601,
      "data augmentation regularization": 25643,
      "research previous works": 100586,
      "existing neural dialogue": 40227,
      "automatic evaluation dialogue": 10526,
      "evaluation dialogue systems": 39183,
      "neural module networks": 78207,
      "answering compositional questions": 6612,
      "questions require multiple": 95361,
      "interpretability word embeddings": 56235,
      "word embeddings substantially": 133216,
      "embeddings substantially successful": 35960,
      "substantially successful capturing": 114914,
      "successful capturing semantic": 115154,
      "word representations propose": 133468,
      "single task model": 109802,
      "model multitask setting": 71547,
      "modeling experimental results": 72429,
      "multitask learning methods": 75849,
      "order paper propose": 82380,
      "semantic representations sentences": 105238,
      "multiple sentences make": 75691,
      "existing stateoftheart systems": 40294,
      "stateoftheart systems benchmark": 112986,
      "datasets demonstrating effectiveness": 27419,
      "conduct thorough analysis": 20901,
      "entities social media": 37871,
      "paper present analysis": 84081,
      "analysis named entities": 5711,
      "named entities publiclyavailable": 76033,
      "tweet dataset turkish": 126012,
      "learning models relation": 62789,
      "models relation extraction": 73907,
      "representations yielded significant": 100002,
      "yielded significant improvements": 135368,
      "significant improvements nlp": 108797,
      "improvements nlp tasks": 52885,
      "improved relation extraction": 52634,
      "relation extraction using": 98033,
      "pretrained language representation": 90107,
      "applied word embeddings": 7147,
      "replacing static word": 98950,
      "word embeddings contextualized": 133069,
      "embeddings contextualized word": 35615,
      "neural networks translation": 78596,
      "obtained using statistical": 81428,
      "using statistical techniques": 130225,
      "documentlevel machine translation": 33150,
      "current state future": 25344,
      "state future directions": 112496,
      "possible avenues future": 88384,
      "extraction recent work": 42462,
      "standard multitask learning": 112271,
      "multitask learning effective": 75834,
      "current stateoftheart model": 25359,
      "carefully designed baselines": 15521,
      "language model recent": 59386,
      "language models shown": 59658,
      "effectiveness selfsupervised learning": 34947,
      "nlp tasks addition": 79705,
      "semantic nlp tasks": 105127,
      "achieve strong improvements": 2574,
      "largescale language modeling": 61456,
      "knowledge work investigate": 58239,
      "investigate extent pretrained": 56758,
      "pretrained models bert": 90135,
      "models trained new": 74221,
      "trained new objective": 123221,
      "applied downstream tasks": 7068,
      "downstream tasks model": 34045,
      "question answering datasets": 95037,
      "entity typing dataset": 38127,
      "opendomain dialogue generation": 81966,
      "systems attracted attention": 116742,
      "based conversation context": 11606,
      "modeling conversation context": 72407,
      "neural networks process": 78539,
      "structure extensive experiments": 113862,
      "effectively improve quality": 34821,
      "dialogue generation proposed": 30682,
      "dialog policy learning": 30580,
      "models limited training": 73509,
      "training data perform": 123510,
      "perform better baseline": 85954,
      "generative language models": 47730,
      "language model natural": 59353,
      "answer natural language": 6546,
      "modeling objective allows": 72496,
      "improvement classification accuracy": 52691,
      "random majority class": 95502,
      "majority class baselines": 66598,
      "contextualized multilingual word": 22560,
      "annotated data languages": 6161,
      "zeroshot transfer learning": 135533,
      "data scarcity lowresource": 26374,
      "lowresource languages work": 65531,
      "new method creating": 79010,
      "compare multiple baselines": 19262,
      "benchmark datasets approach": 12766,
      "improve summarization quality": 52556,
      "dataset rouge1 score": 27176,
      "perform sentiment analysis": 86064,
      "sentiment analysis extensively": 106595,
      "extensively studied various": 41970,
      "propose general approach": 92696,
      "sentiment analysis data": 106579,
      "texts multiple languages": 121557,
      "results sentiment analysis": 102156,
      "prediction using different": 89151,
      "different deep neural": 31084,
      "proposed ensemble model": 93270,
      "model achieved highest": 70548,
      "result f1 score": 101374,
      "new task called": 79198,
      "given seed term": 48111,
      "dataset 18 million": 26711,
      "demonstrate competitive results": 28692,
      "competitive results achieved": 19680,
      "pretraining text layout": 90343,
      "recent years despite": 96609,
      "despite widespread use": 29744,
      "understanding paper propose": 126915,
      "results downstream tasks": 101741,
      "artificial neural network": 9260,
      "little work exists": 64837,
      "exists natural language": 40347,
      "neural network ann": 78240,
      "essential building block": 38542,
      "language processing word": 59986,
      "embedding methods learn": 35443,
      "handle outofvocabulary oov": 49399,
      "gaussian mixture model": 46546,
      "word embedding learning": 133015,
      "intrinsic extrinsic tasks": 56363,
      "novel supervised model": 80742,
      "model neural architecture": 71567,
      "neural architecture consisting": 77832,
      "benchmark datasets furthermore": 12787,
      "word embeddings demonstrate": 133076,
      "zeroshot language transfer": 135508,
      "adapting deep learning": 3305,
      "deep learning sentiment": 28313,
      "learning sentiment classification": 63008,
      "tasks like sentiment": 119261,
      "sentiment classification work": 106721,
      "propose deep learningbased": 92619,
      "models compared existing": 72937,
      "classification models results": 17281,
      "models results proposed": 73955,
      "computationally efficient training": 20450,
      "embeddings sentiment analysis": 35931,
      "semantic representations words": 105240,
      "employ support vector": 36285,
      "significant margin evaluated": 108825,
      "models english corpora": 73155,
      "conventional word embeddings": 22915,
      "embeddings represent words": 35910,
      "dependencies different words": 29077,
      "existing models model": 40205,
      "models model effectively": 73576,
      "domain previous work": 33617,
      "detection scope resolution": 30043,
      "deep learningbased approaches": 28326,
      "architectures bert xlnet": 8784,
      "bert xlnet roberta": 13262,
      "task publicly available": 118589,
      "sfu review corpus": 107898,
      "substantial improvements previously": 114862,
      "previously reported results": 90620,
      "joint training model": 57328,
      "training model multiple": 123712,
      "negation detection scope": 76912,
      "native nonnative speakers": 76231,
      "binary classification model": 14118,
      "shortterm memory architecture": 108316,
      "model simultaneously predicts": 72052,
      "performance study shows": 86764,
      "promising results serve": 92300,
      "starting point future": 112409,
      "point future research": 87801,
      "research machine learning": 100551,
      "learn high quality": 62063,
      "high quality entity": 50115,
      "quality entity representations": 94646,
      "existing entity linking": 40129,
      "knowledge base domain": 57769,
      "domain training data": 33680,
      "sequencetosequence pretraining paper": 107196,
      "pretraining paper presents": 90302,
      "tokens time step": 122342,
      "summarization question generation": 115557,
      "question generation tasks": 95163,
      "compared models using": 19390,
      "machine transliteration using": 66314,
      "neural networks machine": 78511,
      "networks machine translation": 77654,
      "gained attention recent": 46358,
      "focus translating text": 44834,
      "translating text language": 124608,
      "single large neural": 109751,
      "large neural network": 61170,
      "neural network attention": 78252,
      "network attention mechanism": 77152,
      "attention mechanism sequencetosequence": 9913,
      "despite significant progress": 29728,
      "machine translation translation": 66281,
      "morphological richness language": 74730,
      "translation low resource": 124916,
      "resource language like": 100854,
      "expectation maximization em": 40383,
      "unsupervised language independent": 127648,
      "corpus need train": 23897,
      "models statistical machine": 74099,
      "machine translation include": 66007,
      "lstm transformer model": 65672,
      "hierarchical dirichlet process": 49955,
      "designed overcome problems": 29617,
      "outline future directions": 82631,
      "information electronic health": 54517,
      "domains limited training": 33807,
      "training data compared": 123444,
      "models available research": 72798,
      "source codes paper": 110730,
      "propose new approach": 92834,
      "sequence pair matching": 107025,
      "representations multiple levels": 99761,
      "portuguese word embeddings": 88204,
      "like word2vec glove": 64113,
      "single vector representation": 109812,
      "improve performance nlp": 52476,
      "collect large number": 18384,
      "compared existing approaches": 19365,
      "small set seed": 110199,
      "wide range downstream": 132502,
      "suffer semantic drift": 115243,
      "closely related target": 17864,
      "outperforms strong baseline": 83025,
      "strong baseline methods": 113632,
      "baseline methods significantly": 12252,
      "paper presents unsupervised": 84210,
      "extraction main contribution": 42382,
      "extraction tasks based": 42509,
      "messages social media": 68516,
      "multimodal sentiment analysis": 75458,
      "recent years nlp": 96626,
      "sequence labeling word": 106992,
      "order source language": 82409,
      "performance target languages": 86784,
      "finetuning multilingual bert": 44481,
      "labeling tasks experimental": 58548,
      "experimental results dialogue": 40604,
      "entity recognition tasks": 38080,
      "nlp tasks language": 79730,
      "variational inference enables": 130926,
      "experiment typologically diverse": 40515,
      "comparable better results": 19131,
      "better results stateoftheart": 13706,
      "models opendomain dialogue": 73665,
      "text sentiment classification": 121285,
      "computer vision learning": 20496,
      "novel architecture called": 80497,
      "imdb movie reviews": 51833,
      "analysing human language": 5477,
      "mean standard deviation": 67613,
      "recognition named entity": 96914,
      "neural network method": 78326,
      "bidirectional gated recurrent": 13922,
      "proposed model generates": 93450,
      "bidirectional encoder representation": 13908,
      "encoder representation transformersbert": 36565,
      "information captured hidden": 54406,
      "results model better": 101937,
      "methods mainly rely": 69610,
      "propose novel dynamic": 92914,
      "higher success rate": 50210,
      "introduce new largescale": 56485,
      "significantly improving stateoftheart": 108965,
      "improving stateoftheart methods": 53169,
      "manual annotation paper": 66977,
      "identify relevant entities": 51545,
      "early twentieth century": 34396,
      "text corpora study": 120832,
      "similarity deep learning": 109219,
      "stateoftheart language models": 112686,
      "language models bidirectional": 59521,
      "models bidirectional encoder": 72851,
      "encoder representation transformers": 36563,
      "representation transformers bert": 99452,
      "train test models": 123042,
      "maximum mutual information": 67544,
      "mutual information mmi": 75982,
      "nbest list generated": 76691,
      "decoding experimental results": 28092,
      "appropriate responses yielding": 8432,
      "responses yielding substantive": 101305,
      "yielding substantive gains": 135391,
      "substantive gains bleu": 114921,
      "gains bleu scores": 46387,
      "model training inference": 72220,
      "different pretrained models": 31350,
      "machine translation joint": 66017,
      "source target units": 110844,
      "recent neural machine": 96479,
      "translation tasks bleu": 125349,
      "fewer parameters code": 43890,
      "recommendation sentiment analysis": 97088,
      "using bertbased models": 129392,
      "models pretrained models": 73787,
      "multilingual pretrained models": 75339,
      "nlp tasks high": 79725,
      "high resource languages": 50131,
      "improving pretrained models": 53145,
      "multilingual bert zeroshot": 75211,
      "information extraction introduce": 54572,
      "introduce simple new": 56540,
      "language model finetuned": 59324,
      "model finetuned downstream": 71191,
      "finetuned downstream tasks": 44426,
      "endtoend information extraction": 36908,
      "model achieves superior": 70614,
      "achieves superior performance": 2919,
      "superior performance datasets": 115688,
      "representations human brain": 99679,
      "human concept representations": 50783,
      "models human language": 73342,
      "natural language use": 76595,
      "generation present generative": 47549,
      "present generative model": 89506,
      "shared set latent": 108012,
      "covering broad range": 24533,
      "broad range tasks": 14680,
      "fewshot learning setting": 43907,
      "new task demonstrate": 79199,
      "task demonstrate model": 118061,
      "like human beings": 64047,
      "messages readability instead": 68513,
      "readability instead long": 95948,
      "neural dialogue model": 77894,
      "introducing external knowledge": 56634,
      "word frequency distribution": 133293,
      "pos tagger parser": 88221,
      "using elmo embeddings": 129621,
      "event detection ed": 39508,
      "subtask information extraction": 114946,
      "information extraction aims": 54564,
      "previous studies task": 90500,
      "methods usually ignore": 69837,
      "dependency label information": 29150,
      "useful linguistic knowledge": 128902,
      "conduct experiments widely": 20862,
      "experiments widely used": 41207,
      "significant improvement competitive": 108769,
      "improvement competitive baseline": 52698,
      "framework address task": 45415,
      "learning train neural": 63113,
      "experiments multiple datasets": 41028,
      "transformers pretrained language": 124460,
      "varieties nlp tasks": 130959,
      "hundreds millions parameters": 51152,
      "transformer vaswani et": 124395,
      "al 2017 based": 4620,
      "based pretrained models": 11948,
      "small model student": 110173,
      "plays vital role": 87744,
      "used existing works": 128531,
      "stateoftheart baselines different": 112594,
      "glue benchmark tasks": 48312,
      "obtain competitive results": 81269,
      "distillation multilingual pretrained": 32515,
      "promising results compared": 92291,
      "distributional thesaurus embedding": 32741,
      "processing nlp community": 91729,
      "network embedding distributional": 77235,
      "embedding distributional thesaurus": 35399,
      "dialogue states dialogue": 30757,
      "language generation taskoriented": 59100,
      "crucial component taskoriented": 25136,
      "component taskoriented dialog": 20004,
      "generation nlg module": 47516,
      "typically relies heavily": 126452,
      "limited labelled data": 64247,
      "machine translation existing": 65978,
      "translation existing neural": 124805,
      "neural networks generate": 78488,
      "generate target translation": 47028,
      "training nmt model": 123750,
      "nmt model furthermore": 79898,
      "generation target word": 47655,
      "target word decoding": 117745,
      "experimental results widelyused": 40722,
      "significant improvements strong": 108808,
      "propose statistical model": 93083,
      "novel training procedure": 80755,
      "position embeddings selfattention": 88286,
      "stateoftheart results wide": 112933,
      "results wide range": 102327,
      "widely used benchmarks": 132566,
      "model arabic language": 70680,
      "arabic language understanding": 8521,
      "language morphologically rich": 59694,
      "arabic natural language": 8534,
      "bert based models": 13072,
      "stateoftheart results nlp": 112914,
      "compared multilingual bert": 19395,
      "allows model learn": 5172,
      "datasets various sizes": 27789,
      "brings significant improvements": 14652,
      "learning neural dialogue": 62838,
      "generation current stateoftheart": 47350,
      "neural dialogue systems": 77895,
      "stateoftheart models demonstrate": 112731,
      "processing nlp perspective": 91748,
      "addition previous works": 3462,
      "previous works used": 90580,
      "new evaluation framework": 78906,
      "present baseline results": 89393,
      "finally introduce new": 44200,
      "pretraining language model": 90269,
      "small large scale": 110169,
      "models trained corpus": 74195,
      "achieve excellent performance": 2456,
      "computational cost memory": 20369,
      "facilitate future work": 42773,
      "better performance language": 13659,
      "model based selfattention": 70744,
      "achieves stateoftheart accuracy": 2880,
      "additionally propose new": 3622,
      "translation best knowledge": 124673,
      "adaptation domain adaptation": 3217,
      "available new domain": 11054,
      "deep learning framework": 28273,
      "learning framework paper": 62604,
      "framework paper present": 45648,
      "fast domain adaptation": 43133,
      "model based gpt2": 70731,
      "diverse response generation": 32837,
      "deep learning led": 28277,
      "learning led significant": 62687,
      "led significant performance": 63286,
      "tasks models computationally": 119306,
      "classification tasks using": 17461,
      "7th dialog technology": 1281,
      "paper presents systems": 84209,
      "context documentlevel context": 22068,
      "extraction introduce new": 42359,
      "introduce new task": 56502,
      "crosslingual transfer limited": 25033,
      "deep neural architectures": 28349,
      "large amounts training": 61022,
      "segmentation words subword": 104656,
      "words subword units": 134246,
      "models based deep": 72804,
      "neural networks benefit": 78447,
      "obtains best results": 81456,
      "model performance paper": 71727,
      "languages mandarin chinese": 60720,
      "classes nouns verbs": 17068,
      "verbs adjectives adverbs": 131552,
      "crosslingual semantic similarity": 25004,
      "strong baseline results": 113636,
      "available website encourage": 11145,
      "sentiment analysis order": 106622,
      "widely used approaches": 132562,
      "word2vec fasttext glove": 133674,
      "evaluated using intrinsic": 39022,
      "using intrinsic extrinsic": 129767,
      "lack publicly available": 58738,
      "machine learning support": 65826,
      "learning support vector": 63076,
      "vector machine logistic": 131313,
      "machine logistic regression": 65841,
      "logistic regression naive": 65039,
      "regression naive bayes": 97706,
      "deep learning convolutional": 28264,
      "learning convolutional neural": 62465,
      "network recurrent neural": 77403,
      "attentionbased neural networks": 10091,
      "networks paper explores": 77689,
      "learning methods automatic": 62738,
      "rely manually engineered": 98722,
      "features external knowledge": 43504,
      "knowledge paper presents": 58096,
      "presents endtoend neural": 89847,
      "attention mechanism detect": 9896,
      "sentence pairs relevant": 105971,
      "quality empirical results": 94642,
      "neural model outperforms": 78152,
      "proposes novel framework": 93612,
      "models pretrained large": 73783,
      "pretrained large corpora": 90113,
      "unsupervised topic models": 127740,
      "word embeddings classify": 133065,
      "user generated content": 128987,
      "python natural language": 94482,
      "language processing toolkit": 59978,
      "existing widely used": 40332,
      "universal dependencies treebanks": 127294,
      "stanford corenlp software": 112365,
      "tasks coreference resolution": 119017,
      "systems essential step": 116876,
      "paper propose multilingual": 84267,
      "using automatic human": 129367,
      "human evaluation experimental": 50812,
      "experimental results multilingual": 40656,
      "model multiple languages": 71544,
      "modeling challenging task": 72388,
      "multilingual dialogue systems": 75243,
      "descriptions structured data": 29496,
      "unsupervised neural machine": 127681,
      "human evaluation performed": 50825,
      "participating teams teams": 85360,
      "person organization location": 87103,
      "transformer translation model": 124389,
      "number decoder layers": 80865,
      "translation quality experiments": 125174,
      "using newly collected": 129965,
      "semantic types entity": 105341,
      "existing entity typing": 40130,
      "context entire document": 22080,
      "model entity typing": 71089,
      "resources propose novel": 101030,
      "approach fully unsupervised": 7581,
      "data short supply": 26443,
      "requires large amounts": 100285,
      "learn source domain": 62153,
      "speech natural language": 111714,
      "present neural approach": 89571,
      "maximizing mutual information": 67526,
      "learn text representations": 62167,
      "text matching approach": 121103,
      "compared stateoftheart approaches": 19454,
      "experimental results text": 40713,
      "text matching tasks": 121105,
      "generation nlg models": 47515,
      "utterances work study": 130678,
      "model learn learning": 71433,
      "models able learn": 72644,
      "tasks dialogue response": 119058,
      "neural network rnnbased": 78394,
      "proposed models evaluated": 93494,
      "automatic metric human": 10600,
      "experimental results models": 40654,
      "tasks benchmark datasets": 118959,
      "spelling mistakes typos": 111909,
      "fundamental nlp tasks": 46104,
      "sentiment analysis textual": 106672,
      "identify shortcomings existing": 51554,
      "performance findings suggest": 86385,
      "approaches computational language": 8104,
      "widely spoken language": 132555,
      "choice translation language": 16909,
      "et al 2020": 38737,
      "unsupervised word segmentation": 127756,
      "paper proposes hybrid": 84356,
      "word segmentation neural": 133497,
      "godard et al": 48423,
      "challenging language pairs": 16266,
      "language pairs corpus": 59741,
      "nongoal oriented dialogues": 80210,
      "lack labeled data": 58722,
      "address introduce new": 3685,
      "introduce new labeled": 56484,
      "leveraging pretrained language": 63697,
      "language models model": 59602,
      "reading comprehension datasets": 95980,
      "research focused morphosyntactic": 100519,
      "text encoding initiative": 120906,
      "encoding initiative guidelines": 36700,
      "high resource language": 50129,
      "pairs low resource": 83579,
      "quality mt systems": 94724,
      "assess current state": 9473,
      "bleu score results": 14331,
      "useful research community": 128924,
      "mt systems text": 74996,
      "text representation models": 121248,
      "embeddings pretrained language": 35877,
      "models allow build": 72743,
      "rich representations text": 102778,
      "monolingual pretrained models": 74608,
      "fasttext word embeddings": 43208,
      "bert language models": 13152,
      "models trained larger": 74213,
      "produce better results": 91877,
      "results publicly available": 102101,
      "classification sentiment classification": 17395,
      "sets new stateoftheart": 107689,
      "work publicly available": 134759,
      "study present novel": 114475,
      "present novel methods": 89615,
      "models underresourced languages": 74267,
      "implemented python programming": 51973,
      "python programming language": 94486,
      "natural language toolkits": 76549,
      "measure similarity documents": 67826,
      "language use paper": 60300,
      "datadriven language mapping": 26663,
      "million words language": 70113,
      "paper addition describing": 83700,
      "webcrawled social media": 132281,
      "social media sources": 110420,
      "results best model": 101551,
      "new stateoftheart unsupervised": 79179,
      "large body research": 61043,
      "text generation structured": 121008,
      "generation structured data": 47641,
      "models significantly improved": 74045,
      "judged automatic metrics": 57428,
      "metrics human evaluation": 69967,
      "unseen slot values": 127545,
      "large volume entities": 61326,
      "containing large number": 21799,
      "large number natural": 61181,
      "number natural language": 80919,
      "synthetic data neural": 116620,
      "machine translation compared": 65927,
      "authentic parallel data": 10357,
      "parallel data training": 84658,
      "data training corpus": 26571,
      "parallel data generated": 84635,
      "parallel data noisy": 84643,
      "data noisy generated": 26177,
      "noisy generated imperfect": 80104,
      "large scale monolingual": 61252,
      "improvements experimental results": 52856,
      "parallel data nmt": 84642,
      "reinforcement learning strategy": 97828,
      "research existing methods": 100497,
      "existing methods generate": 40184,
      "experiments datasets proposed": 40871,
      "datasets proposed approach": 27633,
      "uses information retrieval": 129234,
      "desired language style": 29667,
      "train proposed model": 122999,
      "proposed model propose": 93475,
      "benchmark datasets languages": 12788,
      "text generation specifically": 121007,
      "language model gpt2": 59330,
      "experimental results english": 40614,
      "higher bleu score": 50167,
      "social media authors": 110352,
      "large scale dataset": 61245,
      "obtained social media": 81413,
      "generative dialogue systems": 47723,
      "dialogue systems opendomain": 30779,
      "attention past years": 9967,
      "improve correlation human": 52362,
      "proposed evaluation method": 93273,
      "models response generation": 73949,
      "stateoftheart matching models": 112698,
      "matching models proposed": 67417,
      "consistent performance improvements": 21396,
      "wikipedia revision histories": 132673,
      "histories paper propose": 50544,
      "propose method automatically": 92764,
      "dataset used training": 27257,
      "dataset dataset available": 26844,
      "perform test set": 86091,
      "sets contrast sets": 107660,
      "drop reading comprehension": 34171,
      "sentiment analysis contrast": 106577,
      "model performance significantly": 71729,
      "performance significantly lower": 86713,
      "downstream tasks introduce": 34042,
      "extraction paper present": 42422,
      "paper present corpus": 84090,
      "zeroshot learning text": 135511,
      "reinforcement learning algorithms": 97795,
      "siamese neural network": 108655,
      "network architecture novel": 77142,
      "multiple languages work": 75601,
      "existing conversational datasets": 40091,
      "data language models": 26072,
      "biomedical text classification": 14207,
      "zeroshot sequence labeling": 135525,
      "labeling convolutional decomposition": 58489,
      "multilabel classification task": 75152,
      "electronic health record": 35251,
      "effective classification model": 34644,
      "provide indepth analysis": 93851,
      "high school students": 50136,
      "finegrained named entity": 44368,
      "allows training evaluating": 5192,
      "relation extraction systems": 98028,
      "entity types relations": 38122,
      "work present corpus": 134678,
      "corpus english web": 23774,
      "web social media": 132260,
      "documents annotated according": 33181,
      "new benchmark datasets": 78814,
      "benchmark datasets tasks": 12807,
      "languages publicly available": 60828,
      "natural language evaluation": 76294,
      "aspect human language": 9332,
      "language understanding ability": 60202,
      "ability use language": 1576,
      "does worse low": 33411,
      "great progress improving": 49015,
      "showing large improvements": 108411,
      "dynamic data selection": 34303,
      "utilize monolingual data": 130517,
      "monolingual data neural": 74576,
      "target domain achieve": 117606,
      "dynamic curriculum learning": 34301,
      "iterative backtranslation models": 57127,
      "models domain adaptation": 73105,
      "language pairs experimental": 59756,
      "methods achieve improvements": 69278,
      "pretrained transformer language": 90195,
      "transformer language models": 124332,
      "text best knowledge": 120686,
      "languages english japanese": 60542,
      "new model named": 79033,
      "local global semantics": 64924,
      "previous work substantial": 90552,
      "work substantial margin": 134828,
      "transfer learning finetuning": 124093,
      "learning finetuning pretrained": 62586,
      "language models downstream": 59541,
      "paper propose effective": 84230,
      "propose effective way": 92642,
      "experiments diverse language": 40919,
      "sequence labeling sequence": 106982,
      "error correction approach": 38293,
      "automated metrics human": 10458,
      "pretrained masked language": 90123,
      "different tag sets": 31474,
      "recognition ner fundamental": 96929,
      "language understanding pipeline": 60251,
      "resources annotated data": 100943,
      "annotated data model": 6164,
      "different ner architectures": 31284,
      "sets new state": 107687,
      "produce stateoftheart results": 91937,
      "stateoftheart results joint": 112904,
      "embeddings pretrained large": 35879,
      "unlabeled data using": 127392,
      "additional unlabeled data": 3595,
      "unlabeled data train": 127390,
      "data train neural": 26566,
      "train neural model": 122983,
      "selftraining semisupervised technique": 104961,
      "technique incorporating unlabeled": 119794,
      "data sets new": 26430,
      "pretrained contextualized word": 90014,
      "task conduct experiments": 118003,
      "classification task classifying": 17427,
      "proposes novel approach": 93610,
      "improve classification performance": 52350,
      "classification performance evaluation": 17331,
      "evaluation results indicate": 39369,
      "text generation methods": 120989,
      "generation methods tend": 47475,
      "text generation task": 121011,
      "generation task aims": 47657,
      "task aims generate": 117867,
      "web search queries": 132254,
      "based external knowledge": 11709,
      "model make use": 71492,
      "pretrained natural language": 90170,
      "language modeling propose": 59458,
      "state tracking dialogue": 112526,
      "mitigate data scarcity": 70361,
      "data scarcity problem": 26375,
      "utterance used guide": 130619,
      "concrete use case": 20741,
      "reading comprehension questions": 96007,
      "meaning representations amrs": 67680,
      "representations amrs represent": 99500,
      "context information provided": 22145,
      "understanding generation existing": 126852,
      "question answering conversational": 95030,
      "autoregressive language model": 10865,
      "language model large": 59340,
      "generating new text": 47238,
      "stateoftheart results variety": 112928,
      "question generation squad": 95162,
      "processing systems recent": 91808,
      "models like bert": 73504,
      "new method generate": 79011,
      "scientific documents based": 104002,
      "delexicalized parser transfer": 28604,
      "results extensive evaluation": 101787,
      "model outperforms previously": 71663,
      "improve results neural": 52526,
      "results neural machine": 101984,
      "machine translation architecture": 65897,
      "study effect different": 114361,
      "effect different architectural": 34592,
      "level experiment different": 63443,
      "relation extraction present": 98012,
      "relation extraction dataset": 97984,
      "span multiple sentences": 111127,
      "multiple sentences argue": 75689,
      "design new metric": 29553,
      "datadriven machine learning": 26666,
      "compare performance models": 19275,
      "previous work introduced": 90533,
      "conversational agent able": 22986,
      "performance different architectures": 86296,
      "different architectures training": 31006,
      "performance compared models": 86232,
      "high quality word": 50125,
      "embeddings word alignments": 36027,
      "require parallel training": 100186,
      "data available propose": 25679,
      "key idea leverage": 57576,
      "word embeddings static": 133213,
      "relying parallel data": 98765,
      "language pairs compared": 59740,
      "contextualized embeddings achieve": 22545,
      "knowledge graphbased dialogue": 57977,
      "graphbased dialogue generation": 48897,
      "dialogue generation improved": 30675,
      "metalearning knowledge graphbased": 68552,
      "graphbased dialogue systems": 48898,
      "dialogue systems capable": 30768,
      "incompleteness knowledge graph": 53428,
      "paper proposes dynamic": 84353,
      "dialogue generation method": 30676,
      "knowledgeaware dialogue generation": 58245,
      "linguistic diversity world": 64469,
      "performance various downstream": 86841,
      "various downstream tasks": 131088,
      "patterns paper propose": 85768,
      "pretraining finetuning stage": 90260,
      "language modeling indomain": 59434,
      "propose novel selective": 92961,
      "experimental results sentiment": 40695,
      "achieve comparable better": 2440,
      "comparable better performance": 19130,
      "method effective efficient": 68790,
      "monolingual sentence embeddings": 74613,
      "using knowledge distillation": 129777,
      "model generate sentence": 71236,
      "embeddings source language": 35946,
      "source language train": 110785,
      "languages various language": 60964,
      "various language families": 131113,
      "distant supervised relation": 32458,
      "relation extraction relation": 98024,
      "existing approaches model": 40054,
      "solve problem paper": 110605,
      "results large scale": 101883,
      "augment existing relation": 10257,
      "existing relation extraction": 40267,
      "achieved considerable success": 2614,
      "considerable success natural": 21261,
      "transformerbased neural machine": 124433,
      "insights inner workings": 55538,
      "inner workings transformers": 55279,
      "machine translation unsupervised": 66286,
      "translation unsupervised neural": 125408,
      "machine translation unmt": 66284,
      "translation unmt recently": 125405,
      "unmt recently achieved": 127470,
      "recently achieved remarkable": 96651,
      "achieved remarkable results": 2688,
      "remarkable results language": 98846,
      "results language pairs": 101875,
      "single language pair": 109749,
      "multiple language pairs": 75590,
      "propose knowledge distillation": 92737,
      "knowledge distillation methods": 57869,
      "results surpassing strong": 102245,
      "language pairs supervised": 59781,
      "graphemetophoneme g2p conversion": 48925,
      "outperforms previous approaches": 82933,
      "achieves good performance": 2792,
      "generation tasks storytelling": 47674,
      "holtzman et al": 50606,
      "propose model called": 92786,
      "unsupervised approach outperforms": 127603,
      "make use deep": 66742,
      "intents slot labels": 55925,
      "intent classification ic": 55896,
      "slot filling sf": 110054,
      "datasets atis snips": 27321,
      "use pretrained language": 128205,
      "models elmo bert": 73126,
      "disentangled representation learning": 32349,
      "supervised approaches use": 115726,
      "disentangle semantics syntax": 32344,
      "low dimensional vector": 65358,
      "multihead selfattention mechanism": 75130,
      "selfattention mechanism learns": 104885,
      "input prior work": 55402,
      "attention heads pruned": 9852,
      "marginal performance degradation": 67209,
      "present user study": 89760,
      "ted talks english": 120052,
      "classification ctc loss": 17167,
      "training data avoid": 123435,
      "different loss function": 31240,
      "parallel text available": 84688,
      "train neural machine": 122980,
      "sequence models work": 107019,
      "different word segmentation": 31574,
      "word segmentation methods": 133495,
      "nmt model training": 79901,
      "nmt models trained": 79922,
      "language model types": 59414,
      "types language models": 126306,
      "language models gpt": 59562,
      "language model main": 59346,
      "downstream nlu tasks": 34018,
      "machine translation summarization": 66244,
      "summarization tasks experiments": 115578,
      "improving decoding efficiency": 53085,
      "autoencoder vae based": 10411,
      "vae based approaches": 130700,
      "diversity generated responses": 32876,
      "methods usually suffer": 69839,
      "propose novel multimodal": 92947,
      "prior posterior distributions": 90724,
      "experimental results widely": 40720,
      "results widely used": 102332,
      "datasets demonstrate model": 27409,
      "demonstrate model achieves": 28789,
      "improvements compared strong": 52831,
      "compared strong baselines": 19469,
      "learning unlabeled data": 63141,
      "unlabeled data target": 127388,
      "target language better": 117633,
      "tackle named entity": 117296,
      "knowledge learned source": 58049,
      "labeled data source": 58443,
      "train student model": 123030,
      "target language proposed": 117650,
      "proposed method works": 93407,
      "datasets demonstrate method": 27407,
      "method outperforms existing": 69016,
      "recent advances nlp": 96418,
      "largescale pretrained language": 61494,
      "language models gpt2": 59563,
      "lefttoright language modeling": 63308,
      "linguistic quality generated": 64538,
      "quality generated text": 94676,
      "explicit discourse relations": 41317,
      "discourse relations text": 32080,
      "information different modalities": 54488,
      "modalities text images": 70483,
      "generative dialog modeling": 47721,
      "challenging task information": 16321,
      "improves model performance": 52995,
      "measured automatic human": 67834,
      "available project gutenberg": 11081,
      "finally conduct experiments": 44159,
      "conduct experiments showing": 20857,
      "settings training data": 107841,
      "training data larger": 123483,
      "based edit distance": 11668,
      "experiments chinese english": 40821,
      "english german english": 37150,
      "common sense language": 18923,
      "embedding method named": 35440,
      "learn word embedding": 62191,
      "machine translation generate": 65992,
      "supervised pos taggers": 115822,
      "partofspeech pos taggers": 85533,
      "lowresource languages languages": 65515,
      "approaches pos tagging": 8284,
      "tackle problem paper": 117305,
      "position representations model": 88292,
      "latent structure input": 61614,
      "structure input sentence": 113883,
      "demonstrate approach significantly": 28672,
      "approach significantly consistently": 7898,
      "significantly consistently improves": 108900,
      "consistently improves translation": 21422,
      "translation quality strong": 125188,
      "extensive analyses confirm": 41853,
      "analyses confirm performance": 5446,
      "input words improve": 55475,
      "modeling sentences paper": 72543,
      "mutual information mi": 75981,
      "experiments text classification": 41178,
      "languages multilingual bert": 60743,
      "supervised zeroshot crosslingual": 115859,
      "experiments named entity": 41031,
      "largescale benchmark dataset": 61401,
      "text classification fundamental": 120724,
      "fundamental natural language": 46101,
      "produce embeddings unseen": 91887,
      "embeddings unseen words": 36000,
      "benchmark datasets method": 12792,
      "outperforms stateoftheart text": 83019,
      "stateoftheart text classification": 113007,
      "systematic empirical analysis": 116670,
      "language dataset creation": 58936,
      "news topic classification": 79382,
      "topic classification task": 122501,
      "document work propose": 33112,
      "approach data augmentation": 7470,
      "social media popular": 110411,
      "introduce new publicly": 56496,
      "new publicly available": 79092,
      "data set tweets": 26415,
      "learning models automatically": 62773,
      "computational lexical semantics": 20388,
      "paper proposes unifying": 84370,
      "different types semantic": 31539,
      "prior work crosslingual": 90747,
      "language model probing": 59382,
      "set expansion methods": 107439,
      "key challenge entity": 57543,
      "propose novel iterative": 92931,
      "probing pretrained language": 90904,
      "stateoftheart methods significantly": 112715,
      "pretrained models chinese": 90138,
      "representations transformers bert": 99946,
      "performance pretrained language": 86615,
      "chinese pretrained language": 16807,
      "carried extensive experiments": 15539,
      "stateoftheart performances nlp": 112862,
      "performances nlp tasks": 86889,
      "findings help future": 44295,
      "help future research": 49727,
      "research resources available": 100619,
      "available training neural": 11135,
      "good correlation human": 48471,
      "correlation human judgment": 24232,
      "improves quality neural": 53035,
      "modeling long context": 72467,
      "taskoriented dialogue state": 118893,
      "dialogue state generation": 30746,
      "based recently proposed": 11975,
      "model simple effective": 72043,
      "learn better representation": 61994,
      "relative improvement baseline": 98358,
      "establishing new stateoftheart": 38620,
      "stateoftheart joint goal": 112679,
      "joint goal accuracy": 57279,
      "multiwoz 20 dataset": 75953,
      "additional bilingual data": 3495,
      "existing data augmentation": 40100,
      "method effectively boost": 68792,
      "models recent studies": 73875,
      "hidden representations learned": 49907,
      "different time steps": 31500,
      "methods human evaluations": 69536,
      "language models available": 59507,
      "automatic humanbased evaluation": 10563,
      "sentences human evaluation": 106344,
      "language model baseline": 59297,
      "existing automatic evaluation": 40072,
      "opendomain dialogue response": 81967,
      "response selection propose": 101240,
      "propose method construct": 92768,
      "human evaluation compared": 50807,
      "compared widely used": 19490,
      "widely used automatic": 132563,
      "fundamental nlp task": 46103,
      "stateoftheart models achieved": 112724,
      "models achieved high": 72681,
      "achieved high accuracy": 2638,
      "corpora different genres": 23458,
      "stateoftheart tagging model": 112997,
      "model trained different": 72187,
      "data multiple sources": 26146,
      "small amounts indomain": 110130,
      "amounts indomain data": 5340,
      "data order magnitude": 26191,
      "present state art": 89716,
      "data error analysis": 25902,
      "features key success": 43578,
      "word alignment method": 132907,
      "using multilingual bert": 129915,
      "style question answering": 114590,
      "solve problem using": 110611,
      "multilingual bert finetuned": 75202,
      "significantly outperformed previous": 108997,
      "achieved f1 score": 2623,
      "higher previous stateoftheart": 50195,
      "stateoftheart supervised methods": 112980,
      "comprehension paper propose": 20210,
      "pretrained multilingual representations": 90167,
      "using integer linear": 129762,
      "parallel sentence extraction": 84678,
      "parallel corpora language": 84602,
      "proposed method using": 93406,
      "using bidirectional rnn": 129406,
      "particularly distant language": 85479,
      "information extraction multilingual": 54581,
      "transfer learning study": 124134,
      "argument role labeling": 8957,
      "best previously published": 13418,
      "downstream mt performance": 34004,
      "comparable corpora method": 19139,
      "does require human": 33395,
      "single model trained": 109765,
      "dataset encourage research": 26890,
      "generation language models": 47454,
      "generation successfully learned": 47646,
      "largescale language model": 61455,
      "method outperforms prior": 69024,
      "come long way": 18769,
      "evaluation downstream applications": 39195,
      "summarization dialog systems": 115494,
      "dialog systems need": 30600,
      "performance downstream applications": 86317,
      "downstream applications investigate": 33989,
      "translation mt models": 124981,
      "present endtoend neural": 89471,
      "transformer model trained": 124344,
      "model trained supervised": 72211,
      "text simplification ts": 121307,
      "models heavily rely": 73322,
      "quality parallel sentences": 94748,
      "different languages domains": 31197,
      "leverage large amounts": 63595,
      "language modeling automatic": 59425,
      "automatic generation parallel": 10546,
      "generation parallel data": 47533,
      "parallel data iterative": 84637,
      "data iterative backtranslation": 26056,
      "different types sentences": 31540,
      "tackle problem propose": 117307,
      "introduce different types": 56411,
      "noise training process": 80072,
      "performance model trained": 86536,
      "model trained unsupervised": 72216,
      "unsupervised semisupervised manner": 127712,
      "outperforms previous systems": 82955,
      "model perform competitively": 71719,
      "dialogue systems natural": 30777,
      "corresponding natural language": 24297,
      "improve performance dual": 52466,
      "supervised learning framework": 115769,
      "prior work learned": 90751,
      "learning framework effectively": 62600,
      "unsupervised learning algorithms": 127653,
      "paper introduce largescale": 83994,
      "dataset facilitate study": 26927,
      "knowledge graph kg": 57968,
      "models severely suffer": 74018,
      "fully utilize information": 45980,
      "transferability outperforming prior": 124216,
      "outperforming prior best": 82824,
      "prior best model": 90694,
      "adapting new domain": 3320,
      "detection fake news": 29953,
      "learning crosslingual transfer": 62475,
      "crosslingual transfer semantic": 25039,
      "language lowresource language": 59261,
      "language parallel data": 59799,
      "parallel data train": 84657,
      "data train semantic": 26567,
      "train semantic parser": 123011,
      "data make use": 26113,
      "learning framework different": 62599,
      "different multitask learning": 31275,
      "large annotated datasets": 61033,
      "language prohibitively expensive": 59994,
      "commercial machine translation": 18836,
      "english nli datasets": 37225,
      "datasets address core": 27301,
      "training set large": 123849,
      "code models data": 18120,
      "model data augmentation": 70948,
      "generation taskoriented dialogue": 47664,
      "virtual assistants google": 131743,
      "assistants google assistant": 9579,
      "amazon alexa apple": 5265,
      "alexa apple siri": 4670,
      "large number services": 61188,
      "number services apis": 80965,
      "language work investigate": 60347,
      "work investigate methods": 134592,
      "use small number": 128282,
      "human evaluation method": 50820,
      "lexical semantic recognition": 63813,
      "task effective way": 118128,
      "models evaluation metrics": 73178,
      "modeling lexical semantics": 72461,
      "future work area": 46321,
      "data augmentation improve": 25635,
      "nlp neural models": 79654,
      "using data augmentation": 129530,
      "data augmentation generating": 25634,
      "generating training examples": 47282,
      "test hypothesis data": 120461,
      "neural networks significant": 78577,
      "sensitivity syntactic structure": 105755,
      "performance subjectverb agreement": 86766,
      "subjectverb agreement prediction": 114712,
      "agreement prediction task": 4399,
      "indic languages present": 53830,
      "10 indian languages": 136,
      "embeddings significantly outperform": 35939,
      "publicly available pretrained": 94318,
      "finetuning language model": 44469,
      "performance language model": 86481,
      "propose new paradigm": 92878,
      "word prediction models": 133414,
      "sentences high accuracy": 106341,
      "language models monolingual": 59604,
      "models multilingual bert": 73591,
      "question answering using": 95123,
      "question answering genqa": 95050,
      "new metric evaluating": 79020,
      "metric significantly higher": 69902,
      "significantly higher correlation": 108920,
      "higher correlation human": 50172,
      "machine translation day": 65943,
      "german french italian": 47893,
      "recent work using": 96595,
      "neural networks task": 78588,
      "problem introduce new": 91090,
      "human evaluation task": 50839,
      "metrics different datasets": 69952,
      "medical entity linking": 68205,
      "entity linking task": 37963,
      "linking task identifying": 64673,
      "unstructured text existing": 127584,
      "text existing methods": 120928,
      "existing methods adopt": 40176,
      "candidate generation module": 15105,
      "improves entity linking": 52976,
      "benchmark datasets address": 12765,
      "entity linking datasets": 37955,
      "make source code": 66728,
      "source code datasets": 110719,
      "code datasets publicly": 18095,
      "response generation current": 101201,
      "response generation process": 101210,
      "relevance informativeness generated": 98512,
      "transformer based model": 124304,
      "attention mechanism trained": 9914,
      "dataset outperforms strong": 27082,
      "language understanding recent": 60260,
      "understanding recent years": 126936,
      "entity recognition questionanswering": 38062,
      "introduce new sentiment": 56500,
      "new sentiment analysis": 79130,
      "models generalize different": 73276,
      "best average performance": 13308,
      "finally provide extensive": 44228,
      "multilingual transformerbased models": 75393,
      "computational social science": 20433,
      "history social media": 50569,
      "natural language limited": 76369,
      "diverse natural language": 32826,
      "language models plms": 59621,
      "pretrained model finetuning": 90132,
      "outperforms previous state": 82944,
      "high translation quality": 50143,
      "outofdomain dependency parsing": 82653,
      "used training data": 128825,
      "training data time": 123553,
      "recognition ner model": 96933,
      "ner model trained": 77054,
      "trained using annotated": 123326,
      "training ner models": 123730,
      "ner models using": 77056,
      "models trained monolingual": 74215,
      "large number parameters": 61184,
      "processing tasks given": 91815,
      "work attempt explore": 134384,
      "measuring progress natural": 67924,
      "learning training set": 63117,
      "evaluation test set": 39421,
      "training set paradigm": 123851,
      "particular data set": 85399,
      "contextual word representation": 22521,
      "lower higher layers": 65431,
      "finetuning downstream tasks": 44455,
      "unconditional text generation": 126642,
      "generation train model": 47683,
      "training data different": 123450,
      "evaluation understudy bleu": 39430,
      "generative pretrained transformer": 47752,
      "significantly better existing": 108886,
      "architecture training procedure": 8763,
      "minimum risk training": 70213,
      "appear natural language": 6800,
      "sentences paper introduce": 106425,
      "entity types trained": 38124,
      "10 f1 score": 133,
      "f1 score code": 42679,
      "probes supervised models": 90885,
      "supervised models designed": 115808,
      "probe structural probe": 90879,
      "syntactic information encoded": 116419,
      "controllable text generation": 22831,
      "prior work controllable": 90744,
      "work controllable text": 134442,
      "techniques used perform": 120009,
      "analysis advantages disadvantages": 5490,
      "paper focus task": 83952,
      "named entity linking": 76048,
      "entity linking nel": 37960,
      "knowledge base population": 57777,
      "english knowledge bases": 37181,
      "knowledge bases wikidata": 57812,
      "knowledge knowledge base": 58030,
      "present knowledge base": 89524,
      "entity linking module": 37959,
      "information relation extraction": 54914,
      "results demonstrate efficiency": 101681,
      "baselines benchmark datasets": 12361,
      "benchmark datasets make": 12790,
      "analyze pretrained models": 5994,
      "datasets research community": 27678,
      "depends quality quantity": 29293,
      "complexsimple sentence pairs": 19948,
      "sentence alignment quality": 105768,
      "create manually annotated": 24625,
      "text simplification corpora": 121300,
      "model capture semantic": 70808,
      "capture semantic similarity": 15384,
      "hierarchical text classification": 50014,
      "adapted beam search": 3282,
      "wellknown english datasets": 132421,
      "shared tasks 2018": 108111,
      "future research efforts": 46300,
      "speech directed children": 111677,
      "variation human language": 130893,
      "task performance models": 118529,
      "generalize better results": 46804,
      "use computational methods": 127950,
      "models lms trained": 73522,
      "produce stateoftheart performance": 91936,
      "stateoftheart performance little": 112828,
      "improves current stateoftheart": 52969,
      "stateoftheart models trained": 112752,
      "increasing beam size": 53666,
      "reduce exposure bias": 97328,
      "indomain test sets": 53983,
      "robustness domain shift": 103095,
      "syntactic generalization neural": 116411,
      "generalization neural language": 46785,
      "achieve lower perplexity": 2497,
      "proper syntactic generalizations": 92424,
      "tremendously recent years": 125739,
      "machine translation crosslingual": 65934,
      "leverage parallel data": 63611,
      "parallel data automatically": 84633,
      "demonstrate effectiveness framework": 28711,
      "range downstream tasks": 95564,
      "comprehensive experiments demonstrate": 20263,
      "phraselevel sentiment classification": 87404,
      "classification tasks conduct": 17443,
      "active learning approach": 3111,
      "active learning training": 3118,
      "corpora language models": 23511,
      "knowledge fixed number": 57937,
      "retrieval machine reading": 102404,
      "retrieval pretrained language": 102421,
      "facts expressed natural": 42910,
      "work machine reading": 134623,
      "present neural framework": 89573,
      "machine learning solutions": 65825,
      "train deep neural": 122921,
      "perform ablation study": 85940,
      "generation neural conversation": 47505,
      "document previous work": 33055,
      "humor paper describes": 51141,
      "available task organizers": 11123,
      "paper describes architecture": 83810,
      "integrating semantic structural": 55802,
      "structural information graph": 113771,
      "posts social media": 88523,
      "existing methods fail": 40182,
      "overcome limitations propose": 83294,
      "extensive experiments realworld": 41928,
      "datasets demonstrate models": 27411,
      "models outperform existing": 73677,
      "outperform existing methods": 82708,
      "nli task determining": 79553,
      "nli datasets contain": 79541,
      "recent deep learningbased": 96449,
      "deep learningbased models": 28328,
      "performing natural language": 86962,
      "best performance overall": 13394,
      "qualitative analysis shows": 94555,
      "training data text": 123551,
      "data text classification": 26559,
      "word embeddings map": 133158,
      "embeddings map words": 35803,
      "results compared standard": 101603,
      "creative commons license": 24727,
      "case study analyzing": 15613,
      "meaningful semantic representations": 67728,
      "used future studies": 128563,
      "henderson et al": 49839,
      "challenging data set": 16240,
      "previous research demonstrated": 90447,
      "al 2012 paper": 4604,
      "spoken dialogue sds": 111979,
      "evaluate models using": 38871,
      "addition improving performance": 3438,
      "improving performance models": 53136,
      "recent years common": 96603,
      "common topic discussion": 18940,
      "classification tasks propose": 17454,
      "human sentence processing": 50962,
      "networks rnns long": 77746,
      "rnns natural language": 102975,
      "little known ability": 64815,
      "model human language": 71291,
      "largescale multidocument summarization": 61467,
      "multidocument summarization dataset": 75079,
      "multidocument summarization mds": 75082,
      "training supervised models": 123894,
      "work presents new": 134697,
      "presents new dataset": 89875,
      "provide quantitative analysis": 93903,
      "empirical results stateoftheart": 36193,
      "previous work automatic": 90521,
      "timeline summarization tls": 122158,
      "paper compare different": 83765,
      "evaluation present new": 39331,
      "transformer shown outperform": 124380,
      "neural networkbased sequencetosequence": 78432,
      "morphological inflection generation": 74709,
      "historical text normalization": 50536,
      "large batch size": 61036,
      "leads better generalization": 61921,
      "large improvements outofdomain": 61109,
      "baseline future research": 12228,
      "research transfer learning": 100652,
      "processing nlp including": 91736,
      "fields machine translation": 44001,
      "pairs germanenglish chineseenglish": 83555,
      "quality extracted bilingual": 94665,
      "models use knowledge": 74277,
      "knowledge base model": 57774,
      "output test time": 83135,
      "achieved state art": 2700,
      "results morphological tagging": 101962,
      "systems explicitly model": 116895,
      "explicitly model internal": 41379,
      "model internal structure": 71361,
      "labels seen training": 58639,
      "neural models perform": 78189,
      "openended text generation": 81990,
      "standard seq2seq models": 112303,
      "test models ability": 120475,
      "terms automatic human": 120272,
      "human evaluation metrics": 50822,
      "problem sequence labelling": 91220,
      "performance state art": 86740,
      "state art supervised": 112477,
      "multiple language models": 75589,
      "different data sources": 31075,
      "language model finally": 59323,
      "positive unlabeled learning": 88346,
      "learning named entity": 62822,
      "lack annotated datasets": 58681,
      "coverage existing knowledge": 24507,
      "existing knowledge resources": 40149,
      "model achieves average": 70553,
      "language vice versa": 60331,
      "information social media": 54989,
      "adverse drug reactions": 4190,
      "introduce new data": 56478,
      "result test set": 101413,
      "general language model": 46668,
      "language model taskspecific": 59407,
      "paper conduct empirical": 83776,
      "based experimental results": 11700,
      "universal dependencies shared": 127291,
      "dependencies shared task": 29116,
      "shared task iwpt": 108057,
      "task iwpt 2020": 118321,
      "given passage question": 48082,
      "representation models bidirectional": 99335,
      "contextualized language representations": 22557,
      "build powerful language": 14799,
      "powerful language models": 88667,
      "transformerbased models bert": 124426,
      "resources paper proposes": 101027,
      "stateoftheart performance compared": 112814,
      "multilingual models data": 75286,
      "nlp tasks pretraining": 79752,
      "improves stateoftheart performance": 53049,
      "outperforming multilingual bert": 82811,
      "works sentiment analysis": 134976,
      "sentiment analysis text": 106670,
      "analysis text classification": 5896,
      "natural language questions": 76521,
      "natural language adversarial": 76265,
      "attack success rate": 9705,
      "data improve performance": 26026,
      "improve performance robustness": 52483,
      "sentence gives rise": 105894,
      "new dataset annotated": 78858,
      "universal dependency treebank": 127299,
      "partofspeech tags dependency": 85566,
      "tags dependency relations": 117471,
      "parsing models using": 85163,
      "using biaffine attention": 129396,
      "entity recognition french": 38017,
      "related named entities": 97883,
      "powerful neural networks": 88674,
      "semantically annotated corpus": 105363,
      "models learn parse": 73486,
      "language modeling work": 59484,
      "shen et al": 108168,
      "et al 2018a": 38726,
      "unlabeled training data": 127414,
      "classification opendomain conversational": 17314,
      "opendomain conversational systems": 81962,
      "address problem introduce": 3736,
      "information enrich utterance": 54533,
      "amounts labeled training": 5344,
      "training data required": 123526,
      "effective method generating": 34705,
      "method generating synthetic": 68862,
      "generating synthetic training": 47273,
      "synthetic training data": 116650,
      "limited amounts labeled": 64203,
      "knowledge bases generate": 57804,
      "approach previous stateoftheart": 7817,
      "collected amazon alexa": 18397,
      "results detailed analysis": 101719,
      "improve translation accuracy": 52569,
      "paper propose apply": 84220,
      "significantly outperforming strong": 109000,
      "fewshot performance reaching": 43909,
      "performance nlp datasets": 86574,
      "news articles human": 79296,
      "generating dialogue responses": 47213,
      "domain specific resources": 33651,
      "based background knowledge": 11542,
      "new task received": 79207,
      "task received attention": 118607,
      "capture structural information": 15400,
      "capture sequential information": 15393,
      "representations capture semantic": 99531,
      "capture semantic information": 15379,
      "model nlp tasks": 71586,
      "evolving research area": 39706,
      "possible future research": 88405,
      "future research directions": 46298,
      "propose new datasets": 92843,
      "based neural language": 11887,
      "paper present largescale": 84109,
      "types semantic relations": 126355,
      "generated different models": 47074,
      "knowledge bases work": 57814,
      "analysis usually relies": 5932,
      "visual textual information": 131814,
      "using simple effective": 130177,
      "different modalities compared": 31261,
      "experiment results models": 40501,
      "classification task input": 17433,
      "capture longdistance relationships": 15337,
      "paper present systematic": 84142,
      "present systematic study": 89734,
      "bert models languages": 13179,
      "achieves best reported": 2741,
      "classification problem mainly": 17348,
      "high classification accuracy": 50041,
      "high dimensional sparse": 50064,
      "order solve problem": 82406,
      "documents experimental results": 33228,
      "language models human": 59566,
      "human reading behavior": 50949,
      "predict human reading": 88892,
      "eye movement data": 42633,
      "methods controlled psycholinguistic": 69406,
      "results suggest different": 102230,
      "semantic space enables": 105303,
      "sign language modelling": 108679,
      "modelling automatic speech": 72598,
      "speech recognition spoken": 111779,
      "work transfer learning": 134859,
      "lstm models trained": 65646,
      "penn treebank corpus": 85861,
      "functional distributional semantics": 46057,
      "distributional semantics functional": 32726,
      "semantics functional distributional": 105419,
      "corpus previous work": 23940,
      "speech acts framework": 111643,
      "graphs contrast previous": 48945,
      "serve training data": 107308,
      "data various types": 26619,
      "texts different genres": 121499,
      "entity recognition coreference": 38007,
      "recognition coreference resolution": 96848,
      "extensive research focused": 41953,
      "paper conduct comprehensive": 83775,
      "comprehensive empirical evaluation": 20257,
      "representation methods using": 99327,
      "using pretrained language": 130045,
      "state art variety": 112483,
      "machine translation question": 66182,
      "present language models": 89529,
      "model evaluate models": 71107,
      "despite success existing": 29735,
      "information input document": 54690,
      "paper propose improve": 84252,
      "hierarchical attention network": 49933,
      "encoder experimental results": 36525,
      "dataset model significantly": 27034,
      "translation mt shown": 124985,
      "professional human translation": 92079,
      "paper presents comprehensive": 84158,
      "automated human evaluation": 10445,
      "languages use multilingual": 60947,
      "extracted common crawl": 42146,
      "languages partofspeech tagging": 60796,
      "data embeddings trained": 25885,
      "embeddings multilingual bert": 35815,
      "text classification work": 120788,
      "model automatic evaluation": 70710,
      "given context existing": 47999,
      "approach resolve problem": 7862,
      "propose automatic evaluation": 92567,
      "existing evaluation metrics": 40134,
      "number native speakers": 80917,
      "make accurate predictions": 66619,
      "prior studies shown": 90736,
      "better paper explore": 13646,
      "binary classification using": 14122,
      "classification using support": 17493,
      "datasets proposed method": 27639,
      "works usually ignore": 134991,
      "propose approach automatically": 92548,
      "provides natural way": 94054,
      "past work relation": 85654,
      "work relation extraction": 134769,
      "arguments propose new": 8993,
      "approaches neural machine": 8254,
      "human evaluation compare": 50806,
      "nmt systems chinese": 79974,
      "embeddings lowresource languages": 35795,
      "largescale annotated data": 61399,
      "probing task design": 90908,
      "better transfer languages": 13749,
      "choosing transfer languages": 16934,
      "transfer languages crosslingual": 124070,
      "plays significant role": 87742,
      "significant role determining": 108864,
      "hundreds thousands examples": 51156,
      "models supervised learning": 74133,
      "new event types": 78911,
      "fewshot learning models": 43906,
      "learning models extensive": 62780,
      "method improve performance": 68886,
      "experiment different architectures": 40463,
      "employ graph neural": 36263,
      "extensive experiments different": 41910,
      "stateoftheart methods automatic": 112705,
      "methods automatic human": 69329,
      "structures syntactic semantic": 114107,
      "steep learning curve": 113227,
      "model achieved accuracy": 70544,
      "arabic text classification": 8560,
      "morphological analysis word": 74684,
      "handcrafted feature engineering": 49337,
      "using characterlevel features": 129437,
      "features propose novel": 43673,
      "imagebased character encoder": 51805,
      "dataset best knowledge": 26767,
      "text classification present": 120755,
      "standard arabic colloquial": 112188,
      "task aim identify": 117861,
      "entities mentioned text": 37819,
      "largescale eventuality knowledge": 61436,
      "propose scalable approach": 93025,
      "experiments analysis demonstrate": 40766,
      "analysis demonstrate effectiveness": 5562,
      "datasets code available": 27348,
      "training sentence pairs": 123836,
      "englishfrench englishgerman translation": 37364,
      "solve challenging tasks": 110591,
      "tasks word sense": 119606,
      "previous work semantic": 90546,
      "wordnet knowledge graph": 133773,
      "existing semantic similarity": 40277,
      "relations knowledge graph": 98194,
      "different granularity levels": 31171,
      "work tackles problem": 134842,
      "entropy minimization adversarial": 38168,
      "new approach task": 78798,
      "ambiguous word particular": 5299,
      "word particular context": 133400,
      "using neural language": 129948,
      "propose methods combining": 92780,
      "established new stateoftheart": 38606,
      "datasets large margin": 27541,
      "target languages english": 117667,
      "languages english different": 60533,
      "realworld applications including": 96150,
      "neural networks sentiment": 78571,
      "using distant supervision": 129591,
      "distant supervision sentiment": 32479,
      "knowledge base previous": 57779,
      "base previous studies": 11470,
      "showed proposed approach": 108386,
      "approach performed better": 7794,
      "better existing methods": 13581,
      "learning models support": 62792,
      "models support vector": 74135,
      "neural networks framework": 78484,
      "generation model trained": 47481,
      "model trained learn": 72199,
      "evaluation model trained": 39299,
      "model trained generate": 72194,
      "documents paper present": 33264,
      "model manually corrected": 71500,
      "text editing propose": 120897,
      "editing propose novel": 34554,
      "text editing task": 120899,
      "approach address problem": 7334,
      "augmentation language models": 10284,
      "language models text": 59668,
      "relative reduction perplexity": 98370,
      "event coreference task": 39502,
      "task generating informative": 118241,
      "attracted growing attention": 10149,
      "informative coherent fluent": 55130,
      "work focuses building": 134537,
      "reading comprehension systems": 96013,
      "requiring manually annotated": 100350,
      "training development data": 123582,
      "crosslingual transfer approach": 25028,
      "reading comprehension language": 95986,
      "largescale language models": 61457,
      "machine translation english": 65972,
      "corpora target language": 23597,
      "target language work": 117663,
      "levels experimental results": 63534,
      "terms exact match": 120320,
      "exact match score": 39720,
      "model based ensemble": 70730,
      "leverage different types": 63584,
      "models ensemble methods": 73159,
      "approach achieved positive": 7300,
      "place 26 teams": 87600,
      "textual similarity dataset": 121738,
      "downstream applications existing": 33988,
      "sentence pairs collected": 105968,
      "million sentence pairs": 70106,
      "sentence pairs using": 105972,
      "different sampling strategies": 31396,
      "sampling strategies used": 103613,
      "deep language understanding": 28244,
      "work introduces machine": 134587,
      "unigram bigram trigram": 127154,
      "accuracy test set": 2302,
      "text generation present": 120997,
      "open domain semantic": 81898,
      "domain semantic parsing": 33634,
      "recognition algorithm based": 96804,
      "based phrase window": 11928,
      "endtoend method based": 36922,
      "method based supervised": 68667,
      "based supervised learning": 12089,
      "problems method firstly": 91343,
      "components order solve": 20041,
      "order solve problems": 82407,
      "rules based phrase": 103384,
      "start end positions": 112399,
      "easy use ambiguity": 34486,
      "accuracy endtoend method": 2146,
      "endtoend method point": 36925,
      "method point corresponding": 69054,
      "point corresponding method": 87798,
      "corresponding method applied": 24294,
      "studies conducted english": 114199,
      "english language work": 37188,
      "language experimental results": 59028,
      "experimental results stateoftheart": 40707,
      "reason paper propose": 96202,
      "scoring function best": 104227,
      "popular benchmark dataset": 88080,
      "approach coreference resolution": 7463,
      "task coreference resolution": 118026,
      "coreference resolution important": 23365,
      "paper report ongoing": 84402,
      "proposes new approach": 93607,
      "bert model used": 13174,
      "conll 2012 dataset": 21076,
      "preliminary results method": 89277,
      "advantages convolutional neural": 4073,
      "validity domain text": 130750,
      "domain text explored": 33675,
      "text explored paper": 120941,
      "explored paper propose": 41631,
      "propose novel hybrid": 92928,
      "novel hybrid architecture": 80596,
      "recurrent units bigru": 97272,
      "feature extraction layer": 43278,
      "conducted extensive experiments": 20925,
      "dataset yelp review": 27283,
      "yelp review polarity": 135322,
      "classification multilabel classification": 17287,
      "achieves better accuracy": 2745,
      "better accuracy compared": 13500,
      "accuracy compared existing": 2120,
      "compared existing methods": 19368,
      "help external linguistic": 49721,
      "techniques sentiment analysis": 119983,
      "large collection documents": 61050,
      "social networks twitter": 110444,
      "twitter data using": 126106,
      "feature selection model": 43309,
      "learning sentiment analysis": 63007,
      "reinforcement learning greedy": 97812,
      "improve parsing accuracy": 52453,
      "provide baseline model": 93760,
      "paper addresses task": 83717,
      "shaping public opinion": 107944,
      "news articles containing": 79292,
      "train supervised models": 123032,
      "models using data": 74299,
      "attracted research attention": 10160,
      "attention machine learning": 9874,
      "community recent years": 19095,
      "machine translation conduct": 65929,
      "better leverage multiple": 13619,
      "auxiliary tasks boost": 10893,
      "performance main task": 86516,
      "experiments method significantly": 41005,
      "method significantly improves": 69137,
      "training data translation": 123558,
      "leverage unlabeled data": 63631,
      "knowledge distillation evaluate": 57865,
      "finetuning pretrained models": 44498,
      "models bert roberta": 72838,
      "existing methods learning": 40187,
      "learning crosslingual representations": 62472,
      "new pretraining task": 79082,
      "contrastive learning specifically": 22732,
      "dataset machine reading": 27006,
      "reasoning machine reading": 96266,
      "challenging machine reading": 16274,
      "dataset serve benchmark": 27191,
      "dataset freely available": 26943,
      "documentlevel sentiment analysis": 33164,
      "paper study effectively": 84442,
      "evaluate proposed models": 38902,
      "public datasets experimental": 94252,
      "paper validate hypothesis": 84492,
      "novel approach sentiment": 80487,
      "sentiment analysis persian": 106628,
      "semantic information sentiment": 105079,
      "data text audio": 26557,
      "text audio video": 120670,
      "languages recent years": 60835,
      "introduce new methods": 56487,
      "sentiment analysis applied": 106566,
      "information external knowledge": 54558,
      "performance proposed approach": 86630,
      "proposed approach compared": 93186,
      "benchmark dataset experimental": 12756,
      "annotation schema present": 6367,
      "present corpus annotated": 89423,
      "error correction neural": 38298,
      "previous generated target": 90415,
      "discrepancy training inference": 32159,
      "translation accuracy paper": 124618,
      "demonstrate method achieves": 28780,
      "analyses verify effectiveness": 5473,
      "semeval2020 task 12": 105551,
      "approach offensive language": 7750,
      "offensive language identification": 81545,
      "task offensive language": 118475,
      "deep transformer layers": 28429,
      "perform multihead attentions": 86034,
      "outperforms individual models": 82912,
      "development set test": 30423,
      "set test set": 107606,
      "test set achieves": 120495,
      "subtask shared task": 114954,
      "ensemble model significantly": 37607,
      "development set improvement": 30421,
      "set features used": 107445,
      "recent shared task": 96513,
      "characterbased language models": 16492,
      "suffer data sparsity": 115215,
      "pretraining related language": 90314,
      "extensive experiments analyses": 41893,
      "semeval2020 task finetuning": 105553,
      "semeval 2020 task": 105511,
      "graded word similarity": 48591,
      "word similarity context": 133558,
      "pretrained bert models": 89993,
      "existing semantically annotated": 40279,
      "semantically annotated datasets": 105364,
      "annotated datasets propose": 6180,
      "social media paper": 110399,
      "media paper approach": 68139,
      "multilingual offensive language": 75319,
      "language identification shared": 59140,
      "identification shared task": 51442,
      "utilizing pretrained language": 130575,
      "macro averaged f1score": 66357,
      "semeval2020 task using": 105554,
      "analysis recent years": 5797,
      "recent years growing": 96616,
      "media platforms facebook": 68143,
      "textual visual information": 121755,
      "analyze sentiment polarity": 6007,
      "task identifying categorizing": 118265,
      "finetune bidirectional encoder": 44405,
      "used extract features": 128544,
      "better baseline algorithm": 13517,
      "sentiment classification code": 106695,
      "research excellence framework": 100493,
      "excellence framework ref": 39912,
      "framework ref 2014": 45670,
      "graphbased ranking algorithm": 48909,
      "labeled data language": 58434,
      "multilingual representation models": 75350,
      "intent classification diverse": 55895,
      "neural abstractive summarization": 77825,
      "meeting summarization task": 68258,
      "morphologicallyrich languages mrls": 74780,
      "explicitly modeling morphological": 41383,
      "related knowledge base": 97871,
      "document sentence token": 33076,
      "constructing knowledge bases": 21679,
      "knowledge bases enable": 57801,
      "using active learning": 129318,
      "automated text processing": 10472,
      "convolutional network relation": 23165,
      "network relation extraction": 77406,
      "reliance linguistic tools": 98632,
      "process work propose": 91591,
      "convolutional network sgcn": 23167,
      "selfattention mechanism using": 104888,
      "achieve stateoftheart result": 2565,
      "dependency parsing tools": 29204,
      "task shared task": 118707,
      "received 27 submissions": 96356,
      "baseline code available": 12204,
      "progress machine translation": 92161,
      "machine translation written": 66306,
      "availability parallel corpora": 10924,
      "research computational linguistics": 100449,
      "largest widely used": 61542,
      "presents efficient algorithm": 89842,
      "language model adapted": 59290,
      "method originally proposed": 69002,
      "error rate results": 38333,
      "information retrieval tasks": 54949,
      "poses significant challenge": 88275,
      "sophisticated language models": 110671,
      "paper propose efficient": 84231,
      "language model proposed": 59385,
      "results public datasets": 102097,
      "public datasets model": 94255,
      "datasets model achieves": 27570,
      "achieves good balance": 2791,
      "fast accurate neural": 43127,
      "field deep learning": 43948,
      "complexity inefficiency insideoutside": 19918,
      "inefficiency insideoutside algorithm": 54072,
      "approach improve efficiency": 7620,
      "improve parsing performance": 52454,
      "parser achieves new": 84913,
      "release code httpsgithubcomyzhangcscrfpar": 98438,
      "promising results modeling": 92298,
      "based set rules": 12038,
      "trained manually annotated": 123193,
      "bert recent works": 13215,
      "multilingual bert model": 75204,
      "pretrained models using": 90157,
      "data set machine": 26410,
      "set machine translation": 107486,
      "machine translation parallel": 66149,
      "evaluation data sets": 39169,
      "information document context": 54505,
      "lack large scale": 58728,
      "large scale multilingual": 61253,
      "available indian languages": 11015,
      "standard evaluation benchmarks": 112237,
      "state art model": 112449,
      "faster training inference": 43196,
      "multiple layers linguistic": 75603,
      "pos tagging named": 88232,
      "tagging named entities": 117414,
      "publicly available corpus": 94297,
      "models present language": 73771,
      "measuring gender bias": 67911,
      "gender bias coreference": 46584,
      "automatic quality assessment": 10632,
      "text classification major": 120738,
      "text classification mltc": 120741,
      "experimental results validate": 40718,
      "large neural models": 61169,
      "cointeractive relation network": 18345,
      "dialog act recognition": 30547,
      "dialog act sentiment": 30549,
      "existing systems treat": 40308,
      "explicitly modeling mutual": 41384,
      "modeling mutual interaction": 72486,
      "mutual interaction relation": 75985,
      "introducing cointeractive relation": 56630,
      "thoroughly study different": 121908,
      "stateoftheart joint model": 112681,
      "terms f1 score": 120325,
      "act recognition task": 3068,
      "encoder representation transformer": 36560,
      "representation transformer bert": 99449,
      "framework boost performance": 45443,
      "finetuning pretrained transformer": 44500,
      "training inference time": 123661,
      "gpt2 language model": 48563,
      "paper evaluate performance": 83897,
      "biomedical text mining": 14208,
      "paper describes work": 83858,
      "team semeval 2020": 119721,
      "2020 task commonsense": 734,
      "task commonsense validation": 117979,
      "commonsense validation explanation": 19013,
      "based pretrained language": 11945,
      "multilingual models machine": 75291,
      "pretrained sequencetosequence model": 90188,
      "sequencetosequence model bart": 107167,
      "evaluation metrics model": 39291,
      "choosing right answer": 16932,
      "based automatic metrics": 11537,
      "automatic metrics humans": 10608,
      "paper describes submitted": 83850,
      "knowledge bases entity": 57802,
      "large number entities": 61176,
      "information recent work": 54902,
      "recent work suggests": 96590,
      "information dependency trees": 54472,
      "transformer models bert": 124348,
      "models implicitly encode": 73359,
      "implicitly encode syntax": 52028,
      "role labeling srl": 103194,
      "relation extraction propose": 98018,
      "neural network output": 78367,
      "transformer attention layers": 124297,
      "empirical analysis demonstrates": 36150,
      "pretrained models multilingual": 90144,
      "models multilingual models": 73592,
      "pretrained models significantly": 90153,
      "code models available": 18119,
      "context document level": 22066,
      "make extraction decisions": 66669,
      "noun phrase coreference": 80428,
      "dataset model performs": 27033,
      "performs substantially better": 87031,
      "substantially better prior": 114883,
      "better prior work": 13684,
      "contribute model performance": 22757,
      "propose variational approach": 93152,
      "ground truth provided": 49100,
      "provided domain experts": 93964,
      "domain experts use": 33529,
      "experts use targetopinion": 41250,
      "use targetopinion word": 128315,
      "targetopinion word pairs": 117784,
      "word pairs supervision": 133396,
      "word pairs extracted": 133393,
      "pairs extracted using": 83545,
      "extracted using dependency": 42191,
      "using dependency parsers": 129556,
      "dependency parsers simple": 29166,
      "parsers simple rules": 85045,
      "simple rules objective": 109511,
      "predict opinion word": 88911,
      "opinion word given": 82108,
      "word given target": 133301,
      "target word ultimate": 117749,
      "word ultimate goal": 133614,
      "ultimate goal learn": 126555,
      "goal learn sentiment": 48365,
      "introducing latent variable": 56642,
      "latent variable sentiment": 61632,
      "variable sentiment polarity": 130848,
      "sentiment polarity objective": 106775,
      "polarity objective function": 87918,
      "objective function inject": 81081,
      "function inject sentiment": 46028,
      "lower bound learn": 65423,
      "bound learn sentiment": 14509,
      "classifier optimizing lower": 17565,
      "optimizing lower bound": 82244,
      "results method outperform": 101920,
      "baselines sentiment analysis": 12461,
      "supervised method hundreds": 115800,
      "method hundreds labels": 68875,
      "hundreds labels aspect": 51149,
      "neural networkbased model": 78426,
      "created gold standard": 24670,
      "machine translation sentiment": 66208,
      "domains languages work": 33801,
      "languages work propose": 60979,
      "translation mt translate": 124989,
      "conduct experiments using": 20861,
      "discuss advantages drawbacks": 32234,
      "concept extraction using": 20571,
      "using pointergenerator networks": 130023,
      "concept extraction crucial": 20569,
      "significantly outperforms standard": 109028,
      "furthermore model readily": 46192,
      "active learning chinese": 3112,
      "chinese clinical named": 16746,
      "clinical named entity": 17774,
      "entity recognition clinical": 38002,
      "recognition clinical named": 96835,
      "entity recognition cner": 38005,
      "recognition cner aims": 96839,
      "step clinical research": 113256,
      "requires large number": 100289,
      "pairs achieve better": 83482,
      "computational results ccks2017": 20427,
      "results ccks2017 task": 101564,
      "ccks2017 task benchmark": 15865,
      "deep learning relies": 28310,
      "annotation timeconsuming expensive": 6391,
      "annotation low resource": 6333,
      "multitask learning transfer": 75871,
      "incorporating prior knowledge": 53553,
      "pretrained multilingual bert": 90160,
      "external knowledge pretrained": 42017,
      "existing algorithms cases": 40044,
      "15 times faster": 353,
      "extraction free text": 42345,
      "free text using": 45768,
      "relation extraction adapt": 97979,
      "2020 shared task": 730,
      "using small set": 130189,
      "successfully applied task": 115177,
      "official evaluation scores": 81599,
      "using transformerbased models": 130316,
      "task english arabic": 118148,
      "shared task evaluated": 108044,
      "claims social media": 17020,
      "1st 2nd 3rd": 484,
      "paper presents submission": 84206,
      "problem sequence labeling": 91218,
      "contextual embedding models": 22457,
      "embeddings best performing": 35575,
      "best performing architecture": 13400,
      "ensemble different models": 37594,
      "terms parts speech": 120356,
      "parts speech tags": 85590,
      "upb semeval2020 task": 127784,
      "developed team semeval2020": 30313,
      "team semeval2020 task": 119723,
      "multimodal multitask learning": 75445,
      "semantic similarity paper": 105288,
      "similarity paper presents": 109279,
      "simple bag words": 109371,
      "paper describes designed": 83815,
      "place semeval2020 task": 87607,
      "excellent performance task": 39917,
      "pairwise ranking loss": 83681,
      "help improve performance": 49738,
      "performance best model": 86180,
      "achieves highest score": 2804,
      "lowresource polysynthetic languages": 65556,
      "english translation using": 37316,
      "crossdomain data model": 24887,
      "performance language models": 86482,
      "language models explored": 59553,
      "approach experimental results": 7549,
      "results establish new": 101765,
      "selecting optimal parameters": 104754,
      "relation extraction documentlevel": 97989,
      "constructing knowledge graphs": 21680,
      "different entity pairs": 31129,
      "approaches significant margin": 8340,
      "used language internet": 128603,
      "solving arithmetic word": 110640,
      "arithmetic word problems": 9010,
      "language understanding reasoning": 60259,
      "capabilities nlp systems": 15173,
      "recent works use": 96600,
      "work explore novel": 134514,
      "proposed method consists": 93360,
      "compared previous stateoftheart": 19418,
      "require complex reasoning": 100122,
      "word vector representation": 133628,
      "nmt models concatenation": 79912,
      "bleu score higher": 14326,
      "higher baseline model": 50163,
      "bleu score achieved": 14324,
      "task existing studies": 118167,
      "various neural architectures": 131150,
      "approaches overlook potential": 8273,
      "paper propose learning": 84258,
      "models specifically introduce": 74079,
      "model auxiliary tasks": 70716,
      "auxiliary tasks multitask": 10895,
      "means auxiliary tasks": 67758,
      "achieve better local": 2429,
      "better local optimum": 13623,
      "results benchmarks indicate": 101548,
      "application text mining": 6891,
      "bagofwords bow model": 11372,
      "dataset deep learning": 26848,
      "named entity recognizer": 76125,
      "proper nouns named": 92421,
      "nouns named entities": 80444,
      "crowdsourced human evaluation": 25112,
      "generation opendomain dialogue": 47522,
      "response generation dialogue": 101202,
      "kullbackleibler divergence kld": 58361,
      "prevents model learning": 90375,
      "evaluation human evaluation": 39233,
      "methods yield significant": 69858,
      "significant improvements baselines": 108786,
      "release code data": 98437,
      "architecture effectively model": 8649,
      "significantly boosts performance": 108896,
      "based event extraction": 11693,
      "event extraction model": 39515,
      "stateoftheart performance ace": 112804,
      "performance ace 2005": 86112,
      "absolute f1 points": 1727,
      "method neural semantic": 68986,
      "neural semantic parsing": 78664,
      "parsing high accuracy": 85123,
      "transitionbased parser dependency": 124511,
      "poorly human judgments": 88069,
      "alleviate issue propose": 5037,
      "evaluating quality generated": 39089,
      "correlates better human": 24215,
      "better human judgments": 13601,
      "different neural network": 31295,
      "approaches proven effective": 8299,
      "entities knowledge graph": 37806,
      "translation nmt work": 125064,
      "quality automatically generated": 94603,
      "task recently gained": 118613,
      "promising results english": 92294,
      "transformer decoder approach": 124319,
      "monolingual multilingual settings": 74601,
      "stateoftheart approaches english": 112583,
      "make heavy use": 66683,
      "inputs question answering": 55493,
      "question answering recent": 95092,
      "answering recent work": 6686,
      "context question answering": 22241,
      "calibration neural models": 15047,
      "model learns attend": 71440,
      "achieving significantly higher": 2976,
      "robustness results suggest": 103129,
      "based pretrained transformers": 11949,
      "question answering problem": 95074,
      "term extraction problem": 120212,
      "language processing chinese": 59851,
      "word segmentation cws": 133492,
      "recognition ner dependency": 96924,
      "ner dependency parsing": 77039,
      "multitask model based": 75876,
      "stateoftheart sota performance": 112962,
      "dependency parsing ner": 29187,
      "achieving sota performance": 2979,
      "similarity measure compare": 109257,
      "report experimental results": 98997,
      "language model performance": 59366,
      "model performance tasks": 71730,
      "highresource languages particular": 50429,
      "languages particular english": 60793,
      "transformerbased language model": 124413,
      "language inference dataset": 59170,
      "dataset natural language": 27050,
      "realworld nlp applications": 96172,
      "investigate extent models": 56757,
      "training multilingual machine": 123720,
      "pretrained language modeling": 90060,
      "models different types": 73073,
      "translation edit rate": 124774,
      "performance wmt 2019": 86856,
      "shared task demonstrate": 108040,
      "transferring knowledge neural": 124239,
      "transfer knowledge large": 124065,
      "data recent advances": 26324,
      "recent advances opendomain": 96419,
      "training opendomain dialogue": 123764,
      "manual evaluation indicates": 66994,
      "method produce highquality": 69074,
      "points f1 score": 87857,
      "f1 score respectively": 42687,
      "using disentangled representations": 129588,
      "tasks despite success": 119054,
      "model trained domain": 72188,
      "domain paper present": 33607,
      "present methods improve": 89555,
      "methods improve robustness": 69542,
      "make final prediction": 66673,
      "propose different methods": 92625,
      "supervised learning tasks": 115781,
      "naturally converse humans": 76642,
      "downstream tasks demonstrate": 34034,
      "dependency treebank universal": 29239,
      "automatic method converting": 10592,
      "outperforms stateoftheart method": 83005,
      "taskoriented spoken dialogue": 118905,
      "belief state tracking": 12690,
      "markov decision process": 67257,
      "decision process pomdp": 27941,
      "reinforcement learning approaches": 97797,
      "learning approaches used": 62365,
      "reinforcement learning drl": 97806,
      "neural networks dialogue": 78465,
      "important challenging problem": 52114,
      "achieved good performance": 2629,
      "paper address problems": 83709,
      "taskoriented dialog recent": 118884,
      "learning dialog policy": 62499,
      "multilingual knowledge base": 75265,
      "evaluation method task": 39272,
      "task language pairs": 118335,
      "research release dataset": 100613,
      "release dataset containing": 98450,
      "method sentence compression": 69124,
      "sentence compression sentence": 105800,
      "compression sentence compression": 20323,
      "models suffer poor": 74127,
      "pretrained bidirectional encoder": 89995,
      "model times faster": 72172,
      "large scale pretraining": 61261,
      "existing models usually": 40212,
      "performance drops substantially": 86327,
      "dataset code publicly": 26789,
      "neural architecture search": 77835,
      "challenging machine translation": 16275,
      "ende enfr datasets": 36845,
      "previous studies usually": 90504,
      "data difficult obtain": 25850,
      "order make use": 82366,
      "models achieves stateoftheart": 72704,
      "work step propose": 134819,
      "semantically related words": 105373,
      "wordlevel language model": 133740,
      "language model size": 59401,
      "segmentation partofspeech tagging": 104613,
      "semantic parsing semantic": 105163,
      "knowledge distillation method": 57868,
      "multitask model surpass": 75878,
      "model surpass singletask": 72122,
      "reasoning commonsense knowledge": 96235,
      "commonsense knowledge graph": 18997,
      "generative pretrained language": 47751,
      "external commonsense knowledge": 41995,
      "controlled language generation": 22839,
      "lack diversity generated": 58703,
      "diversity generated texts": 32878,
      "text generation datasets": 120983,
      "corresponding target response": 24308,
      "compared stateoftheart models": 19460,
      "developed deep learning": 30266,
      "learning model specifically": 62765,
      "question answering text": 95114,
      "answering text summarization": 6713,
      "text summarization information": 121346,
      "recognizing text entailment": 97077,
      "using external knowledge": 129659,
      "natural language toolkit": 76547,
      "models data augmentation": 73010,
      "best performance using": 13396,
      "performance using 10": 86829,
      "10 training data": 154,
      "way experimental results": 132079,
      "outperform stateoftheart method": 82757,
      "extraction paper presents": 42423,
      "subtasks named entity": 114972,
      "main challenges building": 66402,
      "issue proposing new": 57028,
      "metric takes account": 69906,
      "information entity mentions": 54537,
      "different parts document": 31321,
      "use graphbased neural": 128078,
      "f1 percentage points": 42666,
      "networks representation learning": 77735,
      "response generation using": 101214,
      "engaging dialogue systems": 37020,
      "method outperforms competitive": 69011,
      "mbert xlm xlmr": 67572,
      "propose multihop reasoning": 92796,
      "reasoning knowledge graph": 96259,
      "proposed model works": 93487,
      "evaluation proposed method": 39340,
      "proposed method generate": 93368,
      "method generate informative": 68856,
      "named entity knowledge": 76047,
      "target language approach": 117631,
      "training corpusbased machine": 123409,
      "different kinds information": 31189,
      "information named entities": 54784,
      "results automatic evaluation": 101530,
      "obtain better performance": 81264,
      "able capture local": 1617,
      "datasets snips atis": 27719,
      "achieves competitive result": 2774,
      "recent years number": 96627,
      "number approaches proposed": 80842,
      "approaches proposed various": 8296,
      "text generation approaches": 120979,
      "neural topic modeling": 78712,
      "graph graph neural": 48817,
      "neural topic model": 78711,
      "survey semantic parsing": 116194,
      "study semantic parsing": 114510,
      "given context knowledge": 48000,
      "field semantic parsing": 43980,
      "meaning representation using": 67678,
      "semantic parser using": 105136,
      "different benchmark datasets": 31027,
      "used evaluate semantic": 128524,
      "evaluate semantic parser": 38921,
      "answer complex questions": 6520,
      "contextualised embeddings paper": 22532,
      "evaluate performance different": 38883,
      "tasks entity recognition": 119100,
      "entity recognition semantic": 38068,
      "addition propose novel": 3466,
      "human evaluation model": 50823,
      "general purpose text": 46699,
      "compared prior work": 19426,
      "prior work benchmark": 90742,
      "word embeddings increased": 133134,
      "stateoftheart models natural": 112742,
      "pretrained contextual language": 90006,
      "contextual language models": 22479,
      "language models generate": 59557,
      "representations words context": 99990,
      "come high computational": 18765,
      "high computational cost": 50045,
      "computational cost work": 20370,
      "cost work explore": 24378,
      "work explore simple": 134516,
      "strong baseline negligible": 113635,
      "multiple pretrained embeddings": 75644,
      "concatenation technique works": 20544,
      "different pretrained embeddings": 31349,
      "provide open source": 93886,
      "open source implementations": 81934,
      "reverse dictionary task": 102526,
      "models different word": 73074,
      "different word embeddings": 31570,
      "word embeddings learn": 133146,
      "dialogue generation natural": 30679,
      "generate informative responses": 46957,
      "previous research focused": 90448,
      "processing paper propose": 91767,
      "paper propose compare": 84224,
      "based word overlap": 12166,
      "discontinuous constituent parsing": 32030,
      "constituent parsing sequence": 21546,
      "parsing sequence labeling": 85233,
      "language models existing": 59548,
      "control text generation": 22822,
      "model generates fluent": 71239,
      "representation word embeddings": 99465,
      "artificial intelligence systems": 9257,
      "representations words different": 99991,
      "english monolingual data": 37211,
      "monolingual data language": 74574,
      "webbased annotation tool": 132275,
      "integrate machine learning": 55763,
      "proposed models learn": 93496,
      "models learn characteristics": 73479,
      "task 2018 dataset": 117822,
      "time model outperforms": 122050,
      "natural language interactions": 76363,
      "financial services industry": 44260,
      "modeling natural language": 72489,
      "knowledge natural language": 58076,
      "require large quantities": 100166,
      "need labeled data": 76826,
      "strong baselines bert": 113643,
      "baselines bert roberta": 12363,
      "knowledge bases wordnet": 57813,
      "gives best performance": 48181,
      "sentence document levels": 105827,
      "propose new learning": 92858,
      "shows approach outperforms": 108551,
      "reasoning paragraph effects": 96287,
      "paper propose sequential": 84319,
      "reasoning process neural": 96296,
      "process neural network": 91537,
      "learned endtoend manner": 62215,
      "models simply concatenate": 74054,
      "turns dialogue history": 125993,
      "study problem using": 114484,
      "constituency parse trees": 21524,
      "controlled text generation": 22847,
      "existing approaches problem": 40056,
      "guide text generation": 49248,
      "attention mechanism allows": 9881,
      "outperforms sota models": 82987,
      "effective unsupervised domain": 34768,
      "trained language models": 123171,
      "contextualised embedding models": 22530,
      "language models mlms": 59601,
      "domain adaptation tasks": 33456,
      "entity recognition method": 38022,
      "performance results suggest": 86682,
      "imitation learning framework": 51839,
      "previous semisupervised methods": 90457,
      "introduce new public": 56495,
      "stateofthe art deep": 112559,
      "art deep learning": 9048,
      "models task additionally": 74160,
      "inference time complexity": 54239,
      "loss function guide": 65272,
      "guide attention heads": 49230,
      "downstream tasks compared": 34033,
      "tasks compared baselines": 118991,
      "achieving state art": 2981,
      "results lowresource settings": 101896,
      "language modeling performance": 59455,
      "unsupervised constituency parsing": 127614,
      "constituency parsing models": 21529,
      "methods significant margin": 69749,
      "labeled data used": 58450,
      "entailment discourse relations": 37661,
      "large scale labeled": 61249,
      "task aimed testing": 117863,
      "trained raw text": 123258,
      "use online social": 128182,
      "paper present text": 84143,
      "humans use language": 51117,
      "models able capture": 72642,
      "need manually labeled": 76835,
      "target language leveraging": 117639,
      "languages existing approaches": 60565,
      "require expensive crosslingual": 100133,
      "crosslingual representation learning": 24999,
      "supervision target language": 115921,
      "language using minimal": 60313,
      "methods based multilingual": 69335,
      "based multilingual bert": 11865,
      "generation pretrained language": 47554,
      "lacking parallel data": 58772,
      "work explore challenging": 134507,
      "explore challenging task": 41530,
      "task pretrained language": 118554,
      "tasks end introduce": 119095,
      "empirical studies public": 36197,
      "datasets indicate model": 27522,
      "indicate model significantly": 53842,
      "nlp tasks question": 79756,
      "relevant spans text": 98587,
      "translation models modern": 124957,
      "problem empirically investigate": 91023,
      "08 bleu points": 75,
      "lowlevel lexical information": 65479,
      "maximum posteriori map": 67548,
      "posteriori map decoding": 88491,
      "results stateoftheart results": 102210,
      "stateoftheart results language": 112905,
      "uniform information density": 127138,
      "multiinstance multilabel learning": 75147,
      "aspectcategory sentiment analysis": 9362,
      "sentiment analysis acsa": 106562,
      "analysis acsa aims": 5486,
      "acsa aims predict": 3057,
      "aims predict sentiment": 4556,
      "predict sentiment polarities": 88933,
      "respect given aspect": 101080,
      "detect sentiment particular": 29818,
      "sentiment particular aspect": 106767,
      "aspect category predict": 9322,
      "predict sentiment polarity": 88934,
      "suboptimal performance paper": 114791,
      "aspect categories mentioned": 9320,
      "public datasets demonstrate": 94249,
      "models capture linguistic": 72879,
      "language processing involve": 59876,
      "labeling semantic role": 58537,
      "sequence labeling including": 106971,
      "models empirically demonstrate": 73140,
      "empirically demonstrate approach": 36224,
      "demonstrate approach achieves": 28667,
      "integral natural language": 55741,
      "pairs paper proposes": 83601,
      "annotated discourse relations": 6187,
      "questions answers based": 95274,
      "present baseline algorithms": 89391,
      "novel method generating": 80635,
      "neural sequencetosequence seq2seq": 78681,
      "seq2seq models typically": 106904,
      "highly correlate human": 50307,
      "human judgments semantic": 50887,
      "train language models": 122944,
      "natural language recent": 76523,
      "languageagnostic sentence representations": 60355,
      "tasks perform experiments": 119378,
      "languages english chinese": 60532,
      "outperforms recently proposed": 82974,
      "large margin detailed": 61143,
      "detailed analysis reveals": 29759,
      "knowledge distillation technique": 57873,
      "power neural machine": 88642,
      "knowledge distillation kd": 57867,
      "propose novel alternative": 92895,
      "provide overview different": 93890,
      "nlp systems focus": 79697,
      "pretrained models like": 90142,
      "classification question answering": 17362,
      "models real systems": 73862,
      "hinton et al": 50507,
      "efficacy proposed approach": 35015,
      "analysis open information": 5727,
      "neural open information": 78622,
      "information extraction openie": 54587,
      "comes significant computational": 18782,
      "significant computational cost": 108741,
      "task improve performance": 118279,
      "improve performance applying": 52456,
      "establishing new state": 38617,
      "large annotated dataset": 61032,
      "expensive work propose": 40440,
      "work propose framework": 134720,
      "translation existing approaches": 124804,
      "approaches shown promise": 8338,
      "summarization study problem": 115572,
      "learning transfer knowledge": 63119,
      "nlp datasets contain": 79598,
      "paper investigate impact": 84017,
      "training data uses": 123563,
      "training objective model": 123758,
      "extractive question answering": 42557,
      "question answering training": 95119,
      "answering training data": 6717,
      "representations word level": 99984,
      "level document level": 63437,
      "best performing systems": 13406,
      "create release largescale": 24639,
      "english language corpus": 37185,
      "neural models predict": 78190,
      "question answering stateoftheart": 95106,
      "answering qa relies": 6672,
      "relies large amounts": 98654,
      "annotating qa datasets": 6272,
      "reduces human effort": 97386,
      "text sentiment analysis": 121284,
      "transformer models natural": 124349,
      "french spanish german": 45826,
      "sequence generation models": 106950,
      "machine translation benchmark": 65911,
      "news headlines using": 79344,
      "proposed machine learning": 93328,
      "headline body text": 49594,
      "news headline body": 79341,
      "question answering instead": 95054,
      "answering instead using": 6639,
      "predicting masked words": 88994,
      "textual entailment tasks": 121700,
      "hotpotqa fullwiki setting": 50674,
      "crosslingual nlp multilingual": 24987,
      "annotation expensive hard": 6306,
      "multilingual bert models": 75205,
      "learning rank synthetic": 62934,
      "function maps input": 46033,
      "existing attention methods": 40069,
      "optimal paper propose": 82166,
      "experiments datasets different": 40869,
      "lample et al": 58797,
      "accuracy increasing model": 2187,
      "increasing model capacity": 53677,
      "computational overhead empirical": 20416,
      "success pretrained language": 115114,
      "current evaluation metrics": 25283,
      "logic natural way": 64993,
      "correlations human judgements": 24258,
      "largescale human evaluation": 61444,
      "reduces human efforts": 97387,
      "total number annotations": 122699,
      "prior work mainly": 90755,
      "issue paper proposes": 57018,
      "great potential practical": 49011,
      "architecture search nas": 8739,
      "language modeling natural": 59445,
      "contrast prior work": 22708,
      "tasks results mixed": 119473,
      "advances neural language": 4003,
      "analysis reveals significant": 5820,
      "challenging problem paper": 16297,
      "paper describes contribution": 83811,
      "shared task main": 108064,
      "based transfer learning": 12126,
      "performance empirical results": 86335,
      "empirical results models": 36190,
      "results models achieve": 101954,
      "models achieve competitive": 72663,
      "shared task indicating": 108054,
      "phenomenon natural language": 87256,
      "ende translation tasks": 36847,
      "translation case study": 124687,
      "recent work natural": 96577,
      "processing nlp focused": 91735,
      "case studies demonstrate": 15611,
      "language model solve": 59402,
      "task sentence level": 118689,
      "candidate sentence token": 15122,
      "achieve comparable performance": 2441,
      "method outperforms supervised": 69035,
      "future directions task": 46273,
      "use realworld applications": 128225,
      "applications paper perform": 6985,
      "paper perform indepth": 84068,
      "perform indepth error": 86012,
      "discuss remaining challenges": 32283,
      "task despite recent": 118073,
      "contextualized language models": 22555,
      "language models various": 59687,
      "work proposes new": 134754,
      "surface realization using": 116076,
      "task automatic human": 117911,
      "anger fear sadness": 6110,
      "multiple languages using": 75598,
      "text classification systems": 120771,
      "languages spanish chinese": 60886,
      "require manual annotation": 100175,
      "new target language": 79195,
      "approaches based pretrained": 8082,
      "additional performance improvements": 3556,
      "demonstrate benefits approach": 28679,
      "training large scale": 123678,
      "models question answering": 73849,
      "assessment shared task": 9529,
      "shared task wmt": 108105,
      "implement evaluate different": 51932,
      "different neural architectures": 31291,
      "proposed methods achieve": 93411,
      "ensemble data augmentation": 37589,
      "wmt20 biomedical translation": 132866,
      "biomedical translation task": 14214,
      "distinct style vocabulary": 32545,
      "vocabulary models trained": 131885,
      "models trained data": 74196,
      "models lms shown": 73521,
      "knowledge form knowledge": 57940,
      "datasets snli mnli": 27721,
      "dataset using stateoftheart": 27263,
      "using stateoftheart pretrained": 130218,
      "poses significant hurdle": 88277,
      "highquality neural machine": 50401,
      "approach transfer learning": 7982,
      "semantic andor syntactic": 104985,
      "tackle problem using": 117308,
      "knowledge distillation propose": 57872,
      "knowledge distillation approach": 57864,
      "effectiveness approach achieving": 34867,
      "semantic interpretation natural": 105086,
      "interpretation natural language": 56270,
      "natural language characterized": 76272,
      "standard nlp models": 112285,
      "recent work studied": 96589,
      "identify factual errors": 51502,
      "difficult obtain propose": 31650,
      "obtain propose method": 81312,
      "identify factual inconsistencies": 51503,
      "based question generation": 11967,
      "learning approaches natural": 62360,
      "language processing mainly": 59885,
      "experiments demonstrate approach": 40877,
      "generation work propose": 47702,
      "based publicly available": 11959,
      "available pretrained language": 11077,
      "language models approach": 59504,
      "trained english dataset": 123136,
      "effectiveness approach zeroshot": 34872,
      "generation experiments manual": 47404,
      "role labeling english": 103184,
      "dialogue models recent": 30713,
      "recent years suffers": 96639,
      "structures language modeling": 114081,
      "response generation paper": 101209,
      "response generation experiments": 101203,
      "model shows significant": 72014,
      "multiturn dialogue systems": 75917,
      "question answering deep": 95038,
      "based question answering": 11965,
      "answering qa models": 6669,
      "using causal inference": 129433,
      "models propose novel": 73820,
      "extensive experiments using": 41935,
      "stateoftheart sota models": 112961,
      "number publicly available": 80947,
      "dialogue state annotations": 30744,
      "leading suboptimal performance": 61913,
      "dataset paper introduce": 27086,
      "understanding dialogue state": 126828,
      "significant improvements previous": 108801,
      "words multiple senses": 134068,
      "different senses word": 31412,
      "ability discriminate word": 1510,
      "knowledge bases kb": 57806,
      "dialogue context existing": 30649,
      "context existing works": 22089,
      "translation task translating": 125343,
      "task translating texts": 118810,
      "major tasks natural": 66588,
      "processing computational linguistics": 91645,
      "fundamental text processing": 46127,
      "language modeling generation": 59433,
      "linguistic properties words": 64533,
      "address propose new": 3755,
      "word prediction task": 133416,
      "prediction task designed": 89134,
      "language models performance": 59618,
      "english language models": 37186,
      "models using proposed": 74305,
      "existing studies limited": 40302,
      "paper propose attentive": 84222,
      "leveraging contextual information": 63668,
      "demonstrate approach outperforms": 28670,
      "outperforms previous studies": 82954,
      "previous studies terms": 90502,
      "generation challenging task": 47332,
      "heterogeneous information networks": 49862,
      "research attention natural": 100419,
      "attention natural language": 9941,
      "language processing existing": 59863,
      "heterogeneous information network": 49861,
      "corpus improve performance": 23832,
      "improve performance proposed": 52479,
      "proposed method conduct": 93355,
      "method conduct extensive": 68724,
      "paper describes development": 83817,
      "machine translation provides": 66178,
      "thousands language pairs": 121924,
      "language pairs covering": 59742,
      "stateoftheart translation models": 113016,
      "transfer learning technique": 124136,
      "suffer data scarcity": 115213,
      "use transfer learning": 128340,
      "transfer learning tl": 124138,
      "learning tl techniques": 63108,
      "models proven effective": 73833,
      "hate speech classification": 49531,
      "speech classification task": 111655,
      "lowresource setting using": 65561,
      "existing evaluation benchmarks": 40132,
      "binary classification problem": 14119,
      "series experiments determine": 107275,
      "target language models": 117641,
      "models trained solely": 74229,
      "competitive performance task": 19670,
      "different meanings word": 31249,
      "key natural language": 57589,
      "language understanding propose": 60257,
      "understanding propose novel": 126930,
      "information encoded pretrained": 54524,
      "texttospeech automatic speech": 121659,
      "speech recognition applications": 111735,
      "little work addressed": 64836,
      "sentiment analysis used": 106676,
      "used social media": 128768,
      "social media previous": 110415,
      "media previous work": 68156,
      "used machine learning": 128618,
      "features recently deep": 43690,
      "neural networks widely": 78604,
      "networks widely used": 77815,
      "english language paper": 37187,
      "various unsupervised word": 131235,
      "unsupervised word representations": 127755,
      "features experimental results": 43497,
      "transformer achieved great": 124287,
      "like bert gpt": 64022,
      "contexts paper propose": 22415,
      "effectively improve performance": 34820,
      "improve performance tasks": 52491,
      "neural narrative generation": 78222,
      "nlp task model": 79701,
      "model generates story": 71241,
      "response generation chatbots": 101199,
      "methods neural response": 69639,
      "new approach extracting": 78787,
      "texts based linguistic": 121464,
      "models existing approaches": 73192,
      "learn sentence representations": 62144,
      "neural network sentence": 78395,
      "datasets approach significantly": 27317,
      "significantly improves strong": 108957,
      "multiple input sources": 75579,
      "models achieved outstanding": 72685,
      "achieved outstanding performance": 2663,
      "adapted pretrained language": 3288,
      "practical use cases": 88721,
      "training data finetuning": 123465,
      "finetuning target task": 44520,
      "target task achieve": 117721,
      "present simple efficient": 89704,
      "tasks compared original": 118992,
      "performance multitask learning": 86554,
      "multitask learning negation": 75856,
      "targeted sentiment classification": 117771,
      "auxiliary tasks including": 10894,
      "transfer learning language": 124097,
      "performance challenge datasets": 86199,
      "room improvement release": 103264,
      "improvement release datasets": 52757,
      "datasets source code": 27725,
      "metoo timesup movements": 69866,
      "use resulting model": 128249,
      "augmentation natural language": 10289,
      "language understanding data": 60214,
      "challenging paper propose": 16290,
      "data augmentation framework": 25632,
      "models wide range": 74338,
      "tasks glue benchmark": 119148,
      "commonly used neural": 18971,
      "networks text generation": 77796,
      "words experimental results": 133948,
      "based lexical similarity": 11804,
      "form natural language": 45108,
      "present detailed description": 89445,
      "language understanding intent": 60227,
      "understanding intent detection": 126864,
      "slot filling dialogue": 110040,
      "filling dialogue state": 44037,
      "knowledgegrounded dialogue generation": 58276,
      "empirical results benchmarks": 36181,
      "benchmarks indicate model": 12909,
      "methods automatic evaluation": 69328,
      "automatic evaluation human": 10528,
      "evaluation human judgment": 39235,
      "documentlevel transformer models": 33169,
      "metrics including bleu": 69973,
      "indicators human evaluation": 53882,
      "hindienglish code mixed": 50481,
      "human annotated corpus": 50740,
      "annotated corpus hindienglish": 6148,
      "corpus hindienglish codemixed": 23825,
      "best knowledge corpus": 13345,
      "corpus obtain word": 23912,
      "terminology constraints inference": 120257,
      "using largescale pretrained": 129794,
      "nlp tasks potential": 79751,
      "automatically generate large": 10774,
      "linear contextfree rewriting": 64342,
      "contextfree rewriting systems": 22365,
      "accuracy parsing speed": 2231,
      "level subword level": 63508,
      "new model improves": 79031,
      "model improves performance": 71318,
      "tasks time producing": 119565,
      "improves performance models": 53015,
      "parsing recurrent neural": 85216,
      "transformer architecture model": 124293,
      "representation amr parsing": 99167,
      "contextual affective analysis": 22442,
      "work presents extension": 134696,
      "russian spanish results": 103502,
      "handlabeled training data": 49371,
      "limited size makes": 64279,
      "causal relations events": 15812,
      "previous methods large": 90428,
      "automatically labeled training": 10793,
      "english chinese languages": 37089,
      "available website research": 11146,
      "website research purposes": 132297,
      "research purposes addition": 100602,
      "bayes decision tree": 12568,
      "decision tree random": 27947,
      "conditional language modeling": 20761,
      "internal workings nmt": 56180,
      "training process models": 123794,
      "social media focused": 110384,
      "achieving macro f1": 2959,
      "replaced token detection": 98938,
      "information way model": 55092,
      "benchmark demonstrate effectiveness": 12812,
      "improvements strong transformer": 52923,
      "international phonetic alphabet": 56185,
      "phonetic alphabet ipa": 87288,
      "jointly models word": 57368,
      "models word segmentation": 74351,
      "extracting relevant information": 42234,
      "endtoend fashion introduce": 36901,
      "capture contextual information": 15282,
      "translation source sentence": 125279,
      "biomedical entity linking": 14181,
      "biomedical knowledge bases": 14191,
      "labelled training data": 58563,
      "entity linking models": 37958,
      "using learned representations": 129805,
      "paper introduce model": 83996,
      "knowledge base entity": 57771,
      "30 points accuracy": 906,
      "sentence semantic similarity": 106064,
      "mean word embeddings": 67616,
      "offtheshelf pretrained language": 81630,
      "better zeroshot performance": 13774,
      "research area natural": 100412,
      "namedentity recognition model": 76137,
      "recent advances language": 96405,
      "advances language modeling": 3993,
      "language models present": 59624,
      "reinforcement learning framework": 97811,
      "proposed approach generate": 93197,
      "nlp recent years": 79677,
      "paper present contributions": 84089,
      "nli benchmark datasets": 79536,
      "alleviate resource scarcity": 5051,
      "simple effective data": 109406,
      "processing tasks inspired": 91821,
      "tokenlevel sequence labeling": 122295,
      "experiments data sets": 40864,
      "small training sets": 110221,
      "language modeling response": 59464,
      "report stateoftheart performance": 99048,
      "domains data sets": 33757,
      "dataset transfer learning": 27246,
      "transfer learning present": 124121,
      "knowledge base queries": 57783,
      "domain transfer learning": 33682,
      "demonstrate effectiveness models": 28716,
      "multilingual question answering": 75347,
      "questions tydi qa": 95371,
      "dataset introduce new": 26986,
      "crosslingual document retrieval": 24944,
      "crosslingual pretrained models": 24994,
      "experimental results suggest": 40708,
      "qa challenging task": 94499,
      "question answering data": 95034,
      "results wide variety": 102330,
      "demonstrate stateoftheart performance": 28869,
      "code model checkpoints": 18117,
      "representations emerging neural": 99621,
      "models spoken language": 74084,
      "language identification deep": 59131,
      "identification paper present": 51417,
      "language understanding largescale": 60233,
      "setting new state": 107763,
      "state art countless": 112436,
      "code data models": 18079,
      "publicly available unsupervised": 94327,
      "available unsupervised crosslingual": 11140,
      "adaptation multilingual pretrained": 3236,
      "multilingual pretrained language": 75337,
      "better adaptation performance": 13503,
      "results target languages": 102252,
      "dataset used train": 27256,
      "used train test": 128822,
      "investigate possible ways": 56797,
      "parsing recent years": 85213,
      "abstract syntax tree": 1783,
      "work propose alternative": 134711,
      "zeroshot semantic parsing": 135523,
      "695 exact match": 1202,
      "crosslingual eventcentric knowledge": 24955,
      "underresourced languages article": 126728,
      "languages article presents": 60413,
      "eventcentric knowledge processing": 39559,
      "availability language resources": 10914,
      "human judgments results": 50886,
      "models improving robustness": 73374,
      "predicateargument structures existing": 88859,
      "approaches improve robustness": 8189,
      "robustness paper propose": 103122,
      "meaning help model": 67639,
      "recognize important parts": 97054,
      "important parts sentences": 52203,
      "bias training data": 13852,
      "training data does": 123452,
      "propose simple method": 93063,
      "multilingual contextual embeddings": 75220,
      "downstream language understanding": 34001,
      "data difficult expensive": 25848,
      "difficult expensive obtain": 31627,
      "gaussian process preference": 46550,
      "process preference learning": 91549,
      "preference learning gppl": 89240,
      "training data performance": 123511,
      "knowledge graph based": 57957,
      "datatotext generation task": 27814,
      "benchmark datasets paper": 12794,
      "seamlessly integrated existing": 104291,
      "existing language models": 40153,
      "model evaluate approach": 71104,
      "model showing significant": 72008,
      "showing significant improvements": 108432,
      "various dimensions including": 131081,
      "perform series analyses": 86066,
      "embeddings multitask training": 35818,
      "insights propose simple": 55544,
      "stateoftheart results terms": 112923,
      "embedding space using": 35501,
      "language models provides": 59642,
      "representations characters subword": 99542,
      "used dropin replacement": 128504,
      "performance social media": 86726,
      "linguistic codeswitching evaluation": 64439,
      "codeswitching evaluation lince": 18236,
      "distant supervision propose": 32475,
      "strong baseline systems": 113637,
      "conversations work propose": 23079,
      "work propose integrate": 134723,
      "propose new models": 92871,
      "human evaluations compared": 50846,
      "knowledge commonsense knowledge": 57830,
      "new dataset named": 78868,
      "prior work focuses": 90749,
      "questionanswering natural language": 95254,
      "enabling machines understand": 36412,
      "graph optimization problem": 48852,
      "graph train model": 48877,
      "previous best published": 90395,
      "multilingual models multilingual": 75292,
      "little known models": 64816,
      "stateoftheart multilingual models": 112761,
      "models mbert xlmr": 73558,
      "models text editing": 74185,
      "language models transfer": 59675,
      "large raw data": 61234,
      "largescale multilingual language": 61476,
      "benefit transfer learning": 12994,
      "translation nat models": 125000,
      "autoregressive machine translation": 10867,
      "models knowledge distillation": 73441,
      "encoders capture different": 36637,
      "performance experimental results": 86357,
      "results wmt14 englishgerman": 102338,
      "wmt16 englishromanian datasets": 132831,
      "proposed method addition": 93347,
      "results demonstrate multitask": 101695,
      "paper describes neural": 83833,
      "network based approaches": 77155,
      "based approaches process": 11517,
      "approaches process formation": 8291,
      "process formation splitting": 91491,
      "morphological analysis sanskrit": 74681,
      "analysis sanskrit texts": 5828,
      "sequence sequence prediction": 107063,
      "human reading comprehension": 50950,
      "comprehension machine comprehension": 20191,
      "used commonsense knowledge": 128435,
      "machine comprehension models": 65752,
      "investigate machine comprehension": 56779,
      "question answer dataset": 95016,
      "squad 20 dataset": 112075,
      "did perform better": 30919,
      "reading comprehension contextualized": 95977,
      "word embeddings encode": 133097,
      "word sense knowledge": 133535,
      "human language comprehension": 50898,
      "bert embedding space": 13110,
      "approaches paper present": 8275,
      "transformerbased models various": 124428,
      "models various nlu": 74324,
      "various nlu tasks": 131157,
      "event argument extraction": 39497,
      "argument extraction eae": 8942,
      "syntactic structures sentences": 116488,
      "addition introduce novel": 3442,
      "extensive experiments performed": 41918,
      "demonstrate benefits proposed": 28680,
      "benefits proposed model": 13021,
      "model leading stateoftheart": 71424,
      "leading stateoftheart performance": 61911,
      "translation nmt approaches": 125029,
      "quality automatic translation": 94601,
      "paper investigate challenges": 84014,
      "gained popularity recent": 46370,
      "use unlabeled data": 128349,
      "architecture achieves similar": 8606,
      "achieves similar performance": 2870,
      "convolution neural networks": 23145,
      "neural networks gcn": 78486,
      "vectors graphbased models": 131441,
      "overall contextual importance": 83223,
      "contextual importance scores": 22468,
      "obtained dependency tree": 81360,
      "using graph neural": 129717,
      "neural network predicts": 78376,
      "parser penn treebank": 84983,
      "penn treebank ptb": 85862,
      "treebank ptb chinese": 125646,
      "performs par stateoftheart": 87014,
      "use dependency trees": 127995,
      "parser establishes new": 84950,
      "wet lab protocols": 132463,
      "task wnut 2020": 118850,
      "ner task 13": 77089,
      "data annotation process": 25606,
      "learning introduce new": 62660,
      "dataset training evaluating": 27244,
      "language models data": 59532,
      "models data collected": 73012,
      "distributional semantics models": 32730,
      "avenues future work": 11169,
      "selfsupervised learning unsupervised": 104943,
      "work explore unsupervised": 134517,
      "model builds recent": 70793,
      "builds recent work": 14909,
      "word embeddings bert": 133055,
      "data augmentation training": 25655,
      "systems typically rely": 117212,
      "large amounts highquality": 61010,
      "amounts highquality training": 5336,
      "existing datasets limited": 40106,
      "test sets previous": 120529,
      "language pairs previous": 59775,
      "automated speech recognition": 10469,
      "tackling natural language": 117325,
      "require manual effort": 100176,
      "obtained results competitive": 81404,
      "results competitive state": 101613,
      "systems sentiment analysis": 117136,
      "roman urdu text": 103247,
      "usage social media": 127870,
      "detection twitter using": 30090,
      "traditional machine learning": 122828,
      "models previous studies": 73795,
      "model multitask learning": 71546,
      "hierarchical bilstm model": 49940,
      "hierarchical structure model": 50006,
      "loss function helps": 65273,
      "stateoftheart machine learning": 112694,
      "pretraining neural network": 90298,
      "representations sentences experimental": 99881,
      "recently graph neural": 96693,
      "received increasing attention": 96363,
      "increasing attention research": 53662,
      "attention research community": 9994,
      "documents address issues": 33177,
      "representation learning extensive": 99288,
      "learning extensive experiments": 62572,
      "extensive experiments various": 41936,
      "various benchmark datasets": 131052,
      "levels natural languages": 63555,
      "neural networks capture": 78451,
      "multihead attention mechanism": 75125,
      "demonstrate proposed architecture": 28837,
      "simulated multiple reference": 109640,
      "multiple reference training": 75659,
      "reference training smrt": 97541,
      "al 2020 use": 4651,
      "conversational question answering": 23022,
      "based importance sampling": 11766,
      "method able improve": 68580,
      "dialog neural models": 30576,
      "neural models shown": 78197,
      "exhibit strong performance": 40006,
      "answering vqa tasks": 6722,
      "singleturn question answering": 109856,
      "error analysis indicates": 38282,
      "spoken language world": 112010,
      "information flows input": 54621,
      "model performance previous": 71728,
      "translation nonautoregressive translation": 125067,
      "nonautoregressive translation nat": 80162,
      "significantly accelerates inference": 108877,
      "accelerates inference process": 1875,
      "entire target sequence": 37718,
      "datasets approach consistently": 27314,
      "does bert look": 33331,
      "analysis berts attention": 5517,
      "models bert shown": 72839,
      "different attention heads": 31016,
      "attention heads transformer": 9853,
      "labeled data study": 58444,
      "stateoftheart data augmentation": 112628,
      "tasks including text": 119191,
      "including text classification": 53395,
      "machine translation study": 66243,
      "reduce computational cost": 97319,
      "quality generated data": 94673,
      "data propose methods": 26289,
      "propose methods adapt": 92777,
      "language models case": 59526,
      "models achieve high": 72666,
      "performance various tasks": 86847,
      "finegrained linguistic knowledge": 44364,
      "models performance results": 73734,
      "linguistic knowledge capture": 64501,
      "transformerbased neural language": 124431,
      "models shown stateoftheart": 74032,
      "shown stateoftheart performance": 108528,
      "models represent abstract": 73928,
      "models performance tasks": 73735,
      "performance tasks require": 86789,
      "despite high performance": 29693,
      "effective method incorporate": 34707,
      "use contextual embeddings": 127961,
      "machine translation limited": 66032,
      "learning solve problem": 63038,
      "input sentence way": 55425,
      "experimental results translation": 40714,
      "results translation tasks": 102287,
      "multiple strong baselines": 75707,
      "crosslingual text representations": 25024,
      "answer question present": 6562,
      "propose bertbased dual": 92574,
      "experiments recently released": 41112,
      "test dataset proposed": 120442,
      "performs better existing": 86986,
      "existing state art": 40289,
      "contribute improvement performance": 22755,
      "describes contribution semeval": 29389,
      "compare results produced": 19292,
      "trained predict word": 123243,
      "alexa prize 2018": 4674,
      "graph allows utilize": 48757,
      "named entities provided": 76032,
      "plenty room improvement": 87750,
      "poses significant challenges": 88276,
      "trained transformer models": 123316,
      "models different perspectives": 73072,
      "different perspectives experimental": 31330,
      "perspectives experimental results": 87176,
      "experimental results language": 40627,
      "language pairs training": 59783,
      "number interesting findings": 80900,
      "interesting findings help": 56074,
      "findings help humans": 44297,
      "help humans better": 49733,
      "humans better analyze": 51070,
      "better analyze understand": 13509,
      "analyze understand improve": 6014,
      "improve transformer models": 52567,
      "models based observations": 72809,
      "observations propose new": 81182,
      "propose new training": 92889,
      "nlg critical component": 79516,
      "paper present approaches": 84084,
      "hybrid approach combining": 51174,
      "data ground truth": 25990,
      "translation nmt aims": 125028,
      "tasks recently proposed": 119444,
      "recently proposed approach": 96739,
      "proposed approach uses": 93219,
      "approach uses attention": 8002,
      "representation work propose": 99474,
      "nmt model called": 79896,
      "improves existing models": 52978,
      "gated linear unit": 46509,
      "linear unit glu": 64386,
      "multiple translation tasks": 75736,
      "models realworld applications": 73868,
      "annotated data resources": 6165,
      "data available large": 25676,
      "sequential transfer learning": 107260,
      "minimal training data": 70181,
      "machine answer questions": 65741,
      "answer questions based": 6564,
      "questions based given": 95279,
      "based given context": 11740,
      "previous studies tend": 90501,
      "graph experimental results": 48807,
      "reading comprehension dataset": 95979,
      "key component dialogue": 57552,
      "use active learning": 127880,
      "expressions natural language": 41758,
      "inference nli fundamental": 54178,
      "inner workings neural": 55278,
      "neural network interpretability": 78309,
      "propose novel strategy": 92967,
      "instance attention mechanism": 55594,
      "neural model makes": 78151,
      "demonstrate usefulness approach": 28898,
      "machine translation adversarial": 65888,
      "adversarial text generation": 4168,
      "latent space representation": 61609,
      "generate fluent sentences": 46944,
      "loss text generation": 65302,
      "text generation demonstrate": 120984,
      "trained monolingual corpora": 123205,
      "monolingual corpora using": 74560,
      "generates fluent sentences": 47162,
      "compared monolingual baselines": 19392,
      "applied different languages": 7062,
      "paper explores possibility": 83932,
      "explores possibility improving": 41650,
      "data different related": 25845,
      "model joint partofspeech": 71380,
      "joint partofspeech pos": 57307,
      "dependency parsing shown": 29201,
      "universal dependency ud": 127301,
      "deep language models": 28242,
      "language modelling partofspeech": 59490,
      "modelling partofspeech tagging": 72615,
      "multitask model outperforms": 75877,
      "offer indepth analysis": 81565,
      "model bilingual lexicon": 70779,
      "approach leads substantial": 7673,
      "summarization task generating": 115575,
      "conduct experiments dataset": 20849,
      "highlight key challenges": 50265,
      "existing summarization models": 40305,
      "present evaluation results": 89478,
      "performance compared systems": 86241,
      "ground truth data": 49097,
      "challenge baseline approach": 16000,
      "source code publicly": 110723,
      "knowledge base paper": 57775,
      "systems generative models": 116922,
      "dialog systems using": 30605,
      "al 2016 proposed": 4615,
      "propose generalized framework": 92700,
      "helps achieve better": 49808,
      "systems wide range": 117235,
      "novel model architectures": 80644,
      "framework developing nlp": 45495,
      "model achieves score": 70588,
      "models trained labeled": 74207,
      "make best use": 66626,
      "learning task text": 63086,
      "task text classification": 118787,
      "semisupervised learning methods": 105608,
      "significant performance boosts": 108834,
      "dataset yields better": 27286,
      "semisupervised learning strategies": 105611,
      "50 training data": 1078,
      "training data points": 123512,
      "entity embedding methods": 37929,
      "methods unified framework": 69823,
      "entity linking tasks": 37965,
      "different model architectures": 31263,
      "model pretraining proven": 71780,
      "pretraining proven useful": 90310,
      "understanding tasks paper": 126989,
      "making best use": 66826,
      "learning framework based": 62594,
      "pseudo labels supervise": 94181,
      "training large amounts": 123677,
      "dataset relatively small": 27153,
      "wmt20 paper describes": 132868,
      "translation task main": 125334,
      "approach jointly training": 7656,
      "current approaches use": 25261,
      "paper explore ways": 83928,
      "question answering extensive": 95045,
      "learning approach sentiment": 62349,
      "sentiment analysis deals": 106581,
      "positive negative polarities": 88332,
      "comparative analysis proposed": 19203,
      "knowledge stored parameters": 58185,
      "external knowledge sources": 42020,
      "knowledge pretrained language": 58116,
      "language models outperform": 59609,
      "outperform stateoftheart model": 82759,
      "word embeddings various": 133238,
      "computational linguistic methods": 20391,
      "domain adaptation pretrained": 33445,
      "language models largescale": 59578,
      "yields significant performance": 135430,
      "language models proven": 59639,
      "method automatically extract": 68655,
      "source language proposed": 110778,
      "language proposed model": 60002,
      "achieves significant performance": 2862,
      "creation lexical resources": 24720,
      "processing techniques analyze": 91835,
      "study aims develop": 114313,
      "machine learningbased approach": 65838,
      "temporal information topic": 120109,
      "topic received attention": 122565,
      "extract informative features": 42090,
      "existing neural approaches": 40225,
      "approaches explicitly use": 8157,
      "dependency parsing dp": 29179,
      "pos tags universal": 88245,
      "lstmbased models ner": 65684,
      "informationseeking conversation systems": 55120,
      "information recent studies": 54901,
      "based heuristic rules": 11751,
      "task propose reinforced": 118576,
      "learning method trained": 62731,
      "trained endtoend manner": 123131,
      "human annotations specifically": 50748,
      "performance extensive experiments": 86367,
      "experiments standard benchmarks": 41151,
      "methods case studies": 69364,
      "achieving best results": 2933,
      "best results compared": 13439,
      "results compared baselines": 101599,
      "variety evaluation metrics": 130977,
      "significant improvement existing": 108771,
      "improvement existing online": 52707,
      "recent years existing": 96612,
      "years existing approaches": 135262,
      "overcome limitations existing": 83293,
      "propose method automatic": 92763,
      "experiments demonstrate superiority": 40895,
      "demonstrate superiority approach": 28880,
      "approaches mainly based": 8227,
      "used training test": 128826,
      "datasets best knowledge": 27334,
      "learning models designed": 62779,
      "traditional natural language": 122849,
      "models plms like": 73750,
      "plms like bert": 87757,
      "extraction methods different": 42389,
      "commonsense factual knowledge": 18985,
      "stateoftheart architecture neural": 112587,
      "aspectbased sentiment analysis": 9356,
      "online product reviews": 81792,
      "different aspects sentence": 31013,
      "sentiment analysis absa": 106561,
      "speakers dataset consists": 111312,
      "learning model achieved": 62757,
      "aspect term extraction": 9349,
      "various downstream nlp": 131086,
      "nlp tasks challenging": 79710,
      "training data best": 123436,
      "pretraining data augmentation": 90245,
      "data augmentation strategies": 25648,
      "simple efficient way": 109421,
      "transformer model using": 124345,
      "language models question": 59644,
      "answering recent works": 6687,
      "shown language models": 108489,
      "capture different types": 15291,
      "different types knowledge": 31531,
      "paper ask question": 83742,
      "range datasets demonstrate": 95554,
      "based neural architectures": 11886,
      "neural architectures models": 77840,
      "high accuracy standard": 50035,
      "prior state art": 90731,
      "multiple languages method": 75594,
      "evaluate method english": 38853,
      "large monolingual corpora": 61157,
      "small bilingual dictionary": 110139,
      "bilingual dictionary pair": 14033,
      "types event triggers": 126282,
      "injecting word information": 55266,
      "benchmark datasets model": 12793,
      "novel event types": 80566,
      "attempts solve problem": 9776,
      "propagation paper present": 92395,
      "models work better": 74359,
      "achieves best results": 2742,
      "best results benchmark": 13438,
      "new dataset interpersonal": 78866,
      "crucial step understanding": 25173,
      "relation extraction named": 98003,
      "extraction named entities": 42401,
      "named entities texts": 76038,
      "propose task relation": 93108,
      "task relation classification": 118624,
      "relation classification tasks": 97959,
      "challenging existing models": 16252,
      "training data annotated": 123427,
      "dialogue knowledge transfer": 30698,
      "knowledge transfer network": 58212,
      "address problem robustness": 3749,
      "using indomain data": 129752,
      "information form unstructured": 54624,
      "using world knowledge": 130391,
      "ondevice mobile phone": 81669,
      "novel ranking algorithm": 80697,
      "ranking algorithm extract": 95663,
      "using dependency parsing": 129559,
      "dependency parsing codeswitching": 29173,
      "syntactic structure english": 116482,
      "assess quality generated": 9488,
      "generated codeswitched data": 47061,
      "codeswitched data using": 18218,
      "finetuning large pretrained": 44472,
      "unidirectional language model": 127095,
      "policy optimization endtoend": 87964,
      "domains limited data": 33806,
      "discuss advantages disadvantages": 32233,
      "wnut 2020 shared": 132882,
      "2020 shared task1": 731,
      "language processing introduce": 59875,
      "knowledge graph completion": 57958,
      "parallel corpus data": 84616,
      "models best model": 72843,
      "gave best results": 46558,
      "size training data": 109948,
      "dependencies natural language": 29098,
      "pretrained large unannotated": 90114,
      "pretrained models use": 90155,
      "nlu tasks glue": 79816,
      "glue benchmark approach": 48310,
      "architecture neural machine": 8708,
      "machine translation encoderdecoder": 65969,
      "widely used neural": 132580,
      "used neural machine": 128654,
      "address issue present": 3695,
      "issue present novel": 57021,
      "present novel architecture": 89600,
      "language models successful": 59665,
      "pretrained models new": 90145,
      "generated gpt2 model": 47082,
      "encode contextual information": 36424,
      "language model predicts": 59372,
      "model predicts word": 71763,
      "future work focus": 46322,
      "real applications especially": 96053,
      "content paper describes": 21917,
      "paper describes simple": 83846,
      "diverse set domains": 32845,
      "scenarios training data": 103869,
      "translation mt reached": 124983,
      "create evaluation set": 24614,
      "distributed word embeddings": 32635,
      "labels using graph": 58654,
      "standard benchmark datasets": 112205,
      "surpass strong baselines": 116098,
      "compile larger corpus": 19713,
      "finitestate transducers fsts": 44552,
      "yield better performance": 135333,
      "embeddings text classification": 35980,
      "task consisting assigning": 118010,
      "word embeddings text": 133225,
      "exploited word embeddings": 41457,
      "word embeddings inherently": 133136,
      "text representations based": 121253,
      "representations based word": 99513,
      "word embeddings incorporating": 133133,
      "distributions training data": 32773,
      "data propose use": 26293,
      "novel weighting scheme": 80777,
      "performance gap increases": 86409,
      "sentences plays important": 106446,
      "sequence paper propose": 107027,
      "model takes advantage": 72137,
      "based assumption source": 11529,
      "language pairs model": 59766,
      "pairs model outperforms": 83585,
      "outperforms previous unsupervised": 82956,
      "text generation recently": 121005,
      "text generation problem": 121001,
      "learning framework using": 62605,
      "text corpora generating": 120826,
      "domains tackle problem": 33869,
      "propose principled method": 92999,
      "adding small perturbations": 3394,
      "small perturbations input": 110186,
      "positive negative pairs": 88331,
      "generated using method": 47149,
      "text summarization question": 121350,
      "reading comprehension question": 96006,
      "question dialog history": 95150,
      "human annotators write": 50752,
      "experiments conducted dataset": 40840,
      "dataset results proposed": 27170,
      "model brings substantial": 70787,
      "brings substantial improvements": 14654,
      "strong baselines tasks": 113655,
      "release largescale dataset": 98457,
      "recently deep learning": 96676,
      "consistently outperforms conventional": 21431,
      "present fake news": 89491,
      "neural networks nlp": 78525,
      "news detection task": 79332,
      "obtained average f1score": 81348,
      "transfer learning using": 124143,
      "learning using multilingual": 63151,
      "inflected forms verbs": 54281,
      "improve stateoftheart performance": 52550,
      "stateoftheart performance sentiment": 112844,
      "tasks deep learning": 119030,
      "model achieves accuracy": 70552,
      "achieves accuracy 60": 2734,
      "blackbox nature neural": 14253,
      "nature neural models": 76665,
      "aims generate natural": 4532,
      "paper investigate extent": 84016,
      "reason natural language": 96200,
      "train generative models": 122939,
      "generative models capable": 47742,
      "dataset neural approach": 27053,
      "dataset containing million": 26827,
      "context experimental results": 22093,
      "methods address issue": 69297,
      "experiments conducted realworld": 40843,
      "conducted realworld datasets": 20939,
      "f1 score 9090": 42676,
      "performances downstream tasks": 86880,
      "deep transformerbased models": 28433,
      "different forms communication": 31154,
      "user work propose": 129059,
      "demonstrate significant improvements": 28864,
      "binary classification tasks": 14121,
      "outperform baselines achieve": 82694,
      "extensive experiment results": 41880,
      "results developed participation": 101722,
      "shared task consists": 108030,
      "task consists subtasks": 118013,
      "computer science domain": 20489,
      "obtained f1 scores": 81372,
      "large number labeled": 61178,
      "training data proposed": 123519,
      "millions parallel sentences": 70120,
      "different granularities text": 31169,
      "representations source target": 99899,
      "paper study intrinsic": 84444,
      "transformerbased masked language": 124422,
      "downstream tasks results": 34054,
      "transferred natural language": 124232,
      "language downstream tasks": 58987,
      "dialog systems present": 30601,
      "qualitative human evaluations": 94565,
      "draw meaningful conclusions": 34118,
      "general question answering": 46702,
      "existing approaches consider": 40051,
      "simultaneously paper propose": 109688,
      "paper propose cointeractive": 84223,
      "jointly perform tasks": 57378,
      "model successfully captures": 72108,
      "information achieve stateoftheart": 54356,
      "word representations bert": 133445,
      "language text using": 60165,
      "text using universal": 121400,
      "pos morphological annotations": 88211,
      "languages neural machine": 60763,
      "nmt systems use": 79983,
      "improvements neural machine": 52881,
      "based knowledge distillation": 11788,
      "deep transformer model": 28430,
      "benchmarks validate effectiveness": 12940,
      "relation extraction predict": 98009,
      "extraction predict relation": 42438,
      "predict relation type": 88924,
      "relation type entities": 98080,
      "type entities mentioned": 126197,
      "dialogue paper propose": 30718,
      "english question answering": 37257,
      "question answering model": 95060,
      "inner product search": 55276,
      "analysis shows approach": 5850,
      "training data target": 123547,
      "methods machine learning": 69606,
      "conditional masked language": 20765,
      "inference nli tasks": 54183,
      "multilingual models large": 75289,
      "models large margin": 73462,
      "learned representations propose": 62252,
      "representations propose simple": 99830,
      "improving lexical choice": 53110,
      "demonstrate effectiveness universality": 28724,
      "effectiveness universality proposed": 34972,
      "universality proposed approach": 127340,
      "proposed approach extensive": 93195,
      "improves performance reducing": 53017,
      "bleu points respectively": 14310,
      "points respectively source": 87873,
      "respectively source code": 101167,
      "source code released": 110725,
      "dynamic early exiting": 34305,
      "instead propose new": 55675,
      "mechanism experimental results": 67982,
      "language understanding aims": 60203,
      "mechanism extensive experiments": 67985,
      "stateoftheart results benchmarks": 112895,
      "models trained random": 74226,
      "realworld scenarios propose": 96177,
      "significantly improves model": 108948,
      "model performance various": 71731,
      "various evaluation metrics": 131091,
      "multiple languages paper": 75596,
      "broad range topics": 14681,
      "high quality texts": 50123,
      "inspire future work": 55563,
      "language identification important": 59132,
      "text processing pipelines": 121211,
      "research field paper": 100512,
      "automatic language identification": 10574,
      "languages lexical level": 60688,
      "language identification systems": 59144,
      "based supervised machine": 12090,
      "machine learning deep": 65792,
      "success pretrained models": 115116,
      "use beam search": 127919,
      "beam search neural": 12607,
      "model synthetic real": 72132,
      "models event temporal": 73180,
      "language models ptlms": 59643,
      "nlp tasks struggle": 79771,
      "event temporal relations": 39548,
      "selfsupervised learning objectives": 104942,
      "framework event temporal": 45532,
      "performances relation extraction": 86893,
      "representation pretrained language": 99383,
      "approach shown outperform": 7891,
      "language nl explanations": 59719,
      "sequence sequence seq2seq": 107064,
      "sequence seq2seq models": 107051,
      "large labeled data": 61124,
      "settings neural machine": 107825,
      "language processing aims": 59833,
      "aims translate natural": 4582,
      "recent years endtoend": 96610,
      "sampleefficient pretraining task": 103576,
      "instead training model": 55688,
      "pretraining masked language": 90281,
      "evaluate model multiple": 38866,
      "model multiple arabic": 71542,
      "smaller model size": 110238,
      "language modeling understanding": 59480,
      "trained large corpus": 123179,
      "applications language generation": 6951,
      "generation models paper": 47485,
      "models paper develop": 73698,
      "different tasks including": 31480,
      "synthetic news generation": 116635,
      "model achieves perplexity": 70583,
      "propose novel algorithm": 92894,
      "single pass experimental": 109781,
      "pass experimental results": 85606,
      "novel model structure": 80647,
      "comparable performance strong": 19163,
      "achieves absolute improvements": 2732,
      "different languages existing": 31201,
      "existing multilingual machine": 40216,
      "recent success language": 96534,
      "success language model": 115084,
      "multilingual parallel data": 75325,
      "significant improvements wmt": 108816,
      "performance multilingual language": 86547,
      "pretrained multilingual language": 90163,
      "set typologically diverse": 107623,
      "readily available pretrained": 95968,
      "important role downstream": 52240,
      "performance multilingual model": 86549,
      "models achieve impressive": 72668,
      "achieve impressive performance": 2488,
      "large monolingual parallel": 61159,
      "lowresource languages paper": 65524,
      "new training method": 79227,
      "encourages model align": 36769,
      "multiple languages monolingual": 75595,
      "pseudoparallel sentence pairs": 94194,
      "outperforms existing crosslingual": 82890,
      "existing crosslingual models": 40098,
      "delivers new stateoftheart": 28625,
      "results various crosslingual": 102311,
      "processing long documents": 91702,
      "stateoftheart language modeling": 112685,
      "understanding tasks text": 126991,
      "text classification question": 120759,
      "question answering understanding": 95122,
      "advances pretrained language": 4012,
      "temporal relationships events": 120127,
      "model able generate": 70516,
      "cognitive annotation ucca": 18264,
      "annotation ucca abend": 6400,
      "ucca abend rappoport": 126524,
      "abend rappoport 2013": 1476,
      "annotation scheme based": 6370,
      "al 2020 using": 4652,
      "better language modeling": 13615,
      "training time surprisingly": 123928,
      "relative position embeddings": 98363,
      "absolute position embeddings": 1747,
      "word embeddings efficiently": 133093,
      "dataset diverse text": 26878,
      "training largescale language": 123681,
      "performance downstream evaluations": 86318,
      "publicly available code": 94295,
      "models perform experiments": 73726,
      "language proficiency levels": 59992,
      "covid19 fake news": 24563,
      "information fake news": 54613,
      "models bert albert": 72832,
      "model trained evaluated": 72193,
      "2021 shared task": 742,
      "shared task covid19": 108031,
      "task covid19 fake": 118031,
      "news detection english": 79330,
      "language understanding evaluation": 60220,
      "language inference question": 59191,
      "inference question answering": 54206,
      "question answering benchmark": 95025,
      "diverse nlu tasks": 32830,
      "covering text classification": 24548,
      "text classification sequence": 120767,
      "language understanding benchmark": 60207,
      "stateoftheart results outperforming": 112916,
      "training nlp models": 123748,
      "generation methods rely": 47474,
      "training evaluation different": 123619,
      "ner tasks natural": 77091,
      "language modeling introduce": 59435,
      "introduce new pretraining": 56493,
      "new general language": 78935,
      "extensive analysis shows": 41857,
      "methods different tasks": 69448,
      "deep bidirectional transformers": 28208,
      "new stateoftheart english": 79157,
      "processing nlp provides": 91751,
      "use large pretrained": 128115,
      "multilingual pretrained transformer": 75340,
      "models code publicly": 72917,
      "joint representation learning": 57317,
      "relation classification relation": 97955,
      "classification relation classification": 17375,
      "classification rc task": 17366,
      "fundamental tasks information": 46123,
      "relation information entity": 98042,
      "information entity pairs": 54538,
      "distant supervision methods": 32473,
      "data supervised learning": 26519,
      "introduce noise data": 56506,
      "order enable neural": 82315,
      "training set given": 123848,
      "better feature representation": 13587,
      "best knowledge attempt": 13344,
      "introduce loss function": 56452,
      "experiments conducted fewrel": 40842,
      "achieved significant improvements": 2695,
      "detection shared task": 30051,
      "paper present contribution": 84088,
      "propose novel technique": 92969,
      "word prediction nwp": 133415,
      "data set composed": 26402,
      "lowlevel features character": 65474,
      "features character ngrams": 43393,
      "character ngrams based": 16455,
      "commonsense causal reasoning": 18983,
      "adversarial training data": 4171,
      "training data augmentation": 123429,
      "commonsense reasoning task": 19009,
      "task requires complex": 118641,
      "training data task": 123548,
      "dataset paper presents": 27088,
      "paper presents number": 84195,
      "adversarial training generating": 4174,
      "data augmentation using": 25657,
      "boost model performance": 14433,
      "significant improvement performance": 108773,
      "improvement performance robustness": 52740,
      "training paper propose": 123771,
      "empirical experiments using": 36167,
      "news detection paper": 79331,
      "main goal work": 66426,
      "language inference present": 59188,
      "models perform worse": 73731,
      "dataset challenging english": 26783,
      "stress tests models": 113583,
      "model using graph": 72280,
      "models proposed recently": 73826,
      "dependency relations context": 29221,
      "relations address problem": 98107,
      "neural networks experiments": 78479,
      "state tracking using": 112532,
      "forms core component": 45240,
      "data work explore": 26630,
      "model existing stateoftheart": 71121,
      "nlp tasks especially": 79716,
      "network models propose": 77351,
      "knowledge evaluate proposed": 57907,
      "proposed approach natural": 93200,
      "approach consistently outperforms": 7450,
      "outperforms random baseline": 82970,
      "significant gains observed": 108758,
      "tasks tabletotext generation": 119544,
      "hope investigation help": 50630,
      "neural networks transformers": 78595,
      "recognition text classification": 97030,
      "text classification language": 120732,
      "resource languages like": 100859,
      "represent linguistic features": 99119,
      "corpus experiments conducted": 23793,
      "experiments conducted text": 40847,
      "conducted text classification": 20953,
      "stateoftheart publicly available": 112876,
      "embeddings downstream tasks": 35671,
      "best models achieve": 13379,
      "learning sequence tagging": 63013,
      "annotating training data": 6275,
      "sequence tagging task": 107083,
      "models active learning": 72710,
      "active learning framework": 3113,
      "better computational performance": 13547,
      "deep active learning": 28196,
      "word alignment parallel": 132911,
      "crosslingual transfer language": 25031,
      "word embeddings derived": 133078,
      "competitive results word": 19686,
      "training parallel data": 123775,
      "parallel data paper": 84646,
      "paper examine methods": 83902,
      "consistently outperform previous": 21425,
      "performance different language": 86299,
      "generalize new domains": 46814,
      "new domains unseen": 78894,
      "unseen slot types": 127544,
      "conditional language model": 20760,
      "improves joint goal": 52989,
      "zeroshot domain adaptation": 135498,
      "multiwoz 21 dataset": 75955,
      "present transfer learning": 89747,
      "uses transformer model": 129294,
      "cnn model text": 18012,
      "model text classification": 72163,
      "terms extraction using": 120323,
      "words phrases given": 134118,
      "paper focuses unsupervised": 83955,
      "modeling unstructured knowledge": 72580,
      "unstructured knowledge access": 127574,
      "dialogue systems restricted": 30787,
      "knowledge selection knowledgegrounded": 58166,
      "neural baseline models": 77853,
      "different largescale pretrained": 31219,
      "generalization unseen data": 46800,
      "bert transformer model": 13243,
      "internet social media": 56195,
      "text written language": 121421,
      "written language models": 135133,
      "model able detect": 70512,
      "using twitter api": 130327,
      "model obtained accuracy": 71602,
      "language model results": 59390,
      "performance bert model": 86177,
      "novel twostage framework": 80759,
      "news articles paper": 79302,
      "articles paper presents": 9206,
      "sentences given news": 106336,
      "given news article": 48070,
      "spoken word recognition": 112023,
      "language models yielded": 59690,
      "models make use": 73549,
      "previous work work": 90565,
      "work propose train": 134746,
      "translation systems using": 125321,
      "source language words": 110787,
      "words target language": 134257,
      "results human evaluation": 101833,
      "human evaluation indicate": 50818,
      "absolute improvement previous": 1736,
      "improvement previous work": 52750,
      "training models data": 123714,
      "new annotation tool": 78777,
      "annotation tool designed": 6393,
      "multimodal machine learning": 75440,
      "using twitter data": 130328,
      "dirichlet allocation topic": 31924,
      "gives promising results": 48205,
      "mean squared error": 67611,
      "training testing data": 123918,
      "models aspectbased sentiment": 72774,
      "aspects given document": 9387,
      "network cnn conditional": 77190,
      "cnn conditional random": 17997,
      "interactive attention network": 56014,
      "attention network based": 9945,
      "network based bidirectional": 77158,
      "identify sentiment polarity": 51551,
      "dataset results indicate": 27169,
      "deep generative models": 28236,
      "semisupervised learning deep": 105605,
      "different areas nlp": 31008,
      "document classification task": 32968,
      "supervised learning baselines": 115766,
      "widely used nlp": 132584,
      "machine translation documentlevel": 65956,
      "translation documentlevel machine": 124764,
      "alignment different languages": 4954,
      "embedding spaces different": 35503,
      "spaces different languages": 111094,
      "parallel data target": 84655,
      "languages benchmark data": 60425,
      "f1 scores compared": 42694,
      "introduce corpus chinese": 56401,
      "models generally perform": 73283,
      "different types information": 31530,
      "new approach based": 78786,
      "sentence representations experiments": 106037,
      "stateoftheart performance results": 112843,
      "performance results demonstrate": 86680,
      "neural networks integrate": 78493,
      "parsing lowresource languages": 85148,
      "model based multilingual": 70736,
      "unsupervised crosslingual transfer": 127618,
      "access source language": 1923,
      "source language data": 110772,
      "empirical study crosslingual": 36200,
      "dialogue systems benefit": 30767,
      "systems lowresource languages": 116987,
      "training dataset paper": 123569,
      "using multilingual pretrained": 129917,
      "chinese named entity": 16789,
      "pos tagging chinese": 88226,
      "language process nlp": 59831,
      "transformers bert model": 124446,
      "based attention mechanism": 11531,
      "function graph attention": 46020,
      "attention mechanism proposed": 9910,
      "using artificial neural": 129348,
      "answering tasks including": 6710,
      "numerical reasoning text": 81011,
      "endtoend reinforcement learning": 36960,
      "models syntactic dependency": 74147,
      "use composition functions": 127948,
      "small significant improvements": 110205,
      "improvements parsing accuracy": 52894,
      "accuracy analysis reveals": 2100,
      "challenge present solution": 16083,
      "different transformerbased pretrained": 31514,
      "f1score test dataset": 42720,
      "relation extraction multilingual": 98002,
      "xlmr conneau et": 135201,
      "models structured knowledge": 74106,
      "models achieved state": 72691,
      "art performance various": 9083,
      "current language models": 25291,
      "coreference resolution cdcr": 23362,
      "concepts text documents": 20646,
      "media propose new": 68158,
      "task aims identify": 117868,
      "inspire future research": 55562,
      "code dataset publicly": 18091,
      "manually labeled data": 67067,
      "improve classification accuracy": 52349,
      "experiments english datasets": 40933,
      "jointly predicting target": 57383,
      "performance current models": 86267,
      "language processing use": 59981,
      "news articles dataset": 79293,
      "semiautomated data collection": 105568,
      "dataset news article": 27057,
      "news articles based": 79289,
      "based syntactic structures": 12103,
      "property natural language": 92503,
      "finally propose method": 44222,
      "linguistic characteristics language": 64436,
      "generation task requires": 47662,
      "task requires reasoning": 118645,
      "text human language": 121037,
      "framework using combination": 45735,
      "language models better": 59520,
      "analysis emotion recognition": 5594,
      "emotion recognition paper": 36111,
      "model modern hebrew": 71527,
      "performance language tasks": 86484,
      "sentiment entire sentence": 106738,
      "bertbased language model": 13273,
      "language model outperforms": 59362,
      "data collection annotation": 25741,
      "models emotion detection": 73135,
      "knowledge selection knowledge": 58165,
      "approaches text classification": 8375,
      "text classification classification": 120713,
      "computer vision cv": 20495,
      "supervised learning setting": 115780,
      "classification task especially": 17430,
      "highquality labeled data": 50388,
      "labeled data required": 58440,
      "data required training": 26346,
      "required training new": 100227,
      "training new domain": 123746,
      "transfer knowledge source": 124067,
      "domain target domain": 33668,
      "work evaluate existing": 134495,
      "models propose approach": 73818,
      "approach bridge gap": 7398,
      "class imbalance issue": 17039,
      "systems recent advances": 117101,
      "generation process producing": 47568,
      "modern neural networks": 74405,
      "output beam search": 83055,
      "bleu points vanilla": 14315,
      "proposed method outperformed": 93381,
      "downstream tasks word": 34060,
      "sense disambiguation semantic": 105666,
      "similarity information retrieval": 109243,
      "fasttext models trained": 43206,
      "word analogy tasks": 132920,
      "models word analogy": 74348,
      "understanding language models": 126874,
      "models recent progress": 73874,
      "performance gains nlp": 86403,
      "helps improve performance": 49818,
      "downstream tasks finetuning": 34038,
      "language models commonly": 59530,
      "popular pretrained language": 88119,
      "present recent advances": 89661,
      "phrases sentences paragraphs": 87458,
      "topological data analysis": 122681,
      "analysis test approach": 5894,
      "language processing detect": 59861,
      "task using handcrafted": 118832,
      "crucial machine translation": 25148,
      "human labelled data": 50893,
      "training data train": 123554,
      "models achieve comparable": 72661,
      "important research problem": 52230,
      "applied various fields": 7141,
      "inefficient work propose": 54077,
      "based model uses": 11851,
      "semantic information encoded": 105075,
      "model different datasets": 70994,
      "different datasets different": 31078,
      "detailed ablation studies": 29751,
      "code datasets used": 18097,
      "ninth dialog technology": 79492,
      "task generate responses": 118237,
      "task divided subtasks": 118111,
      "existing approaches rely": 40061,
      "supervised training data": 115848,
      "important information context": 52172,
      "tackle issues propose": 117293,
      "finally conduct extensive": 44160,
      "rely synthetic datasets": 98751,
      "generation models produce": 47486,
      "models produce synthetic": 73811,
      "data release code": 26333,
      "propose sequencetosequence seq2seq": 93043,
      "transformer model generate": 124342,
      "goes long way": 48426,
      "study dependency parsing": 114350,
      "rich languages neural": 102760,
      "neural dependency parsing": 77889,
      "remarkable performance domains": 98835,
      "massive labeled data": 67332,
      "labeled data limits": 58437,
      "languages work focus": 60976,
      "rich languages mrls": 102759,
      "distribution natural language": 32669,
      "typically trained large": 126464,
      "amounts data resulting": 5333,
      "large models require": 61155,
      "rate wer reduction": 95808,
      "dialogue generation opendomain": 30680,
      "generation existing methods": 47395,
      "datasets automatic human": 27323,
      "evaluations model significantly": 39474,
      "heavily training data": 49662,
      "hindienglish codemixed data": 50485,
      "years emotion detection": 135256,
      "huge amounts data": 50715,
      "attention paper present": 9964,
      "hinglish dataset labelled": 50500,
      "code mixed tweets": 18112,
      "mixed tweets using": 70404,
      "tweets using bilingual": 126070,
      "using bilingual word": 129412,
      "embeddings derived fasttext": 35642,
      "derived fasttext word2vec": 29350,
      "fasttext word2vec approaches": 43210,
      "learning models including": 62781,
      "models including cnns": 73379,
      "including cnns lstms": 53272,
      "cnns lstms bidirectional": 18039,
      "lstms bidirectional lstms": 65704,
      "bidirectional lstms attention": 13960,
      "like bert roberta": 64024,
      "based bert model": 11547,
      "model outperforms models": 71654,
      "giving best performance": 48214,
      "curriculum learning sentiment": 25424,
      "curriculum learning cl": 25423,
      "suffer catastrophic forgetting": 115209,
      "based ensemble learning": 11683,
      "submitted shared task": 114773,
      "based model built": 11847,
      "topic models latent": 122551,
      "models latent dirichlet": 73469,
      "features model results": 43611,
      "using majority voting": 129854,
      "outperforms existing state": 82897,
      "f1 score validation": 42691,
      "training data selection": 123533,
      "web machine translation": 132239,
      "monolingual target data": 74621,
      "results models outperform": 101958,
      "sentence selection as2": 106060,
      "modern question answering": 74413,
      "english languages rich": 37190,
      "translated target language": 124567,
      "documents proposed method": 33278,
      "task sufficient data": 118761,
      "general language understanding": 46669,
      "translate source language": 124543,
      "language neural model": 59715,
      "target language task": 117655,
      "translation models learn": 124956,
      "data learning models": 26084,
      "current neural networks": 25314,
      "systems models use": 117005,
      "investigate choice subword": 56733,
      "tasks morphological probing": 119309,
      "morphological probing pos": 74721,
      "probing pos tagging": 90901,
      "massively multilingual models": 67342,
      "best results obtained": 13441,
      "strategy works best": 113547,
      "publicly release code": 94330,
      "using fasttext embeddings": 129668,
      "exist english language": 40022,
      "neglected long time": 76972,
      "using skipgram model": 130185,
      "language modeling transfer": 59477,
      "modeling transfer learning": 72574,
      "challenge multilingual models": 16057,
      "languages used perform": 60949,
      "used perform crosslingual": 128686,
      "language transfer learning": 60182,
      "language modeling used": 59482,
      "modeling mlm task": 72476,
      "nmt systems translate": 79982,
      "previous work limited": 90534,
      "develop better systems": 30182,
      "compare different methods": 19238,
      "improving machine translation": 53114,
      "different note types": 31305,
      "identifying protected health": 51614,
      "protected health information": 93661,
      "health information phi": 49620,
      "data sources medical": 26476,
      "dialogue systems using": 30794,
      "adapt new unseen": 3188,
      "using minimal data": 129885,
      "previous stateoftheart dialogue": 90471,
      "codeswitched data paper": 18217,
      "transfer learning recent": 124128,
      "costly timeconsuming process": 24392,
      "pretraining multilingual bert": 90291,
      "machine translation reinforcement": 66196,
      "translation reinforcement learning": 125209,
      "propose address problem": 92532,
      "heterogeneous data sources": 49853,
      "model training propose": 72222,
      "training propose novel": 123800,
      "specifically jointly train": 111562,
      "jointly train modules": 57393,
      "different inductive biases": 31178,
      "bert model trained": 13173,
      "automated quality assessment": 10464,
      "deep contextualized language": 28215,
      "support quality assurance": 115986,
      "supervision work propose": 115932,
      "work propose bertbased": 134716,
      "hope speech detection": 50640,
      "speech detection using": 111675,
      "using logistic regression": 129829,
      "logistic regression random": 65041,
      "random forest svm": 95496,
      "majority voting ensemble": 66612,
      "obtained finetuning pretrained": 81374,
      "output layer second": 83087,
      "english tamil malayalam": 37300,
      "weighted f1 score": 132346,
      "vietnamese natural language": 131685,
      "word word left": 133659,
      "word left right": 133336,
      "neural model joint": 78150,
      "chinese benchmark datasets": 16738,
      "model achieves higher": 70572,
      "information text data": 55038,
      "broad range applications": 14675,
      "range applications information": 95542,
      "applications information retrieval": 6943,
      "summarization question answering": 115556,
      "information extraction sentiment": 54601,
      "extraction sentiment analysis": 42483,
      "f1 score test": 42689,
      "embedding learning word": 35433,
      "embedding learning methods": 35431,
      "learning methods require": 62744,
      "methods require large": 69729,
      "simple classifier trained": 109383,
      "classifier trained predict": 17587,
      "wide variety models": 132524,
      "intelligence ai machine": 55823,
      "ai machine learning": 4423,
      "prior work pretrained": 90761,
      "text representations useful": 121254,
      "language time paper": 60170,
      "report experiments languages": 99000,
      "knowledge base completion": 57765,
      "make datasets pretrained": 66651,
      "datasets pretrained models": 27629,
      "bert shown promising": 13227,
      "effectiveness neural network": 34922,
      "cui et al": 25224,
      "pretrain language model": 89975,
      "behavior different models": 12654,
      "different neural networks": 31297,
      "trained data set": 123107,
      "performance different models": 86300,
      "experiment shows performance": 40512,
      "model terms accuracy": 72154,
      "accuracy training efficiency": 2307,
      "brought great deal": 14712,
      "great deal models": 48999,
      "14 probing tasks": 319,
      "representations transformers achieved": 99945,
      "language processing model": 59887,
      "model based distilbert": 70728,
      "obtain good performance": 81287,
      "improve model explainability": 52416,
      "highly inflectional morphology": 50329,
      "inflectional morphology paper": 54296,
      "language models varying": 59688,
      "demonstrate methods effectiveness": 28786,
      "detection stance detection": 30063,
      "language models need": 59606,
      "language models significantly": 59659,
      "language modeling objectives": 59450,
      "linguistic information word": 64491,
      "results various experiments": 102312,
      "learning slot tagging": 63033,
      "proposed method extends": 93367,
      "processing applications general": 91617,
      "pretrained contextual embeddings": 90004,
      "bert using attention": 13249,
      "using attention mechanism": 129354,
      "data proposed method": 26296,
      "translating natural language": 124599,
      "power natural language": 88639,
      "task metrics data": 118392,
      "solutions lessons learned": 110578,
      "models able provide": 72647,
      "provide significant improvement": 93918,
      "sentiment analysis models": 106611,
      "progressive selfsupervised attention": 92192,
      "selfsupervised attention learning": 104935,
      "equipped attention mechanism": 38240,
      "frequent words sentiment": 45865,
      "words sentiment polarities": 134203,
      "propose progressive selfsupervised": 93004,
      "attention learning approach": 9870,
      "useful attention supervision": 128862,
      "attention supervision information": 10020,
      "based attention weights": 11533,
      "activemisleading influence correctincorrect": 3127,
      "influence correctincorrect prediction": 54306,
      "correctincorrect prediction instance": 24135,
      "masked subsequent iterations": 67305,
      "augment conventional training": 10252,
      "conventional training objective": 22908,
      "training objective regularization": 123759,
      "objective regularization term": 81112,
      "extracted active context": 42136,
      "active context words": 3107,
      "approach yields better": 8023,
      "yields better attention": 135400,
      "source code trained": 110726,
      "code trained models": 18159,
      "labels natural language": 58623,
      "rules natural language": 103422,
      "language making use": 59272,
      "words rarely seen": 134156,
      "challenging natural language": 16280,
      "language inference models": 59174,
      "neural networks understand": 78597,
      "ability neural networks": 1552,
      "neural networks generalize": 78487,
      "generalize unseen examples": 46822,
      "propose hierarchical model": 92711,
      "conversational model learning": 23017,
      "building humanlike conversational": 14846,
      "overcome issue propose": 83287,
      "weights encoder decoder": 132377,
      "model able leverage": 70519,
      "model train using": 72181,
      "empathetic response generation": 36134,
      "previous stateoftheart model": 90476,
      "standard arabic msa": 112189,
      "use sentence embeddings": 128266,
      "tags dependency trees": 117472,
      "best results achieved": 13436,
      "results achieved using": 101495,
      "bidirectional context propose": 13904,
      "lexically constrained decoding": 63857,
      "generation human evaluation": 47430,
      "present method generating": 89549,
      "evaluation using rouge": 39437,
      "human evaluation using": 50842,
      "text summarization model": 121348,
      "pretraining synthetic data": 90331,
      "require large labeled": 100162,
      "issues propose new": 57068,
      "semisupervised learning asr": 105604,
      "recognition asr models": 96814,
      "contrastive representation learning": 22740,
      "computer vision speech": 20500,
      "positive negative examples": 88326,
      "social media videos": 110433,
      "core component taskoriented": 23315,
      "component taskoriented dialogue": 20006,
      "contextual information multiturn": 22472,
      "prediction experimental results": 89059,
      "relations biomedical text": 98118,
      "unsupervised machine learning": 127659,
      "existing knowledge base": 40147,
      "leverage large unlabeled": 63596,
      "recent success deep": 96530,
      "objective taskspecific objective": 81122,
      "propose novel selfsupervised": 92962,
      "novel selfsupervised learning": 80713,
      "improves performance existing": 53013,
      "based models using": 11855,
      "data source code": 26471,
      "source code work": 110728,
      "aspect category sentiment": 9323,
      "category sentiment analysis": 15793,
      "review rating prediction": 102566,
      "improve overall performance": 52450,
      "scenarios public datasets": 103862,
      "predefined aspect categories": 88822,
      "hope release dataset": 50638,
      "dataset shed light": 27197,
      "sentiment analysis propose": 106636,
      "joint model outperforms": 57291,
      "paper explore effects": 83918,
      "language models build": 59523,
      "arabic modern standard": 8526,
      "arabic msa dialectal": 8530,
      "msa dialectal arabic": 74943,
      "compare different models": 19239,
      "multilingual training neural": 75387,
      "machine translation mnmt": 66046,
      "improving translation performance": 53179,
      "translation performance language": 125113,
      "highresource language pairs": 50424,
      "paper explores data": 83930,
      "explores data augmentation": 41645,
      "techniques improve performance": 119907,
      "answering qa dataset": 6667,
      "knowledge graphs kg": 57985,
      "question answering kgs": 95058,
      "dataset baseline models": 26759,
      "neural discourse segmentation": 77897,
      "graph structure neural": 48871,
      "active research area": 3121,
      "research area field": 100411,
      "generate correct sequence": 46920,
      "simultaneous speech translation": 109663,
      "artificial intelligence especially": 9251,
      "paper present results": 84131,
      "perform manual evaluation": 86031,
      "human machine performances": 50911,
      "automatic speech translation": 10672,
      "finetuning pretrained multilingual": 44499,
      "brings consistent improvements": 14646,
      "set annotated examples": 107359,
      "annotated examples results": 6193,
      "provide strong baseline": 93927,
      "multitask learning mtl": 75852,
      "model obtain stateoftheart": 71600,
      "results machine reading": 101899,
      "high interannotator agreement": 50080,
      "semantic annotation task": 104988,
      "using new manually": 129963,
      "convolutional networks existing": 23169,
      "semeval2021 task toxic": 105560,
      "task toxic spans": 118796,
      "toxic spans detection": 122712,
      "extract features based": 42084,
      "features based word": 43377,
      "sequencetosequence models trained": 107184,
      "cat got tongue": 15687,
      "error neural machine": 38311,
      "state art multilingual": 112452,
      "multilingual graphemetophoneme conversion": 75258,
      "translation language pairs": 124891,
      "resulting dataset available": 101442,
      "substantially effective resourcelean": 114887,
      "effective resourcelean scenarios": 34740,
      "current stateoftheart multilingual": 25361,
      "significantly outperforms multilingual": 109018,
      "outperforms multilingual bert": 82921,
      "2020 present results": 728,
      "controlling text generation": 22862,
      "text generation pretrained": 120998,
      "models benefit training": 72830,
      "large unlabeled text": 61315,
      "models text generation": 74186,
      "language model generate": 59328,
      "language model parameters": 59365,
      "words occur frequently": 134090,
      "frequently training set": 45879,
      "nmt results corpus": 79957,
      "corpus corpus freely": 23729,
      "corpus freely available": 23809,
      "freely available noncommercial": 45783,
      "research best knowledge": 100431,
      "capture longterm dependencies": 15342,
      "yield performance improvements": 135348,
      "performance improvements especially": 86447,
      "improvements especially sequencelevel": 52849,
      "especially sequencelevel metrics": 38508,
      "paper presents attempt": 84151,
      "annotated corpus available": 6146,
      "corpus available online": 23674,
      "information extraction provide": 54596,
      "emergence social media": 36058,
      "social media effective": 110376,
      "attentionbased deep neural": 10066,
      "neural network framework": 78301,
      "outperforms baselines tasks": 82852,
      "set manually annotated": 107488,
      "manually annotated test": 67034,
      "present qualitative analysis": 89657,
      "model language representation": 71410,
      "language representation paper": 60047,
      "improve reasoning ability": 52514,
      "generated language model": 47091,
      "significantly improves zeroshot": 108961,
      "improves zeroshot performance": 53066,
      "task model uses": 118399,
      "explore different ways": 41541,
      "languages resources scarce": 60848,
      "dependencies paper present": 29104,
      "resulting crosslingual embeddings": 101438,
      "sentiment analysis model": 106610,
      "word embeddings evaluation": 133104,
      "analysis model achieved": 5699,
      "word embeddings sentiment": 133207,
      "generated natural language": 47097,
      "networks work propose": 77819,
      "new method named": 79014,
      "representations using multiple": 99967,
      "representations paper presents": 99796,
      "monolingual bert model": 74552,
      "follow recent trend": 44940,
      "monolingual bert models": 74553,
      "output classification layer": 83061,
      "predicted label experiments": 88960,
      "logical reasoning questions": 65015,
      "current approaches focus": 25260,
      "elementary discourse units": 35273,
      "discourse units edus": 32100,
      "downstream qa tasks": 34023,
      "learning algorithms used": 62327,
      "document representation learning": 33064,
      "improve embedding quality": 52380,
      "representation learning general": 99294,
      "methods benchmark datasets": 69341,
      "approaches document classification": 8133,
      "neural networks outperform": 78528,
      "bridge gap propose": 14594,
      "hierarchical graph neural": 49966,
      "evaluate method newly": 38854,
      "dataset verify effectiveness": 27272,
      "using crossdomain learning": 129516,
      "present submission eacl": 89726,
      "previous studies demonstrated": 90493,
      "maintain competitive performance": 66502,
      "single word embedding": 109815,
      "classification tasks method": 17446,
      "tasks method efficient": 119294,
      "resourcerich languages like": 100932,
      "lot data available": 65322,
      "data lowresource languages": 26100,
      "lowresource languages data": 65511,
      "models african languages": 72734,
      "language toolkit nltk": 60172,
      "systems knowledge graph": 116963,
      "knowledge grounded responses": 57996,
      "abstraction real world": 1796,
      "grounded responses integrating": 49112,
      "generation process endtoend": 47565,
      "process endtoend manner": 91472,
      "task paper proposes": 118511,
      "proposes novel architecture": 93611,
      "novel architecture integrating": 80498,
      "training bert model": 123375,
      "empirical evaluation suggests": 36161,
      "f1 score compared": 42680,
      "elmo models provided": 35318,
      "context words input": 22330,
      "text generation systems": 121010,
      "systems remains challenge": 117109,
      "especially multilingual settings": 38483,
      "multilingual settings paper": 75368,
      "settings paper propose": 107829,
      "translating morphologicallyrich languages": 124597,
      "mt models used": 74974,
      "finetuning data augmentation": 44453,
      "task semeval2020 task": 118682,
      "provide ablation study": 93749,
      "results error analysis": 101762,
      "performance lowresource setting": 86510,
      "model achieved promising": 70549,
      "natural language interaction": 76362,
      "languages morphological segmentation": 60737,
      "morphologicallyrich agglutinative languages": 74777,
      "train sequencetosequence models": 123017,
      "f1 score 725": 42675,
      "better random baseline": 13687,
      "south african languages": 110962,
      "language models lowresource": 59597,
      "using bytepair encoding": 129425,
      "rich morphology languages": 102774,
      "networks rnns transformers": 77748,
      "multilingual training improves": 75386,
      "improves performance datasets": 53012,
      "open new avenues": 81918,
      "dialogue datasets focus": 30658,
      "morphological inflection learning": 74710,
      "makes difficult determine": 66771,
      "semeval 2021 task": 105515,
      "et al 2021": 38741,
      "al 2020 task": 4650,
      "shared task training": 108100,
      "training set contains": 123847,
      "chen et al": 16714,
      "make inferences underlying": 66688,
      "semantic relation extraction": 105206,
      "relation extraction scientific": 98026,
      "data paper presents": 26215,
      "classification relation extraction": 17376,
      "significantly improved quality": 108937,
      "unsupervised opinion summarization": 127687,
      "summarization models generate": 115539,
      "models generate summary": 73286,
      "corpus algerian dialect": 23651,
      "approach hate speech": 7598,
      "languages english best": 60530,
      "english best knowledge": 37080,
      "network bidirectional lstm": 77171,
      "bidirectional lstm bilstm": 13953,
      "performance cnn model": 86210,
      "cnn model achieved": 18011,
      "model based unsupervised": 70750,
      "experiments seven datasets": 41131,
      "consistently boosts performance": 21413,
      "strong baselines achieves": 113640,
      "test data sets": 120439,
      "knowledge graph representations": 57974,
      "entities knowledge base": 37804,
      "large number candidate": 61175,
      "neural reranking model": 78653,
      "adding linguistic information": 3387,
      "contextual word embedding": 22517,
      "method incorporating linguistic": 68899,
      "linguistic information helps": 64490,
      "baselines stateoftheart approaches": 12471,
      "architecture enables learn": 8651,
      "language learning materials": 59244,
      "learning ml algorithms": 62750,
      "models including deep": 73382,
      "deep learning frameworks": 28274,
      "facilitate downstream tasks": 42768,
      "real applications existing": 96054,
      "incorporate new concepts": 53481,
      "dataset validate effectiveness": 27267,
      "alignment source target": 4989,
      "source target sequence": 110840,
      "sequence previous work": 107035,
      "machine translation aim": 65890,
      "achieved superior performance": 2711,
      "thorough comparative study": 121879,
      "use contextual information": 127962,
      "paper investigates different": 84029,
      "upb semeval2021 task": 127786,
      "task extracting semantic": 118195,
      "extracting semantic information": 42238,
      "multiturn question answering": 75921,
      "language models prlms": 59634,
      "large model size": 61151,
      "extensive natural language": 41945,
      "higher quality data": 50199,
      "language models high": 59565,
      "testbed evaluating models": 120557,
      "models reasoning capabilities": 73871,
      "achieves great success": 2796,
      "models specifically propose": 74080,
      "play different roles": 87694,
      "empirically conduct experiments": 36222,
      "conduct experiments neural": 20853,
      "classification benchmark datasets": 17135,
      "benchmark datasets extensive": 12784,
      "semantics syntax sentence": 105470,
      "models pretrained language": 73780,
      "contextual representations pretrained": 22493,
      "representations pretrained models": 99818,
      "syntax work present": 116571,
      "sentence embedding model": 105835,
      "embedding model learns": 35449,
      "model learns disentangle": 71443,
      "learns disentangle semantics": 63205,
      "sentence embeddings obtained": 105841,
      "based source sentence": 12061,
      "robustness syntactic variation": 103133,
      "new languages domains": 78982,
      "language generation human": 59073,
      "evaluate performance natural": 38885,
      "models multitask learning": 73595,
      "multitask learning setting": 75867,
      "case study use": 15625,
      "human evaluation generated": 50816,
      "discuss future research": 32254,
      "pretrained bert embedding": 89987,
      "embedding pretrained language": 35471,
      "nlp tasks research": 79764,
      "downstream tasks measure": 34044,
      "word embedding propose": 133030,
      "similarity tasks word": 109319,
      "baseline methods method": 12251,
      "language model present": 59374,
      "present ongoing work": 89629,
      "diverse set topics": 32847,
      "use recent years": 128227,
      "traditional methods recent": 122839,
      "explore multiple model": 41564,
      "rich syntactic semantic": 102795,
      "syntactic semantic annotations": 116465,
      "domain paper explore": 33605,
      "make predictions based": 66710,
      "having retrain model": 49566,
      "consider problem learning": 21223,
      "neural network predictions": 78375,
      "models neural language": 73625,
      "language models evaluate": 59547,
      "impressive performance nlp": 52319,
      "tasks contextualized embeddings": 119012,
      "given question context": 48097,
      "al 2016 word": 4616,
      "pimentel et al": 87514,
      "semeval2021 task multilingual": 105557,
      "multilingual crosslingual wordincontext": 75231,
      "crosslingual wordincontext disambiguation": 25048,
      "wordincontext disambiguation mclwic": 133720,
      "experiment pretrained language": 40484,
      "language models investigate": 59571,
      "models investigate impact": 73428,
      "investigate impact different": 56764,
      "approach recent years": 7841,
      "recent years word": 96644,
      "years word embeddings": 135315,
      "word embeddings widely": 133245,
      "embeddings widely used": 36025,
      "metrics based word": 69927,
      "word embeddings capturing": 133059,
      "ones native language": 81699,
      "language propose novel": 59999,
      "given literal sentence": 48060,
      "model able effectively": 70514,
      "competitive baseline models": 19638,
      "baseline models text": 12265,
      "language models plm": 59620,
      "examples different language": 39824,
      "paper investigate zeroshot": 84025,
      "investigate zeroshot performance": 56832,
      "dialectal arabic da": 30528,
      "significant performance drop": 108836,
      "context named entity": 22189,
      "recognition ner partofspeech": 96936,
      "accuracy pos tagging": 2238,
      "report new stateoftheart": 99015,
      "models research community": 73942,
      "expensive work present": 40439,
      "work present new": 134683,
      "present new stateoftheart": 89591,
      "unsupervised method based": 127666,
      "sentence embeddings significantly": 105846,
      "like masked language": 64064,
      "sentences coherent text": 106241,
      "generation problem present": 47562,
      "pretrained transformerbased model": 90204,
      "order given set": 82331,
      "additionally perform experiments": 3618,
      "factual knowledge stored": 42936,
      "instead structural knowledge": 55685,
      "structural knowledge base": 113776,
      "queries masked sentences": 94926,
      "masked sentences paris": 67299,
      "sentences paris capital": 106437,
      "paris capital mask": 84866,
      "capital mask used": 15240,
      "mask used probes": 67285,
      "simple nearest neighbor": 109478,
      "nearest neighbor matching": 76721,
      "effects natural language": 34997,
      "characters words sentence": 16632,
      "taskoriented semantic parsing": 118902,
      "utterance ontology tokens": 130606,
      "task text generation": 118788,
      "components natural language": 20036,
      "current sequence sequence": 25336,
      "processing long sequences": 91703,
      "translation tasks tasks": 125364,
      "hierarchical attention transformerbased": 49936,
      "achieves stateoftheart rouge": 2905,
      "stateoftheart rouge scores": 112939,
      "rouge scores summarization": 103301,
      "performance classification tasks": 86204,
      "parsing recent work": 85212,
      "recent work crosslingual": 96569,
      "highquality machine translation": 50395,
      "new languages propose": 78984,
      "asking answering questions": 9307,
      "stateoftheart correlations human": 112625,
      "correlations human judgment": 24259,
      "representations pretrained language": 99816,
      "units like morphemes": 127254,
      "propose new type": 92890,
      "switchboard dialog act": 116286,
      "dialog act corpus": 30546,
      "use pointwise mutual": 128198,
      "improve overall quality": 52451,
      "automatic speech recognition model": 10666,
      "paper presents new approach": 84185,
      "natural language processing nlp": 76439,
      "language processing nlp modules": 59909,
      "natural language processing present": 76474,
      "robust approach linking existing": 103018,
      "approach linking existing lexicalsemantic": 7692,
      "linking existing lexicalsemantic hierarchies": 64660,
      "constraint satisfaction algorithm relaxation": 21585,
      "machine translation present approach": 66161,
      "formal framework linguistic annotation": 45161,
      "machine translation paper present": 66142,
      "using surface expressions examples": 130252,
      "language understanding machine translation": 60237,
      "using information obtained recall": 129757,
      "information obtained recall rate": 54813,
      "language models speech recognition": 59662,
      "language modeling speech recognition": 59468,
      "word error rate wer": 133271,
      "paper describes new approach": 83835,
      "processing machine learning techniques": 91706,
      "conventional ngram language models": 22895,
      "performance machine learning algorithms": 86513,
      "exploiting diversity natural language": 41466,
      "natural language processing systems": 76487,
      "natural language dialogue interfaces": 76286,
      "dialogue interfaces semiautonomous systems": 30695,
      "executable program simple scripting": 39965,
      "program simple scripting language": 92114,
      "surface natural language generation": 116070,
      "natural language generation present": 76317,
      "select set candidates node": 104713,
      "set candidates node target": 107385,
      "candidates node target taxonomy": 15139,
      "node target taxonomy bests": 80028,
      "target taxonomy bests matches": 117727,
      "taxonomy bests matches node": 119663,
      "bests matches node source": 13488,
      "matches node source taxonomy": 67383,
      "wordnet 15 wordnet 16": 133765,
      "wall street journal corpus": 132003,
      "natural language processing approach": 76398,
      "present novel machine learning": 89612,
      "background information data sets": 11302,
      "present general overview systems": 89503,
      "general overview systems taken": 46692,
      "speech recognition word error": 111788,
      "posterior probabilities word lattices": 88486,
      "effectiveness method way experiments": 34911,
      "referential properties noun phrases": 97566,
      "noun phrases japanese language": 80432,
      "phrases japanese language articles": 87434,
      "natural language generation nlg": 76309,
      "machine learning algorithms automatically": 65768,
      "training data somewhat surprisingly": 123542,
      "speech recognition paper introduces": 111766,
      "word error rate reduction": 133270,
      "size word error rate": 109953,
      "wall street journal wsj": 132004,
      "relative reduction word error": 98372,
      "reduction word error rate": 97468,
      "baseline word error rate": 12338,
      "word sense disambiguation task": 133522,
      "new model achieves improvement": 79026,
      "paper describes experiments carried": 83823,
      "spanish english lexical sample": 111155,
      "english lexical sample tasks": 37194,
      "natural language processing paper": 76469,
      "research natural language processing": 100565,
      "shared task languageindependent named": 108061,
      "task languageindependent named entity": 118338,
      "languageindependent named entity recognition": 60367,
      "named entity recognition background": 76053,
      "entity recognition background information": 37994,
      "recognition background information data": 96823,
      "evaluation method present general": 39270,
      "method present general overview": 69065,
      "overview systems taken task": 83379,
      "systems taken task discuss": 117183,
      "taken task discuss performance": 117518,
      "challenge nlp applications machine": 16069,
      "nlp applications machine translation": 79570,
      "history natural language processing": 50564,
      "previous works mainly focus": 90574,
      "source language target language": 110780,
      "statistical machine translation smt": 113112,
      "machine translation smt systems": 66231,
      "natural language texts paper": 76546,
      "available online research purposes": 11063,
      "textual entailment rte task": 121697,
      "representation natural language semantics": 99349,
      "latent semantic analysis lsa": 61604,
      "semantic similarities paragraph paragraph": 105278,
      "important tasks natural language": 52276,
      "tasks natural language processing": 119330,
      "natural language processing information": 76425,
      "language processing information retrieval": 59873,
      "information retrieval machine translation": 54939,
      "context multidocument summarization evolving": 22184,
      "multidocument summarization evolving events": 75081,
      "detection coreference resolution task": 29911,
      "effective natural language processing": 34719,
      "work word sense disambiguation": 134881,
      "current natural language processing": 25306,
      "language processing nlp like": 59904,
      "used natural language processing": 128648,
      "language processing nlp applications": 59892,
      "various natural language processing": 131142,
      "text mining natural language": 121122,
      "mining natural language processing": 70245,
      "languages english spanish french": 60550,
      "natural language sentences formulas": 76533,
      "questions given natural language": 95313,
      "english natural language understanding": 37217,
      "problem natural language understanding": 91143,
      "knowledge domain consideration social": 57880,
      "domain consideration social behavior": 33485,
      "applications natural language processing": 6973,
      "evaluate method text classification": 38856,
      "natural language understanding based": 76552,
      "language understanding based objectoriented": 60205,
      "understanding based objectoriented semantics": 126801,
      "based objectoriented semantics algorithms": 11907,
      "processing text information presented": 91839,
      "systems closely connected text": 116784,
      "closely connected text processing": 17853,
      "connected text processing criminology": 21131,
      "text processing criminology operation": 121206,
      "processing criminology operation business": 91649,
      "criminology operation business medicine": 24779,
      "operation business medicine document": 82055,
      "business medicine document systems": 14978,
      "wsj section penn treebank": 135181,
      "existing natural language processing": 40223,
      "natural language processing methods": 76436,
      "paper presents novel approach": 84191,
      "problem mapping natural language": 91124,
      "mapping natural language sentences": 67142,
      "natural language interfaces databases": 76366,
      "neural probabilistic language model": 78634,
      "statistical machine translation neural": 113108,
      "probabilistic language model nplm": 90814,
      "rapid development natural language": 95717,
      "natural language understanding module": 76566,
      "information extraction natural language": 54583,
      "reducing human effort required": 97421,
      "word sense disambiguation techniques": 133524,
      "machine translation partofspeech tagging": 66152,
      "machine translation indian languages": 66009,
      "source language text target": 110783,
      "language text target language": 60164,
      "representation natural language sentences": 99350,
      "results statistical machine translation": 102213,
      "application natural language processing": 6870,
      "natural language processing including": 76424,
      "machine translation information retrieval": 66011,
      "statistical machine translation systems": 113120,
      "machine translation systems paper": 66252,
      "extraction named entity recognition": 42403,
      "named entity recognition ner": 76079,
      "improving quality mt output": 53152,
      "entities proper names location": 37845,
      "using statistical machine translation": 130223,
      "statistical machine translation toolkit": 113121,
      "natural language applications question": 76268,
      "language applications question answering": 58835,
      "paper propose new task": 84285,
      "model paper presents novel": 71696,
      "model significantly improves performance": 72023,
      "significantly improves performance stateoftheart": 108953,
      "phrasebased statistical machine translation": 87393,
      "developing natural language processing": 30353,
      "natural language processing applications": 76393,
      "paper present new approach": 84119,
      "experimental evaluation shows proposed": 40533,
      "evaluation shows proposed method": 39401,
      "shows proposed method outperforms": 108618,
      "predicateargument structure sentence plays": 88856,
      "structure sentence plays critical": 113950,
      "sentence plays critical role": 105990,
      "plays critical role deep": 87724,
      "critical role deep processing": 24830,
      "role deep processing natural": 103166,
      "deep processing natural language": 28398,
      "natural language understanding generation": 76560,
      "pointwise mutual information pmi": 87893,
      "log likelihood ratio llr": 64981,
      "american sign language asl": 5312,
      "natural language processing tasks": 76491,
      "language natural language processing": 59706,
      "area natural language processing": 8876,
      "using rule based approach": 130129,
      "machine translation mt research": 66069,
      "paper presents novel deep": 84192,
      "novel deep learning architecture": 80537,
      "data named entity recognition": 26151,
      "named entity recognition named": 76075,
      "improve named entity recognition": 52429,
      "entity recognition ner using": 38053,
      "large scale text corpora": 61264,
      "training statistical machine translation": 123884,
      "explicit semantic analysis esa": 41346,
      "machine translation natural language": 66079,
      "training corpus statistical machine": 123405,
      "corpus statistical machine translation": 24020,
      "natural language processing tools": 76508,
      "fully automatic multilanguage translation": 45949,
      "swiss avalanche bulletin swiss": 116272,
      "avalanche bulletin swiss avalanche": 11156,
      "bulletin swiss avalanche bulletin": 14960,
      "swiss avalanche bulletin produced": 116270,
      "avalanche bulletin produced twice": 11153,
      "bulletin produced twice day": 14956,
      "produced twice day languages": 91978,
      "twice day languages lack": 126084,
      "day languages lack time": 27843,
      "languages lack time available": 60667,
      "lack time available manual": 58760,
      "time available manual translation": 121987,
      "available manual translation fully": 11043,
      "manual translation fully automated": 67018,
      "translation fully automated translation": 124831,
      "fully automated translation employed": 45944,
      "automated translation employed based": 10478,
      "translation employed based catalogue": 124784,
      "employed based catalogue predefined": 36294,
      "based catalogue predefined phrases": 11564,
      "catalogue predefined phrases predetermined": 15704,
      "predefined phrases predetermined rules": 88831,
      "phrases predetermined rules phrases": 87445,
      "predetermined rules phrases combined": 88843,
      "rules phrases combined produce": 103429,
      "phrases combined produce sentences": 87414,
      "able automatically translate sentences": 1612,
      "automatically translate sentences german": 10833,
      "translate sentences german target": 124539,
      "sentences german target languages": 106333,
      "german target languages french": 47917,
      "target languages french italian": 117669,
      "languages french italian english": 60598,
      "french italian english subsequent": 45815,
      "italian english subsequent proofreading": 57090,
      "english subsequent proofreading correction": 37296,
      "catalogue phrases limited small": 15697,
      "phrases limited small sublanguage": 87438,
      "reduction daily translation costs": 97445,
      "daily translation costs expected": 25516,
      "translation costs expected offset": 124719,
      "costs expected offset initial": 24400,
      "expected offset initial development": 40396,
      "offset initial development costs": 81620,
      "initial development costs years": 55212,
      "operational winter seasons assess": 82060,
      "winter seasons assess quality": 132734,
      "seasons assess quality produced": 104378,
      "assess quality produced texts": 9490,
      "quality produced texts based": 94763,
      "origins catalogue phrases versus": 82582,
      "catalogue phrases versus manually": 15700,
      "phrases versus manually written": 87466,
      "versus manually written translated": 131629,
      "manually written translated texts": 67085,
      "written translated texts mean": 135155,
      "translated texts mean recognition": 124573,
      "texts mean recognition rate": 121550,
      "mean recognition rate 55": 67607,
      "recognition rate 55 users": 96980,
      "rate 55 users hardly": 95780,
      "55 users hardly distinguish": 1114,
      "types texts similar ratings": 126371,
      "texts similar ratings respect": 121609,
      "similar ratings respect language": 109133,
      "ratings respect language quality": 95827,
      "respect language quality overall": 101084,
      "language quality overall output": 60011,
      "quality overall output catalogue": 94742,
      "overall output catalogue considered": 83243,
      "output catalogue considered virtually": 83058,
      "catalogue considered virtually equivalent": 15693,
      "considered virtually equivalent text": 21312,
      "virtually equivalent text written": 131751,
      "equivalent text written avalanche": 38255,
      "text written avalanche forecasters": 121419,
      "written avalanche forecasters manually": 135117,
      "avalanche forecasters manually translated": 11160,
      "forecasters manually translated professional": 45033,
      "manually translated professional translators": 67081,
      "translated professional translators furthermore": 124559,
      "professional translators furthermore forecasters": 92083,
      "translators furthermore forecasters declared": 125515,
      "furthermore forecasters declared relevant": 46177,
      "forecasters declared relevant situations": 45029,
      "declared relevant situations captured": 27977,
      "relevant situations captured sufficient": 98583,
      "situations captured sufficient accuracy": 109897,
      "text classification text classification": 120782,
      "text classification widely studied": 120787,
      "like natural language processing": 64074,
      "information extraction machine translation": 54578,
      "languages named entity recognition": 60751,
      "language processing tasks information": 59959,
      "processing tasks information extraction": 91819,
      "rulebased machine learning approaches": 103366,
      "named entity recognition paper": 76098,
      "approaches named entity recognition": 8245,
      "representations named entity recognition": 99765,
      "named entity recognition social": 76113,
      "entity recognition social media": 38073,
      "detection dialogue act tagging": 29924,
      "language processing nlp tools": 59927,
      "commonly used english language": 18970,
      "crfbased named entity recognition": 24765,
      "open source morphological analyzer": 81936,
      "important natural language processing": 52189,
      "natural language processing task": 76489,
      "experiments approach achieves significant": 40779,
      "approach achieves significant improvements": 7320,
      "achieves significant improvements stateoftheart": 2860,
      "distributed word representations natural": 32639,
      "tasks remains open question": 119456,
      "tasks named entity recognition": 119323,
      "social media texts paper": 110428,
      "improving named entity recognition": 53121,
      "named entity recognition performance": 76102,
      "classified positive negative neutral": 17513,
      "advancement natural language processing": 3967,
      "topological analysis complex networks": 122679,
      "important task natural language": 52265,
      "task natural language processing": 118438,
      "natural language processing used": 76509,
      "based deep neural network": 11631,
      "deep neural network model": 28361,
      "word representations learnt using": 133463,
      "neural machine translation neural": 78037,
      "neural language models learn": 77947,
      "language models learn word": 59581,
      "models learn word representations": 73490,
      "capture rich linguistic conceptual": 15373,
      "rich linguistic conceptual information": 102766,
      "linguistic conceptual information investigate": 64448,
      "conceptual information investigate embeddings": 20665,
      "information investigate embeddings learned": 54704,
      "investigate embeddings learned neural": 56753,
      "embeddings learned neural machine": 35783,
      "learned neural machine translation": 62237,
      "neural machine translation models": 78029,
      "language modelling machine translation": 59489,
      "machine translation paper presents": 66143,
      "paper presents indepth investigation": 84174,
      "neural language models difficult": 77942,
      "challenge machine translation mt": 16052,
      "machine translation mt systems": 66071,
      "alleviate data sparsity problem": 5032,
      "significantly improve translation quality": 108934,
      "word representations capture rich": 133447,
      "linguistics natural language processing": 64618,
      "tools natural language processing": 122459,
      "natural language processing important": 76423,
      "downstream tasks information retrieval": 34041,
      "tasks information retrieval machine": 119204,
      "combinatory categorial grammar ccg": 18613,
      "model performs significantly better": 71743,
      "named entity recognition neural": 76097,
      "stateoftheart named entity recognition": 112765,
      "entity recognition ner systems": 38049,
      "tasks partofspeech pos tagging": 119375,
      "dataset achieve stateoftheart results": 26720,
      "recently shown promising results": 96767,
      "bidirectional long shortterm memory": 13939,
      "long shortterm memory lstm": 65111,
      "shortterm memory lstm neural": 108337,
      "memory lstm neural networks": 68343,
      "comparable state art english": 19177,
      "neural networks vector representations": 78603,
      "recently introduced neural network": 96700,
      "improving statistical machine translation": 53172,
      "annotated using amazon mechanical": 6259,
      "using amazon mechanical turk": 129334,
      "neural network takes input": 78403,
      "task natural language understanding": 118443,
      "paper present simple approach": 84134,
      "model trained endtoend requires": 72190,
      "statistical methods widely employed": 113130,
      "despite large studies devoted": 29704,
      "large studies devoted represent": 61283,
      "studies devoted represent texts": 114211,
      "devoted represent texts physical": 30478,
      "represent texts physical models": 99150,
      "performance natural language processing": 86560,
      "language processing tasks paper": 59963,
      "stateoftheart performance word similarity": 112858,
      "similarity word sense disambiguation": 109337,
      "word sense disambiguation tasks": 133523,
      "controlled natural language cnl": 22842,
      "goal article provide common": 48330,
      "tasks demonstrate effectiveness proposed": 119037,
      "vector space word representations": 131390,
      "approach performs significantly better": 7799,
      "performs significantly better previous": 87024,
      "shortterm memory lstm recurrent": 108340,
      "memory lstm recurrent neural": 68345,
      "lstm recurrent neural networks": 65663,
      "recurrent neural networks learn": 97233,
      "experiments morphologically rich languages": 41025,
      "words word sense disambiguation": 134319,
      "word sense disambiguation wsd": 133527,
      "bengali text corpus developed": 13036,
      "text corpus developed tdil": 120836,
      "corpus developed tdil project": 23754,
      "developed tdil project govt": 30310,
      "tdil project govt india": 119680,
      "extraction word sense disambiguation": 42540,
      "automatic machine translation evaluation": 10583,
      "machine translation evaluation metrics": 65976,
      "word embeddings distributed representations": 133088,
      "embeddings distributed representations words": 35662,
      "word similarity analogy tasks": 133554,
      "natural language processing models": 76438,
      "models natural language processing": 73610,
      "named entity recognition classification": 76058,
      "based multinomial naive bayes": 11869,
      "multinomial naive bayes mnb": 75475,
      "term frequency inverse document": 120217,
      "frequency inverse document frequency": 45846,
      "language processing applications paper": 59840,
      "received little attention despite": 96367,
      "neural network models natural": 78347,
      "network models natural language": 77349,
      "recently neural network models": 96722,
      "perspective natural language processing": 87168,
      "natural language processing research": 76482,
      "feedforward networks convolutional networks": 43848,
      "networks convolutional networks recurrent": 77554,
      "convolutional networks recurrent networks": 23174,
      "model achieves stateoftheart performance": 70598,
      "achieves stateoftheart performance multilingual": 2886,
      "special case machine translation": 111352,
      "machine translation machine translation": 66037,
      "statistical machine translation used": 113122,
      "year nara institute science": 135236,
      "nara institute science technology": 76159,
      "workshop asian translation based": 135000,
      "long shortterm memory recurrent": 65127,
      "shortterm memory recurrent neural": 108347,
      "memory recurrent neural network": 68373,
      "neural network bidirectional long": 78267,
      "recurrent neural network blstmrnn": 97201,
      "neural network blstmrnn shown": 78269,
      "network blstmrnn shown effective": 77177,
      "sequential data speech utterances": 107219,
      "data speech utterances handwritten": 26495,
      "speech utterances handwritten documents": 111846,
      "study propose use blstmrnn": 114489,
      "partofspeech pos tagging task": 85539,
      "rational speech acts model": 95836,
      "rational speech acts rsa": 95837,
      "speech acts rsa model": 111646,
      "language use recursive process": 60302,
      "use natural language processing": 128164,
      "integer linear programming ilp": 55736,
      "logistic regression convolutional neural": 65034,
      "regression convolutional neural network": 97699,
      "convolutional neural network cnn": 23182,
      "using naive bayes nb": 129930,
      "machine learning methods proposed": 65801,
      "stateoftheart neural language models": 112777,
      "articles achieve stateoftheart performance": 9168,
      "information paper propose method": 54826,
      "work propose novel method": 134737,
      "propose novel method incorporate": 92940,
      "language model based long": 59295,
      "model based long shortterm": 70734,
      "based long shortterm memory": 11814,
      "long shortterm memory units": 65129,
      "range natural language processing": 95584,
      "characterbased neural machine translation": 16497,
      "neural machine translation introduce": 78009,
      "neural machine translation model": 78025,
      "present neural network architecture": 89576,
      "learning vector representations words": 63159,
      "processing tasks information retrieval": 91820,
      "partofspeech tagging named entity": 85556,
      "tagging named entity recognition": 117416,
      "problem machine translation especially": 91116,
      "sourcetopivot pivottotarget translation models": 110951,
      "using english pivot language": 129629,
      "target word context words": 117744,
      "shortterm memory lstm models": 108332,
      "models achieve stateoftheart performances": 72674,
      "using syntactic parser external": 130258,
      "syntactic parser external sentiment": 116440,
      "parser external sentiment lexicons": 84957,
      "performance benchmark datasets measuring": 86172,
      "hidden markov model based": 49896,
      "experiments carried jadavpur university": 40814,
      "carried jadavpur university participation": 15544,
      "tool developed task based": 122381,
      "developed task based trigram": 30305,
      "task based trigram hidden": 117923,
      "based trigram hidden markov": 12140,
      "trigram hidden markov model": 125798,
      "hidden markov model utilizes": 49899,
      "markov model utilizes information": 67264,
      "pos tag word level": 88216,
      "word level features enhance": 133341,
      "level features enhance observation": 63447,
      "features enhance observation probabilities": 43486,
      "enhance observation probabilities known": 37480,
      "observation probabilities known tokens": 81168,
      "probabilities known tokens unknown": 90846,
      "known tokens unknown tokens": 58326,
      "tokens unknown tokens submitted": 122348,
      "unknown tokens submitted runs": 127362,
      "trained tested datasets released": 123309,
      "attentionbased neural machine translation": 10086,
      "neural machine translation attentional": 77974,
      "machine translation attentional mechanism": 65903,
      "endtoend neural machine translation": 36940,
      "capture partial aspects attentional": 15358,
      "partial aspects attentional regularities": 85287,
      "neural machine translation instead": 78008,
      "sourcetotarget targettosource translation models": 110957,
      "improve performance downstream tasks": 52465,
      "attentionbased convolutional neural network": 10063,
      "based convolutional neural network": 11610,
      "achieve stateoftheart results tasks": 2569,
      "shows model outperforms previous": 108601,
      "address problem propose new": 3744,
      "recurrent neural network based": 97198,
      "neural network based language": 78258,
      "network based language model": 77161,
      "word embeddings weighted contexts": 133243,
      "recurrent neural network rnn": 97213,
      "architectures long shortterm memory": 8819,
      "component natural language understanding": 19993,
      "natural language understanding using": 76591,
      "achieve state art results": 2555,
      "word alignment paper propose": 132910,
      "various machine learning techniques": 131129,
      "using various machine learning": 130356,
      "various machine learning algorithms": 131126,
      "machine learning algorithms like": 65769,
      "entropy support vector machine": 38174,
      "challenges applications sentiment analysis": 16130,
      "support vector machines svm": 116014,
      "deep neural networks dnn": 28371,
      "pointwise mutual information matrix": 87891,
      "mutual information matrix stochastic": 75979,
      "information matrix stochastic gradient": 54755,
      "matrix stochastic gradient descent": 67484,
      "entity recognition ner entity": 38035,
      "neural network language model": 78314,
      "recurrent neural network lstm": 97209,
      "performance answer sentence selection": 86140,
      "answer sentence selection task": 6578,
      "natural language processing machine": 76432,
      "language processing machine learning": 59882,
      "present neural network based": 89577,
      "convolution neural network cnn": 23144,
      "neural network cnn architecture": 78279,
      "language generation spoken dialogue": 59096,
      "generation spoken dialogue systems": 47632,
      "performance terms bleu score": 86794,
      "articles written different languages": 9232,
      "disambiguation named entity disambiguation": 31975,
      "named entity disambiguation ned": 76044,
      "shortterm memory lstm language": 108330,
      "model language generation model": 71406,
      "proposed method outperforms traditional": 93390,
      "recurrent neural network architectures": 97197,
      "experimental results attention mechanism": 40557,
      "neural networks natural language": 78521,
      "networks natural language understanding": 77672,
      "models natural language understanding": 73614,
      "models generalize unseen words": 73279,
      "automatic speech recognition asr": 10657,
      "speech recognition asr machine": 111738,
      "recognition asr machine translation": 96812,
      "asr machine translation mt": 9435,
      "translation morphologically rich languages": 124974,
      "tree adjoining grammar tag": 125577,
      "question answering information retrieval": 95052,
      "answering information retrieval information": 6636,
      "information retrieval information extraction": 54934,
      "bidirectional longshort term memory": 13948,
      "achieves stateoftheart performance standard": 2892,
      "tuning statistical machine translation": 125943,
      "statistical machine translation using": 113123,
      "long shortterm memory bilstm": 65105,
      "shortterm memory bilstm networks": 108320,
      "research natural language understanding": 100566,
      "weston et al 2015": 132460,
      "common european framework reference": 18877,
      "work studies comparatively typical": 134823,
      "studies comparatively typical sentence": 114196,
      "classification tasks textual entailment": 17460,
      "tasks textual entailment te": 119562,
      "textual entailment te answer": 121702,
      "entailment te answer selection": 37681,
      "te answer selection observing": 119685,
      "prior work limitations phrase": 90753,
      "work limitations phrase generation": 134614,
      "limitations phrase generation representation": 64187,
      "phrase generation representation ii": 87351,
      "generation representation ii conducts": 47598,
      "representation ii conducts alignment": 99255,
      "ii conducts alignment word": 51703,
      "conducts alignment word phrase": 20969,
      "alignment word phrase levels": 5004,
      "word phrase levels handcrafted": 133405,
      "phrase levels handcrafted features": 87357,
      "levels handcrafted features iii": 63540,
      "handcrafted features iii utilizes": 49344,
      "features iii utilizes single": 43546,
      "considering characteristics specific tasks": 21318,
      "characteristics specific tasks limits": 16535,
      "effectiveness tasks propose architecture": 34963,
      "tasks propose architecture based": 119407,
      "propose architecture based gated": 92555,
      "architecture based gated recurrent": 8626,
      "based gated recurrent unit": 11732,
      "gated recurrent unit supports": 46517,
      "recurrent unit supports representation": 97267,
      "unit supports representation learning": 127213,
      "supports representation learning phrases": 116041,
      "representation learning phrases arbitrary": 99306,
      "learning phrases arbitrary granularity": 62892,
      "phrases arbitrary granularity ii": 87409,
      "arbitrary granularity ii taskspecific": 8584,
      "experimental results te match": 40711,
      "results te match observation": 102261,
      "significantly outperforms stateoftheart approaches": 109032,
      "neural machine translation order": 78069,
      "phenomena neural machine translation": 87238,
      "model experimental results largescale": 71130,
      "vocabulary neural machine translation": 131888,
      "time unlike previous work": 122131,
      "training data test data": 123550,
      "large amounts annotated data": 61008,
      "machine translation mt evaluation": 66066,
      "language models deep learning": 59534,
      "models deep learning models": 73023,
      "neural machine translation investigate": 78010,
      "machine translation investigate use": 66016,
      "neural machine translation nmt": 78040,
      "neural network text classification": 78407,
      "learning neural network based": 62843,
      "neural network based methods": 78260,
      "variety natural language processing": 131001,
      "language processing tasks previous": 59966,
      "processing tasks previous works": 91826,
      "multitask learning framework jointly": 75840,
      "learning framework jointly learn": 62603,
      "jointly learn multiple related": 57349,
      "learn multiple related tasks": 62098,
      "based recurrent neural network": 11979,
      "neural networks existing methods": 78478,
      "datasets demonstrate efficacy proposed": 27406,
      "grammar error correction gec": 48634,
      "error correction gec tasks": 38296,
      "results conll2014 test set": 101627,
      "paper novel approach proposed": 84056,
      "translation experiments proposed method": 124814,
      "experiments proposed method achieve": 41073,
      "machine translation models neural": 66055,
      "translation models neural machine": 124959,
      "models neural machine translation": 73628,
      "source sentence paper propose": 110813,
      "model introduces continuous latent": 71373,
      "introduces continuous latent variable": 56608,
      "latent variable explicitly model": 61628,
      "neural machine translation achieves": 77968,
      "neural machine translation baselines": 77978,
      "vanilla recurrent neural network": 130820,
      "relations events time expressions": 98159,
      "generation recent neural models": 47584,
      "using policy gradient methods": 130029,
      "achieves stateoftheart performance multiple": 2887,
      "zero pronoun resolution existing": 135477,
      "pronoun resolution existing approaches": 92343,
      "propose simple novel approach": 93067,
      "experimental results proposed approach": 40671,
      "results proposed approach significantly": 102064,
      "proposed approach significantly outperforms": 93217,
      "approach significantly outperforms stateoftheart": 7913,
      "significantly outperforms stateoftheart systems": 109037,
      "systems neural machine translation": 117020,
      "knowledge neural machine translation": 58083,
      "lexicons neural machine translation": 63933,
      "machine translation neural machine": 66082,
      "translation neural machine translation": 125009,
      "time neural machine translation": 122058,
      "neural machine translation simultaneous": 78102,
      "machine translation simultaneous translation": 66224,
      "existing neural machine translation": 40229,
      "based neural machine translation": 11889,
      "apply neural machine translation": 7198,
      "neural machine translation task": 78113,
      "neural machine translation significantly": 78101,
      "neural machine translation systems": 78110,
      "shared news translation task": 107996,
      "corpus additional training data": 23648,
      "improve neural machine translation": 52439,
      "neural machine translation recently": 78091,
      "recently achieved impressive results": 96650,
      "datasets morphologically rich languages": 27583,
      "text classification convolutional neural": 120716,
      "classification convolutional neural networks": 17163,
      "convolutional neural networks cnns": 23201,
      "results method outperforms baseline": 101922,
      "embeddings word embeddings trained": 36031,
      "parallel training data available": 84695,
      "training data available training": 123434,
      "neural network joint model": 78311,
      "consider incorporating topic information": 21207,
      "obtained pretrained lda model": 81396,
      "automatic evaluation metrics human": 10538,
      "machine translation quality estimation": 66181,
      "quality estimation machine translation": 94649,
      "approach biomedical information extraction": 7394,
      "social media text messages": 110422,
      "machine learning approaches work": 65777,
      "rely hand crafted features": 98703,
      "neural network based models": 78262,
      "languages paper propose neural": 60783,
      "paper propose neural network": 84272,
      "propose neural network based": 92826,
      "neural network based model": 78261,
      "representation neural machine translation": 99355,
      "networks neural machine translation": 77676,
      "word embedding vectors using": 133042,
      "translation quality neural machine": 125180,
      "quality neural machine translation": 94730,
      "paper introduce novel approach": 84001,
      "improves neural machine translation": 52999,
      "complexity neural machine translation": 19930,
      "machine translation nmt systems": 66122,
      "recurrent convolutional neural networks": 97170,
      "achieving new stateoftheart performance": 2964,
      "natural language question answering": 76519,
      "parsing paper present approach": 85183,
      "demonstrate effectiveness proposed approach": 28719,
      "coreference resolution named entities": 23369,
      "word2vec mikolov et al": 133679,
      "mikolov et al 2013a": 70082,
      "using pretrained word embeddings": 130051,
      "models discriminating similar languages": 73087,
      "semantics pragmatics paper present": 105446,
      "paper present novel approach": 84125,
      "approach natural language understanding": 7731,
      "lstm recurrent neural network": 65662,
      "critical task natural language": 24838,
      "task natural language generation": 118435,
      "phrase based statistical machine": 87337,
      "based statistical machine translation": 12073,
      "new word analogy corpus": 79251,
      "consistent improvements translation quality": 21391,
      "spoken language understanding paper": 111998,
      "bidirectional long short term": 13936,
      "long short term memory": 65094,
      "short term memory networks": 108254,
      "experiments proposed model robust": 41083,
      "language identification social media": 59143,
      "using latent dirichlet allocation": 129799,
      "representations natural language processing": 99770,
      "natural language processing using": 76510,
      "challenges natural language processing": 16179,
      "natural language processing architectures": 76399,
      "machine translation smt neural": 66229,
      "sentiment analysis social media": 106658,
      "publicly available research purposes": 94322,
      "generation neural machine translation": 47508,
      "generates target sentence word": 47184,
      "target sentence word word": 117701,
      "paper proposed novel approach": 84349,
      "neural network language models": 78317,
      "improves correlation human judgments": 52965,
      "different word embedding methods": 31568,
      "noise contrastive estimation nce": 80047,
      "language processing tasks word": 59972,
      "objective function experimental results": 81080,
      "natural language processing work": 76514,
      "languages universal dependencies treebank": 60940,
      "recurrent neural networks encoder": 97230,
      "bidirectional recurrent neural network": 13971,
      "vector representation decoder generates": 131343,
      "attentionbased recurrent neural network": 10096,
      "tts automatic speech recognition": 125916,
      "models achieved impressive results": 72684,
      "training data paper propose": 123508,
      "data paper propose deep": 26217,
      "machine translation nmt heavily": 66098,
      "translation nmt heavily relies": 125035,
      "languages natural word delimiters": 60756,
      "natural word delimiters chinese": 76627,
      "languages spoken language resources": 60892,
      "spoken language resources likely": 111991,
      "language resources likely annotated": 60055,
      "resources likely annotated translations": 101003,
      "likely annotated translations transcriptions": 64136,
      "using dynamic time warping": 129613,
      "task recurrent neural networks": 118617,
      "neural networks achieved stateoftheart": 78437,
      "present empirical study using": 89468,
      "model achieves best performance": 70555,
      "achieves best performance datasets": 2739,
      "extensive experiments demonstrate proposed": 41908,
      "experiments demonstrate proposed methods": 40893,
      "demonstrate proposed methods significantly": 28846,
      "using bidirectional recurrent neural": 129404,
      "bidirectional recurrent neural networks": 13973,
      "recurrent neural networks work": 97255,
      "using recurrent neural networks": 130093,
      "recurrent neural networks rnn": 97240,
      "parallel corpus source target": 84624,
      "corpus source target language": 24006,
      "strategies neural machine translation": 113479,
      "based natural language processing": 11880,
      "tweets labeled using hashtags": 126035,
      "smt neural machine translation": 110289,
      "efficient neural machine translation": 35099,
      "current neural machine translation": 25309,
      "neural machine translation used": 78129,
      "evaluation machine translation mt": 39260,
      "word embeddings word embeddings": 133249,
      "language statistical machine translation": 60125,
      "performance standard test set": 86737,
      "sequential recurrent neural networks": 107250,
      "achieves stateoftheart results tasks": 2904,
      "power recurrent neural networks": 88652,
      "model achieves f1 score": 70571,
      "lstm networks model trained": 65652,
      "using recurrent neural network": 130090,
      "recurrent neural network paper": 97212,
      "level quality estimation qe": 63495,
      "sentence paper propose novel": 105976,
      "paper propose novel approach": 84288,
      "recurrent neural network language": 97206,
      "network language model rnnlm": 77296,
      "language model rnnlm architecture": 59392,
      "different natural language processing": 31279,
      "machine translation nmt new": 66109,
      "translation nmt new approach": 125046,
      "nmt new approach machine": 79933,
      "new approach machine translation": 78790,
      "great progress recent years": 49017,
      "conventional statistical machine translation": 22903,
      "endtoend manner experimental results": 36918,
      "experimental results chineseenglish translation": 40568,
      "approach achieves significant consistent": 7317,
      "achieves significant consistent improvements": 2854,
      "significant consistent improvements stateoftheart": 108746,
      "machine translation nmt significantly": 66117,
      "translation nmt significantly improved": 125055,
      "statistical machine translation shown": 113111,
      "target language neural machine": 117643,
      "language neural machine translation": 59714,
      "phrasebased machine translation pbmt": 87384,
      "english german translation task": 37158,
      "demonstration neural machine translation": 28989,
      "target language source language": 117653,
      "using byte pair encoding": 129423,
      "based recurrent neural networks": 11981,
      "recurrent neural networks rnns": 97241,
      "experimental results indicate proposed": 40625,
      "sequencetosequence neural network models": 107191,
      "news articles labels derived": 79298,
      "introduce novel neural architecture": 56518,
      "language processing nlp tasks": 59919,
      "using term frequencyinverse document": 130282,
      "term frequencyinverse document frequency": 120221,
      "frequencyinverse document frequency tfidf": 45857,
      "classification paper proposes novel": 17327,
      "language model rnnlm long": 59393,
      "model rnnlm long shortterm": 71939,
      "rnnlm long shortterm memory": 102956,
      "using long short term": 129833,
      "short term memory lstm": 108250,
      "term memory lstm network": 120233,
      "unified framework experimental results": 127110,
      "experimental results proposed method": 40674,
      "results proposed method significantly": 102074,
      "method significantly outperform baseline": 69141,
      "significantly outperform baseline methods": 108981,
      "results proposed method improves": 102071,
      "dependency parsing paper presents": 29190,
      "novel neural network architecture": 80671,
      "neural network architecture task": 78248,
      "data neural machine translation": 26165,
      "machine translation nmt achieved": 66087,
      "translation nmt achieved remarkable": 125025,
      "nmt achieved remarkable progress": 79827,
      "remarkable progress past years": 98841,
      "alleviate problem propose novel": 5045,
      "significantly improves adequacy nmt": 108942,
      "detection natural language processing": 30001,
      "using natural language processing": 129937,
      "conditional random field crf": 20778,
      "experiments model outperforms stateoftheart": 41016,
      "model outperforms stateoftheart methods": 71673,
      "chinese word segmentation pos": 16840,
      "word segmentation pos tagging": 133504,
      "neural network named entity": 78361,
      "network named entity recognition": 77367,
      "named entity recognition chinese": 76056,
      "entity recognition chinese social": 37999,
      "recognition chinese social media": 96832,
      "focus named entity recognition": 44794,
      "entity recognition ner chinese": 38032,
      "ner chinese social media": 77029,
      "gated recurrent units gru": 46522,
      "processed convolutional neural network": 91596,
      "experiments public data sets": 41091,
      "previous works mainly focused": 90575,
      "paper propose endtoend neural": 84233,
      "experimental results effectiveness proposed": 40611,
      "results effectiveness proposed model": 101747,
      "embeddings based neural network": 35567,
      "rapid development deep learning": 95715,
      "natural language processing widely": 76511,
      "language processing widely used": 59985,
      "word embeddings natural language": 133167,
      "paper propose new neural": 84283,
      "propose new neural network": 92877,
      "corpus experimental results proposed": 23790,
      "widely used speech recognition": 132589,
      "integrated crisis early warning": 55775,
      "paper address problem complementary": 83705,
      "address problem complementary entity": 3730,
      "problem complementary entity recognition": 90973,
      "complementary entity recognition cer": 19743,
      "solve problem propose novel": 110609,
      "proposed method does require": 93362,
      "characterlevel neural machine translation": 16577,
      "neural machine translation attractive": 77976,
      "wmt englishgerman translation task": 132789,
      "proven effective natural language": 93724,
      "approaches use external resources": 8392,
      "set new state art": 107507,
      "multilingual natural language processing": 75304,
      "using continuous vector representations": 129490,
      "forum information retrieval evaluation": 45307,
      "teams participated shared task": 119730,
      "applied natural language processing": 7098,
      "paper propose general methodology": 84247,
      "multiple nlp tasks including": 75631,
      "present novel neural network": 89619,
      "model outperforms previous work": 71662,
      "rapid growth social media": 95721,
      "experimental results demonstrate proposed": 40599,
      "results demonstrate proposed model": 101703,
      "demonstrate proposed model obtains": 28851,
      "compared previous state art": 19417,
      "state art neural machine": 112459,
      "art neural machine translation": 9071,
      "based sequence sequence learning": 12027,
      "machine translation language directions": 66021,
      "introduced neural machine translation": 56589,
      "stateoftheart neural machine translation": 112779,
      "novel neural network model": 80673,
      "use deep neural networks": 127988,
      "deep neural networks learn": 28374,
      "model neural network joint": 71574,
      "natural language understanding nlu": 76569,
      "language understanding nlu tasks": 60248,
      "deep neural network dnn": 28360,
      "sequence labeling problem word": 106980,
      "paper propose alternative approach": 84219,
      "conventional machine learning methods": 22883,
      "stateoftheart deep learning models": 112635,
      "sentiment analysis sarcasm detection": 106649,
      "sarcasm detection cognitive features": 103642,
      "statistical classification using enhanced": 113084,
      "classification using enhanced feature": 17491,
      "using enhanced feature set": 129632,
      "recent advances machine learning": 96408,
      "machine learning ml natural": 65806,
      "learning ml natural language": 62754,
      "ml natural language processing": 70438,
      "machine translation nmt models": 66103,
      "distillation neural machine translation": 32518,
      "neural machine translation knowledge": 78012,
      "machine translation knowledge distillation": 66019,
      "processing deep neural networks": 91659,
      "field natural language processing": 43969,
      "language processing nlp convolutional": 59895,
      "processing nlp convolutional neural": 91731,
      "neural network cnn recurrent": 78287,
      "network cnn recurrent neural": 77198,
      "cnn recurrent neural network": 18020,
      "state art nlp tasks": 112462,
      "model neural network models": 71576,
      "named entity recognition text": 76121,
      "make datasets publicly available": 66654,
      "representation words word embeddings": 99472,
      "representation natural language text": 99351,
      "specific natural language processing": 111470,
      "neural machine translation previous": 78079,
      "machine translation previous work": 66171,
      "neural machine translation approach": 77969,
      "languages english german italian": 60539,
      "supervised learning procedure using": 115777,
      "paper propose novel framework": 84297,
      "achieves better performance compared": 2747,
      "realworld datasets different domains": 96164,
      "translation machine translation mt": 124922,
      "model recurrent neural network": 71873,
      "shortterm memory lstm cells": 108329,
      "information extraction scientific literature": 54600,
      "train recurrent neural network": 123006,
      "recurrent neural network model": 97210,
      "features recurrent neural network": 43693,
      "training data improve generalization": 123477,
      "machine learning models using": 65814,
      "training deep neural networks": 123577,
      "deep neural networks dnns": 28372,
      "performance natural language inference": 86559,
      "natural language inference nli": 76334,
      "language inference nli task": 59183,
      "based distributed word representation": 11652,
      "dependency parsing dependency parsing": 29177,
      "parsing dependency parsing important": 85099,
      "parsing important nlp task": 85128,
      "usual vector space model": 130402,
      "word segmentation word segmentation": 133509,
      "problem natural language processing": 91140,
      "memory recurrent neural networks": 68376,
      "recurrent neural networks use": 97254,
      "achieve new stateoftheart results": 2505,
      "problems natural language processing": 91347,
      "experiments publicly available dataset": 41098,
      "compared strong supervised baselines": 19471,
      "discovery massive text corpora": 32144,
      "experiments demonstrate proposed framework": 40891,
      "speech recognition asr systems": 111742,
      "paper proposes novel technique": 84366,
      "leads improved performance tasks": 61938,
      "improving neural machine translation": 53128,
      "experimental results proposed model": 40680,
      "proposed model consistently outperforms": 93441,
      "englishgerman chineseenglish translation tasks": 37374,
      "crosslingual semantic textual similarity": 25006,
      "semantic textual similarity sts": 105334,
      "informative sentences maintaining grammaticality": 55155,
      "hierarchical recurrent neural network": 49996,
      "social media twitter facebook": 110430,
      "evaluation metric machine translation": 39280,
      "topics natural language processing": 122642,
      "language processing nlp literature": 59905,
      "features achieve good performance": 43350,
      "support vector machine svm": 116006,
      "word representations paper propose": 133466,
      "novel crosslingual transfer method": 80520,
      "sentiment analysis important task": 106598,
      "important task scientific paper": 52271,
      "task scientific paper analysis": 118666,
      "formed feature space examined": 45227,
      "sentence mapped features input": 105941,
      "mapped features input classifiers": 67125,
      "supervised classification using 10crossvalidation": 115739,
      "classification using 10crossvalidation scheme": 17484,
      "using 10crossvalidation scheme evaluation": 129309,
      "10crossvalidation scheme evaluation conducted": 204,
      "important problems machine translation": 52217,
      "datasets experimental results proposed": 27473,
      "experimental results proposed methods": 40678,
      "results proposed methods outperform": 102077,
      "proposed methods outperform previous": 93415,
      "applications information extraction question": 6942,
      "information extraction question answering": 54598,
      "tai et al 2015": 117497,
      "word embedding models using": 133027,
      "word embeddings capture semantic": 133058,
      "social media websites electronic": 110435,
      "media websites electronic newspapers": 68184,
      "websites electronic newspapers internet": 132302,
      "electronic newspapers internet forums": 35258,
      "newspapers internet forums allow": 79411,
      "internet forums allow visitors": 56192,
      "forums allow visitors leave": 45313,
      "allow visitors leave comments": 5101,
      "visitors leave comments read": 131780,
      "leave comments read interact": 63261,
      "comments read interact exchange": 18823,
      "read interact exchange free": 95934,
      "interact exchange free participants": 55933,
      "exchange free participants malicious": 39943,
      "free participants malicious intentions": 45763,
      "referring expression generation using": 97574,
      "proposed model significantly outperforms": 93482,
      "model significantly outperforms existing": 72030,
      "evaluate robustness different models": 38918,
      "convolutional neural networks recurrent": 23206,
      "neural networks recurrent neural": 78552,
      "networks recurrent neural networks": 77729,
      "characters byte pair encoding": 16599,
      "larger vocabulary training complexity": 61390,
      "vocabulary training complexity decoding": 131912,
      "training complexity decoding complexity": 123390,
      "using neural machine translation": 129954,
      "abstract meaning representation parsing": 1773,
      "meaning representation parsing abstract": 67676,
      "representation parsing abstract meaning": 99371,
      "parsing abstract meaning representation": 85059,
      "abstract meaning representation amr": 1770,
      "method based annotation projection": 68662,
      "language target language using": 60148,
      "neural machine translation present": 78076,
      "machine translation present simple": 66166,
      "present simple effective approach": 89702,
      "incorporating syntactic structure neural": 53565,
      "neural network models including": 78346,
      "tree adjoining grammars tags": 125579,
      "text classification neural network": 120750,
      "classification neural network models": 17308,
      "neural network models shown": 78353,
      "adversarial multitask learning framework": 4147,
      "different text classification tasks": 31490,
      "text classification tasks demonstrates": 120776,
      "significantly improve state art": 108931,
      "state art semantic dependency": 112473,
      "art semantic dependency parsing": 9091,
      "achieving new state art": 2962,
      "new state art code": 79148,
      "state art code opensource": 112434,
      "art code opensource available": 9042,
      "model achieves stateoftheart results": 70605,
      "knowledge graph embeddings study": 57965,
      "poses new challenges existing": 88271,
      "automatic human evaluations model": 10559,
      "skipgram negative sampling sgns": 109991,
      "negative sampling sgns word": 76951,
      "support vector machine superior": 116005,
      "paper describes duluth systems": 83819,
      "describes duluth systems participated": 29401,
      "language models pretrained word": 59632,
      "models pretrained word embeddings": 73790,
      "pretrained word embeddings learned": 90221,
      "relatively little labeled data": 98401,
      "standard datasets named entity": 112226,
      "datasets named entity recognition": 27589,
      "recent advances gpu hardware": 96404,
      "propose novel neural network": 92954,
      "neural network model joint": 78331,
      "lowresource neural machine translation": 65545,
      "neural machine translation quality": 78085,
      "propose novel data augmentation": 92909,
      "method improves translation quality": 68893,
      "uses convolutional neural networks": 129215,
      "learning universal sentence representations": 63139,
      "representations natural language inference": 99769,
      "natural language inference data": 76326,
      "stanford natural language inference": 112372,
      "natural language inference datasets": 76328,
      "transfer learning nlp tasks": 124117,
      "approach sequence sequence learning": 7882,
      "recurrent neural networks introduce": 97232,
      "based entirely convolutional neural": 11687,
      "entirely convolutional neural networks": 37732,
      "accuracy deep lstm setup": 2133,
      "englishgerman wmt14 englishfrench translation": 37393,
      "short message service sms": 108220,
      "use machine learning techniques": 128134,
      "neural networks proven effective": 78546,
      "achieve significantly better performance": 2545,
      "latent variables experimental results": 61638,
      "pretrained word vectors paper": 90228,
      "model performs comparably stateoftheart": 71739,
      "event trigger identification using": 39552,
      "existing work natural language": 40339,
      "model convolutional neural networks": 70924,
      "morphologically rich languages present": 74773,
      "uses convolutional neural network": 129213,
      "zero anaphora resolution zar": 135473,
      "training neural machine translation": 123735,
      "wmt multimodal machine translation": 132803,
      "large amounts monolingual data": 61018,
      "translation statistical machine translation": 125294,
      "convolutional neural networks present": 23203,
      "based convolutional neural networks": 11612,
      "convolutional neural networks cnn": 23200,
      "partofspeech tagging morphological tagging": 85554,
      "neural networks rnns proved": 78565,
      "tasks spoken language understanding": 119520,
      "spoken language understanding slu": 112000,
      "neural network architectures measure": 78251,
      "results verify effectiveness approach": 102319,
      "standard neural network architectures": 112282,
      "partofspeech taggers dependency parsers": 85546,
      "natural language processing techniques": 76505,
      "information paper propose novel": 54827,
      "neural machine translation training": 78118,
      "translation nmt models usually": 125044,
      "parser dyer et al": 84946,
      "dyer et al 2015": 34292,
      "vinyals et al 2015": 131731,
      "propose novel framework called": 92921,
      "experimental results prove effectiveness": 40684,
      "machine translation nmt recently": 66115,
      "machine translation nmt suffers": 66121,
      "et al 2017 proposed": 38713,
      "bahdanau et al 2015": 11390,
      "datadriven natural language generation": 26673,
      "natural language generation systems": 76318,
      "dataset poses new challenges": 27099,
      "widely used machine translation": 132573,
      "metrics correlate strongly human": 69945,
      "using deep neural networks": 129549,
      "achieve similar results compared": 2549,
      "neural natural language generation": 78225,
      "extensive experiments benchmark datasets": 41896,
      "conll 2017 ud shared": 21081,
      "2017 ud shared task": 644,
      "stateoftheart natural language processing": 112769,
      "positive negative training examples": 88336,
      "improve performance compared stateoftheart": 52462,
      "speech recognition natural language": 111763,
      "natural language understanding language": 76563,
      "news translation task paper": 79389,
      "translation task paper describes": 125338,
      "neural machine translation paper": 78070,
      "machine translation paper proposes": 66147,
      "text datasets visualizing relevant": 120858,
      "datasets visualizing relevant words": 27796,
      "advanced machine learning algorithms": 3947,
      "texts summarize contents documents": 121627,
      "summarize contents documents belonging": 115601,
      "new york times article": 79259,
      "york times article snippets": 135455,
      "significant improvements translation quality": 108813,
      "propose new method based": 92864,
      "based bidirectional long shortterm": 11551,
      "memory lstm neural network": 68342,
      "word embeddings improve performance": 133130,
      "pretrained word embeddings using": 90225,
      "word similarity word analogy": 133574,
      "naive bayes logistic regression": 76018,
      "problem named entity recognition": 91137,
      "word embeddings word sense": 133251,
      "embeddings word sense disambiguation": 36034,
      "paper propose novel method": 84300,
      "use word sense disambiguation": 128365,
      "proposed method compare performance": 93352,
      "task experimental results proposed": 118176,
      "word embeddings paper propose": 133177,
      "embeddings paper propose novel": 35852,
      "pretrained word embeddings model": 90222,
      "model yields better performance": 72338,
      "parsing semantic role labeling": 85227,
      "sentiment analysis paper describes": 106624,
      "workshop building linguistically generalizable": 135004,
      "linguistically generalizable nlp systems": 64587,
      "address problem propose paper": 3746,
      "achieves comparable performance stateoftheart": 2760,
      "comparable performance stateoftheart supervised": 19161,
      "neural machine translation examine": 77999,
      "numerous natural language processing": 81025,
      "processing nlp tasks recent": 91758,
      "mikolov et al 2013": 70081,
      "use recurrent neural networks": 128232,
      "recurrent neural networks lstm": 97236,
      "dependency grammar induction neural": 29143,
      "neural network model based": 78330,
      "models sutskever et al": 74144,
      "sutskever et al 2014": 116228,
      "resources neural machine translation": 101020,
      "neural machine translation using": 78130,
      "used statistical machine translation": 128785,
      "fundamental component nlp tasks": 46091,
      "nlp tasks named entity": 79743,
      "using convolutional neural network": 129498,
      "convolutional neural network word": 23197,
      "significantly outperforms existing models": 109014,
      "play important role natural": 87697,
      "important role natural language": 52244,
      "proposed model achieves significant": 93430,
      "model achieves significant improvements": 70591,
      "translation nmt achieved notable": 125023,
      "nmt achieved notable success": 79824,
      "achieved notable success recent": 2661,
      "aspect extraction sentiment classification": 9329,
      "deep learning based natural": 28258,
      "learning based natural language": 62394,
      "recently variety model designs": 96776,
      "variety model designs methods": 130996,
      "model designs methods blossomed": 70976,
      "designs methods blossomed context": 29657,
      "context natural language processing": 22193,
      "language processing nlp paper": 59910,
      "improve machine translation quality": 52413,
      "machine translation smt framework": 66227,
      "learning word sense embeddings": 63169,
      "discourse relations challenging task": 32078,
      "challenging task natural language": 16325,
      "labeled data target domain": 58447,
      "domain work propose approach": 33700,
      "reductions word error rate": 97475,
      "word embedding features model": 133011,
      "model achieves state art": 70594,
      "achieves state art performance": 2875,
      "endtoend trainable neural network": 36982,
      "responses successfully complete taskoriented": 101299,
      "second dialog state tracking": 104403,
      "dialog state tracking challenge": 30589,
      "state tracking challenge dstc2": 112522,
      "neural network models using": 78356,
      "question answering question answering": 95089,
      "answering question answering qa": 6681,
      "question answering qa systems": 95085,
      "using questionanswer pairs supervision": 130077,
      "experimental results datasets framework": 40582,
      "consistently improves performance achieving": 21420,
      "improves performance achieving competitive": 53008,
      "performance achieving competitive results": 86119,
      "achieving competitive results despite": 2942,
      "competitive results despite use": 19683,
      "natural language sentences machine": 76534,
      "machine translation image captioning": 66001,
      "leverages pretrained language model": 63655,
      "experiments language pairs demonstrate": 40974,
      "improve performance nmt systems": 52478,
      "neural network model outperforms": 78336,
      "architecture named entity recognition": 8703,
      "named entity recognition disambiguation": 76065,
      "crucial natural language processing": 25153,
      "perform extensive error analysis": 86001,
      "recurrent neural network models": 97211,
      "word vectors word vectors": 133654,
      "intrinsic evaluation metric word": 56355,
      "pretrained word embeddings input": 90220,
      "gu et al 2016": 49211,
      "model outperformed baseline model": 71630,
      "neural machine translation work": 78133,
      "language model training data": 59413,
      "endtoend neural network architectures": 36945,
      "conditional random fields crfs": 20785,
      "decoder neural machine translation": 28036,
      "encoderdecoder neural machine translation": 36611,
      "paper presents approach task": 84150,
      "multilayer recurrent neural network": 75185,
      "recurrent neural network approach": 97195,
      "approach substantially outperforms previous": 7941,
      "outperforms previous work terms": 82958,
      "previous work terms bleu": 90558,
      "rules stored exception structure": 103444,
      "stored exception structure new": 113386,
      "exception structure new rules": 39926,
      "structure new rules added": 113920,
      "new rules added correct": 79116,
      "approach outperforms previous stateoftheart": 7774,
      "paper presents empirical study": 84167,
      "based neural network architecture": 11894,
      "named entity recognition syntactic": 76115,
      "entity recognition syntactic parsing": 38075,
      "word embedding methods word2vec": 133020,
      "embedding methods word2vec glove": 35446,
      "method learning domainspecific word": 68938,
      "performance neural machine translation": 86566,
      "mitigating impact speech recognition": 70387,
      "impact speech recognition errors": 51893,
      "machine learning based method": 65782,
      "vector machine svm model": 131319,
      "multisource neural machine translation": 75788,
      "train machine translation systems": 122954,
      "language processing nlp techniques": 59924,
      "intent classification slot filling": 55898,
      "shortterm memory lstm networks": 108334,
      "memory lstm networks model": 68339,
      "method outperforms baseline models": 69009,
      "learning language understanding lu": 62675,
      "language understanding lu dialogue": 60235,
      "semantic textual similarity natural": 105332,
      "textual similarity natural language": 121741,
      "similarity natural language inference": 109275,
      "natural language inference tasks": 76356,
      "natural language natural language": 76375,
      "natural language visual reasoning": 76605,
      "nlp convolutional neural networks": 79593,
      "representation word taking account": 99468,
      "16th early 20th century": 397,
      "results natural language processing": 101976,
      "analysis social media twitter": 5862,
      "model outperforms state art": 71666,
      "natural language inference neural": 76333,
      "models natural language inference": 73608,
      "lexical syntactic semantic information": 63831,
      "multilayer neural network models": 75179,
      "wmt metrics shared task": 132800,
      "convolutional recurrent neural networks": 23215,
      "optical character recognition ocr": 82158,
      "quality machine translation paper": 94713,
      "machine translation paper propose": 66144,
      "propose new deep learning": 92846,
      "model outperforms stateoftheart baseline": 71670,
      "recurrent neural network translation": 97222,
      "neural network translation models": 78412,
      "learning crosslingual word embeddings": 62477,
      "monolingual word embedding spaces": 74630,
      "pos tagging dependency parsing": 88229,
      "task neural machine translation": 118454,
      "machine translation nmt struggles": 66120,
      "sentiment analysis spanish tweets": 106662,
      "analysis paper describes systems": 5737,
      "shared task customer feedback": 108036,
      "task customer feedback analysis": 118043,
      "baseline model best performing": 12257,
      "model best performing model": 70772,
      "proposing multitask learning approach": 93626,
      "approach leads significant improvements": 7672,
      "significant improvements baseline model": 108785,
      "neural network paper propose": 78369,
      "network paper propose novel": 77383,
      "paper propose novel deep": 84293,
      "convolutional neural network architecture": 23178,
      "experimental results demonstrate effectiveness": 40585,
      "best accuracy stanford dependencies": 13300,
      "support vector machine classifier": 116002,
      "languages english french japanese": 60536,
      "english french japanese spanish": 37145,
      "grammar induction task learning": 48642,
      "learning neural machine translation": 62841,
      "semantic syntactic information words": 105316,
      "including long shortterm memory": 53315,
      "results demonstrate effectiveness proposed": 101677,
      "problem generating natural language": 91066,
      "traditional statistical machine translation": 122870,
      "translation smt neural machine": 125273,
      "corpora neural machine translation": 23539,
      "translation nmt new paradigm": 125048,
      "learning natural language processing": 62827,
      "language processing nlp models": 59908,
      "created new word analogy": 24680,
      "original english word2vec word": 82518,
      "english word2vec word analogy": 37346,
      "word2vec word analogy corpus": 133688,
      "corpora basic evaluation word": 23425,
      "basic evaluation word similarities": 12522,
      "quality resulting word embeddings": 94780,
      "latent dirichlet allocation lda": 61582,
      "vietnamese partofspeech pos tagging": 131691,
      "neural sequence sequence models": 78674,
      "new state art results": 79152,
      "nlpcc 2017 shared task": 79796,
      "present language independent unsupervised": 89527,
      "model handles problem data": 71275,
      "yields improved word embeddings": 135418,
      "alignment paper propose novel": 4978,
      "introduce new evaluation method": 56483,
      "paraphrase identification natural language": 84820,
      "identification natural language inference": 51408,
      "bring significant improvements existing": 14637,
      "large scale human evaluation": 61248,
      "widely used natural language": 132577,
      "different word embedding models": 31569,
      "translation nmt models generally": 125043,
      "natural language understanding models": 76565,
      "supervised machine learning algorithm": 115789,
      "significantly outperform stateoftheart methods": 108992,
      "slot filling spoken language": 110058,
      "filling spoken language understanding": 44057,
      "slot filling task spoken": 110061,
      "filling task spoken language": 44060,
      "task spoken language understanding": 118737,
      "finite dimensional vector spaces": 44535,
      "model outperforms stateoftheart baselines": 71672,
      "neural machine translation based": 77977,
      "translation based neural network": 124666,
      "sequence sequence learning problem": 107057,
      "domain neural machine translation": 33596,
      "models trained proposed approach": 74225,
      "natural language processing recently": 76480,
      "machine learning neural networks": 65820,
      "learning neural networks methods": 62845,
      "achieves performance comparable stateoftheart": 2832,
      "used natural language generation": 128647,
      "natural language generation tasks": 76321,
      "experiments proposed model outperforms": 41081,
      "proposed model outperforms current": 93463,
      "model outperforms current stateoftheart": 71644,
      "knowledgebased word sense disambiguation": 58260,
      "data available social media": 25682,
      "areas natural language processing": 8896,
      "natural language processing sentiment": 76483,
      "language processing sentiment analysis": 59944,
      "different domains different languages": 31112,
      "using machine learning methods": 129847,
      "stateoftheart speech recognition systems": 112968,
      "recognition systems rely heavily": 97021,
      "propose novel deep learning": 92912,
      "subcorpus gutenberg english poetry": 114659,
      "gutenberg english poetry corpus": 49289,
      "recurrent neural machine translation": 97190,
      "experiments chineseenglish englishgerman translation": 40825,
      "chineseenglish englishgerman translation tasks": 16851,
      "englishgerman translation tasks demonstrate": 37390,
      "tasks demonstrate proposed model": 119042,
      "demonstrate proposed model achieves": 28848,
      "models achieved great success": 72680,
      "comparable performance stateoftheart methods": 19160,
      "principal component analysis pca": 90673,
      "results support vector machine": 102242,
      "stateoftheart recurrent neural network": 112883,
      "tasks information extraction knowledge": 119202,
      "information extraction knowledge base": 54574,
      "extraction knowledge base construction": 42364,
      "order promote research area": 82390,
      "neural machine translation partofspeech": 78075,
      "tasks neural machine translation": 119340,
      "based neural network models": 11895,
      "using feedforward neural network": 129675,
      "goaloriented dialogue systems colloquially": 48414,
      "dialogue systems colloquially known": 30771,
      "systems colloquially known goal": 116789,
      "colloquially known goal oriented": 18527,
      "known goal oriented chatbots": 58300,
      "goal oriented chatbots help": 48375,
      "oriented chatbots help users": 82489,
      "chatbots help users achieve": 16674,
      "help users achieve predefined": 49787,
      "users achieve predefined goal": 129092,
      "achieve predefined goal book": 2513,
      "predefined goal book movie": 88826,
      "goal book movie ticket": 48335,
      "book movie ticket closed": 14410,
      "movie ticket closed domain": 74909,
      "ticket closed domain step": 121959,
      "closed domain step understand": 17845,
      "domain step understand users": 33660,
      "step understand users goal": 113309,
      "understand users goal using": 126788,
      "users goal using natural": 129126,
      "goal using natural language": 48397,
      "using natural language understanding": 129940,
      "natural language understanding techniques": 76588,
      "language understanding techniques goal": 60288,
      "understanding techniques goal known": 126995,
      "techniques goal known bot": 119897,
      "goal known bot manage": 48362,
      "known bot manage dialogue": 58291,
      "bot manage dialogue achieve": 14483,
      "manage dialogue achieve goal": 66898,
      "dialogue achieve goal conducted": 30627,
      "achieve goal conducted respect": 2465,
      "goal conducted respect learnt": 48340,
      "conducted respect learnt policy": 20942,
      "respect learnt policy success": 101088,
      "learnt policy success dialogue": 63252,
      "policy success dialogue depends": 87967,
      "success dialogue depends quality": 115075,
      "dialogue depends quality policy": 30662,
      "depends quality policy turn": 29291,
      "quality policy turn reliant": 94755,
      "policy turn reliant availability": 87972,
      "turn reliant availability highquality": 125978,
      "reliant availability highquality training": 98636,
      "availability highquality training data": 10911,
      "highquality training data policy": 50412,
      "training data policy learning": 123514,
      "data policy learning method": 26245,
      "policy learning method instance": 87960,
      "learning method instance deep": 62723,
      "method instance deep reinforcement": 68914,
      "instance deep reinforcement learning": 55597,
      "deep reinforcement learning domain": 28411,
      "reinforcement learning domain specificity": 97804,
      "learning domain specificity available": 62521,
      "domain specificity available data": 33655,
      "specificity available data typically": 111616,
      "available data typically low": 10973,
      "data typically low allow": 26589,
      "typically low allow training": 126443,
      "low allow training good": 65349,
      "allow training good dialogue": 5095,
      "training good dialogue policies": 123644,
      "introduce transfer learning method": 56558,
      "transfer learning method mitigate": 124102,
      "learning method mitigate effects": 62727,
      "method mitigate effects low": 68965,
      "mitigate effects low indomain": 70365,
      "effects low indomain data": 34992,
      "low indomain data availability": 65368,
      "indomain data availability transfer": 53962,
      "data availability transfer learning": 25668,
      "availability transfer learning based": 10930,
      "transfer learning based approach": 124083,
      "learning based approach improves": 62385,
      "based approach improves bots": 11513,
      "approach improves bots success": 7625,
      "improves bots success rate": 52957,
      "bots success rate 20": 14491,
      "success rate 20 relative": 115121,
      "rate 20 relative terms": 95775,
      "20 relative terms distant": 524,
      "relative terms distant domains": 98381,
      "terms distant domains double": 120310,
      "distant domains double close": 32448,
      "domains double close domains": 33770,
      "double close domains compared": 33963,
      "close domains compared model": 17813,
      "domains compared model transfer": 33747,
      "compared model transfer learning": 19386,
      "model transfer learning transfer": 72226,
      "transfer learning transfer learning": 124141,
      "learning transfer learning chatbots": 63121,
      "transfer learning chatbots learn": 124086,
      "learning chatbots learn policy": 62428,
      "chatbots learn policy 10": 16678,
      "learn policy 10 times": 62114,
      "policy 10 times faster": 87946,
      "10 times faster finally": 151,
      "times faster finally transfer": 122175,
      "faster finally transfer learning": 43176,
      "finally transfer learning approach": 44243,
      "transfer learning approach complementary": 124079,
      "learning approach complementary additional": 62344,
      "approach complementary additional processing": 7439,
      "complementary additional processing warmstarting": 19735,
      "additional processing warmstarting joint": 3561,
      "processing warmstarting joint application": 91854,
      "warmstarting joint application gives": 132028,
      "joint application gives best": 57260,
      "application gives best outcomes": 6854,
      "endtoend deep neural network": 36886,
      "deep neural network models": 28362,
      "pretrained word embedding vectors": 90215,
      "evaluate effectiveness proposed method": 38829,
      "work present novel approach": 134686,
      "combine convolutional neural networks": 18619,
      "use convolutional neural network": 127970,
      "data set automatically extracted": 26401,
      "knearest neighbors knn model": 57719,
      "deep contextualized word representations": 28220,
      "deep bidirectional language model": 28204,
      "bidirectional language model bilm": 13932,
      "added existing models significantly": 3360,
      "existing models significantly improve": 40209,
      "models significantly improve state": 74043,
      "question answering textual entailment": 95118,
      "open information extraction oie": 81906,
      "relation extraction question answering": 98022,
      "problem area natural language": 90942,
      "language processing nlp domain": 59897,
      "experimental evaluations proposed approach": 40540,
      "recent advances artificial intelligence": 96395,
      "neural machine translation toolkit": 78116,
      "improve natural language understanding": 52432,
      "natural language understanding present": 76576,
      "sentiment analysis arabic challenging": 106570,
      "recent years deep neural": 96607,
      "years deep neural networks": 135252,
      "sentiment classification natural language": 106704,
      "classification natural language processing": 17302,
      "language processing applications word": 59841,
      "different machine learning algorithms": 31243,
      "dataset extensive experiments proposed": 26921,
      "supervised deep learning models": 115750,
      "deep learning models based": 28286,
      "semeval2018 task 11 machine": 105533,
      "task 11 machine comprehension": 117813,
      "11 machine comprehension using": 222,
      "machine comprehension using commonsense": 65755,
      "comprehension using commonsense knowledge": 20242,
      "data code publicly available": 25737,
      "natural language processing area": 76400,
      "makes difficult train model": 66773,
      "propose endtoend neural model": 92653,
      "vaswani et al 2017": 131289,
      "stateoftheart results machine translation": 112908,
      "approach yields improvements 13": 8029,
      "yields improvements 13 bleu": 135422,
      "systems achieve good performance": 116713,
      "text remains challenging task": 121242,
      "propose new language modeling": 92857,
      "classification natural language inference": 17301,
      "natural language inference sentence": 76349,
      "paper describes results shared": 83844,
      "describes results shared task": 29430,
      "word sense induction wsi": 133533,
      "propose neural network architecture": 92825,
      "semantic natural language processing": 105120,
      "rajpurkar et al 2016": 95468,
      "research multilingual crosslingual sentiment": 100560,
      "multilingual crosslingual sentiment analysis": 75230,
      "text classification sentiment analysis": 120765,
      "classification sentiment analysis tasks": 17394,
      "propose novel approach called": 92897,
      "neural network architecture learns": 78246,
      "matter neural machine translation": 67489,
      "bilingual word embeddings based": 14071,
      "success natural language processing": 115103,
      "representations work present simple": 99998,
      "neural machine translation framework": 78005,
      "codemixed social media text": 18185,
      "social media platforms twitter": 110409,
      "media platforms twitter facebook": 68150,
      "noisy social media text": 80122,
      "using distributed representations words": 129598,
      "propose novel method called": 92939,
      "network architecture sentiment analysis": 77145,
      "representations resourcepoor languages jointly": 99859,
      "resourcepoor languages jointly training": 100913,
      "languages jointly training resourcerich": 60658,
      "jointly training resourcerich languages": 57398,
      "training resourcerich languages using": 123817,
      "resourcerich languages using siamese": 100935,
      "languages using siamese network": 60955,
      "model consists twin bidirectional": 70897,
      "consists twin bidirectional long": 21502,
      "twin bidirectional long shortterm": 126090,
      "recurrent neural networks bilstm": 97227,
      "neural networks bilstm rnn": 78449,
      "networks bilstm rnn shared": 77527,
      "bilstm rnn shared parameters": 14105,
      "rnn shared parameters joined": 102934,
      "shared parameters joined contrastive": 108001,
      "parameters joined contrastive loss": 84759,
      "joined contrastive loss function": 57252,
      "contrastive loss function based": 22735,
      "loss function based similarity": 65268,
      "function based similarity metric": 46001,
      "based similarity metric model": 12046,
      "similarity metric model learns": 109269,
      "representations resourcepoor resourcerich language": 99862,
      "resourcepoor resourcerich language common": 100918,
      "space using similarity metric": 111077,
      "using similarity metric based": 130175,
      "model projects sentences similar": 71804,
      "datasets resourcerich languages english": 27681,
      "resourcerich languages english spanish": 100930,
      "languages english spanish resourcepoor": 60551,
      "english spanish resourcepoor languages": 37286,
      "spanish resourcepoor languages hindi": 111169,
      "resourcepoor languages hindi telugu": 100910,
      "languages hindi telugu reveal": 60622,
      "approaches based distributional semantics": 8077,
      "based distributional semantics semantic": 11656,
      "distributional semantics semantic rules": 32734,
      "semantics semantic rules lexicon": 105462,
      "semantic rules lexicon lists": 105264,
      "rules lexicon lists deep": 103418,
      "lexicon lists deep neural": 63895,
      "lists deep neural network": 64717,
      "deep neural network representations": 28364,
      "sense disambiguation wsd tasks": 105673,
      "yield significant performance gains": 135354,
      "paper propose simple fast": 84324,
      "machine learning natural language": 65816,
      "neural machine translation significant": 78100,
      "neural machine translation achieved": 77967,
      "paper describes process building": 83842,
      "information paper addresses problem": 54823,
      "named entity recognition using": 76123,
      "random field crf model": 95480,
      "experimental results approach outperforms": 40552,
      "texts russian national corpus": 121599,
      "space paper propose novel": 111039,
      "tasks demonstrate proposed method": 119041,
      "experimental results demonstrate models": 40596,
      "parser achieves stateoftheart performance": 84917,
      "stateoftheart performance downstream tasks": 112822,
      "report results preliminary experiments": 99038,
      "semeval2018 task irony detection": 105539,
      "task irony detection english": 118317,
      "irony detection english tweets": 56940,
      "models based recurrent neural": 72814,
      "capture semantic syntactic information": 15386,
      "word2vec word embeddings pretrained": 133691,
      "550 million english tweets": 1118,
      "models achieve stateoftheart results": 72675,
      "downstream natural language understanding": 34010,
      "natural language understanding tasks": 76584,
      "recognizing textual entailment rte": 97080,
      "fake news challenge dataset": 43014,
      "community question answering cqa": 19092,
      "task natural language inference": 118436,
      "natural language inference task": 76354,
      "extensive experiments proposed model": 41922,
      "experiments proposed model achieves": 41080,
      "pretrained language models lms": 90086,
      "better robustness experiments benchmark": 13714,
      "experiments benchmark datasets demonstrate": 40796,
      "benchmark datasets demonstrate effectiveness": 12773,
      "datasets demonstrate effectiveness method": 27400,
      "transfer transfer learning sentence": 124201,
      "transfer learning sentence embeddings": 124131,
      "sentence embeddings semantic similarity": 105845,
      "natural language understanding work": 76594,
      "received relatively little attention": 96374,
      "little attention paper propose": 64797,
      "generating text structured data": 47277,
      "experiments model outperforms strong": 41018,
      "model outperforms strong baselines": 71680,
      "intelligent personal digital assistants": 55854,
      "personal digital assistants ipdas": 87118,
      "application spoken language understanding": 6887,
      "natural language understanding task": 76583,
      "effectiveness approach extensive experiments": 34869,
      "proposed data augmentation methods": 93249,
      "standard rnn language model": 112297,
      "endtoend taskoriented dialog systems": 36978,
      "usually suffer challenge incorporating": 130458,
      "argument reasoning comprehension task": 8952,
      "given small size dataset": 48131,
      "representations neural machine translation": 99776,
      "neural machine translation use": 78128,
      "graph convolutional networks gcns": 48789,
      "present machine learning approach": 89540,
      "using support vector machines": 130248,
      "support vector machines svms": 116015,
      "translation neural encoderdecoder models": 125007,
      "using natural language inference": 129936,
      "natural language inference propose": 76345,
      "open information extraction systems": 81911,
      "different domains demonstrate effectiveness": 31109,
      "unsupervised word sense disambiguation": 127758,
      "word sense disambiguation underresourced": 133525,
      "sense disambiguation underresourced languages": 105671,
      "native language identification nli": 76226,
      "nist chinesetoenglish wmt englishtogerman": 79503,
      "syntax neural machine translation": 116549,
      "model achieves new stateoftheart": 70577,
      "achieves new stateoftheart performance": 2822,
      "new stateoftheart performance benchmarks": 79163,
      "deep neural network architectures": 28358,
      "computational linguistics artificial intelligence": 20394,
      "word embedding models trained": 133026,
      "achieves state art results": 2877,
      "paper proposes new task": 84362,
      "hierarchical convolutional neural network": 49947,
      "results proposed model outperforms": 102081,
      "quality estimation qe machine": 94651,
      "estimation qe machine translation": 38664,
      "models require large amounts": 73934,
      "datasets penn treebank wikitext2": 27618,
      "experimental results approach significantly": 40553,
      "results approach significantly outperforms": 101519,
      "dialogue systems paper presents": 30782,
      "models promising text generation": 73815,
      "human behavior computer scientists": 50765,
      "build automatic methods infer": 14743,
      "automatic methods infer social": 10595,
      "methods infer social contexts": 69552,
      "semantic relation classification task": 105204,
      "machine translation paper provides": 66148,
      "paper provides comparative analysis": 84383,
      "distributional semantic models dsms": 32721,
      "comprehension present new dataset": 20217,
      "present novel neural architecture": 89618,
      "based universal sentence representations": 12146,
      "machine translation experimental results": 65984,
      "dataset proposed method achieves": 27120,
      "proposed method achieves stateoftheart": 93344,
      "method achieves stateoftheart performance": 68609,
      "data models publicly available": 26140,
      "models use recurrent neural": 74280,
      "tasks paraphrase identification semantic": 119369,
      "paraphrase identification semantic textual": 84825,
      "identification semantic textual similarity": 51438,
      "language modeling machine translation": 59442,
      "models pretrained word embedding": 73789,
      "large amounts parallel data": 61020,
      "model significantly outperforms stateoftheart": 72033,
      "significantly outperforms stateoftheart methods": 109035,
      "bootstrapping natural language understanding": 14462,
      "natural language understanding paper": 76575,
      "embeddings word sense induction": 36035,
      "weighted average word embeddings": 132339,
      "neural machine translation improve": 78006,
      "machine translation improve performance": 66004,
      "word order source target": 133383,
      "order source target languages": 82411,
      "experiments proposed method achieves": 41074,
      "proposed method achieves comparable": 93342,
      "machine translation speech recognition": 66237,
      "tasks including neural machine": 119185,
      "including neural machine translation": 53338,
      "machine translation automatic speech": 65907,
      "translation automatic speech recognition": 124660,
      "speech recognition speech synthesis": 111777,
      "introduce neural network architecture": 56469,
      "various neural network models": 131152,
      "recurrent convolutional neural network": 97169,
      "machine translation grammatical error": 65996,
      "translation grammatical error correction": 124845,
      "natural language processing community": 76409,
      "high number oov words": 50091,
      "neural sequence labeling models": 78671,
      "2019 shared task crosslingual": 710,
      "transfer learning approach based": 124078,
      "approach achieves stateoftheart performance": 7324,
      "model outperforms existing stateoftheart": 71652,
      "outperforms existing stateoftheart models": 82901,
      "open domain suggestion mining": 81901,
      "machine translation present challenge": 66162,
      "translation present challenge set": 125132,
      "held concert annual conference": 49683,
      "summarize research trends papers": 115611,
      "research trends papers presented": 100655,
      "trends papers presented proceedings": 125753,
      "multilingual neural machine translation": 75308,
      "attention models simple effective": 9936,
      "models shown promising results": 74031,
      "promising results text classification": 92304,
      "short term memory bilstm": 108249,
      "classification tasks sentiment analysis": 17458,
      "achieve new state art": 2502,
      "applications natural language generation": 6972,
      "baseline recurrent neural network": 12294,
      "taskoriented dialogue dialogue state": 118892,
      "work propose novel framework": 134736,
      "multidomain taskoriented dialogue dataset": 75099,
      "model neural machine translation": 71569,
      "order handle issue propose": 82335,
      "nist chineseenglish translation tasks": 79498,
      "model achieves substantial improvements": 70611,
      "development natural language processing": 30403,
      "natural language processing language": 76429,
      "using conditional random fields": 129472,
      "natural language processing algorithms": 76391,
      "uses conditional random field": 129209,
      "englishhindi codemixed social media": 37403,
      "codemixed social media content": 18183,
      "social media content corpus": 110362,
      "media content corpus baseline": 68094,
      "posting social media called": 88500,
      "social media called codemixing": 110356,
      "linear mixed effects models": 64354,
      "wide variety downstream tasks": 132521,
      "capture syntactic semantic information": 15407,
      "syntactic semantic information words": 116468,
      "language understanding slot filling": 60267,
      "test effectiveness proposed method": 120450,
      "fundamental task natural language": 46120,
      "knowledge graph natural language": 57972,
      "experimental results model outperforms": 40649,
      "results model outperforms stateoftheart": 101946,
      "natural language user utterances": 76598,
      "language user utterances semantic": 60308,
      "user utterances semantic frames": 129055,
      "dialogue state tracker dst": 30749,
      "neural network rnn based": 78387,
      "experiments demonstrate effectiveness approach": 40883,
      "long shortterm memory blstm": 65108,
      "goaloriented dialogue systems based": 48413,
      "language understanding nlu module": 60246,
      "dependency parsing syntactic dependency": 29203,
      "recurrent neural networks recurrent": 97238,
      "deep neural networks shown": 28379,
      "wide range application areas": 132498,
      "memory lstm networks proven": 68340,
      "processing nlp tasks sentiment": 91759,
      "applications neural machine translation": 6978,
      "word analogy task languages": 132919,
      "languages different language families": 60503,
      "successful natural language processing": 115164,
      "embeddings natural language processing": 35824,
      "direction power pairs participants": 31826,
      "power pairs participants email": 88648,
      "language understanding nlu models": 60245,
      "achieves better performance stateoftheart": 2749,
      "tu et al 2016": 125923,
      "automatic evaluation metrics bleu": 10537,
      "experimental results approach achieves": 40551,
      "benchmark datasets compared previous": 12771,
      "used text classification tasks": 128810,
      "standard neural machine translation": 112279,
      "collected cleaned processed illustrate": 18408,
      "cleaned processed illustrate use": 17710,
      "processed illustrate use zeroresource": 91600,
      "computational language documentation experiments": 20385,
      "use computational techniques identify": 127952,
      "named entity recognition techniques": 76120,
      "deep learning natural language": 28299,
      "loss neural machine translation": 65286,
      "distances word embedding space": 32444,
      "data sets proposed method": 26432,
      "interactive neural machine translation": 56026,
      "neural machine translation data": 77984,
      "adapting neural machine translation": 3318,
      "attention mechanism neural machine": 9908,
      "mechanism neural machine translation": 68019,
      "neural machine translation experiments": 78002,
      "language processing nlp systems": 59917,
      "indian languages hindi telugu": 53823,
      "recent progress endtoend speech": 96497,
      "progress endtoend speech recognition": 92149,
      "endtoend speech recognition systems": 36968,
      "switchboard conversational speech corpus": 116282,
      "pretrained word embeddings used": 90224,
      "domain limited labeled data": 33574,
      "model trained source domain": 72210,
      "multichannel convolutional neural network": 75044,
      "model achieves competitive performance": 70567,
      "achieves competitive performance stateoftheart": 2773,
      "language processing tasks partofspeech": 59965,
      "partofspeech pos tagging dependency": 85535,
      "trained language modeling task": 123170,
      "challenge natural language processing": 16062,
      "fundamental shortcoming sequence generation": 46116,
      "address issue propose novel": 3699,
      "proposed method significantly outperforms": 93401,
      "method significantly outperforms stateoftheart": 69148,
      "significantly outperforms stateoftheart baselines": 109033,
      "dataset experimental results method": 26912,
      "experimental results method significantly": 40642,
      "results method significantly outperform": 101928,
      "graph neural networks knowledge": 48849,
      "knowledge base question answering": 57785,
      "semimarkov conditional random fields": 105583,
      "paper propose novel endtoend": 84296,
      "semimarkov conditional random field": 105582,
      "deep recurrent neural networks": 28406,
      "sentiment positive negative neutral": 106781,
      "sentiment analysis limited work": 106604,
      "deep convolutional neural networks": 28227,
      "shallow multilayer perceptron mlp": 107926,
      "wellstudied problem natural language": 132438,
      "question answering named entity": 95063,
      "answering named entity recognition": 6650,
      "tasks semantic role labeling": 119483,
      "semantic role labeling dependency": 105247,
      "minimum semantic units human": 70216,
      "semantic units human languages": 105347,
      "novel soft loss function": 80734,
      "sememe knowledge base hownet": 105489,
      "important research topic natural": 52232,
      "research topic natural language": 100644,
      "topic natural language processing": 122556,
      "performance various natural language": 86845,
      "approach learning graph embeddings": 7679,
      "structural measures pairwise node": 113782,
      "measures pairwise node similarities": 67886,
      "shortest path distance distance": 108305,
      "path distance distance measures": 85671,
      "distance distance measures information": 32415,
      "distance measures information graph": 32426,
      "measures information graph structure": 67874,
      "information graph structure account": 54648,
      "semantic similarity word sense": 105295,
      "approach yields competitive results": 8027,
      "machine translation systems translate": 66254,
      "use reinforcement learning rl": 128239,
      "reinforcement learning rl finetune": 97825,
      "slot filling f1 score": 110045,
      "prediction neural machine translation": 89089,
      "neural machine translation text": 78115,
      "processing nlp applications information": 91728,
      "popular word embedding methods": 88147,
      "paper introduce new corpus": 83999,
      "abusive language detection models": 1846,
      "different pretrained word embeddings": 31352,
      "effectively reduce gender bias": 34844,
      "unsupervised way previous work": 127749,
      "propose novel method automatically": 92938,
      "simple data augmentation strategy": 109395,
      "words source sentence target": 134231,
      "source sentence target sentence": 110816,
      "stateoftheart results various text": 112931,
      "results various text classification": 102315,
      "various text classification tasks": 131222,
      "text classification tasks arabic": 120774,
      "classification tasks arabic dialect": 17440,
      "tasks arabic dialect identification": 118937,
      "arabic dialect identification native": 8505,
      "dialect identification native language": 30522,
      "identification native language identification": 51405,
      "wide range nlp tasks": 132511,
      "learning deep neural network": 62489,
      "yielding new stateoftheart results": 135382,
      "experimental results automatic human": 40559,
      "results automatic human evaluations": 101532,
      "automatic human evaluations demonstrate": 10558,
      "training neural language models": 123733,
      "neural language models trained": 77956,
      "results wordlevel language modeling": 102345,
      "text representation paper present": 121250,
      "especially morphologically rich languages": 38481,
      "using noisy channel model": 129975,
      "using lstm language model": 129843,
      "recent years natural language": 96621,
      "years natural language processing": 135276,
      "paper proposes simple effective": 84369,
      "model uses convolutional neural": 72266,
      "task mapping natural language": 118378,
      "neural models natural language": 78184,
      "work recurrent neural networks": 134765,
      "trained large amounts data": 123175,
      "error reduction benchmark datasets": 38341,
      "machine translation systems wmt18": 66257,
      "wmt18 news translation task": 132846,
      "narayan et al 2017": 76163,
      "work natural language processing": 134647,
      "named entity recognition languages": 76072,
      "substantial improvements strong baselines": 114866,
      "supervised learning reinforcement learning": 115779,
      "neural machine translation propose": 78083,
      "resulting new state art": 101459,
      "models namedentity recognition ner": 73605,
      "address issues paper propose": 3704,
      "question answering qa datasets": 95079,
      "statistical neural mt systems": 113145,
      "natural language processing greatly": 76421,
      "embeddings neural machine translation": 35830,
      "parsing converts natural language": 85090,
      "machine interpretable meaning representations": 65763,
      "proposed approach able achieve": 93181,
      "approach able achieve stateoftheart": 7290,
      "able achieve stateoftheart performance": 1604,
      "wordlevel quality estimation qe": 133749,
      "dual conditional crossentropy filtering": 34230,
      "shared task parallel corpus": 108082,
      "task parallel corpus filtering": 118514,
      "information extraction pipeline automatically": 54593,
      "recent advances neural machine": 96414,
      "advances neural machine translation": 4005,
      "machine translation nmt different": 66095,
      "experimental results demonstrate model": 40595,
      "results demonstrate model significantly": 101693,
      "publicly available parallel corpora": 94317,
      "consisting noisy comments reddit": 21459,
      "network machine reading comprehension": 77310,
      "machine reading comprehension paper": 65864,
      "reading comprehension paper introduce": 96001,
      "machine reading comprehension model": 65856,
      "model extends existing approaches": 71150,
      "extends existing approaches perspectives": 41827,
      "model outperforms stateoftheart models": 71674,
      "neural conversation models tend": 77876,
      "address challenge propose simple": 3657,
      "challenge propose simple effective": 16090,
      "propose simple effective approach": 93055,
      "et al 2016 evaluate": 38703,
      "using automatic metrics human": 129370,
      "automatic metrics human judgments": 10607,
      "machine translation nmt lowresource": 66101,
      "language understanding slu essential": 60270,
      "understanding slu essential component": 126961,
      "slu essential component conversational": 110115,
      "essential component conversational systems": 38548,
      "provide informative cues better": 93855,
      "recent utterances important recent": 96562,
      "utterances important recent ones": 130650,
      "experiments benchmark dialogue state": 40802,
      "benchmark dialogue state tracking": 12816,
      "dialogue state tracking challenge": 30751,
      "state tracking challenge dstc4": 112524,
      "tracking challenge dstc4 dataset": 122751,
      "learning present novel approach": 62905,
      "zhang et al 2018": 135540,
      "conll 2018 shared task": 21085,
      "2018 shared task universal": 683,
      "data augmentation spoken language": 25646,
      "augmentation spoken language understanding": 10305,
      "sentences paper propose novel": 106430,
      "require manually annotated data": 100179,
      "sentence planning surface realization": 105987,
      "previous work shown neural": 90550,
      "participated open track parseme": 85340,
      "open track parseme shared": 81944,
      "track parseme shared task": 122734,
      "et al 2018 modern": 38723,
      "nist chinesetoenglish translation tasks": 79501,
      "recurrent neural networks paper": 97237,
      "machine translation despite recent": 65949,
      "machine translation nmt model": 66102,
      "experimental results datasets method": 40583,
      "achieved recurrent neural network": 2680,
      "vector space capturing semantic": 131371,
      "capturing semantic relations words": 15488,
      "paper propose novel model": 84302,
      "recognizing lexical semantic relations": 97072,
      "lexical semantic relations recognizing": 63815,
      "paper propose novel methods": 84301,
      "pairs experimental results demonstrate": 83541,
      "based hidden markov models": 11754,
      "hidden markov models hmms": 49903,
      "machine translation work investigates": 66304,
      "models based transformer architecture": 72818,
      "based transformer architecture demonstrate": 12129,
      "propose novel model called": 92944,
      "experimental results dataset proposed": 40576,
      "dataset proposed model outperforms": 27124,
      "proposed model outperforms competitive": 93460,
      "model outperforms competitive baselines": 71641,
      "recent years deep learning": 96605,
      "years deep learning methods": 135249,
      "deep learning methods achieved": 28281,
      "methods achieved great success": 69287,
      "premise hypothesis paper propose": 89287,
      "paper propose simple effective": 84322,
      "word embeddings experimental results": 133108,
      "supervised machine learning approaches": 115791,
      "performance paper propose new": 86599,
      "paper propose new framework": 84279,
      "electronic health records ehrs": 35253,
      "training neural networks learn": 123743,
      "neural models terms bleu": 78202,
      "task paper presents nicts": 118508,
      "paper presents nicts participation": 84188,
      "presents nicts participation wmt18": 89878,
      "nicts participation wmt18 shared": 79480,
      "machine translation nmt using": 66126,
      "wmt18 shared news translation": 132849,
      "experimental results model achieves": 40647,
      "results model achieves stateoftheart": 101935,
      "large number training examples": 61192,
      "establish baseline results dataset": 38580,
      "large amounts unlabeled text": 61026,
      "introduce convolutional neural network": 56399,
      "convolutional neural network structure": 23195,
      "utility natural language processing": 130487,
      "results compared stateoftheart methods": 101607,
      "using significantly fewer parameters": 130172,
      "order magnitude larger previous": 82363,
      "novel natural language generation": 80663,
      "natural language generation task": 76319,
      "experiment results proposed method": 40504,
      "results proposed method outperforms": 102073,
      "model shows superior performance": 72016,
      "large amounts labelled data": 61016,
      "global phenomenon multilingual communities": 48262,
      "codeswitching automatic speech recognition": 18228,
      "spoken language understanding spoken": 112006,
      "language understanding spoken language": 60276,
      "understanding spoken language understanding": 126973,
      "language understanding slu systems": 60274,
      "acoustic model language model": 3016,
      "winograd schema challenge wsc": 132729,
      "achieves stateoftheart performance benchmark": 2882,
      "stateoftheart performance benchmark datasets": 112810,
      "important natural language understanding": 52193,
      "datatotext natural language generation": 27817,
      "natural language understanding traditional": 76590,
      "proposed model outperforms baseline": 93459,
      "language processing nlp research": 59916,
      "processing morphologically rich languages": 91718,
      "information retrieval natural language": 54942,
      "natural language generation paper": 76316,
      "pronouns frequently omitted prodrop": 92354,
      "frequently omitted prodrop languages": 45876,
      "omitted prodrop languages chinese": 81657,
      "prodrop languages chinese generally": 91868,
      "languages chinese generally leading": 60442,
      "chinese generally leading significant": 16770,
      "generally leading significant challenges": 46868,
      "leading significant challenges respect": 61906,
      "significant challenges respect production": 108736,
      "challenges respect production complete": 16207,
      "respect production complete translations": 101099,
      "et al 2018 proposed": 38725,
      "novel reconstructionbased approach alleviating": 80701,
      "problems neural machine translation": 91350,
      "approach significantly improves translation": 7905,
      "significantly improves translation performance": 108959,
      "paper propose new architecture": 84275,
      "propose new architecture based": 92836,
      "introducing interlingual loss additional": 56638,
      "interlingual loss additional training": 56126,
      "loss additional training objective": 65254,
      "additional training objective adding": 3591,
      "training objective adding forcing": 123755,
      "objective adding forcing interlingual": 81063,
      "adding forcing interlingual loss": 3380,
      "forcing interlingual loss able": 45021,
      "interlingual loss able train": 56123,
      "loss able train multiple": 65248,
      "able train multiple encoders": 1698,
      "train multiple encoders decoders": 122971,
      "multiple encoders decoders language": 75551,
      "encoders decoders language sharing": 36641,
      "decoders language sharing common": 28067,
      "classification problem image classification": 17346,
      "problem image classification problem": 91078,
      "vector representations experimental results": 131353,
      "neural machine translation large": 78014,
      "corpus nus sms corpus": 23909,
      "able achieve significantly better": 1599,
      "achieve significantly better results": 2546,
      "named entity recognition work": 76124,
      "named entity recognition models": 76074,
      "approaches leverage machine translation": 8216,
      "bilingual contextual word similarity": 14025,
      "contextual word similarity bcws": 22524,
      "paper propose new model": 84282,
      "standard datasets demonstrate effectiveness": 112223,
      "datasets demonstrate effectiveness approach": 27399,
      "important natural language applications": 52188,
      "speech recognition machine translation": 111760,
      "information neural machine translation": 54801,
      "machine translation previous studies": 66170,
      "annotation experimental results various": 6310,
      "task oriented dialog systems": 118485,
      "outperforms strong baselines terms": 83028,
      "dialogue systems conversational agents": 30774,
      "proven useful nlp tasks": 93737,
      "including named entity recognition": 53331,
      "named entity recognition partofspeech": 76100,
      "entity recognition partofspeech tagging": 38058,
      "obtains new stateoftheart performance": 81470,
      "based deep neural networks": 11632,
      "deep neural networks state": 28380,
      "neural networks state art": 78584,
      "language pairs english french": 59748,
      "pairs english french english": 83528,
      "different neural machine translation": 31293,
      "information different representation subspaces": 54491,
      "different representation subspaces different": 31382,
      "representation subspaces different positions": 99427,
      "et al 2017 work": 38714,
      "experimental results wmt14 englishtogerman": 40726,
      "demonstrate proposed approach outperforms": 28834,
      "outperforms strong transformer baseline": 83030,
      "machine learning models task": 65813,
      "training data paper present": 123507,
      "multilingual contextual word representations": 75222,
      "amounts target language data": 5359,
      "publicly available state art": 94325,
      "available state art natural": 11113,
      "state art natural language": 112455,
      "art natural language generator": 9067,
      "integer linear programming model": 55737,
      "present alternative approach based": 89365,
      "generation propose new task": 47572,
      "data different languages english": 25844,
      "languages english lowresource languages": 60545,
      "experimental results existing datasets": 40617,
      "propose new neural language": 92874,
      "new neural language model": 79053,
      "results demonstrate effectiveness approach": 101674,
      "language models recent years": 59649,
      "et al 2014 work": 38700,
      "guo et al 2019": 49284,
      "et al 2019 dataset": 38729,
      "automatic metrics human evaluations": 10605,
      "neural word sense disambiguation": 78732,
      "reduce number different sense": 97345,
      "number different sense tags": 80872,
      "disambiguate words lexical database": 31944,
      "leads state art results": 61963,
      "results significantly outperform state": 102186,
      "significantly outperform state art": 108986,
      "consistently outperforms strong baselines": 21436,
      "short social media posts": 108238,
      "tokens extensive experiments datasets": 122310,
      "compared standard beam search": 19450,
      "crosslingual transfer natural language": 25036,
      "collect release large dataset": 18392,
      "evaluated automatic metrics human": 38954,
      "external language model lm": 42025,
      "slot filling paper describes": 110052,
      "feature decay algorithms fda": 43260,
      "systems require large amounts": 117114,
      "speech commands dataset shows": 111659,
      "make good use relevant": 66680,
      "experimental results natural language": 40659,
      "proposed method achieve better": 93337,
      "better performance existing methods": 13658,
      "natural language generation dialogue": 76301,
      "sources including social media": 110896,
      "social media feeds news": 110381,
      "media feeds news blogs": 68115,
      "feeds news blogs online": 43859,
      "news blogs online newspapers": 79312,
      "various deep learning models": 131077,
      "outperforms current state art": 82876,
      "model opinion target extraction": 71614,
      "conduct extensive experiments benchmark": 20871,
      "benchmark datasets framework achieves": 12786,
      "previous work bridging anaphora": 90523,
      "bridging anaphora resolution poesio": 14605,
      "anaphora resolution poesio et": 6070,
      "resolution poesio et al": 100778,
      "poesio et al 2004": 87784,
      "et al 2004 hou": 38686,
      "al 2004 hou et": 4595,
      "2004 hou et al": 548,
      "hou et al 2013b": 50678,
      "results bridging anaphora resolution": 101560,
      "relies recurrent neural networks": 98658,
      "word level phrase level": 133345,
      "translation task paper present": 125340,
      "challenge neural machine translation": 16066,
      "used various natural language": 128843,
      "named entity recognition entity": 76067,
      "problem twophase task detecting": 91275,
      "pairs followed organizing pairs": 83550,
      "datasets different domains demonstrate": 27426,
      "dialogue systems shed light": 30789,
      "improve quality generated responses": 52509,
      "diversitypromoting objective function neural": 32904,
      "maximum likelihood estimation mle": 67538,
      "context paper propose new": 22211,
      "function encourages model generate": 46014,
      "graph neural networks gnns": 48848,
      "tagging natural language processing": 117426,
      "neural machine translation lowresource": 78019,
      "machine translation lowresource languages": 66035,
      "machine translation mt tasks": 66072,
      "language pairs parallel corpora": 59774,
      "neural networks attention mechanism": 78442,
      "deep neural network architecture": 28357,
      "based sequencetosequence neural network": 12033,
      "slotfilling paradigm user refer": 110079,
      "paradigm user refer slots": 84553,
      "user refer slots context": 129035,
      "refer slots context conversation": 97501,
      "slots context conversation goal": 110086,
      "context conversation goal contextual": 22041,
      "conversation goal contextual understanding": 22954,
      "goal contextual understanding resolve": 48344,
      "contextual understanding resolve referring": 22509,
      "understanding resolve referring expressions": 126945,
      "resolve referring expressions appropriate": 100807,
      "referring expressions appropriate slots": 97577,
      "expressions appropriate slots context": 41749,
      "propose endtoend neural network": 92654,
      "enhanced sequential inference model": 37521,
      "address challenge outofvocabulary oov": 3653,
      "challenge outofvocabulary oov words": 16075,
      "outofvocabulary oov words second": 82677,
      "oov words second attentive": 81873,
      "words second attentive hierarchical": 134183,
      "second attentive hierarchical recurrent": 104391,
      "attentive hierarchical recurrent encoder": 10113,
      "hierarchical recurrent encoder ahre": 49993,
      "program using natural language": 92119,
      "different machine learning models": 31244,
      "models taskoriented dialogue systems": 74163,
      "language understanding nlu systems": 60247,
      "taskoriented dialog systems slot": 118886,
      "dialog systems slot filling": 30603,
      "neural machine translation explore": 78003,
      "context neural machine translation": 22198,
      "continuous latent variable model": 22623,
      "interactive speech recognition systems": 56032,
      "improves word error rate": 53064,
      "conditional random fields crf": 20783,
      "random fields crf model": 95485,
      "experiments neural machine translation": 41037,
      "neural machine translation tasks": 78114,
      "wmt14 englishgerman wmt17 chineseenglish": 132819,
      "englishgerman wmt17 chineseenglish translation": 37399,
      "advances natural language processing": 4000,
      "natural language processing help": 76422,
      "paper present new corpus": 84120,
      "natural language understanding recently": 76580,
      "neural networkbased natural language": 78429,
      "networkbased natural language understanding": 77491,
      "representations recurrent neural networks": 99845,
      "neural networks rnns learn": 78563,
      "learn continuous vector representations": 62011,
      "deep learning based nlp": 28260,
      "deep neural networks natural": 28375,
      "networks natural language processing": 77670,
      "language processing nlp interpretability": 59902,
      "processing nlp interpretability remains": 91738,
      "nlp interpretability remains challenge": 79626,
      "dialog technology challenges dstc6": 30612,
      "approaches based neural networks": 8081,
      "tagging dependency parsing propose": 117385,
      "propose multitask learning model": 92811,
      "word segmentation partofspeech pos": 133499,
      "segmentation partofspeech pos tagging": 104612,
      "bist graphbased dependency parser": 14234,
      "graphbased dependency parser kiperwasser": 48892,
      "dependency parser kiperwasser goldberg": 29162,
      "parser kiperwasser goldberg 2016": 84969,
      "benchmark datasets experimental results": 12783,
      "dialog technology challenges dstc7": 30613,
      "previous stateoftheart models use": 90480,
      "proposed model outperforms previous": 93467,
      "model outperforms previous models": 71656,
      "outperforms previous models including": 82940,
      "models achieves new stateoftheart": 72703,
      "achieves new stateoftheart performances": 2823,
      "language model trained text": 59410,
      "model trained text corpus": 72215,
      "corpus linguistic acceptability cola": 23869,
      "bert devlin et al": 13097,
      "devlin et al 2018": 30470,
      "et al 2018 gpt": 38722,
      "gpt radford et al": 48556,
      "radford et al 2018": 95435,
      "amie automatedvehicle multimodal incabin": 5321,
      "automatedvehicle multimodal incabin experience": 10485,
      "topic field natural language": 122521,
      "nlp tasks sentence classification": 79766,
      "training data study propose": 123545,
      "tasks including machine translation": 119180,
      "crosslingual language model pretraining": 24966,
      "language model pretraining recent": 59381,
      "improving previous state art": 53148,
      "obtain new state art": 81304,
      "code pretrained models publicly": 18135,
      "pretrained models publicly available": 90149,
      "word embeddings propose novel": 133191,
      "different text classification datasets": 31489,
      "task word sense disambiguation": 118853,
      "word given context using": 133300,
      "according official evaluation results": 1993,
      "machine translation language modeling": 66022,
      "wmt14 englishgerman test set": 132816,
      "natural language understanding natural": 76567,
      "language understanding natural language": 60241,
      "relatively small amounts training": 98416,
      "small amounts training data": 110133,
      "publicly available data sets": 94299,
      "present novel semantic framework": 89623,
      "novel semantic framework modeling": 80716,
      "covering entirety universal dependencies": 24541,
      "entirety universal dependencies english": 37739,
      "universal dependencies english web": 127286,
      "dependencies english web treebank": 29081,
      "english web treebank use": 37337,
      "web treebank use dataset": 132268,
      "crosslingual transfer nlp models": 25038,
      "little parallel data available": 64824,
      "retrieval natural language processing": 102415,
      "popular word embedding models": 88148,
      "state art performance benchmark": 112464,
      "art performance benchmark datasets": 9078,
      "unstructured natural language text": 127578,
      "reading comprehension answer reranking": 95976,
      "recent advances deep neural": 96400,
      "neural networks language modeling": 78500,
      "neural network rnn model": 78392,
      "recurrent neural network used": 97223,
      "biomedical natural language processing": 14199,
      "natural language processing despite": 76415,
      "despite recent advances natural": 29720,
      "recent advances natural language": 96410,
      "natural language processing statistical": 76485,
      "neural named entity recognition": 78220,
      "entity recognition ner important": 38040,
      "short text classification knowledge": 108257,
      "text classification knowledge powered": 120730,
      "classification knowledge powered attention": 17244,
      "knowledge external knowledge source": 57927,
      "purpose measuring importance knowledge": 94434,
      "measuring importance knowledge introduce": 67915,
      "importance knowledge introduce attention": 52064,
      "knowledge introduce attention mechanisms": 58025,
      "information unlike traditional approaches": 55071,
      "conduct extensive experiments public": 20875,
      "extensive experiments public datasets": 41926,
      "experiments public datasets different": 41093,
      "able achieve stateoftheart results": 1605,
      "similar performance models trained": 109120,
      "achieve competitive results compared": 2449,
      "results compared current stateoftheart": 101601,
      "representation paper propose novel": 99368,
      "word embeddings word embedding": 133248,
      "tasks machine translation mt": 119278,
      "entity discovery linking edl": 37926,
      "address issue propose new": 3698,
      "text classification datasets demonstrate": 120718,
      "evaluation shows proposed model": 39402,
      "datasets proposed model significantly": 27645,
      "significantly outperforms existing stateoftheart": 109015,
      "paraphrase identification question answering": 84823,
      "graph convolutional network gcn": 48782,
      "experimental results datasets demonstrate": 40578,
      "english chinese penn treebanks": 37091,
      "optimization neural machine translation": 82200,
      "use reinforcement learning learn": 128238,
      "highquality datasets digital humanities": 50379,
      "fantasy novel book series": 43065,
      "resources natural language processing": 101017,
      "multilingual word sense disambiguation": 75405,
      "f1 score heldout test": 42683,
      "score heldout test set": 104078,
      "pretrained word embedding models": 90214,
      "entity recognition ner task": 38050,
      "task achieve new stateoftheart": 117833,
      "new stateoftheart results task": 79172,
      "nlp tasks source code": 79770,
      "source code data available": 110717,
      "entity recognition task named": 38078,
      "recognition task named entity": 97024,
      "task named entity recognition": 118431,
      "factors contribute effective pretraining": 42886,
      "generation tasks machine translation": 47670,
      "open domain question answering": 81897,
      "question answering current approaches": 95033,
      "approaches natural language generation": 8248,
      "opendomain question answering qa": 81978,
      "question answering qa based": 95077,
      "achieving best accuracy date": 2930,
      "best accuracy date english": 13297,
      "accuracy date english ptb": 2128,
      "date english ptb dataset": 27827,
      "dependency parsing semantic role": 29199,
      "semantic role labeling named": 105251,
      "role labeling named entity": 103188,
      "labeling named entity recognition": 58516,
      "named entity recognition natural": 76077,
      "entity recognition natural language": 38029,
      "recognition natural language inference": 96919,
      "languages universal dependencies project": 60939,
      "process natural language processing": 91535,
      "languages english french german": 60535,
      "experiments realworld datasets demonstrate": 41108,
      "realworld datasets demonstrate proposed": 96162,
      "demonstrate proposed method outperforms": 28841,
      "developers train custom models": 30332,
      "language modeling text generation": 59475,
      "highquality embeddings rare words": 50383,
      "learning spoken language understanding": 63048,
      "language understanding slu models": 60272,
      "japanese predicate argument structure": 57197,
      "predicate argument structure analysis": 88849,
      "argument structure analysis pasa": 8962,
      "models achieved stateoftheart results": 72696,
      "relations expressed single sentences": 98169,
      "noisy labels distant supervision": 80111,
      "elmo embeddings peters et": 35315,
      "embeddings peters et al": 35862,
      "peters et al 2018": 87210,
      "performance downstream nlp tasks": 86321,
      "suggestion mining online reviews": 115359,
      "mining online reviews forums": 70251,
      "recurrent neural network architecture": 97196,
      "results large room improvement": 101882,
      "promising directions future research": 92274,
      "iteratively refine span representations": 57146,
      "framework significantly outperforms stateoftheart": 45687,
      "stringtotree neural machine translation": 113610,
      "neural networks paper propose": 78533,
      "evaluate performance proposed model": 38889,
      "dataset model achieves stateoftheart": 27029,
      "neural sequencetosequence models currently": 78679,
      "approach natural language processing": 7730,
      "language processing tasks require": 59969,
      "model does require parallel": 71022,
      "representations neural language models": 99774,
      "neural language models nlm": 77951,
      "address problem propose method": 3743,
      "reduces word error rate": 97405,
      "model abstractive text summarization": 70524,
      "model achieves strong performance": 70609,
      "machine translation training data": 66275,
      "minimum description length paradigm": 70209,
      "complex word identification cwi": 19896,
      "shared task released data": 108089,
      "neural networks multitask learning": 78519,
      "learning models achieve stateoftheart": 62770,
      "models achieve stateoftheart performance": 72673,
      "using word movers distance": 130383,
      "approach natural language generation": 7729,
      "surface realization shared task": 116074,
      "language model lm based": 59345,
      "generating pun sentence given": 47250,
      "pun sentence given pair": 94391,
      "demonstrate method significantly outperforms": 28783,
      "speech recognition asr errors": 111737,
      "model learn domaininvariant features": 71430,
      "demonstrate effectiveness proposed model": 28723,
      "method achieves consistent improvements": 68599,
      "novel language representation model": 80613,
      "methods achieving new stateoftheart": 69292,
      "achieving new stateoftheart results": 2965,
      "chinese natural language processing": 16794,
      "language processing tasks including": 59957,
      "including natural language inference": 53334,
      "natural language inference semantic": 76348,
      "named entity recognition sentiment": 76111,
      "entity recognition sentiment analysis": 38070,
      "recognition sentiment analysis question": 96996,
      "sentiment analysis question answering": 106641,
      "data work propose new": 26633,
      "work propose new task": 134732,
      "propose simple data augmentation": 93053,
      "past future neural machine": 85645,
      "future neural machine translation": 46285,
      "shown neural machine translation": 108498,
      "translation nmt models benefit": 125041,
      "results model effectively predict": 101940,
      "using contextualized word embeddings": 129487,
      "news articles using neural": 79305,
      "shortterm memory bilstm network": 108319,
      "using ngram language model": 129969,
      "method achieves high performance": 68602,
      "superior performance compared stateoftheart": 115687,
      "performance compared stateoftheart baselines": 86239,
      "evaluation neural machine translation": 39309,
      "propose deep learning based": 92617,
      "deep learning based model": 28256,
      "sentence encoders language modeling": 105853,
      "corpus open information extraction": 23918,
      "information extraction oie systems": 54585,
      "task binary classification task": 117940,
      "shortterm memory lstm network": 108333,
      "contextualized word embeddings present": 22578,
      "biomedical translation shared task": 14213,
      "translation shared task paper": 125249,
      "shared task paper describes": 108078,
      "paper describes machine translation": 83826,
      "machine translation systems developed": 66249,
      "machine translation using moses": 66295,
      "natural language generation neural": 76308,
      "natural language generation models": 76305,
      "language understanding generation tasks": 60225,
      "employing shared transformer network": 36322,
      "achieves new stateoftheart results": 2825,
      "new stateoftheart results natural": 79169,
      "stateoftheart results natural language": 112913,
      "code pretrained models available": 18134,
      "language generation tasks including": 59102,
      "experimental results demonstrate framework": 40589,
      "achieves stateoftheart performance unsupervised": 2893,
      "stateoftheart performance unsupervised approach": 112853,
      "using twodimensional word embedding": 130331,
      "recent work super characters": 96592,
      "work super characters method": 134836,
      "stateoftheart results text classification": 112925,
      "results text classification tasks": 102272,
      "interactive demo ready workshop": 56017,
      "learning natural language understanding": 62829,
      "generation natural language understanding": 47501,
      "language generation nlg critical": 59083,
      "field natural language understanding": 43971,
      "sentiment analysis sentiment analysis": 106654,
      "sentiment analysis opinion mining": 106620,
      "waikato environment knowledge analysis": 131986,
      "environment knowledge analysis weka": 38188,
      "paper propose novel techniques": 84310,
      "language representation models bert": 60043,
      "various nlp tasks existing": 131155,
      "existing pretrained language models": 40257,
      "knowledge graphs kgs provide": 57987,
      "common nlp tasks source": 18902,
      "source code paper obtained": 110722,
      "pretrained bidirectional language models": 89999,
      "stateoftheart performance wide range": 112856,
      "reading comprehension natural language": 95997,
      "comprehension natural language inference": 20205,
      "natural language inference sentiment": 76350,
      "language inference sentiment analysis": 59196,
      "transfer learning multilingual model": 124110,
      "address data sparsity issue": 3673,
      "experimental results largescale datasets": 40632,
      "significantly outperforms stateoftheart models": 109036,
      "training neural network models": 123741,
      "machine translation image video": 66002,
      "implemented following clientserver architecture": 51964,
      "developed website communicates neural": 30322,
      "graph neural network gnn": 48843,
      "using semantic similarity measures": 130142,
      "better fit data better": 13592,
      "machine translation nmt shown": 66116,
      "experimental results benchmark dataset": 40562,
      "dataset demonstrate model significantly": 26857,
      "model significantly outperform stateoftheart": 72026,
      "machine translation systems built": 66248,
      "data augmentation neural machine": 25640,
      "augmentation neural machine translation": 10294,
      "machine translation data augmentation": 65939,
      "present novel data augmentation": 89603,
      "novel data augmentation method": 80526,
      "neural machine translation different": 77990,
      "machine translation datasets demonstrate": 65942,
      "superiority method strong baselines": 115704,
      "neural networks rnns widely": 78567,
      "networks rnns widely used": 77750,
      "incremental dialog state tracker": 53726,
      "tracking challenge dstc2 dataset": 122749,
      "model achieve better performance": 70537,
      "entities knowledge graphs kgs": 37809,
      "achieves stateoftheart performance datasets": 2884,
      "word sense induction word": 133531,
      "sense induction word sense": 105683,
      "induction word sense induction": 54042,
      "sense induction wsi task": 105686,
      "release new largescale dataset": 98464,
      "achieved great success various": 2635,
      "results demonstrate proposed method": 101701,
      "texts paper propose new": 121572,
      "masked language modeling mlm": 67293,
      "natural language processing artificial": 76402,
      "language processing artificial intelligence": 59847,
      "generation recent years seen": 47587,
      "open domain dialogue systems": 81893,
      "important application natural language": 52100,
      "proposed method outperforms stateoftheart": 93389,
      "method outperforms stateoftheart approaches": 69031,
      "natural language nl utterances": 76378,
      "universal conceptual cognitive annotation": 127278,
      "wiseman et al 2017": 132742,
      "achieve better bleu scores": 2427,
      "sentiment analysis tasks using": 106669,
      "lowresource named entity recognition": 65539,
      "named entity recognition recent": 76106,
      "entity recognition recent years": 38064,
      "entity recognition ner especially": 38036,
      "achieves new state art": 2820,
      "machine translation work present": 66305,
      "model consistently improves performance": 70892,
      "model vaswani et al": 72302,
      "encoder neural machine translation": 36548,
      "improve translation performance experiment": 52572,
      "translation performance experiment results": 125109,
      "large amounts web data": 61028,
      "natural language inference paraphrase": 76342,
      "language inference paraphrase identification": 59187,
      "knowledge external knowledge bases": 57926,
      "success deep learning models": 115069,
      "datasets text classification experimental": 27759,
      "text classification experimental results": 120720,
      "systems natural language processing": 117014,
      "shows significant performance gains": 108629,
      "novel hierarchical attention mechanism": 80592,
      "machine learning paper presents": 65823,
      "present novel endtoend neural": 89607,
      "novel endtoend neural network": 80559,
      "endtoend neural network model": 36947,
      "approach achieves significant improvement": 7319,
      "largescale multilabel text classification": 61472,
      "multilabel text classification lmtc": 75160,
      "current state art methods": 25342,
      "elmo embeddings improve performance": 35313,
      "outperforms best reported results": 82861,
      "establish new stateoftheart results": 38590,
      "morphological tagging text classification": 74749,
      "performance range nlp tasks": 86650,
      "wide range natural language": 132508,
      "achieve stateoftheart performance standard": 2563,
      "translation terms adequacy fluency": 125374,
      "approach recurrent neural network": 7846,
      "task reach high performance": 118604,
      "workshop noisy usergenerated text": 135013,
      "visual question answering vqa": 131804,
      "multiturn response selection retrievalbased": 75925,
      "matching model response selection": 67414,
      "model response selection retrievalbased": 71921,
      "noisy training data propose": 80130,
      "results public data sets": 102095,
      "public data sets indicate": 94246,
      "plays important role identifying": 87734,
      "recent developments natural language": 96456,
      "english french german spanish": 37143,
      "french german spanish russian": 45812,
      "trained natural language inference": 123214,
      "attention simultaneous machine translation": 10013,
      "simultaneous machine translation simultaneous": 109657,
      "machine translation simultaneous machine": 66222,
      "translation simultaneous machine translation": 125263,
      "elastic weight consolidation ewc": 35233,
      "spanishenglish speech translation corpus": 111181,
      "proposed model outperforms existing": 93464,
      "model outperforms existing research": 71651,
      "demonstrate proposed method significantly": 28843,
      "proposed method significantly improve": 93399,
      "generate diverse set candidate": 46933,
      "paper propose novel neural": 84303,
      "propose novel neural approach": 92951,
      "achieves significantly better performance": 2865,
      "text recurrent neural networks": 121233,
      "responses given previous utterances": 101273,
      "experimental results model significantly": 40652,
      "machine learning techniques improve": 65834,
      "semisupervised sequence labeling model": 105623,
      "experimental results showed proposed": 40699,
      "results showed proposed method": 102175,
      "showed proposed method improved": 108389,
      "neural machine translation despite": 77989,
      "train test machine learning": 123040,
      "test machine learning models": 120469,
      "machine translation neural networks": 66084,
      "multiple neural network architectures": 75627,
      "data best knowledge work": 25700,
      "use deep neural network": 127987,
      "attention neural machine translation": 9956,
      "neural machine translation sequencetosequence": 78098,
      "paper presents empirical evaluation": 84166,
      "dialogue systems paper introduce": 30781,
      "achieve new stateoftheart performance": 2504,
      "specifically propose novel hierarchical": 111585,
      "experiments proposed approach significantly": 41069,
      "approach significantly outperforms existing": 7908,
      "significantly outperforms existing methods": 109013,
      "written modern version documents": 135137,
      "sequences recurrent neural networks": 107136,
      "deep learning techniques using": 28321,
      "using transfer learning approach": 130311,
      "application machine learning ml": 6861,
      "natural language processing automatic": 76404,
      "deep neural network proposed": 28363,
      "word level sentence level": 133348,
      "measures accuracy precision recall": 67851,
      "accuracy precision recall f1score": 2242,
      "novel graphbased neural network": 80588,
      "submission wmt19 robustness task": 114743,
      "bias natural language inference": 13823,
      "language inference nli datasets": 59179,
      "language models experimental results": 59550,
      "approach bring significant improvement": 7401,
      "documentlevel neural machine translation": 33155,
      "machine translation paper describes": 66141,
      "submissions wmt19 news translation": 114761,
      "news translation shared task": 79387,
      "neural machine translation deep": 77987,
      "naver labs europes systems": 76680,
      "according automatic metrics bleu": 1970,
      "fields natural language processing": 44004,
      "relation detection knowledge base": 97969,
      "detection knowledge base question": 29980,
      "base question answering relation": 11478,
      "question answering relation detection": 95097,
      "stateoftheart code data available": 112612,
      "train neural network model": 122987,
      "evaluation natural language understanding": 39306,
      "training data multitask learning": 123498,
      "knowledge plays critical role": 58105,
      "significantly outperform competitive baselines": 108984,
      "models experimental results demonstrate": 73199,
      "wmt 2019 shared task": 132777,
      "shown correlate poorly human": 108459,
      "propose new hierarchical attention": 92853,
      "new hierarchical attention model": 78944,
      "pretrained language model lm": 90054,
      "language processing nlp task": 59918,
      "finetune pretrained language model": 44415,
      "finetuning pretrained language model": 44495,
      "neural language models supervised": 77955,
      "compare performance current stateoftheart": 19272,
      "training natural language generation": 123727,
      "experiments language modeling machine": 40971,
      "language understanding recently pretrained": 60263,
      "achieved stateoftheart results various": 2708,
      "play crucial role natural": 87691,
      "crucial role natural language": 25164,
      "role natural language processing": 103207,
      "natural language processing current": 76413,
      "source codes pretrained models": 110734,
      "neural machine translation trained": 78117,
      "bert bidirectional encoder representations": 13080,
      "bidirectional encoder representations transformers": 13916,
      "devlin et al 2019": 30472,
      "systems based transformer model": 116753,
      "conversational machine comprehension conversational": 23014,
      "machine comprehension conversational machine": 65748,
      "comprehension conversational machine comprehension": 20171,
      "graph neural network based": 48842,
      "compared existing stateoftheart methods": 19370,
      "proposed deep learning model": 93257,
      "amr abstract meaning representation": 5383,
      "representing meaning natural language": 100059,
      "techniques natural language processing": 119937,
      "nlpcc 2019 shared task": 79799,
      "task semisupervised domain adaptation": 118685,
      "named entity recognition present": 76103,
      "neural network rnn models": 78393,
      "machine reading comprehension machine": 65854,
      "reading comprehension machine reading": 95988,
      "comprehension machine reading comprehension": 20193,
      "pretrained language model bert": 90049,
      "evaluation machine translation systems": 39261,
      "heavily rely humanannotated data": 49658,
      "tackle training data bottleneck": 117314,
      "training data experimental results": 123461,
      "experimental results commonly used": 40571,
      "achieve competitive performance compared": 2447,
      "competitive performance compared previous": 19660,
      "significantly outperforms previous methods": 109024,
      "neural noisy channel modeling": 78618,
      "language models trained billions": 59670,
      "abductive natural language inference": 1472,
      "deep pretrained language models": 28393,
      "classification machine learning models": 17259,
      "language named entity recognition": 59701,
      "different languages like english": 31208,
      "data transfer learning approach": 26581,
      "outperforming previous best model": 82818,
      "model terms bleu entity": 72156,
      "terms bleu entity f1": 120285,
      "bleu entity f1 scores": 14282,
      "wmt19 news shared task": 132856,
      "statistical machine translation pbsmt": 113110,
      "dataset demonstrate effectiveness proposed": 26854,
      "demonstrate effectiveness proposed method": 28720,
      "models achieved stateoftheart performance": 72694,
      "performance code publicly available": 86213,
      "large pretrained language models": 61218,
      "deep contextualized word embeddings": 28219,
      "tagging lemmatization dependency parsing": 117399,
      "udpipe 20 bestperforming systems": 126540,
      "20 bestperforming systems conll": 491,
      "bestperforming systems conll 2018": 13480,
      "systems conll 2018 shared": 116804,
      "contextualized word embedding methods": 22574,
      "natural language processing computational": 76411,
      "monolingual data demonstrated helpful": 74568,
      "data demonstrated helpful improving": 25825,
      "demonstrated helpful improving translation": 28918,
      "helpful improving translation quality": 49799,
      "improve translation quality experimental": 52575,
      "translation quality experimental results": 125172,
      "quality experimental results chineseenglish": 94662,
      "chineseenglish germanenglish machine translation": 16855,
      "germanenglish machine translation tasks": 47936,
      "machine translation tasks proposed": 66263,
      "deep neural networks developed": 28369,
      "deep neural networks trained": 28383,
      "conduct experiments text generation": 20859,
      "experiments text generation tasks": 41181,
      "text generation tasks abstractive": 121013,
      "generation tasks abstractive summarization": 47667,
      "neural machine translation sequence": 78097,
      "englishgerman englishfrench translation tasks": 37378,
      "prediction auxiliary task multitask": 89035,
      "auxiliary task multitask learning": 10891,
      "machine translation present experiments": 66164,
      "automatic speech recognition especially": 10665,
      "stateoftheart models named entity": 112740,
      "models named entity recognition": 73601,
      "large amounts labeled data": 61014,
      "extensive experiments proposed method": 41920,
      "experiments proposed method performs": 41075,
      "proposed method performs better": 93393,
      "method performs better stateoftheart": 69050,
      "performs better stateoftheart baselines": 86988,
      "finetuning neural machine translation": 44484,
      "neural machine translation machine": 78021,
      "machine translation models trained": 66058,
      "using transductive data selection": 130308,
      "improve performance sentiment classification": 52485,
      "method achieves significant improvements": 68607,
      "inventors conceive better inventions": 56694,
      "liu et al 2019": 64843,
      "model named entity recognition": 71555,
      "named entity recognition propose": 76104,
      "named entity recognition based": 76055,
      "information named entity recognition": 54786,
      "lowresource natural language understanding": 65542,
      "fundamental problem natural language": 46107,
      "modelagnostic metalearning algorithm maml": 72348,
      "models outperform strong baselines": 73687,
      "using pointwise mutual information": 130026,
      "datasets demonstrate proposed model": 27413,
      "vector space models word": 131376,
      "space models word meaning": 111028,
      "syntax neural language models": 116547,
      "neural language models recurrent": 77952,
      "language models recurrent neural": 59652,
      "models recurrent neural networks": 73895,
      "empirical results benchmark datasets": 36179,
      "results benchmark datasets demonstrate": 101542,
      "crosslingual masked language model": 24975,
      "masked language model cmlm": 67289,
      "publicly available annotated datasets": 94291,
      "language understanding pretrained language": 60254,
      "understanding pretrained language models": 126925,
      "pretrained language models achieved": 90062,
      "language models achieved great": 59497,
      "various natural language understanding": 131147,
      "tasks including named entity": 119182,
      "natural language inference xnli": 76358,
      "chain conditional random fields": 15973,
      "representational similarity analysis rsa": 99480,
      "massively multilingual neural machine": 67344,
      "machine translation recently proposed": 66195,
      "languages english single model": 60548,
      "classification sequence labeling tasks": 17399,
      "orthogonal transformations embedding space": 82590,
      "contextaware neural machine translation": 22344,
      "propose new evaluation metric": 92850,
      "model using generative adversarial": 72279,
      "improvements downstream nlp tasks": 52845,
      "stateoftheart pretrained language models": 112870,
      "currently available different languages": 25399,
      "simultaneous translation simultaneous translation": 109668,
      "machine translation task proposed": 66260,
      "natural language processing based": 76405,
      "establish new state art": 38587,
      "new state art datasets": 79150,
      "jointly learning align translate": 57355,
      "state art machine translation": 112445,
      "art machine translation mt": 9057,
      "machine translation model training": 66048,
      "grammatical error correction gec": 48698,
      "latest developments natural language": 61654,
      "developments natural language processing": 30445,
      "range natural language understanding": 95588,
      "knowledgebased question answering kbqa": 58256,
      "require lots training data": 100173,
      "selfattention neural machine translation": 104899,
      "machine translation models rely": 66057,
      "essential natural language understanding": 38561,
      "language understanding tasks natural": 60282,
      "understanding tasks natural language": 126987,
      "tasks natural language inference": 119327,
      "natural language inference machine": 76330,
      "method consistently improves performance": 68730,
      "entity recognition ner tasks": 38051,
      "20 report stateoftheart results": 528,
      "information extraction tasks named": 54607,
      "extraction tasks named entity": 42512,
      "named entity recognition relation": 76108,
      "entity recognition relation extraction": 38067,
      "relation extraction event extraction": 97993,
      "framework achieves stateoftheart results": 45412,
      "entity recognition ner models": 38042,
      "recent work shown promising": 96586,
      "work shown promising results": 134804,
      "shown promising results crosslingual": 108516,
      "crosslingual transfer highresource languages": 25030,
      "transfer highresource languages lowresource": 124060,
      "highresource languages lowresource languages": 50428,
      "paper propose simple efficient": 84323,
      "shed light future research": 108152,
      "work pretrained language models": 134702,
      "pretrained language models bert": 90064,
      "language models bert devlin": 59515,
      "models bert devlin et": 72834,
      "approach achieves stateoftheart results": 7325,
      "recent success transfer learning": 96542,
      "language modeling language models": 59439,
      "language models lms predominantly": 59594,
      "word representations contextual word": 133451,
      "representations contextual word representations": 99568,
      "language model neural machine": 59356,
      "language model experimental results": 59321,
      "text generation tasks machine": 121016,
      "tasks machine translation text": 119281,
      "machine translation text summarization": 66269,
      "trained using maximum likelihood": 123329,
      "using maximum likelihood estimation": 129867,
      "intent detection slot filling": 55904,
      "address issue paper propose": 3693,
      "issue paper propose novel": 57017,
      "dataset publicly available research": 27135,
      "publicly available research community": 94321,
      "scarcity labeled training data": 103809,
      "baselines achieve new stateoftheart": 12347,
      "task oriented dialogue systems": 118488,
      "dialogue state tracking dst": 30754,
      "art natural language processing": 9068,
      "model achieve sota results": 70540,
      "neural networks text classification": 78590,
      "graph neural networks gnn": 48847,
      "graph convolutional networks gcn": 48788,
      "language inference nli models": 59182,
      "embedding language models elmo": 35422,
      "improving natural language processing": 53124,
      "implicit discourse relation recognition": 52007,
      "conditional variational autoencoder cvae": 20794,
      "paper explore new approach": 83921,
      "disambiguation using deep learning": 31995,
      "deep learning approach automatically": 28248,
      "text named entity recognition": 121138,
      "language generation nlg tasks": 59088,
      "generation abstractive summarization model": 47290,
      "abstractive summarization model outperforms": 1815,
      "learn word concept embeddings": 62190,
      "experiments proposed model significantly": 41084,
      "model significantly outperforms strong": 72038,
      "deep neural network acoustic": 28355,
      "neural network acoustic model": 78239,
      "neural network cnn followed": 78284,
      "simple transfer learning method": 109536,
      "lowresource languages recent work": 65527,
      "morphological tagging named entity": 74745,
      "recently pretrained language models": 96732,
      "language models achieved remarkable": 59498,
      "models achieved remarkable success": 72689,
      "broad range natural language": 14678,
      "tasks including partofspeech tagging": 119188,
      "including partofspeech tagging named": 53348,
      "meaning representations natural language": 67683,
      "representations natural language utterances": 99771,
      "model semantic parsing datasets": 71966,
      "architecture achieves stateoftheart results": 8608,
      "training machine translation systems": 123698,
      "systems low resource language": 116985,
      "neural networks shown perform": 78576,
      "entity recognition ner speech": 38047,
      "recognition ner speech pos": 96940,
      "ner speech pos tagging": 77082,
      "random field crf output": 95481,
      "field crf output layer": 43946,
      "using finite state transducers": 129682,
      "universal decompositional semantics uds": 127282,
      "language models paper investigate": 59612,
      "propose neural machine translation": 92820,
      "approach obtains promising results": 7748,
      "achieved various natural language": 2723,
      "language processing tasks using": 59971,
      "models automatic speech recognition": 72793,
      "speech recognition asr task": 111743,
      "substantially improve performance compared": 114894,
      "20 relative error reduction": 522,
      "model outperforms existing models": 71650,
      "spoken language understanding using": 112009,
      "adversarial learning domain generalization": 4133,
      "underlying question study adversarial": 126695,
      "question study adversarial learning": 95223,
      "study adversarial learning used": 114309,
      "train models higher level": 122962,
      "models higher level abstraction": 73329,
      "higher level abstraction order": 50188,
      "level abstraction order increase": 63414,
      "abstraction order increase robustness": 1793,
      "order increase robustness lexical": 82346,
      "increase robustness lexical stylistic": 53615,
      "automatic speech recognition errors": 10664,
      "strategy evaluated french corpus": 113511,
      "evaluated french corpus encyclopedic": 38978,
      "adversarial learning increases models": 4136,
      "learning increases models generalization": 62648,
      "increases models generalization capabilities": 53648,
      "parallel data language pairs": 84640,
      "propose new data augmentation": 92841,
      "datasets natural language inference": 27592,
      "natural language inference using": 76357,
      "natural language generation natural": 76306,
      "language generation natural language": 59078,
      "method achieve competitive results": 68586,
      "natural language processing recent": 76478,
      "progress natural language processing": 92164,
      "taskoriented dialogue systems work": 118895,
      "existing methods best accuracy": 40178,
      "transformer network vaswani et": 124357,
      "network vaswani et al": 77472,
      "data paper present novel": 26214,
      "transfer learning natural language": 124114,
      "language processing paper present": 59929,
      "achieve stateoftheart performance natural": 2561,
      "stateoftheart performance natural language": 112834,
      "language processing tasks text": 59970,
      "text classification machine translation": 120737,
      "obtaining new stateoftheart results": 81446,
      "new stateoftheart results tasks": 79173,
      "evaluation germanenglish machine translation": 39225,
      "107 phenomena organized 14": 197,
      "phenomena organized 14 categories": 87241,
      "outperforms previous stateoftheart methods": 82948,
      "work liu et al": 134619,
      "using neural networks model": 129958,
      "model achieves performance comparable": 70581,
      "representations experimental results approach": 99646,
      "approach achieves new stateoftheart": 7310,
      "framework named entity recognition": 45622,
      "named entity recognition task": 76117,
      "paper propose unified framework": 84338,
      "sequence labeling problem propose": 106978,
      "formulate machine reading comprehension": 45275,
      "machine reading comprehension mrc": 65858,
      "reading comprehension mrc task": 95995,
      "question person mentioned text": 95200,
      "additionally query encodes informative": 3629,
      "query encodes informative prior": 94958,
      "encodes informative prior knowledge": 36672,
      "informative prior knowledge strategy": 55148,
      "prior knowledge strategy facilitates": 90713,
      "knowledge strategy facilitates process": 58189,
      "strategy facilitates process entity": 113516,
      "facilitates process entity extraction": 42805,
      "process entity extraction leading": 91475,
      "entity extraction leading better": 37936,
      "extraction leading better performances": 42372,
      "datasets experimental results demonstrate": 27469,
      "downstream applications machine translation": 33991,
      "code data publicly available": 18087,
      "pretrained word embeddings characterlevel": 90217,
      "word embeddings characterlevel word": 133064,
      "uses deep neural network": 129221,
      "evaluation automatic human shows": 39127,
      "shows significant improvements existing": 108627,
      "sequence generation tasks machine": 106953,
      "image captioning machine translation": 51769,
      "paper propose new metric": 84281,
      "metrics like bleu meteor": 69981,
      "deep learning models semantics": 28294,
      "sentence source language text": 106085,
      "improve automatic evaluation metrics": 52341,
      "machine translation models generate": 66052,
      "models generate target words": 73288,
      "problem paper propose novel": 91159,
      "compared stateoftheart transformer model": 19465,
      "nonautoregressive machine translation nonautoregressive": 80157,
      "improving grammatical error correction": 53102,
      "statistical machine translation model": 113107,
      "data experimental results demonstrate": 25923,
      "pretrained language models used": 90106,
      "paper propose novel semantic": 84308,
      "downstream tasks text classification": 34058,
      "text classification natural language": 120748,
      "classification tasks natural language": 17450,
      "robustness neural machine translation": 103114,
      "neural machine translation translating": 78121,
      "machine translation nmt paper": 66112,
      "adaptation neural machine translation": 3240,
      "success neural machine translation": 115107,
      "simultaneous neural machine translation": 109661,
      "neural machine translation snmt": 78104,
      "neural machine translation translate": 78120,
      "neural language generation models": 77936,
      "models deep neural networks": 73026,
      "method natural language generation": 68983,
      "progress pretrained language models": 92175,
      "pretrained language models led": 90082,
      "methods large margin addition": 69578,
      "supervised relation classification rc": 115828,
      "syntax semantic role labeling": 116558,
      "aspect based sentiment analysis": 9318,
      "language model pretrained language": 59376,
      "model pretrained language models": 71775,
      "ubiquitous natural language processing": 126512,
      "partofspeech tagging dependency parsing": 85552,
      "dependency parsing named entity": 29185,
      "parsing named entity recognition": 85168,
      "nmt systems language pairs": 79977,
      "distant languages pairs like": 32455,
      "workshop asian translation wat": 135001,
      "entity recognition ner existing": 38037,
      "target language paper propose": 117646,
      "outperforms existing stateoftheart methods": 82900,
      "focuses applying endtoend dialog": 44889,
      "applying endtoend dialog technologies": 7242,
      "audio visual sceneaware dialog": 10245,
      "paper describes task definition": 83853,
      "describes task definition provided": 29440,
      "task definition provided datasets": 118055,
      "evaluation setup track summarize": 39391,
      "setup track summarize results": 107861,
      "track summarize results submitted": 122738,
      "summarize results submitted systems": 115615,
      "results submitted systems highlight": 102222,
      "submitted systems highlight overall": 114777,
      "systems highlight overall trends": 116933,
      "highlight overall trends stateoftheart": 50269,
      "overall trends stateoftheart technologies": 83267,
      "trends stateoftheart technologies tasks": 125757,
      "reading comprehension reading comprehension": 96010,
      "guide decoder generate coherent": 49233,
      "stateoftheart performance human evaluation": 112825,
      "compared distractors generated baselines": 19361,
      "difficulties natural language processing": 31685,
      "achieve high performance terms": 2477,
      "recurrent neural network conduct": 97203,
      "neural network conduct extensive": 78293,
      "proposed framework outperforms stateoftheart": 93294,
      "framework outperforms stateoftheart methods": 45644,
      "outperforms stateoftheart methods large": 83007,
      "stateoftheart methods large margin": 112711,
      "task experimental results indicate": 118174,
      "experimental results indicate model": 40624,
      "models automatic human evaluations": 72791,
      "using support vector machine": 130247,
      "support vector machine algorithm": 116001,
      "recent neural language models": 96478,
      "features downstream applications summarization": 43469,
      "significantly outperforms strong baselines": 109040,
      "enabling neural machine translation": 36416,
      "demonstrate superiority proposed model": 28885,
      "using connectionist temporal classification": 129476,
      "neural machine translation method": 78023,
      "connectionist temporal classification ctc": 21143,
      "translation quality language pairs": 125177,
      "language modeling language modeling": 59438,
      "representation learning natural language": 99301,
      "sentiment analysis emotion analysis": 106590,
      "word representations learned large": 133459,
      "continuous bag words cbow": 22614,
      "leads significant improvements task": 61956,
      "neural sequencetosequence models successfully": 78680,
      "neural machine translation pretraining": 78078,
      "achieved great success natural": 2633,
      "great success natural language": 49030,
      "english german german english": 37153,
      "machine translation tasks model": 66262,
      "translation tasks model outperforms": 125357,
      "tasks model outperforms strong": 119301,
      "neural machine translation review": 78096,
      "field machine translation mt": 43966,
      "machine translation mt automatic": 66064,
      "training taskoriented dialogue systems": 123907,
      "approaches achieves stateoftheart results": 8045,
      "anger disgust fear surprise": 6108,
      "automatic evaluation dialogue systems": 10527,
      "word embeddings substantially successful": 133217,
      "embeddings substantially successful capturing": 35961,
      "substantially successful capturing semantic": 114915,
      "successful capturing semantic relations": 115155,
      "model achieves stateoftheart result": 70604,
      "order paper propose novel": 82381,
      "deep learning models relation": 28292,
      "learning models relation extraction": 62790,
      "models relation extraction task": 73908,
      "representations yielded significant improvements": 100003,
      "yielded significant improvements nlp": 135369,
      "significant improvements nlp tasks": 108798,
      "pretrained language representation models": 90109,
      "replacing static word embeddings": 98951,
      "static word embeddings contextualized": 113069,
      "word embeddings contextualized word": 133070,
      "embeddings contextualized word representations": 35616,
      "current state future directions": 25345,
      "significantly outperforms current stateoftheart": 109011,
      "pretrained language models shown": 90101,
      "dialogue systems attracted attention": 30764,
      "language model natural language": 59354,
      "random majority class baselines": 95503,
      "lack annotated data languages": 58680,
      "experiments benchmark datasets approach": 40795,
      "paper propose general approach": 84246,
      "results sentiment analysis language": 102157,
      "different deep neural network": 31085,
      "artificial neural network ann": 9261,
      "methods natural language processing": 69633,
      "natural language processing word": 76513,
      "proposed model performs better": 93472,
      "employ support vector machines": 36286,
      "word embeddings represent words": 133200,
      "architectures bert xlnet roberta": 8785,
      "negation detection scope resolution": 76913,
      "long shortterm memory architecture": 65103,
      "high quality entity representations": 50116,
      "sequencetosequence pretraining paper presents": 107197,
      "summarization question generation tasks": 115558,
      "new stateoftheart results datasets": 79168,
      "neural networks machine translation": 78512,
      "gained attention recent years": 46359,
      "single large neural network": 109752,
      "neural network attention mechanism": 78253,
      "low resource language like": 65387,
      "based expectation maximization em": 11698,
      "models statistical machine translation": 74100,
      "information electronic health records": 54518,
      "domains limited training data": 33808,
      "models available research community": 72799,
      "paper propose new approach": 84274,
      "radford et al 2019": 95436,
      "given small set seed": 48129,
      "word order source language": 133382,
      "sequence labeling tasks experimental": 106986,
      "labeling tasks experimental results": 58549,
      "named entity recognition tasks": 76119,
      "convolutional neural networks text": 23210,
      "entity recognition named entity": 38026,
      "bidirectional encoder representation transformersbert": 13913,
      "datasets experimental results model": 27472,
      "significantly improving stateoftheart methods": 108966,
      "language models bidirectional encoder": 59522,
      "bidirectional encoder representation transformers": 13911,
      "encoder representation transformers bert": 36564,
      "experimental results model able": 40646,
      "maximum mutual information mmi": 67545,
      "appropriate responses yielding substantive": 8433,
      "responses yielding substantive gains": 101306,
      "yielding substantive gains bleu": 135392,
      "substantive gains bleu scores": 114922,
      "neural machine translation joint": 78011,
      "recent neural machine translation": 96480,
      "pretrained language models pretrained": 90095,
      "downstream natural language processing": 34007,
      "using publicly available datasets": 130068,
      "model achieves superior performance": 70615,
      "models human language processing": 73343,
      "generation present generative model": 47550,
      "messages readability instead long": 68514,
      "paper propose novel architecture": 84290,
      "conduct experiments widely used": 20863,
      "significant improvement competitive baseline": 108770,
      "information natural language processing": 54789,
      "transformer vaswani et al": 124396,
      "et al 2017 based": 38711,
      "outperforms stateoftheart baselines different": 83002,
      "shows promising results compared": 108614,
      "language processing nlp community": 59894,
      "network embedding distributional thesaurus": 77236,
      "natural language generation taskoriented": 76320,
      "crucial component taskoriented dialog": 25137,
      "component taskoriented dialog systems": 20005,
      "systems natural language generation": 117012,
      "language generation nlg module": 59086,
      "neural machine translation existing": 78000,
      "machine translation existing neural": 65980,
      "translation existing neural machine": 124806,
      "propose simple effective method": 93057,
      "achieves significant improvements strong": 2861,
      "new stateoftheart results wide": 79176,
      "stateoftheart results wide range": 112934,
      "arabic natural language processing": 8535,
      "nlp tasks like sentiment": 79734,
      "tasks like sentiment analysis": 119262,
      "stateoftheart results nlp tasks": 112915,
      "learning neural dialogue generation": 62839,
      "language processing nlp perspective": 59912,
      "machine translation best knowledge": 65913,
      "domain adaptation domain adaptation": 33433,
      "recent advances deep learning": 96398,
      "advances deep learning led": 3982,
      "deep learning led significant": 28278,
      "7th dialog technology challenges": 1282,
      "introduce new task named": 56503,
      "large amounts training data": 61023,
      "segmentation words subword units": 104657,
      "models based deep neural": 72806,
      "proposed approach significantly improves": 93215,
      "approach significantly improves performance": 7903,
      "nouns verbs adjectives adverbs": 80448,
      "evaluated using intrinsic extrinsic": 39023,
      "machine learning support vector": 65827,
      "learning support vector machine": 63077,
      "support vector machine logistic": 116003,
      "vector machine logistic regression": 131314,
      "logistic regression naive bayes": 65040,
      "deep learning convolutional neural": 28265,
      "learning convolutional neural network": 62466,
      "convolutional neural network recurrent": 23193,
      "neural network recurrent neural": 78381,
      "network recurrent neural network": 77404,
      "neural networks paper explores": 78530,
      "paper presents endtoend neural": 84169,
      "paper proposes novel framework": 84365,
      "language models pretrained large": 59629,
      "models pretrained large corpora": 73784,
      "python natural language processing": 94483,
      "natural language processing toolkit": 76507,
      "using automatic human evaluation": 129368,
      "human evaluation experimental results": 50813,
      "unsupervised neural machine translation": 127682,
      "speech natural language processing": 111715,
      "paper present neural approach": 84116,
      "generation natural language generation": 47499,
      "language generation nlg models": 59085,
      "tasks dialogue response generation": 119059,
      "recurrent neural network rnnbased": 97220,
      "approaches computational language documentation": 8105,
      "leveraging pretrained language models": 63698,
      "pretrained language models model": 90087,
      "text encoding initiative guidelines": 120907,
      "high resource language pairs": 50130,
      "pairs low resource languages": 83580,
      "embeddings pretrained language models": 35878,
      "downstream nlp tasks including": 34016,
      "implemented python programming language": 51974,
      "text generation structured data": 121009,
      "judged automatic metrics human": 57429,
      "automatic metrics human evaluation": 10604,
      "large number natural language": 61182,
      "synthetic data neural machine": 116621,
      "neural machine translation compared": 77980,
      "synthetic parallel data noisy": 116639,
      "parallel data noisy generated": 84644,
      "data noisy generated imperfect": 26178,
      "datasets proposed approach outperforms": 27634,
      "approach outperforms strong baselines": 7780,
      "proposed model propose new": 93476,
      "demonstrate proposed approach significantly": 28835,
      "pretrained language model gpt2": 90053,
      "improve correlation human judgments": 52363,
      "paper propose method automatically": 84262,
      "neural network architecture novel": 78247,
      "datasets proposed approach yields": 27635,
      "finegrained named entity recognition": 44369,
      "understanding natural language inference": 126898,
      "tasks natural language understanding": 119333,
      "monolingual data neural machine": 74577,
      "language pairs experimental results": 59757,
      "pretrained transformer language models": 90196,
      "sennrich et al 2016": 105646,
      "previous work substantial margin": 90553,
      "transfer learning finetuning pretrained": 124094,
      "automated metrics human evaluation": 10459,
      "pretrained masked language models": 90125,
      "entity recognition ner fundamental": 38038,
      "sets new state art": 107688,
      "word embeddings pretrained large": 133188,
      "unlabeled data train neural": 127391,
      "pretrained contextualized word representations": 90015,
      "new task proposed model": 79206,
      "paper proposes novel approach": 84364,
      "evaluation results indicate proposed": 39370,
      "text generation methods tend": 120990,
      "dialogue state tracking dialogue": 30753,
      "mitigate data scarcity problem": 70362,
      "abstract meaning representations amrs": 1776,
      "language understanding generation existing": 60224,
      "language processing systems recent": 59949,
      "language models like bert": 59587,
      "model based transformer architecture": 70749,
      "model outperforms previously published": 71664,
      "results neural machine translation": 101985,
      "current stateoftheart neural machine": 25363,
      "neural machine translation architecture": 77970,
      "performance compared models trained": 86233,
      "knowledge graphbased dialogue generation": 57978,
      "metalearning knowledge graphbased dialogue": 68553,
      "knowledge graphbased dialogue systems": 57979,
      "model significantly outperforms baselines": 72029,
      "performance various downstream tasks": 86842,
      "experimental results sentiment analysis": 40696,
      "languages various language families": 60965,
      "distant supervised relation extraction": 32459,
      "solve problem paper propose": 110606,
      "augment existing relation extraction": 10258,
      "achieved considerable success natural": 2615,
      "considerable success natural language": 21262,
      "transformerbased neural machine translation": 124434,
      "neural machine translation unsupervised": 78126,
      "machine translation unsupervised neural": 66287,
      "translation unsupervised neural machine": 125409,
      "neural machine translation unmt": 78124,
      "machine translation unmt recently": 66285,
      "unmt recently achieved remarkable": 127471,
      "recently achieved remarkable results": 96653,
      "achieved remarkable results language": 2689,
      "performance lowresource language pairs": 86508,
      "holtzman et al 2019": 50607,
      "use pretrained language models": 128206,
      "pretrained language models elmo": 90074,
      "language models elmo bert": 59544,
      "recurrent neural networks trained": 97252,
      "temporal classification ctc loss": 120098,
      "train neural machine translation": 122981,
      "variational autoencoder vae based": 130915,
      "autoencoder vae based approaches": 10412,
      "experimental results widely used": 40721,
      "achieves significant improvements compared": 2859,
      "significant improvements compared strong": 108790,
      "improvements compared strong baselines": 52832,
      "tackle named entity recognition": 117297,
      "benchmark datasets demonstrate method": 12777,
      "datasets demonstrate method outperforms": 27408,
      "largescale pretrained language models": 61496,
      "linguistic quality generated text": 64539,
      "various machine learning tasks": 131128,
      "measured automatic human evaluation": 67835,
      "latent structure input sentence": 61615,
      "translation tasks demonstrate approach": 125351,
      "approach significantly consistently improves": 7899,
      "significantly consistently improves translation": 108901,
      "extensive analyses confirm performance": 41854,
      "experiments text classification task": 41179,
      "experiments named entity recognition": 41032,
      "produce embeddings unseen words": 91888,
      "experiments benchmark datasets method": 40798,
      "outperforms stateoftheart text classification": 83020,
      "introduce new publicly available": 56497,
      "supervised machine learning models": 115794,
      "machine learning models automatically": 65810,
      "previous stateoftheart methods significantly": 90475,
      "encoder representations transformers bert": 36570,
      "performance pretrained language models": 86616,
      "pretrained language models paper": 90089,
      "chinese pretrained language models": 16808,
      "stateoftheart performances nlp tasks": 112863,
      "findings help future research": 44296,
      "stateoftheart joint goal accuracy": 112680,
      "data paper propose novel": 26219,
      "paper propose novel data": 84291,
      "existing data augmentation methods": 40101,
      "existing automatic evaluation metrics": 40073,
      "opendomain dialogue response generation": 81968,
      "small amounts indomain data": 110131,
      "data order magnitude larger": 26192,
      "data error analysis results": 25903,
      "using integer linear programming": 129763,
      "parallel corpora language pairs": 84603,
      "particularly distant language pairs": 85480,
      "machine translation mt models": 66067,
      "automatic generation parallel data": 10547,
      "generation parallel data iterative": 47534,
      "parallel data iterative backtranslation": 84638,
      "model outperforms previous systems": 71661,
      "dialogue systems natural language": 30778,
      "experiments demonstrate proposed approach": 40890,
      "extensive experiments demonstrate effectiveness": 41906,
      "transferability outperforming prior best": 124217,
      "outperforming prior best model": 82825,
      "data train semantic parser": 26568,
      "task multitask learning framework": 118427,
      "multitask learning framework different": 75839,
      "virtual assistants google assistant": 131744,
      "amazon alexa apple siri": 5266,
      "large number services apis": 61189,
      "performance subjectverb agreement prediction": 86767,
      "generative question answering genqa": 47757,
      "metric significantly higher correlation": 69903,
      "significantly higher correlation human": 108921,
      "higher correlation human judgments": 50173,
      "deep neural networks task": 28382,
      "problem paper propose new": 91158,
      "entity linking task identifying": 37964,
      "unstructured text existing methods": 127585,
      "text existing methods adopt": 120929,
      "code datasets publicly available": 18096,
      "language understanding recent years": 60261,
      "named entity recognition questionanswering": 76105,
      "introduce new sentiment analysis": 56501,
      "pretrained language model pretrained": 90055,
      "pretrained language models plms": 90092,
      "outperforms previous state art": 82945,
      "multilingual named entity recognition": 75301,
      "namedentity recognition ner model": 76139,
      "models trained monolingual data": 74216,
      "language processing tasks given": 59956,
      "measuring progress natural language": 67925,
      "progress natural language understanding": 92165,
      "bilingual evaluation understudy bleu": 14039,
      "sentences paper introduce new": 106426,
      "prior work controllable text": 90745,
      "work controllable text generation": 134443,
      "named entity linking nel": 76049,
      "proposed approach outperforms previous": 93207,
      "neural language models lms": 77949,
      "language models lms trained": 59596,
      "neural machine translation standard": 78106,
      "syntactic generalization neural language": 116412,
      "generalization neural language models": 46786,
      "neural language models stateoftheart": 77954,
      "information retrieval machine reading": 54938,
      "facts expressed natural language": 42911,
      "train deep neural network": 122922,
      "response generation neural conversation": 101207,
      "generation neural conversation models": 47506,
      "empirical results model outperforms": 36188,
      "results model outperforms previous": 101943,
      "model outperforms previous stateoftheart": 71657,
      "extensive experiments realworld datasets": 41929,
      "models outperform existing methods": 73678,
      "challenge natural language inference": 16061,
      "inference nli task determining": 54182,
      "achieves best performance overall": 2740,
      "training data text classification": 123552,
      "word embeddings map words": 133159,
      "et al 2012 paper": 38695,
      "neural networks rnns long": 78564,
      "largescale multidocument summarization dataset": 61468,
      "language processing nlp including": 59901,
      "explicitly model internal structure": 41380,
      "terms automatic human evaluation": 120273,
      "automatic human evaluation metrics": 10553,
      "problem sequence labelling task": 91221,
      "learning named entity recognition": 62823,
      "introduce new data set": 56479,
      "universal dependencies shared task": 127292,
      "shared task iwpt 2020": 108058,
      "language representation models bidirectional": 60044,
      "representation models bidirectional encoder": 99336,
      "models bidirectional encoder representations": 72852,
      "automatic speech recognition systems": 10668,
      "build powerful language models": 14800,
      "sentiment analysis text classification": 106671,
      "generating natural language adversarial": 47235,
      "named entity recognition french": 76068,
      "classification opendomain conversational agents": 17315,
      "amounts labeled training data": 5345,
      "collected amazon alexa prize": 18398,
      "stateoftheart deep learning methods": 112634,
      "possible future research directions": 88406,
      "processing tasks paper present": 91823,
      "using naive bayes classifier": 129929,
      "neural language models human": 77944,
      "predict human reading behavior": 88893,
      "modelling automatic speech recognition": 72599,
      "automatic speech recognition spoken": 10667,
      "work transfer learning methods": 134860,
      "functional distributional semantics functional": 46058,
      "distributional semantics functional distributional": 32727,
      "semantics functional distributional semantics": 105420,
      "functional distributional semantics provides": 46060,
      "rational speech acts framework": 95835,
      "entity recognition coreference resolution": 38008,
      "models widely used natural": 74341,
      "tasks machine translation question": 119279,
      "machine translation question answering": 66183,
      "machine translation mt shown": 66070,
      "contextual embeddings multilingual bert": 22460,
      "classification using support vector": 17494,
      "past work relation extraction": 85655,
      "approaches neural machine translation": 8255,
      "neural machine translation mt": 78035,
      "choosing transfer languages crosslingual": 16935,
      "employ graph neural networks": 36264,
      "stateoftheart methods automatic human": 112706,
      "methods automatic human evaluations": 69330,
      "modern standard arabic colloquial": 74417,
      "experiments analysis demonstrate effectiveness": 40767,
      "tasks word sense disambiguation": 119607,
      "unlabeled data target domain": 127389,
      "ambiguous word particular context": 5300,
      "using neural language models": 129949,
      "using distant supervision sentiment": 129592,
      "knowledge base previous studies": 57780,
      "results showed proposed approach": 102174,
      "machine learning models support": 65811,
      "learning models support vector": 62793,
      "text editing propose novel": 120898,
      "new neural network architecture": 79057,
      "model experimental results datasets": 71129,
      "semantic textual similarity dataset": 105330,
      "different sampling strategies used": 31397,
      "open domain semantic parsing": 81899,
      "present natural language processing": 89569,
      "endtoend method based supervised": 36923,
      "method based supervised learning": 68668,
      "accuracy endtoend method point": 2147,
      "endtoend method point corresponding": 36926,
      "method point corresponding method": 69055,
      "point corresponding method applied": 87799,
      "results proposed approach able": 102063,
      "approach coreference resolution task": 7464,
      "text classification tasks sentiment": 120778,
      "advantages convolutional neural networks": 4074,
      "validity domain text explored": 130751,
      "domain text explored paper": 33676,
      "paper propose novel hybrid": 84298,
      "propose novel hybrid architecture": 92930,
      "bidirectional gated recurrent units": 13924,
      "gated recurrent units bigru": 46521,
      "conducted extensive experiments benchmark": 20926,
      "help external linguistic knowledge": 49722,
      "analysis social media data": 5861,
      "attention machine learning community": 9875,
      "experiments method significantly improves": 41006,
      "method significantly improves performance": 69138,
      "leverage unlabeled data target": 63632,
      "pretrained models bert roberta": 90136,
      "propose new pretraining task": 92880,
      "dataset machine reading comprehension": 27007,
      "public datasets experimental results": 94253,
      "novel approach sentiment analysis": 80488,
      "data text audio video": 26558,
      "benchmark dataset experimental results": 12757,
      "effectiveness proposed approach compared": 34927,
      "meaning natural language text": 67652,
      "approach offensive language identification": 7752,
      "offensive language identification social": 81548,
      "document classification models using": 32964,
      "task offensive language identification": 118476,
      "development set test set": 30424,
      "model significantly improves accuracy": 72022,
      "graded word similarity context": 48592,
      "existing semantically annotated datasets": 40280,
      "social media paper approach": 110400,
      "multilingual offensive language identification": 75320,
      "language identification shared task": 59141,
      "utilizing pretrained language models": 130576,
      "pretrained language models downstream": 90072,
      "language models downstream tasks": 59542,
      "social media platforms facebook": 110404,
      "convolutional neural network models": 23191,
      "research excellence framework ref": 100494,
      "excellence framework ref 2014": 39913,
      "document sentence token level": 33077,
      "automated text processing systems": 10473,
      "graph convolutional network relation": 48783,
      "convolutional network relation extraction": 23166,
      "process work propose novel": 91592,
      "work propose novel model": 134738,
      "graph convolutional network sgcn": 48785,
      "machine translation written text": 66307,
      "language models paper presents": 59613,
      "complexity inefficiency insideoutside algorithm": 19919,
      "parser achieves new stateoftheart": 84914,
      "data set machine translation": 26411,
      "indian language machine translation": 53820,
      "segmentation pos tagging named": 104620,
      "extensive experimental results proposed": 41887,
      "experimental results proposed framework": 40673,
      "proposed framework significantly outperforms": 93299,
      "multilabel text classification mltc": 75161,
      "explicitly modeling mutual interaction": 41385,
      "modeling mutual interaction relation": 72487,
      "experimental results public datasets": 40686,
      "dialog act recognition task": 30548,
      "bidirectional encoder representation transformer": 13909,
      "encoder representation transformer bert": 36561,
      "semeval 2020 task commonsense": 105512,
      "2020 task commonsense validation": 735,
      "task commonsense validation explanation": 117980,
      "language models paper describes": 59611,
      "semantic role labeling srl": 105257,
      "data natural language processing": 26156,
      "models significantly better performance": 74040,
      "performs substantially better prior": 87032,
      "substantially better prior work": 114884,
      "sentiment analysis paper propose": 106626,
      "paper propose variational approach": 84343,
      "provided domain experts use": 93965,
      "domain experts use targetopinion": 33530,
      "experts use targetopinion word": 41251,
      "use targetopinion word pairs": 128316,
      "targetopinion word pairs supervision": 117785,
      "word pairs extracted using": 133394,
      "pairs extracted using dependency": 83546,
      "extracted using dependency parsers": 42192,
      "using dependency parsers simple": 129557,
      "dependency parsers simple rules": 29167,
      "parsers simple rules objective": 85046,
      "predict opinion word given": 88912,
      "opinion word given target": 82109,
      "word given target word": 133302,
      "given target word ultimate": 48148,
      "target word ultimate goal": 117750,
      "word ultimate goal learn": 133615,
      "ultimate goal learn sentiment": 126556,
      "introducing latent variable sentiment": 56643,
      "latent variable sentiment polarity": 61633,
      "variable sentiment polarity objective": 130849,
      "sentiment polarity objective function": 106776,
      "polarity objective function inject": 87919,
      "objective function inject sentiment": 81082,
      "lower bound learn sentiment": 65424,
      "classifier optimizing lower bound": 17566,
      "experiment results method outperform": 40496,
      "supervised method hundreds labels": 115801,
      "method hundreds labels aspect": 68876,
      "sentiment classification sentiment classification": 106713,
      "machine translation mt translate": 66074,
      "chinese clinical named entity": 16747,
      "clinical named entity recognition": 17775,
      "named entity recognition clinical": 76059,
      "entity recognition clinical named": 38003,
      "recognition clinical named entity": 96836,
      "named entity recognition cner": 76061,
      "entity recognition cner aims": 38006,
      "computational results ccks2017 task": 20428,
      "results ccks2017 task benchmark": 101565,
      "ccks2017 task benchmark dataset": 15866,
      "multitask learning transfer learning": 75872,
      "problem sequence labeling task": 91219,
      "developed team semeval2020 task": 30314,
      "approach achieves good performance": 7306,
      "results establish new stateoftheart": 101766,
      "solving arithmetic word problems": 110641,
      "based pretrained language models": 11947,
      "pretrained language models specifically": 90102,
      "model achieve better local": 70535,
      "achieve better local optimum": 2430,
      "proper nouns named entities": 92422,
      "methods yield significant improvements": 69859,
      "stateoftheart performance ace 2005": 112805,
      "method neural semantic parsing": 68987,
      "correlate poorly human judgments": 24199,
      "correlates better human judgments": 24216,
      "entities knowledge graph kg": 37807,
      "machine translation nmt work": 66127,
      "question answering recent work": 95093,
      "natural language processing chinese": 76407,
      "chinese word segmentation cws": 16836,
      "partofspeech pos tagging named": 85537,
      "pos tagging named entity": 88233,
      "entity recognition ner dependency": 38033,
      "recognition ner dependency parsing": 96925,
      "highresource languages particular english": 50430,
      "transformerbased language model pretrained": 124414,
      "natural language inference dataset": 76327,
      "dataset natural language inference": 27051,
      "languages paper present new": 60779,
      "paper present new dataset": 84121,
      "training multilingual machine translation": 123721,
      "tagging natural language understanding": 117427,
      "dialogue systems paper propose": 30783,
      "processing nlp tasks despite": 91756,
      "nlp tasks despite success": 79714,
      "performance model trained domain": 86537,
      "markov decision process pomdp": 67258,
      "reinforcement learning approaches used": 97798,
      "deep reinforcement learning drl": 28413,
      "sentence compression sentence compression": 105801,
      "pretrained bidirectional encoder representations": 89996,
      "dataset code publicly available": 26790,
      "pretrained language model based": 90048,
      "models achieves stateoftheart results": 72705,
      "chinese word segmentation partofspeech": 16838,
      "word segmentation partofspeech tagging": 133501,
      "segmentation partofspeech tagging named": 104615,
      "dependency parsing semantic parsing": 29198,
      "multitask model surpass singletask": 75879,
      "text generation paper propose": 120996,
      "model outperforms existing baselines": 71647,
      "applications question answering text": 6998,
      "question answering text summarization": 95116,
      "using external knowledge bases": 129660,
      "systems paper presents novel": 117047,
      "information extraction paper presents": 54589,
      "subtasks named entity recognition": 114973,
      "neural networks representation learning": 78557,
      "method outperforms competitive baselines": 69012,
      "graph graph neural networks": 48818,
      "stateoftheart models natural language": 112743,
      "come high computational cost": 18766,
      "high computational cost work": 50046,
      "computational cost work explore": 20371,
      "cost work explore simple": 24379,
      "improves performance strong baseline": 53020,
      "performance strong baseline negligible": 86758,
      "provide open source implementations": 93887,
      "constituent parsing sequence labeling": 21547,
      "different neural network architectures": 31296,
      "shared task 2018 dataset": 108018,
      "languages english german spanish": 60540,
      "knowledge natural language inference": 58077,
      "strong baselines bert roberta": 113644,
      "reasoning process neural network": 96297,
      "multiturn response selection task": 75926,
      "effective unsupervised domain adaptation": 34769,
      "language models recent work": 59647,
      "models recent work shown": 73878,
      "masked language models mlms": 67297,
      "named entity recognition method": 76073,
      "stateofthe art deep learning": 112560,
      "achieving state art results": 2983,
      "use online social networks": 128183,
      "languages paper propose new": 60784,
      "generation pretrained language models": 47556,
      "work explore challenging task": 134508,
      "task pretrained language models": 118555,
      "various natural language tasks": 131146,
      "datasets indicate model significantly": 27523,
      "indicate model significantly outperform": 53843,
      "nlp tasks question answering": 79757,
      "machine translation models modern": 66054,
      "maximum posteriori map decoding": 67549,
      "aspectcategory sentiment analysis acsa": 9363,
      "sentiment analysis acsa aims": 106563,
      "analysis acsa aims predict": 5487,
      "acsa aims predict sentiment": 3058,
      "aims predict sentiment polarities": 4557,
      "detect sentiment particular aspect": 29819,
      "suboptimal performance paper propose": 114792,
      "results public datasets demonstrate": 102098,
      "public datasets demonstrate effectiveness": 94250,
      "models capture linguistic information": 72880,
      "natural language processing involve": 76428,
      "labeling semantic role labeling": 58538,
      "neural sequencetosequence seq2seq models": 78682,
      "highly correlate human judgments": 50308,
      "power neural machine translation": 88643,
      "paper provide overview different": 84380,
      "pretrained models like bert": 90143,
      "hinton et al 2015": 50508,
      "demonstrate efficacy proposed approach": 28731,
      "analysis open information extraction": 5728,
      "neural open information extraction": 78623,
      "open information extraction openie": 81909,
      "comes significant computational cost": 18783,
      "establishing new state art": 38618,
      "new state art task": 79153,
      "machine translation existing approaches": 65979,
      "question answering training data": 95120,
      "question answering qa relies": 95083,
      "transformer models natural language": 124350,
      "question answering instead using": 95055,
      "extensive experiments datasets different": 41904,
      "accuracy increasing model capacity": 2188,
      "success pretrained language models": 115115,
      "prior work mainly focused": 90756,
      "address issue paper proposes": 3694,
      "datasets demonstrate effectiveness proposed": 27401,
      "neural architecture search nas": 77836,
      "language modeling natural language": 59446,
      "modeling natural language inference": 72490,
      "recent advances neural language": 96413,
      "results models achieve competitive": 101955,
      "models achieve competitive results": 72664,
      "recent work natural language": 96578,
      "language processing nlp focused": 59900,
      "perform indepth error analysis": 86013,
      "task automatic human evaluation": 117912,
      "pretrained multilingual bert mbert": 90161,
      "proposed methods achieve stateoftheart": 93412,
      "wmt biomedical translation task": 132781,
      "language models lms shown": 59595,
      "pretrained language models capture": 90068,
      "highquality neural machine translation": 50402,
      "paper tackle problem using": 84467,
      "demonstrate effectiveness approach achieving": 28707,
      "semantic interpretation natural language": 105087,
      "generation pretrained language model": 47555,
      "learning approaches natural language": 62361,
      "approaches natural language processing": 8250,
      "natural language processing mainly": 76435,
      "natural language paper propose": 76380,
      "experiments demonstrate approach achieve": 40878,
      "based publicly available pretrained": 11960,
      "available pretrained language models": 11078,
      "dataset demonstrate effectiveness approach": 26853,
      "demonstrate effectiveness approach zeroshot": 28708,
      "semantic role labeling english": 105248,
      "deep neural network based": 28359,
      "based question answering qa": 11966,
      "question answering qa models": 95080,
      "conduct extensive experiments using": 20877,
      "natural language understanding dialogue": 76557,
      "language understanding dialogue state": 60216,
      "understanding dialogue state tracking": 126829,
      "results significant improvements previous": 102181,
      "paper present new neural": 84123,
      "major tasks natural language": 66589,
      "language processing computational linguistics": 59856,
      "demonstrate approach outperforms previous": 28671,
      "research attention natural language": 100420,
      "attention natural language processing": 9942,
      "natural language processing existing": 76418,
      "verify effectiveness proposed method": 131584,
      "effectiveness proposed method conduct": 34933,
      "proposed method conduct extensive": 93356,
      "method conduct extensive experiments": 68725,
      "approach outperforms stateoftheart methods": 7778,
      "use transfer learning tl": 128341,
      "transfer learning tl techniques": 124139,
      "natural language understanding propose": 76577,
      "language understanding propose novel": 60258,
      "texttospeech automatic speech recognition": 121660,
      "social media previous work": 110416,
      "deep neural networks widely": 28386,
      "neural networks widely used": 78605,
      "methods neural response generation": 69640,
      "approach significantly improves strong": 7904,
      "models achieved outstanding performance": 72686,
      "adapted pretrained language model": 3289,
      "room improvement release datasets": 103265,
      "augmentation natural language understanding": 10290,
      "natural language understanding data": 76556,
      "natural language understanding intent": 76562,
      "language understanding intent detection": 60228,
      "understanding intent detection slot": 126865,
      "slot filling dialogue state": 110041,
      "filling dialogue state tracking": 44038,
      "benchmarks indicate model significantly": 12910,
      "automatic evaluation human judgment": 10530,
      "annotated corpus hindienglish codemixed": 6149,
      "simple data augmentation technique": 109396,
      "using largescale pretrained language": 129795,
      "linear contextfree rewriting systems": 64343,
      "new model improves performance": 79032,
      "meaning representation amr parsing": 67671,
      "experimental results benchmark datasets": 40563,
      "method outperforms previous methods": 69018,
      "outperforms previous methods large": 82937,
      "previous methods large margin": 90429,
      "available website research purposes": 11147,
      "naive bayes decision tree": 76016,
      "pretrained neural language models": 90173,
      "previous stateoftheart methods large": 90473,
      "significant improvements strong transformer": 108809,
      "improvements strong transformer baseline": 52924,
      "international phonetic alphabet ipa": 56186,
      "jointly models word segmentation": 57369,
      "dataset method achieves stateoftheart": 27022,
      "using pretrained language models": 130047,
      "generation paper propose novel": 47531,
      "research area natural language": 100413,
      "recent advances language modeling": 96406,
      "simple effective data augmentation": 109407,
      "language processing tasks inspired": 59962,
      "stateoftheart machine translation systems": 112696,
      "models experimental results suggest": 73201,
      "stateoftheart results wide variety": 112935,
      "language identification paper present": 59137,
      "natural language understanding largescale": 76564,
      "setting new state art": 107764,
      "code data models publicly": 18080,
      "publicly available unsupervised crosslingual": 94328,
      "multilingual pretrained language models": 75338,
      "underresourced languages article presents": 126729,
      "data difficult expensive obtain": 25849,
      "gaussian process preference learning": 46551,
      "process preference learning gppl": 91550,
      "language model evaluate approach": 59318,
      "model showing significant improvements": 72009,
      "linguistic codeswitching evaluation lince": 64440,
      "questionanswering natural language inference": 95255,
      "pretrained language models present": 90094,
      "language models present method": 59625,
      "previous best published result": 90396,
      "language models transfer learning": 59676,
      "largescale multilingual language models": 61477,
      "experimental results wmt14 englishgerman": 40725,
      "englishgerman wmt16 englishromanian datasets": 37396,
      "experimental results demonstrate multitask": 40597,
      "neural network based approaches": 78255,
      "network based approaches process": 77156,
      "based approaches process formation": 11518,
      "approaches process formation splitting": 8292,
      "morphological analysis sanskrit texts": 74682,
      "event argument extraction eae": 39498,
      "demonstrate benefits proposed model": 28681,
      "machine translation nmt approaches": 66093,
      "large amounts unlabeled data": 61025,
      "overall contextual importance scores": 83224,
      "using graph neural network": 129718,
      "penn treebank ptb chinese": 85863,
      "model builds recent work": 70794,
      "achieves competitive performance compared": 2771,
      "contextualized word embeddings bert": 22576,
      "large amounts highquality training": 61011,
      "amounts highquality training data": 5337,
      "results competitive state art": 101614,
      "model achieves better performance": 70560,
      "increasing attention research community": 53663,
      "representation learning extensive experiments": 99289,
      "benchmark datasets demonstrate efficacy": 12776,
      "simulated multiple reference training": 109641,
      "multiple reference training smrt": 75660,
      "et al 2020 use": 38739,
      "conversational question answering systems": 23023,
      "question answering vqa tasks": 95125,
      "significantly accelerates inference process": 108878,
      "improve translation quality strong": 52577,
      "pretrained models bert shown": 90137,
      "using pretrained language model": 130046,
      "processing tasks including text": 91817,
      "tasks including text classification": 119192,
      "models achieve high performance": 72667,
      "transformerbased neural language models": 124432,
      "models performance tasks require": 73736,
      "simple effective method incorporate": 109412,
      "cloze test dataset proposed": 17897,
      "method performs better existing": 69049,
      "paper describes contribution semeval": 83812,
      "amazon alexa prize 2018": 5268,
      "different perspectives experimental results": 31331,
      "experimental results language pairs": 40628,
      "number interesting findings help": 80901,
      "interesting findings help humans": 56075,
      "findings help humans better": 44298,
      "help humans better analyze": 49734,
      "humans better analyze understand": 51071,
      "better analyze understand improve": 13510,
      "models based observations propose": 72810,
      "based observations propose new": 11911,
      "data natural language generation": 26155,
      "generation nlg critical component": 47514,
      "machine translation nmt aims": 66092,
      "gated linear unit glu": 46510,
      "require large amounts annotated": 100156,
      "machine answer questions based": 65742,
      "answer questions based given": 6565,
      "questions based given context": 95280,
      "language inference nli fundamental": 59180,
      "training data different related": 123451,
      "network model joint partofspeech": 77330,
      "model joint partofspeech pos": 71381,
      "joint partofspeech pos tagging": 57308,
      "present multitask learning framework": 89566,
      "language modelling partofspeech tagging": 59491,
      "source code publicly available": 110724,
      "et al 2016 proposed": 38706,
      "masked language model pretraining": 67291,
      "learning task text classification": 63087,
      "language model pretraining proven": 59379,
      "model pretraining proven useful": 71781,
      "language understanding tasks paper": 60284,
      "method able improve performance": 68581,
      "learning approach sentiment analysis": 62350,
      "knowledge pretrained language models": 58117,
      "unsupervised domain adaptation pretrained": 127631,
      "language models proven effective": 59640,
      "significant improvement existing online": 108772,
      "recent years existing approaches": 96613,
      "paper propose method automatic": 84261,
      "language models plms like": 59622,
      "models plms like bert": 73751,
      "pretrained language models lm": 90085,
      "aspectbased sentiment analysis absa": 9357,
      "various downstream nlp tasks": 131087,
      "question answering recent works": 95094,
      "range datasets demonstrate effectiveness": 95555,
      "datasets model achieves significant": 27571,
      "model achieves significant improvement": 70590,
      "error propagation paper present": 38318,
      "paper propose task relation": 84333,
      "propose task relation classification": 93109,
      "dialogue knowledge transfer network": 30699,
      "wnut 2020 shared task1": 132883,
      "models pretrained large unannotated": 73785,
      "tasks glue benchmark approach": 119149,
      "architecture neural machine translation": 8709,
      "neural machine translation encoderdecoder": 77997,
      "widely used neural machine": 132581,
      "used neural machine translation": 128655,
      "address issue present novel": 3696,
      "language model predicts word": 59373,
      "machine translation mt reached": 66068,
      "learning distributed representations words": 62513,
      "word embeddings text classification": 133226,
      "representations based word embeddings": 99514,
      "training data propose use": 123518,
      "sentences plays important role": 106447,
      "language pairs model outperforms": 59767,
      "obtains new stateoftheart results": 81471,
      "domains tackle problem propose": 33870,
      "proposed method significantly improves": 93400,
      "dataset results proposed model": 27171,
      "model brings substantial improvements": 70788,
      "brings substantial improvements strong": 14655,
      "detection social media using": 30058,
      "fake news detection task": 43018,
      "transfer learning using multilingual": 124144,
      "dataset proposed model achieves": 27123,
      "blackbox nature neural models": 14254,
      "aims generate natural language": 4533,
      "context experimental results demonstrate": 22094,
      "experiments conducted realworld datasets": 40844,
      "extensive experiment results demonstrate": 41881,
      "short term memory network": 108253,
      "spoken language understanding pipeline": 111999,
      "require large number labeled": 100165,
      "simultaneously paper propose cointeractive": 109689,
      "information achieve stateoftheart performance": 54357,
      "contextualized word representations bert": 22581,
      "lowresource languages neural machine": 65522,
      "languages neural machine translation": 60764,
      "improvements neural machine translation": 52882,
      "relation extraction predict relation": 98010,
      "extraction predict relation type": 42439,
      "predict relation type entities": 88925,
      "relation type entities mentioned": 98081,
      "conditional masked language model": 20766,
      "language inference nli tasks": 59185,
      "learned representations propose simple": 62253,
      "nonautoregressive translation nat models": 80163,
      "demonstrate effectiveness universality proposed": 28725,
      "effectiveness universality proposed approach": 34973,
      "bleu points respectively source": 14311,
      "points respectively source code": 87874,
      "based supervised machine learning": 12091,
      "machine learning deep learning": 65793,
      "pretrained language models ptlms": 90099,
      "pretrained language models proven": 90098,
      "natural language nl explanations": 76377,
      "tasks question answering qa": 119429,
      "sequence sequence seq2seq models": 107066,
      "settings neural machine translation": 107826,
      "natural language processing aims": 76390,
      "pretraining masked language modeling": 90282,
      "single pass experimental results": 109782,
      "existing multilingual machine translation": 40217,
      "multilingual machine translation systems": 75277,
      "recent success language model": 96535,
      "success language model pretraining": 115085,
      "performance multilingual language models": 86548,
      "pretrained multilingual language models": 90164,
      "set typologically diverse languages": 107624,
      "models achieve impressive performance": 72669,
      "large monolingual parallel corpora": 61160,
      "outperforms existing crosslingual models": 82891,
      "delivers new stateoftheart results": 28626,
      "new stateoftheart results various": 79174,
      "stateoftheart results various crosslingual": 112930,
      "language understanding tasks text": 60285,
      "understanding tasks text classification": 126992,
      "text classification question answering": 120760,
      "advances pretrained language models": 4013,
      "human evaluation shows model": 50835,
      "conceptual cognitive annotation ucca": 20659,
      "cognitive annotation ucca abend": 18265,
      "annotation ucca abend rappoport": 6401,
      "ucca abend rappoport 2013": 126525,
      "wang et al 2020": 132009,
      "et al 2020 using": 38740,
      "training largescale language models": 123682,
      "covid19 fake news detection": 24564,
      "shared task covid19 fake": 108032,
      "task covid19 fake news": 118032,
      "fake news detection english": 43016,
      "natural language inference question": 76346,
      "language inference question answering": 59192,
      "covering text classification sequence": 24549,
      "text classification sequence labeling": 120768,
      "recognition ner tasks natural": 96945,
      "nlp tasks partofspeech tagging": 79750,
      "integral natural language processing": 55742,
      "pretrained language models task": 90103,
      "language processing nlp provides": 59915,
      "segmentation partofspeech tagging morphological": 104614,
      "models code publicly available": 72918,
      "relation classification relation classification": 97956,
      "relation classification rc task": 97954,
      "fundamental tasks information extraction": 46124,
      "relation information entity pairs": 98043,
      "extensive experiments conducted fewrel": 41902,
      "shared task paper present": 108079,
      "lowlevel features character ngrams": 65475,
      "adversarial training data augmentation": 4172,
      "large pretrained language model": 61217,
      "statistically significant improvement performance": 113186,
      "fake news detection paper": 43017,
      "natural language inference present": 76344,
      "relations address problem propose": 98108,
      "dialogue state tracking using": 30755,
      "neural network models propose": 78349,
      "proposed approach natural language": 93201,
      "language models based recurrent": 59510,
      "recurrent neural networks transformers": 97253,
      "entity recognition text classification": 38083,
      "high resource languages like": 50132,
      "resource languages like english": 100860,
      "experiments conducted text classification": 40848,
      "contextualized word embeddings derived": 22577,
      "dialog state tracking dst": 30590,
      "improves joint goal accuracy": 52990,
      "cnn model text classification": 18013,
      "modeling unstructured knowledge access": 72581,
      "language models paper propose": 59614,
      "pretrained language model results": 90058,
      "news articles paper presents": 79303,
      "new annotation tool designed": 78778,
      "latent dirichlet allocation topic": 61583,
      "neural network cnn conditional": 78281,
      "network cnn conditional random": 77191,
      "cnn conditional random field": 17998,
      "attention network based bidirectional": 9946,
      "nlp natural language understanding": 79652,
      "machine translation documentlevel machine": 65957,
      "translation documentlevel machine translation": 124765,
      "embedding spaces different languages": 35504,
      "parallel data target language": 84656,
      "datasets method outperforms existing": 27567,
      "chinese named entity recognition": 16790,
      "natural language process nlp": 76388,
      "representations transformers bert model": 99947,
      "using artificial neural networks": 129349,
      "question answering tasks including": 95113,
      "xlmr conneau et al": 135202,
      "conneau et al 2020": 21115,
      "transformerbased language models achieved": 124416,
      "language models achieved state": 59499,
      "models achieved state art": 72692,
      "achieved state art performance": 2701,
      "state art performance various": 112467,
      "code dataset publicly available": 18092,
      "longshort term memory lstm": 65189,
      "evaluate performance proposed method": 38888,
      "labeled data required training": 58441,
      "data required training new": 26347,
      "required training new domain": 100228,
      "method significantly outperforms strong": 69150,
      "significantly outperforms strong baseline": 109039,
      "performance natural language generation": 86558,
      "word sense disambiguation semantic": 133521,
      "language models recent progress": 59646,
      "large performance gains nlp": 61208,
      "popular pretrained language models": 88120,
      "natural language processing detect": 76416,
      "task using handcrafted features": 118833,
      "crucial machine translation systems": 25149,
      "models achieve comparable performance": 72662,
      "learning based model uses": 62392,
      "ninth dialog technology challenge": 79493,
      "finally conduct extensive experiments": 44161,
      "experiments public datasets experimental": 41094,
      "demonstrate proposed model outperforms": 28852,
      "generation models produce synthetic": 47487,
      "morphologically rich languages neural": 74772,
      "error rate wer reduction": 38336,
      "datasets automatic human evaluations": 27324,
      "deep learning based approach": 28254,
      "hindienglish code mixed tweets": 50482,
      "code mixed tweets using": 18113,
      "mixed tweets using bilingual": 70405,
      "tweets using bilingual word": 126071,
      "using bilingual word embeddings": 129413,
      "bilingual word embeddings derived": 14072,
      "word embeddings derived fasttext": 133079,
      "embeddings derived fasttext word2vec": 35643,
      "derived fasttext word2vec approaches": 29351,
      "deep learning models including": 28287,
      "learning models including cnns": 62782,
      "models including cnns lstms": 73380,
      "including cnns lstms bidirectional": 53273,
      "cnns lstms bidirectional lstms": 18040,
      "lstms bidirectional lstms attention": 65705,
      "curriculum learning sentiment analysis": 25425,
      "topic models latent dirichlet": 122552,
      "models latent dirichlet allocation": 73470,
      "outperforms existing state art": 82898,
      "answer sentence selection as2": 6577,
      "convolutional neural network model": 23190,
      "machine translation models learn": 66053,
      "tasks morphological probing pos": 119310,
      "morphological probing pos tagging": 74722,
      "probing pos tagging ner": 90902,
      "language modeling transfer learning": 59478,
      "language modeling mlm task": 59444,
      "translation nmt systems translate": 125060,
      "identifying protected health information": 51615,
      "protected health information phi": 93662,
      "adapt new unseen domains": 3189,
      "machine translation reinforcement learning": 66197,
      "specifically jointly train modules": 111563,
      "deep contextualized language models": 28216,
      "hope speech detection using": 50641,
      "using majority voting ensemble": 129855,
      "finetuning pretrained transformer models": 44501,
      "vietnamese natural language processing": 131686,
      "word word left right": 133660,
      "results proposed model achieves": 102080,
      "text summarization question answering": 121351,
      "information extraction sentiment analysis": 54602,
      "word embedding learning methods": 133017,
      "learning methods require large": 62745,
      "artificial intelligence ai machine": 9248,
      "intelligence ai machine learning": 55824,
      "ai machine learning ml": 4424,
      "make datasets pretrained models": 66652,
      "models trained data set": 74197,
      "brought great deal models": 14713,
      "deep learning models using": 28296,
      "encoder representations transformers achieved": 36569,
      "natural language processing model": 76437,
      "traditional machine learning models": 122832,
      "highly inflectional morphology paper": 50330,
      "language processing applications general": 59837,
      "power natural language processing": 88640,
      "progressive selfsupervised attention learning": 92193,
      "frequent words sentiment polarities": 45866,
      "propose progressive selfsupervised attention": 93005,
      "selfsupervised attention learning approach": 104936,
      "useful attention supervision information": 128863,
      "activemisleading influence correctincorrect prediction": 3128,
      "influence correctincorrect prediction instance": 54307,
      "augment conventional training objective": 10253,
      "conventional training objective regularization": 22909,
      "training objective regularization term": 123760,
      "extracted active context words": 42137,
      "approach yields better attention": 8024,
      "source code trained models": 110727,
      "aspectbased sentiment analysis task": 9359,
      "modern natural language understanding": 74401,
      "natural language inference models": 76331,
      "compared previous stateoftheart model": 19419,
      "modern standard arabic msa": 74418,
      "pos tags dependency trees": 88243,
      "best results achieved using": 13437,
      "require large labeled data": 100163,
      "address issues propose new": 3707,
      "speech recognition asr models": 111740,
      "leverage large unlabeled data": 63597,
      "recent success deep learning": 96531,
      "propose novel selfsupervised learning": 92963,
      "aspect category sentiment analysis": 9324,
      "category sentiment analysis acsa": 15794,
      "arabic modern standard arabic": 8527,
      "standard arabic msa dialectal": 112190,
      "arabic msa dialectal arabic": 8531,
      "multilingual training neural machine": 75388,
      "neural machine translation mnmt": 78024,
      "translation performance language pairs": 125114,
      "language pairs paper propose": 59772,
      "paper explores data augmentation": 83931,
      "question answering qa dataset": 95078,
      "training natural language understanding": 123728,
      "representation natural language understanding": 99352,
      "machine reading comprehension natural": 65862,
      "novel deep learning model": 80539,
      "proposed model achieves stateoftheart": 93433,
      "semeval2021 task toxic spans": 105561,
      "task toxic spans detection": 118797,
      "machine translation language pairs": 66023,
      "substantially effective resourcelean scenarios": 114888,
      "significantly outperforms multilingual bert": 109019,
      "outperforms multilingual bert mbert": 82922,
      "text generation pretrained language": 120999,
      "corpus corpus freely available": 23730,
      "corpus freely available noncommercial": 23810,
      "train recurrent neural networks": 123007,
      "yield performance improvements especially": 135349,
      "performance improvements especially sequencelevel": 86448,
      "improvements especially sequencelevel metrics": 52850,
      "attentionbased deep neural network": 10067,
      "neural networks work propose": 78609,
      "networks work propose new": 77820,
      "work propose new method": 134731,
      "elementary discourse units edus": 35274,
      "learning extensive experiments demonstrate": 62573,
      "paper present submission eacl": 84141,
      "datasets demonstrate approach outperforms": 27396,
      "resourcerich languages like english": 100933,
      "using natural language toolkit": 129939,
      "natural language toolkit nltk": 76548,
      "generation process endtoend manner": 47566,
      "average f1 score 725": 11189,
      "neural networks rnns transformers": 78566,
      "et al 2020 task": 38738,
      "establishes new stateoftheart performance": 38611,
      "approach hate speech detection": 7599,
      "languages english best knowledge": 60531,
      "machine learning algorithms used": 65770,
      "deep convolutional neural network": 28225,
      "contextual word embedding models": 22518,
      "machine learning ml algorithms": 65803,
      "nlp tasks sentiment analysis": 79768,
      "pretrained language models prlms": 90097,
      "extensive natural language processing": 41946,
      "language models pretrained language": 59627,
      "models pretrained language models": 73782,
      "model learns disentangle semantics": 71444,
      "learns disentangle semantics syntax": 63206,
      "disentangle semantics syntax sentence": 32345,
      "natural language generation human": 76303,
      "evaluate performance natural language": 38886,
      "natural language generation model": 76304,
      "discuss future research directions": 32255,
      "classification slot filling tasks": 17410,
      "method outperforms previous systems": 69023,
      "models neural language models": 73626,
      "et al 2016 word": 38707,
      "semeval2021 task multilingual crosslingual": 105558,
      "task multilingual crosslingual wordincontext": 118417,
      "multilingual crosslingual wordincontext disambiguation": 75232,
      "crosslingual wordincontext disambiguation mclwic": 25049,
      "experiment pretrained language models": 40485,
      "pretrained language models investigate": 90079,
      "recent years word embeddings": 96645,
      "word embeddings widely used": 133246,
      "finetuning pretrained language models": 44496,
      "pretrained language models plm": 90091,
      "paper investigate zeroshot performance": 84026,
      "context named entity recognition": 22190,
      "entity recognition ner partofspeech": 38044,
      "work present new stateoftheart": 134684,
      "instead structural knowledge base": 55686,
      "queries masked sentences paris": 94927,
      "masked sentences paris capital": 67300,
      "sentences paris capital mask": 106438,
      "paris capital mask used": 84867,
      "capital mask used probes": 15241,
      "model achieves stateoftheart rouge": 70606,
      "achieves stateoftheart rouge scores": 2906,
      "semantic parsing recent work": 105162,
      "representations pretrained language models": 99817,
      "solve problem propose new": 110608,
      "switchboard dialog act corpus": 116287,
      "use pointwise mutual information": 128199,
      "natural language processing nlp modules": 76455,
      "robust approach linking existing lexicalsemantic": 103019,
      "approach linking existing lexicalsemantic hierarchies": 7693,
      "using information obtained recall rate": 129758,
      "executable program simple scripting language": 39966,
      "select set candidates node target": 104714,
      "set candidates node target taxonomy": 107386,
      "candidates node target taxonomy bests": 15140,
      "node target taxonomy bests matches": 80029,
      "target taxonomy bests matches node": 117728,
      "taxonomy bests matches node source": 119664,
      "bests matches node source taxonomy": 13489,
      "present general overview systems taken": 89504,
      "noun phrases japanese language articles": 80433,
      "relative reduction word error rate": 98373,
      "reduction word error rate wer": 97469,
      "spanish english lexical sample tasks": 111156,
      "shared task languageindependent named entity": 108062,
      "task languageindependent named entity recognition": 118339,
      "languageindependent named entity recognition background": 60368,
      "named entity recognition background information": 76054,
      "entity recognition background information data": 37995,
      "recognition background information data sets": 96824,
      "evaluation method present general overview": 39271,
      "method present general overview systems": 69066,
      "general overview systems taken task": 46693,
      "overview systems taken task discuss": 83380,
      "systems taken task discuss performance": 117184,
      "statistical machine translation smt systems": 113117,
      "important tasks natural language processing": 52277,
      "natural language processing information retrieval": 76427,
      "context multidocument summarization evolving events": 22185,
      "natural language processing nlp like": 76450,
      "used natural language processing nlp": 128650,
      "natural language processing nlp applications": 76441,
      "various natural language processing nlp": 131144,
      "text mining natural language processing": 121123,
      "knowledge domain consideration social behavior": 57881,
      "natural language understanding based objectoriented": 76553,
      "language understanding based objectoriented semantics": 60206,
      "understanding based objectoriented semantics algorithms": 126802,
      "systems closely connected text processing": 116785,
      "closely connected text processing criminology": 17854,
      "connected text processing criminology operation": 21132,
      "text processing criminology operation business": 121207,
      "processing criminology operation business medicine": 91650,
      "criminology operation business medicine document": 24780,
      "operation business medicine document systems": 82056,
      "neural probabilistic language model nplm": 78635,
      "source language text target language": 110784,
      "natural language applications question answering": 76269,
      "predicateargument structure sentence plays critical": 88857,
      "structure sentence plays critical role": 113951,
      "sentence plays critical role deep": 105991,
      "plays critical role deep processing": 87725,
      "critical role deep processing natural": 24831,
      "role deep processing natural language": 103167,
      "named entity recognition ner using": 76096,
      "training corpus statistical machine translation": 123406,
      "swiss avalanche bulletin swiss avalanche": 116273,
      "avalanche bulletin swiss avalanche bulletin": 11157,
      "bulletin swiss avalanche bulletin produced": 14961,
      "swiss avalanche bulletin produced twice": 116271,
      "avalanche bulletin produced twice day": 11154,
      "bulletin produced twice day languages": 14957,
      "produced twice day languages lack": 91979,
      "twice day languages lack time": 126085,
      "day languages lack time available": 27844,
      "languages lack time available manual": 60668,
      "lack time available manual translation": 58761,
      "time available manual translation fully": 121988,
      "available manual translation fully automated": 11044,
      "manual translation fully automated translation": 67019,
      "translation fully automated translation employed": 124832,
      "fully automated translation employed based": 45945,
      "automated translation employed based catalogue": 10479,
      "translation employed based catalogue predefined": 124785,
      "employed based catalogue predefined phrases": 36295,
      "based catalogue predefined phrases predetermined": 11565,
      "catalogue predefined phrases predetermined rules": 15705,
      "predefined phrases predetermined rules phrases": 88832,
      "phrases predetermined rules phrases combined": 87446,
      "predetermined rules phrases combined produce": 88844,
      "rules phrases combined produce sentences": 103430,
      "able automatically translate sentences german": 1613,
      "automatically translate sentences german target": 10834,
      "translate sentences german target languages": 124540,
      "sentences german target languages french": 106334,
      "german target languages french italian": 47918,
      "target languages french italian english": 117670,
      "languages french italian english subsequent": 60599,
      "french italian english subsequent proofreading": 45816,
      "italian english subsequent proofreading correction": 57091,
      "catalogue phrases limited small sublanguage": 15698,
      "reduction daily translation costs expected": 97446,
      "daily translation costs expected offset": 25517,
      "translation costs expected offset initial": 124720,
      "costs expected offset initial development": 24401,
      "expected offset initial development costs": 40397,
      "offset initial development costs years": 81621,
      "operational winter seasons assess quality": 82061,
      "winter seasons assess quality produced": 132735,
      "seasons assess quality produced texts": 104379,
      "assess quality produced texts based": 9491,
      "origins catalogue phrases versus manually": 82583,
      "catalogue phrases versus manually written": 15701,
      "phrases versus manually written translated": 87467,
      "versus manually written translated texts": 131630,
      "manually written translated texts mean": 67086,
      "written translated texts mean recognition": 135156,
      "translated texts mean recognition rate": 124574,
      "texts mean recognition rate 55": 121551,
      "mean recognition rate 55 users": 67608,
      "recognition rate 55 users hardly": 96981,
      "rate 55 users hardly distinguish": 95781,
      "types texts similar ratings respect": 126372,
      "texts similar ratings respect language": 121610,
      "similar ratings respect language quality": 109134,
      "ratings respect language quality overall": 95828,
      "respect language quality overall output": 101085,
      "language quality overall output catalogue": 60012,
      "quality overall output catalogue considered": 94743,
      "overall output catalogue considered virtually": 83244,
      "output catalogue considered virtually equivalent": 83059,
      "catalogue considered virtually equivalent text": 15694,
      "considered virtually equivalent text written": 21313,
      "virtually equivalent text written avalanche": 131752,
      "equivalent text written avalanche forecasters": 38256,
      "text written avalanche forecasters manually": 121420,
      "written avalanche forecasters manually translated": 135118,
      "avalanche forecasters manually translated professional": 11161,
      "forecasters manually translated professional translators": 45034,
      "manually translated professional translators furthermore": 67082,
      "translated professional translators furthermore forecasters": 124560,
      "professional translators furthermore forecasters declared": 92084,
      "translators furthermore forecasters declared relevant": 125516,
      "furthermore forecasters declared relevant situations": 46178,
      "forecasters declared relevant situations captured": 45030,
      "declared relevant situations captured sufficient": 27978,
      "relevant situations captured sufficient accuracy": 98584,
      "natural language processing tasks information": 76496,
      "language processing tasks information extraction": 59960,
      "representations named entity recognition ner": 99766,
      "named entity recognition social media": 76114,
      "natural language processing nlp tools": 76468,
      "important natural language processing task": 52191,
      "experiments approach achieves significant improvements": 40780,
      "approach achieves significant improvements stateoftheart": 7321,
      "important task natural language processing": 52266,
      "neural language models learn word": 77948,
      "language models learn word representations": 59582,
      "capture rich linguistic conceptual information": 15374,
      "rich linguistic conceptual information investigate": 102767,
      "linguistic conceptual information investigate embeddings": 64449,
      "conceptual information investigate embeddings learned": 20666,
      "information investigate embeddings learned neural": 54705,
      "investigate embeddings learned neural machine": 56754,
      "embeddings learned neural machine translation": 35784,
      "learned neural machine translation models": 62238,
      "challenge machine translation mt systems": 16053,
      "tasks information retrieval machine translation": 119205,
      "stateoftheart named entity recognition ner": 112766,
      "named entity recognition ner systems": 76093,
      "bidirectional long shortterm memory lstm": 13942,
      "long shortterm memory lstm neural": 65121,
      "shortterm memory lstm neural networks": 108339,
      "annotated using amazon mechanical turk": 6260,
      "important task natural language understanding": 52267,
      "despite large studies devoted represent": 29705,
      "large studies devoted represent texts": 61284,
      "studies devoted represent texts physical": 114212,
      "devoted represent texts physical models": 30479,
      "performance natural language processing tasks": 86562,
      "natural language processing tasks paper": 76498,
      "similarity word sense disambiguation tasks": 109338,
      "long shortterm memory lstm recurrent": 65122,
      "shortterm memory lstm recurrent neural": 108341,
      "memory lstm recurrent neural networks": 68346,
      "bengali text corpus developed tdil": 13037,
      "text corpus developed tdil project": 120837,
      "corpus developed tdil project govt": 23755,
      "developed tdil project govt india": 30311,
      "word embeddings distributed representations words": 133089,
      "tasks natural language processing nlp": 119332,
      "based multinomial naive bayes mnb": 11870,
      "term frequency inverse document frequency": 120218,
      "natural language processing applications paper": 76396,
      "neural network models natural language": 78348,
      "network models natural language processing": 77350,
      "feedforward networks convolutional networks recurrent": 43849,
      "networks convolutional networks recurrent networks": 77555,
      "model achieves stateoftheart performance multilingual": 70601,
      "year nara institute science technology": 135237,
      "bidirectional long shortterm memory recurrent": 13945,
      "long shortterm memory recurrent neural": 65128,
      "shortterm memory recurrent neural network": 108348,
      "memory recurrent neural network blstmrnn": 68374,
      "recurrent neural network blstmrnn shown": 97202,
      "neural network blstmrnn shown effective": 78270,
      "sequential data speech utterances handwritten": 107220,
      "data speech utterances handwritten documents": 26496,
      "rational speech acts rsa model": 95838,
      "logistic regression convolutional neural network": 65035,
      "language model based long shortterm": 59296,
      "model based long shortterm memory": 70735,
      "range natural language processing applications": 95585,
      "language processing tasks information retrieval": 59961,
      "partofspeech tagging named entity recognition": 85557,
      "long shortterm memory lstm models": 65118,
      "using syntactic parser external sentiment": 130259,
      "syntactic parser external sentiment lexicons": 116441,
      "experiments carried jadavpur university participation": 40815,
      "tool developed task based trigram": 122382,
      "developed task based trigram hidden": 30306,
      "task based trigram hidden markov": 117924,
      "based trigram hidden markov model": 12141,
      "trigram hidden markov model utilizes": 125799,
      "hidden markov model utilizes information": 49900,
      "word level features enhance observation": 133342,
      "level features enhance observation probabilities": 63448,
      "features enhance observation probabilities known": 43487,
      "enhance observation probabilities known tokens": 37481,
      "observation probabilities known tokens unknown": 81169,
      "probabilities known tokens unknown tokens": 90847,
      "known tokens unknown tokens submitted": 58327,
      "tokens unknown tokens submitted runs": 122349,
      "attentionbased neural machine translation attentional": 10087,
      "neural machine translation attentional mechanism": 77975,
      "capture partial aspects attentional regularities": 15359,
      "recurrent neural network based language": 97199,
      "neural network based language model": 78259,
      "architectures long shortterm memory lstm": 8820,
      "using various machine learning algorithms": 130357,
      "pointwise mutual information matrix stochastic": 87892,
      "mutual information matrix stochastic gradient": 75980,
      "information matrix stochastic gradient descent": 54756,
      "tasks named entity recognition ner": 119324,
      "named entity recognition ner entity": 76082,
      "performance answer sentence selection task": 86141,
      "natural language processing machine learning": 76433,
      "language processing machine learning techniques": 59883,
      "language generation spoken dialogue systems": 59097,
      "disambiguation named entity disambiguation ned": 31976,
      "long shortterm memory lstm language": 65116,
      "neural networks natural language understanding": 78523,
      "automatic speech recognition asr machine": 10659,
      "speech recognition asr machine translation": 111739,
      "recognition asr machine translation mt": 96813,
      "question answering information retrieval information": 95053,
      "answering information retrieval information extraction": 6637,
      "bidirectional long shortterm memory bilstm": 13940,
      "long shortterm memory bilstm networks": 65107,
      "work studies comparatively typical sentence": 134824,
      "tasks textual entailment te answer": 119563,
      "textual entailment te answer selection": 121703,
      "entailment te answer selection observing": 37682,
      "prior work limitations phrase generation": 90754,
      "work limitations phrase generation representation": 134615,
      "limitations phrase generation representation ii": 64188,
      "phrase generation representation ii conducts": 87352,
      "generation representation ii conducts alignment": 47599,
      "representation ii conducts alignment word": 99256,
      "ii conducts alignment word phrase": 51704,
      "conducts alignment word phrase levels": 20970,
      "alignment word phrase levels handcrafted": 5005,
      "word phrase levels handcrafted features": 133406,
      "phrase levels handcrafted features iii": 87358,
      "levels handcrafted features iii utilizes": 63541,
      "handcrafted features iii utilizes single": 49345,
      "considering characteristics specific tasks limits": 21319,
      "effectiveness tasks propose architecture based": 34964,
      "tasks propose architecture based gated": 119408,
      "propose architecture based gated recurrent": 92556,
      "architecture based gated recurrent unit": 8627,
      "based gated recurrent unit supports": 11733,
      "gated recurrent unit supports representation": 46518,
      "recurrent unit supports representation learning": 97268,
      "unit supports representation learning phrases": 127214,
      "supports representation learning phrases arbitrary": 116042,
      "representation learning phrases arbitrary granularity": 99307,
      "learning phrases arbitrary granularity ii": 62893,
      "phrases arbitrary granularity ii taskspecific": 87410,
      "experimental results te match observation": 40712,
      "endtoend neural machine translation nmt": 36941,
      "variety natural language processing tasks": 131003,
      "natural language processing tasks previous": 76500,
      "language processing tasks previous works": 59967,
      "multitask learning framework jointly learn": 75841,
      "jointly learn multiple related tasks": 57350,
      "neural machine translation models neural": 78032,
      "machine translation models neural machine": 66056,
      "translation models neural machine translation": 124960,
      "model introduces continuous latent variable": 71374,
      "zero pronoun resolution existing approaches": 135478,
      "experimental results proposed approach significantly": 40672,
      "results proposed approach significantly outperforms": 102065,
      "proposed approach significantly outperforms stateoftheart": 93218,
      "approach significantly outperforms stateoftheart systems": 7916,
      "lexicons neural machine translation neural": 63934,
      "neural machine translation neural machine": 78038,
      "machine translation neural machine translation": 66083,
      "translation neural machine translation nmt": 125012,
      "neural machine translation simultaneous translation": 78103,
      "translation neural machine translation recently": 125013,
      "translation quality neural machine translation": 125181,
      "quality neural machine translation systems": 94732,
      "neural machine translation nmt systems": 78064,
      "word2vec mikolov et al 2013a": 133680,
      "phrase based statistical machine translation": 87338,
      "bidirectional long short term memory": 13937,
      "long short term memory networks": 65098,
      "statistical machine translation smt neural": 113116,
      "generates target sentence word word": 47185,
      "use natural language processing nlp": 128165,
      "tts automatic speech recognition asr": 125917,
      "training data paper propose deep": 123509,
      "neural machine translation nmt heavily": 78049,
      "machine translation nmt heavily relies": 66099,
      "languages natural word delimiters chinese": 60757,
      "based recurrent neural network rnn": 11980,
      "languages spoken language resources likely": 60893,
      "spoken language resources likely annotated": 111992,
      "language resources likely annotated translations": 60056,
      "resources likely annotated translations transcriptions": 101004,
      "model achieves best performance datasets": 70556,
      "extensive experiments demonstrate proposed methods": 41909,
      "using bidirectional recurrent neural networks": 129405,
      "using recurrent neural networks rnn": 130094,
      "parallel corpus source target language": 84625,
      "using recurrent neural network language": 130091,
      "recurrent neural network language model": 97207,
      "neural network language model rnnlm": 78315,
      "network language model rnnlm architecture": 77297,
      "different natural language processing nlp": 31280,
      "neural machine translation nmt new": 78054,
      "machine translation nmt new approach": 66110,
      "translation nmt new approach machine": 125047,
      "nmt new approach machine translation": 79934,
      "conventional statistical machine translation smt": 22904,
      "approach achieves significant consistent improvements": 7318,
      "neural machine translation nmt significantly": 78060,
      "machine translation nmt significantly improved": 66118,
      "target language neural machine translation": 117644,
      "based recurrent neural networks rnns": 11982,
      "natural language processing nlp tasks": 76464,
      "using term frequencyinverse document frequency": 130283,
      "term frequencyinverse document frequency tfidf": 120222,
      "language model rnnlm long shortterm": 59394,
      "model rnnlm long shortterm memory": 71940,
      "using long short term memory": 129834,
      "long short term memory lstm": 65096,
      "short term memory lstm network": 108252,
      "experimental results proposed method significantly": 40677,
      "method significantly outperform baseline methods": 69142,
      "experimental results proposed method improves": 40676,
      "neural machine translation nmt achieved": 78041,
      "machine translation nmt achieved remarkable": 66090,
      "translation nmt achieved remarkable progress": 125026,
      "chinese word segmentation pos tagging": 16841,
      "neural network named entity recognition": 78362,
      "named entity recognition chinese social": 76057,
      "entity recognition chinese social media": 38000,
      "focus named entity recognition ner": 44795,
      "named entity recognition ner chinese": 76080,
      "experimental results effectiveness proposed model": 40612,
      "natural language processing widely used": 76512,
      "paper address problem complementary entity": 83706,
      "address problem complementary entity recognition": 3731,
      "problem complementary entity recognition cer": 90974,
      "proven effective natural language processing": 93725,
      "multilingual natural language processing nlp": 75305,
      "experimental results demonstrate proposed model": 40602,
      "state art neural machine translation": 112460,
      "present novel neural network model": 89620,
      "model neural network joint model": 71575,
      "natural language understanding nlu tasks": 76574,
      "statistical classification using enhanced feature": 113085,
      "classification using enhanced feature set": 17492,
      "machine learning ml natural language": 65807,
      "learning ml natural language processing": 62755,
      "ml natural language processing nlp": 70439,
      "neural machine translation nmt models": 78053,
      "neural machine translation knowledge distillation": 78013,
      "field natural language processing nlp": 43970,
      "natural language processing nlp convolutional": 76443,
      "language processing nlp convolutional neural": 59896,
      "convolutional neural network cnn recurrent": 23188,
      "neural network cnn recurrent neural": 78288,
      "network cnn recurrent neural network": 77199,
      "cnn recurrent neural network rnn": 18021,
      "neural machine translation previous work": 78081,
      "machine translation machine translation mt": 66038,
      "long shortterm memory lstm cells": 65115,
      "natural language inference nli task": 76340,
      "dependency parsing dependency parsing important": 29178,
      "automatic speech recognition asr systems": 10662,
      "topics natural language processing nlp": 122643,
      "natural language processing nlp literature": 76451,
      "important task scientific paper analysis": 52272,
      "sentence mapped features input classifiers": 105942,
      "supervised classification using 10crossvalidation scheme": 115740,
      "classification using 10crossvalidation scheme evaluation": 17485,
      "using 10crossvalidation scheme evaluation conducted": 129310,
      "experimental results proposed methods outperform": 40679,
      "results proposed methods outperform previous": 102078,
      "social media websites electronic newspapers": 110436,
      "media websites electronic newspapers internet": 68185,
      "websites electronic newspapers internet forums": 132303,
      "electronic newspapers internet forums allow": 35259,
      "newspapers internet forums allow visitors": 79412,
      "internet forums allow visitors leave": 56193,
      "forums allow visitors leave comments": 45314,
      "allow visitors leave comments read": 5102,
      "visitors leave comments read interact": 131781,
      "leave comments read interact exchange": 63262,
      "comments read interact exchange free": 18824,
      "read interact exchange free participants": 95935,
      "interact exchange free participants malicious": 55934,
      "exchange free participants malicious intentions": 39944,
      "convolutional neural networks recurrent neural": 23207,
      "neural networks recurrent neural networks": 78553,
      "translation neural machine translation model": 125011,
      "larger vocabulary training complexity decoding": 61391,
      "vocabulary training complexity decoding complexity": 131913,
      "abstract meaning representation parsing abstract": 1774,
      "meaning representation parsing abstract meaning": 67677,
      "representation parsing abstract meaning representation": 99372,
      "parsing abstract meaning representation amr": 85060,
      "source language target language using": 110781,
      "neural machine translation present simple": 78077,
      "text classification neural network models": 120751,
      "state art semantic dependency parsing": 112474,
      "new state art code opensource": 79149,
      "state art code opensource available": 112435,
      "skipgram negative sampling sgns word": 109992,
      "paper describes duluth systems participated": 83820,
      "language models pretrained word embeddings": 59633,
      "standard datasets named entity recognition": 112227,
      "propose novel neural network model": 92955,
      "novel neural network model joint": 80674,
      "based entirely convolutional neural networks": 11688,
      "recurrent neural network based models": 97200,
      "neural machine translation systems paper": 78111,
      "recurrent neural networks rnns proved": 97245,
      "tasks spoken language understanding slu": 119521,
      "important natural language processing tasks": 52192,
      "training neural machine translation nmt": 123736,
      "machine translation nmt models usually": 66108,
      "parser dyer et al 2015": 84947,
      "neural machine translation nmt recently": 78058,
      "conll 2017 ud shared task": 21082,
      "neural machine translation paper proposes": 78074,
      "text datasets visualizing relevant words": 120859,
      "texts summarize contents documents belonging": 121628,
      "new york times article snippets": 79260,
      "based bidirectional long shortterm memory": 11552,
      "shortterm memory lstm neural network": 108338,
      "word embeddings word sense disambiguation": 133252,
      "task experimental results proposed model": 118177,
      "achieves comparable performance stateoftheart supervised": 2761,
      "language processing nlp tasks recent": 59922,
      "models sutskever et al 2014": 74145,
      "used statistical machine translation smt": 128786,
      "nlp tasks named entity recognition": 79744,
      "play important role natural language": 87698,
      "current neural machine translation nmt": 25310,
      "phrasebased statistical machine translation smt": 87396,
      "proposed model achieves significant improvements": 93432,
      "machine translation nmt achieved notable": 66089,
      "translation nmt achieved notable success": 125024,
      "nmt achieved notable success recent": 79825,
      "deep learning based natural language": 28259,
      "learning based natural language processing": 62395,
      "recently variety model designs methods": 96777,
      "variety model designs methods blossomed": 130997,
      "model designs methods blossomed context": 70977,
      "context natural language processing nlp": 22194,
      "natural language processing nlp paper": 76456,
      "statistical machine translation smt framework": 113114,
      "challenging task natural language processing": 16326,
      "model achieves state art performance": 70595,
      "second dialog state tracking challenge": 104404,
      "question answering question answering qa": 95090,
      "consistently improves performance achieving competitive": 21421,
      "improves performance achieving competitive results": 53009,
      "performance achieving competitive results despite": 86120,
      "achieving competitive results despite use": 2943,
      "crucial natural language processing nlp": 25154,
      "using pretrained word embeddings input": 130052,
      "outperforms previous work terms bleu": 82959,
      "rules stored exception structure new": 103445,
      "stored exception structure new rules": 113387,
      "exception structure new rules added": 39927,
      "structure new rules added correct": 113921,
      "named entity recognition syntactic parsing": 76116,
      "word embedding methods word2vec glove": 133021,
      "mitigating impact speech recognition errors": 70388,
      "support vector machine svm model": 116008,
      "using natural language processing nlp": 129938,
      "natural language processing nlp techniques": 76465,
      "long shortterm memory lstm networks": 65120,
      "shortterm memory lstm networks model": 108335,
      "learning language understanding lu dialogue": 62676,
      "semantic textual similarity natural language": 105333,
      "textual similarity natural language inference": 121742,
      "results natural language processing tasks": 101977,
      "sentiment analysis social media twitter": 106660,
      "models natural language inference nli": 73609,
      "attentionbased recurrent neural network translation": 10097,
      "task neural machine translation nmt": 118455,
      "neural machine translation nmt struggles": 78062,
      "shared task customer feedback analysis": 108037,
      "neural network paper propose novel": 78370,
      "languages english french japanese spanish": 60537,
      "including long shortterm memory lstm": 53316,
      "experimental results demonstrate effectiveness proposed": 40588,
      "results demonstrate effectiveness proposed approach": 101678,
      "traditional statistical machine translation smt": 122871,
      "machine translation smt neural machine": 66230,
      "translation smt neural machine translation": 125274,
      "smt neural machine translation nmt": 110290,
      "machine translation nmt new paradigm": 66111,
      "learning natural language processing nlp": 62828,
      "natural language processing nlp models": 76454,
      "created new word analogy corpus": 24681,
      "original english word2vec word analogy": 82519,
      "english word2vec word analogy corpus": 37347,
      "corpora basic evaluation word similarities": 23426,
      "paraphrase identification natural language inference": 84821,
      "existing neural machine translation nmt": 40230,
      "machine translation nmt models generally": 66107,
      "slot filling spoken language understanding": 110059,
      "slot filling task spoken language": 110062,
      "filling task spoken language understanding": 44061,
      "machine learning neural networks methods": 65821,
      "widely used natural language generation": 132578,
      "natural language processing sentiment analysis": 76484,
      "subcorpus gutenberg english poetry corpus": 114660,
      "experiments chineseenglish englishgerman translation tasks": 40826,
      "chineseenglish englishgerman translation tasks demonstrate": 16852,
      "demonstrate proposed model achieves significant": 28849,
      "stateoftheart recurrent neural network rnn": 112884,
      "information extraction knowledge base construction": 54575,
      "goaloriented dialogue systems colloquially known": 48415,
      "dialogue systems colloquially known goal": 30772,
      "systems colloquially known goal oriented": 116790,
      "colloquially known goal oriented chatbots": 18528,
      "known goal oriented chatbots help": 58301,
      "goal oriented chatbots help users": 48376,
      "oriented chatbots help users achieve": 82490,
      "chatbots help users achieve predefined": 16675,
      "help users achieve predefined goal": 49788,
      "users achieve predefined goal book": 129093,
      "achieve predefined goal book movie": 2514,
      "predefined goal book movie ticket": 88827,
      "goal book movie ticket closed": 48336,
      "book movie ticket closed domain": 14411,
      "movie ticket closed domain step": 74910,
      "ticket closed domain step understand": 121960,
      "closed domain step understand users": 17846,
      "domain step understand users goal": 33661,
      "step understand users goal using": 113310,
      "understand users goal using natural": 126789,
      "users goal using natural language": 129127,
      "goal using natural language understanding": 48398,
      "using natural language understanding techniques": 129941,
      "natural language understanding techniques goal": 76589,
      "language understanding techniques goal known": 60289,
      "understanding techniques goal known bot": 126996,
      "techniques goal known bot manage": 119898,
      "goal known bot manage dialogue": 48363,
      "known bot manage dialogue achieve": 58292,
      "bot manage dialogue achieve goal": 14484,
      "manage dialogue achieve goal conducted": 66899,
      "dialogue achieve goal conducted respect": 30628,
      "achieve goal conducted respect learnt": 2466,
      "goal conducted respect learnt policy": 48341,
      "conducted respect learnt policy success": 20943,
      "respect learnt policy success dialogue": 101089,
      "learnt policy success dialogue depends": 63253,
      "policy success dialogue depends quality": 87968,
      "success dialogue depends quality policy": 115076,
      "dialogue depends quality policy turn": 30663,
      "depends quality policy turn reliant": 29292,
      "quality policy turn reliant availability": 94756,
      "policy turn reliant availability highquality": 87973,
      "turn reliant availability highquality training": 125979,
      "reliant availability highquality training data": 98637,
      "availability highquality training data policy": 10912,
      "highquality training data policy learning": 50413,
      "training data policy learning method": 123515,
      "data policy learning method instance": 26246,
      "policy learning method instance deep": 87961,
      "learning method instance deep reinforcement": 62724,
      "method instance deep reinforcement learning": 68915,
      "instance deep reinforcement learning domain": 55598,
      "deep reinforcement learning domain specificity": 28412,
      "reinforcement learning domain specificity available": 97805,
      "learning domain specificity available data": 62522,
      "domain specificity available data typically": 33656,
      "specificity available data typically low": 111617,
      "available data typically low allow": 10974,
      "data typically low allow training": 26590,
      "typically low allow training good": 126444,
      "low allow training good dialogue": 65350,
      "allow training good dialogue policies": 5096,
      "introduce transfer learning method mitigate": 56559,
      "transfer learning method mitigate effects": 124103,
      "learning method mitigate effects low": 62728,
      "method mitigate effects low indomain": 68966,
      "mitigate effects low indomain data": 70366,
      "effects low indomain data availability": 34993,
      "low indomain data availability transfer": 65369,
      "indomain data availability transfer learning": 53963,
      "data availability transfer learning based": 25669,
      "availability transfer learning based approach": 10931,
      "transfer learning based approach improves": 124084,
      "learning based approach improves bots": 62386,
      "based approach improves bots success": 11514,
      "approach improves bots success rate": 7626,
      "improves bots success rate 20": 52958,
      "bots success rate 20 relative": 14492,
      "success rate 20 relative terms": 115122,
      "rate 20 relative terms distant": 95776,
      "20 relative terms distant domains": 525,
      "relative terms distant domains double": 98382,
      "terms distant domains double close": 120311,
      "distant domains double close domains": 32449,
      "domains double close domains compared": 33771,
      "double close domains compared model": 33964,
      "close domains compared model transfer": 17814,
      "domains compared model transfer learning": 33748,
      "compared model transfer learning transfer": 19387,
      "model transfer learning transfer learning": 72227,
      "transfer learning transfer learning chatbots": 124142,
      "learning transfer learning chatbots learn": 63122,
      "transfer learning chatbots learn policy": 124087,
      "learning chatbots learn policy 10": 62429,
      "chatbots learn policy 10 times": 16679,
      "learn policy 10 times faster": 62115,
      "policy 10 times faster finally": 87947,
      "10 times faster finally transfer": 152,
      "times faster finally transfer learning": 122176,
      "faster finally transfer learning approach": 43177,
      "finally transfer learning approach complementary": 44244,
      "transfer learning approach complementary additional": 124080,
      "learning approach complementary additional processing": 62345,
      "approach complementary additional processing warmstarting": 7440,
      "complementary additional processing warmstarting joint": 19736,
      "additional processing warmstarting joint application": 3562,
      "processing warmstarting joint application gives": 91855,
      "warmstarting joint application gives best": 132029,
      "joint application gives best outcomes": 57261,
      "added existing models significantly improve": 3361,
      "models significantly improve state art": 74044,
      "problem area natural language processing": 90943,
      "natural language processing nlp domain": 76444,
      "recent years deep neural networks": 96608,
      "sentiment classification natural language processing": 106705,
      "natural language processing applications word": 76397,
      "based convolutional neural network cnn": 11611,
      "semeval2018 task 11 machine comprehension": 105534,
      "task 11 machine comprehension using": 117814,
      "11 machine comprehension using commonsense": 223,
      "machine comprehension using commonsense knowledge": 65756,
      "approach yields improvements 13 bleu": 8030,
      "paper describes results shared task": 83845,
      "research multilingual crosslingual sentiment analysis": 100561,
      "text classification sentiment analysis tasks": 120766,
      "paper propose novel approach called": 84289,
      "success natural language processing nlp": 115104,
      "social media platforms twitter facebook": 110410,
      "representations resourcepoor languages jointly training": 99860,
      "resourcepoor languages jointly training resourcerich": 100914,
      "languages jointly training resourcerich languages": 60659,
      "jointly training resourcerich languages using": 57399,
      "training resourcerich languages using siamese": 123818,
      "resourcerich languages using siamese network": 100936,
      "model consists twin bidirectional long": 70898,
      "consists twin bidirectional long shortterm": 21503,
      "twin bidirectional long shortterm memory": 126091,
      "shortterm memory recurrent neural networks": 108349,
      "memory recurrent neural networks bilstm": 68377,
      "recurrent neural networks bilstm rnn": 97228,
      "neural networks bilstm rnn shared": 78450,
      "networks bilstm rnn shared parameters": 77528,
      "bilstm rnn shared parameters joined": 14106,
      "rnn shared parameters joined contrastive": 102935,
      "shared parameters joined contrastive loss": 108002,
      "parameters joined contrastive loss function": 84760,
      "joined contrastive loss function based": 57253,
      "contrastive loss function based similarity": 22736,
      "loss function based similarity metric": 65269,
      "function based similarity metric model": 46002,
      "based similarity metric model learns": 12047,
      "representations resourcepoor resourcerich language common": 99863,
      "space using similarity metric based": 111078,
      "datasets resourcerich languages english spanish": 27682,
      "resourcerich languages english spanish resourcepoor": 100931,
      "languages english spanish resourcepoor languages": 60552,
      "english spanish resourcepoor languages hindi": 37287,
      "spanish resourcepoor languages hindi telugu": 111170,
      "resourcepoor languages hindi telugu reveal": 100911,
      "approaches based distributional semantics semantic": 8078,
      "based distributional semantics semantic rules": 11657,
      "distributional semantics semantic rules lexicon": 32735,
      "semantics semantic rules lexicon lists": 105463,
      "semantic rules lexicon lists deep": 105265,
      "rules lexicon lists deep neural": 103419,
      "lexicon lists deep neural network": 63896,
      "lists deep neural network representations": 64718,
      "word sense disambiguation wsd tasks": 133528,
      "machine learning natural language processing": 65817,
      "translation neural machine translation achieved": 125010,
      "conditional random field crf model": 20780,
      "semeval2018 task irony detection english": 105540,
      "task irony detection english tweets": 118318,
      "models based recurrent neural networks": 72815,
      "downstream natural language understanding tasks": 34011,
      "extensive experiments proposed model achieves": 41923,
      "benchmark datasets demonstrate effectiveness method": 12775,
      "transfer transfer learning sentence embeddings": 124202,
      "experiments model outperforms strong baselines": 41019,
      "intelligent personal digital assistants ipdas": 55855,
      "using support vector machines svms": 130249,
      "word sense disambiguation underresourced languages": 133526,
      "model achieves new stateoftheart performance": 70578,
      "quality estimation qe machine translation": 94652,
      "experimental results approach significantly outperforms": 40554,
      "results approach significantly outperforms stateoftheart": 101520,
      "build automatic methods infer social": 14744,
      "automatic methods infer social contexts": 10596,
      "dataset proposed method achieves stateoftheart": 27121,
      "proposed method achieves stateoftheart performance": 93345,
      "models use recurrent neural networks": 74281,
      "paraphrase identification semantic textual similarity": 84826,
      "identification semantic textual similarity natural": 51439,
      "model significantly outperforms stateoftheart methods": 72036,
      "neural machine translation improve performance": 78007,
      "word order source target languages": 133384,
      "tasks including neural machine translation": 119186,
      "machine translation automatic speech recognition": 65908,
      "machine translation grammatical error correction": 65997,
      "model outperforms existing stateoftheart models": 71653,
      "machine translation present challenge set": 66163,
      "summarize research trends papers presented": 115612,
      "research trends papers presented proceedings": 100656,
      "long short term memory bilstm": 65095,
      "model significantly outperforms stateoftheart approaches": 72034,
      "model neural machine translation nmt": 71571,
      "stateoftheart natural language processing algorithms": 112770,
      "model performs significantly better previous": 71744,
      "englishhindi codemixed social media content": 37404,
      "codemixed social media content corpus": 18184,
      "social media content corpus baseline": 110363,
      "posting social media called codemixing": 88501,
      "capture syntactic semantic information words": 15408,
      "fundamental task natural language processing": 46121,
      "task natural language processing applications": 118439,
      "various natural language processing tasks": 131145,
      "experimental results model outperforms stateoftheart": 40651,
      "natural language user utterances semantic": 76599,
      "language user utterances semantic frames": 60309,
      "recurrent neural network rnn based": 97215,
      "bidirectional long shortterm memory blstm": 13941,
      "natural language understanding nlu module": 76572,
      "recurrent neural networks recurrent neural": 97239,
      "shortterm memory lstm networks proven": 108336,
      "language processing nlp tasks sentiment": 59923,
      "stateoftheart neural machine translation models": 112780,
      "successful natural language processing tasks": 115165,
      "direction power pairs participants email": 31827,
      "natural language understanding nlu models": 76571,
      "collected cleaned processed illustrate use": 18409,
      "cleaned processed illustrate use zeroresource": 17711,
      "deep learning natural language processing": 28300,
      "loss neural machine translation nmt": 65287,
      "attention mechanism neural machine translation": 9909,
      "natural language processing nlp systems": 76462,
      "recent progress endtoend speech recognition": 96498,
      "model achieves competitive performance stateoftheart": 70568,
      "natural language processing tasks partofspeech": 76499,
      "partofspeech pos tagging dependency parsing": 85536,
      "proposed method significantly outperforms stateoftheart": 93403,
      "method significantly outperforms stateoftheart baselines": 69149,
      "deep convolutional neural networks cnns": 28228,
      "wellstudied problem natural language processing": 132439,
      "problem natural language processing nlp": 91141,
      "question answering named entity recognition": 95064,
      "minimum semantic units human languages": 70217,
      "important research topic natural language": 52233,
      "research topic natural language processing": 100645,
      "performance various natural language processing": 86846,
      "structural measures pairwise node similarities": 113783,
      "shortest path distance distance measures": 108306,
      "path distance distance measures information": 85672,
      "distance distance measures information graph": 32416,
      "distance measures information graph structure": 32427,
      "measures information graph structure account": 67875,
      "semantic similarity word sense disambiguation": 105296,
      "use reinforcement learning rl finetune": 128240,
      "performance natural language processing nlp": 86561,
      "language processing nlp applications information": 59893,
      "words source sentence target sentence": 134232,
      "stateoftheart results various text classification": 112932,
      "results various text classification tasks": 102316,
      "various text classification tasks arabic": 131223,
      "text classification tasks arabic dialect": 120775,
      "classification tasks arabic dialect identification": 17441,
      "tasks arabic dialect identification native": 118938,
      "arabic dialect identification native language": 8506,
      "dialect identification native language identification": 30523,
      "experimental results automatic human evaluations": 40560,
      "results automatic human evaluations demonstrate": 101533,
      "used natural language processing tasks": 128651,
      "recent years natural language processing": 96622,
      "uses convolutional neural network cnn": 129214,
      "neural models natural language processing": 78185,
      "work natural language processing nlp": 134648,
      "using convolutional neural network cnn": 129499,
      "shared task parallel corpus filtering": 108083,
      "recent advances neural machine translation": 96415,
      "neural machine translation nmt different": 78046,
      "machine reading comprehension paper introduce": 65865,
      "model extends existing approaches perspectives": 71151,
      "results model outperforms stateoftheart models": 101947,
      "address challenge propose simple effective": 3658,
      "challenge propose simple effective approach": 16091,
      "using automatic metrics human judgments": 129371,
      "neural machine translation nmt lowresource": 78051,
      "spoken language understanding slu essential": 112002,
      "language understanding slu essential component": 60271,
      "understanding slu essential component conversational": 126962,
      "slu essential component conversational systems": 110116,
      "recent utterances important recent ones": 96563,
      "experiments benchmark dialogue state tracking": 40803,
      "benchmark dialogue state tracking challenge": 12817,
      "dialogue state tracking challenge dstc4": 30752,
      "state tracking challenge dstc4 dataset": 112525,
      "data augmentation spoken language understanding": 25647,
      "participated open track parseme shared": 85341,
      "open track parseme shared task": 81945,
      "neural machine translation nmt model": 78052,
      "recognizing lexical semantic relations recognizing": 97073,
      "neural machine translation work investigates": 78134,
      "proposed model outperforms competitive baselines": 93461,
      "recent years deep learning methods": 96606,
      "years deep learning methods achieved": 135250,
      "task paper presents nicts participation": 118509,
      "paper presents nicts participation wmt18": 84189,
      "presents nicts participation wmt18 shared": 89879,
      "neural machine translation nmt using": 78067,
      "wmt18 shared news translation task": 132850,
      "experimental results model achieves stateoftheart": 40648,
      "results model achieves stateoftheart performance": 101936,
      "experiment results proposed method outperforms": 40505,
      "codeswitching automatic speech recognition asr": 18229,
      "spoken language understanding spoken language": 112007,
      "language understanding spoken language understanding": 60277,
      "understanding spoken language understanding slu": 126974,
      "spoken language understanding slu systems": 112005,
      "model achieves stateoftheart performance benchmark": 70599,
      "achieves stateoftheart performance benchmark datasets": 2883,
      "datatotext natural language generation nlg": 27818,
      "task natural language processing nlp": 118441,
      "natural language processing nlp research": 76461,
      "pronouns frequently omitted prodrop languages": 92355,
      "frequently omitted prodrop languages chinese": 45877,
      "omitted prodrop languages chinese generally": 81658,
      "prodrop languages chinese generally leading": 91869,
      "languages chinese generally leading significant": 60443,
      "chinese generally leading significant challenges": 16771,
      "generally leading significant challenges respect": 46869,
      "leading significant challenges respect production": 61907,
      "significant challenges respect production complete": 108737,
      "challenges respect production complete translations": 16208,
      "approach significantly improves translation performance": 7906,
      "paper propose new architecture based": 84276,
      "introducing interlingual loss additional training": 56639,
      "interlingual loss additional training objective": 56127,
      "loss additional training objective adding": 65255,
      "additional training objective adding forcing": 3592,
      "training objective adding forcing interlingual": 123756,
      "objective adding forcing interlingual loss": 81064,
      "adding forcing interlingual loss able": 3381,
      "forcing interlingual loss able train": 45022,
      "interlingual loss able train multiple": 56124,
      "loss able train multiple encoders": 65249,
      "able train multiple encoders decoders": 1699,
      "train multiple encoders decoders language": 122972,
      "multiple encoders decoders language sharing": 75552,
      "encoders decoders language sharing common": 36642,
      "widely used natural language processing": 132579,
      "classification problem image classification problem": 17347,
      "able achieve significantly better results": 1600,
      "bilingual contextual word similarity bcws": 14026,
      "neural machine translation previous studies": 78080,
      "quality neural machine translation nmt": 94731,
      "named entity recognition partofspeech tagging": 76101,
      "deep neural networks state art": 28381,
      "language pairs english french english": 59749,
      "information different representation subspaces different": 54492,
      "different representation subspaces different positions": 31383,
      "vaswani et al 2017 work": 131291,
      "publicly available state art natural": 94326,
      "available state art natural language": 11114,
      "state art natural language generator": 112456,
      "propose new neural language model": 92875,
      "reduce number different sense tags": 97346,
      "results significantly outperform state art": 102187,
      "social media feeds news blogs": 110382,
      "media feeds news blogs online": 68116,
      "feeds news blogs online newspapers": 43860,
      "conduct extensive experiments benchmark datasets": 20872,
      "bridging anaphora resolution poesio et": 14606,
      "anaphora resolution poesio et al": 6071,
      "resolution poesio et al 2004": 100779,
      "poesio et al 2004 hou": 87785,
      "et al 2004 hou et": 38687,
      "al 2004 hou et al": 4596,
      "2004 hou et al 2013b": 549,
      "used various natural language processing": 128844,
      "datasets different domains demonstrate effectiveness": 27427,
      "improving neural machine translation lowresource": 53129,
      "neural machine translation lowresource languages": 78020,
      "slotfilling paradigm user refer slots": 110080,
      "paradigm user refer slots context": 84554,
      "user refer slots context conversation": 129036,
      "refer slots context conversation goal": 97502,
      "slots context conversation goal contextual": 110087,
      "context conversation goal contextual understanding": 22042,
      "conversation goal contextual understanding resolve": 22955,
      "goal contextual understanding resolve referring": 48345,
      "contextual understanding resolve referring expressions": 22510,
      "understanding resolve referring expressions appropriate": 126946,
      "resolve referring expressions appropriate slots": 100808,
      "referring expressions appropriate slots context": 97578,
      "address challenge outofvocabulary oov words": 3654,
      "challenge outofvocabulary oov words second": 16076,
      "outofvocabulary oov words second attentive": 82678,
      "oov words second attentive hierarchical": 81874,
      "words second attentive hierarchical recurrent": 134184,
      "second attentive hierarchical recurrent encoder": 104392,
      "attentive hierarchical recurrent encoder ahre": 10114,
      "natural language understanding nlu systems": 76573,
      "taskoriented dialog systems slot filling": 118887,
      "context neural machine translation nmt": 22199,
      "conditional random fields crf model": 20784,
      "experiments neural machine translation tasks": 41038,
      "wmt14 englishgerman wmt17 chineseenglish translation": 132820,
      "neural networkbased natural language understanding": 78430,
      "recurrent neural networks rnns learn": 97243,
      "deep neural networks natural language": 28376,
      "neural networks natural language processing": 78522,
      "networks natural language processing nlp": 77671,
      "natural language processing nlp interpretability": 76449,
      "language processing nlp interpretability remains": 59903,
      "processing nlp interpretability remains challenge": 91739,
      "pos tagging dependency parsing propose": 88230,
      "word segmentation partofspeech pos tagging": 133500,
      "bist graphbased dependency parser kiperwasser": 14235,
      "graphbased dependency parser kiperwasser goldberg": 48893,
      "dependency parser kiperwasser goldberg 2016": 29163,
      "language model trained text corpus": 59411,
      "bert devlin et al 2018": 13098,
      "gpt radford et al 2018": 48557,
      "amie automatedvehicle multimodal incabin experience": 5322,
      "topic field natural language processing": 122522,
      "code pretrained models publicly available": 18136,
      "natural language understanding natural language": 76568,
      "relatively small amounts training data": 98417,
      "present novel semantic framework modeling": 89624,
      "covering entirety universal dependencies english": 24542,
      "entirety universal dependencies english web": 37740,
      "universal dependencies english web treebank": 127287,
      "dependencies english web treebank use": 29082,
      "english web treebank use dataset": 37338,
      "information retrieval natural language processing": 54943,
      "achieves state art performance benchmark": 2876,
      "state art performance benchmark datasets": 112465,
      "recurrent neural network rnn model": 97218,
      "despite recent advances natural language": 29721,
      "recent advances natural language processing": 96411,
      "named entity recognition ner important": 76086,
      "short text classification knowledge powered": 108258,
      "text classification knowledge powered attention": 120731,
      "purpose measuring importance knowledge introduce": 94435,
      "measuring importance knowledge introduce attention": 67916,
      "importance knowledge introduce attention mechanisms": 52065,
      "conduct extensive experiments public datasets": 20876,
      "datasets proposed model significantly outperforms": 27646,
      "model significantly outperforms existing stateoftheart": 72031,
      "data neural machine translation nmt": 26166,
      "f1 score heldout test set": 42684,
      "named entity recognition ner task": 76094,
      "task achieve new stateoftheart results": 117834,
      "entity recognition task named entity": 38079,
      "recognition task named entity recognition": 97025,
      "task named entity recognition ner": 118432,
      "approaches natural language generation nlg": 8249,
      "achieving best accuracy date english": 2931,
      "best accuracy date english ptb": 13298,
      "accuracy date english ptb dataset": 2129,
      "dependency parsing semantic role labeling": 29200,
      "parsing semantic role labeling named": 85228,
      "semantic role labeling named entity": 105252,
      "role labeling named entity recognition": 103189,
      "named entity recognition natural language": 76078,
      "entity recognition natural language inference": 38030,
      "experiments realworld datasets demonstrate proposed": 41109,
      "spoken language understanding slu models": 112003,
      "japanese predicate argument structure analysis": 57198,
      "predicate argument structure analysis pasa": 88850,
      "elmo embeddings peters et al": 35316,
      "embeddings peters et al 2018": 35863,
      "suggestion mining online reviews forums": 115360,
      "dataset model achieves stateoftheart results": 27030,
      "learning models achieve stateoftheart performance": 62771,
      "generating pun sentence given pair": 47251,
      "automatic speech recognition asr errors": 10658,
      "natural language processing tasks including": 76495,
      "named entity recognition sentiment analysis": 76112,
      "entity recognition sentiment analysis question": 38071,
      "recognition sentiment analysis question answering": 96997,
      "past future neural machine translation": 85646,
      "shown neural machine translation nmt": 108499,
      "machine translation nmt models benefit": 66105,
      "long shortterm memory bilstm network": 65106,
      "propose deep learning based model": 92618,
      "open information extraction oie systems": 81907,
      "long shortterm memory lstm network": 65119,
      "translation shared task paper describes": 125250,
      "natural language understanding generation tasks": 76561,
      "new stateoftheart results natural language": 79170,
      "achieves stateoftheart performance unsupervised approach": 2894,
      "recent work super characters method": 96593,
      "stateoftheart results text classification tasks": 112926,
      "natural language generation nlg critical": 76311,
      "waikato environment knowledge analysis weka": 131987,
      "common nlp tasks source code": 18903,
      "reading comprehension natural language inference": 95998,
      "natural language inference sentiment analysis": 76351,
      "model significantly outperforms stateoftheart models": 72037,
      "neural machine translation nmt shown": 78059,
      "translation neural machine translation systems": 125014,
      "data augmentation neural machine translation": 25641,
      "neural machine translation data augmentation": 77985,
      "recurrent neural networks rnns widely": 97247,
      "neural networks rnns widely used": 78568,
      "state tracking challenge dstc2 dataset": 112523,
      "word sense induction word sense": 133532,
      "sense induction word sense induction": 105684,
      "induction word sense induction wsi": 54043,
      "word sense induction wsi task": 133534,
      "natural language processing artificial intelligence": 76403,
      "named entity recognition recent years": 76107,
      "named entity recognition ner especially": 76083,
      "model vaswani et al 2017": 72303,
      "encoder neural machine translation neural": 36549,
      "improve translation performance experiment results": 52573,
      "natural language inference paraphrase identification": 76343,
      "datasets text classification experimental results": 27760,
      "present novel endtoend neural network": 89608,
      "largescale multilabel text classification lmtc": 61473,
      "matching model response selection retrievalbased": 67415,
      "results public data sets indicate": 102096,
      "recurrent neural network language models": 97208,
      "simultaneous machine translation simultaneous machine": 109658,
      "machine translation simultaneous machine translation": 66223,
      "experimental results proposed model outperforms": 40682,
      "proposed model outperforms existing research": 93466,
      "using recurrent neural networks rnns": 130095,
      "paper propose novel neural approach": 84304,
      "train test machine learning models": 123041,
      "neural machine translation neural networks": 78039,
      "experiments proposed approach significantly outperforms": 41070,
      "natural language inference nli datasets": 76336,
      "neural machine translation paper describes": 78072,
      "relation detection knowledge base question": 97970,
      "detection knowledge base question answering": 29981,
      "knowledge base question answering relation": 57787,
      "base question answering relation detection": 11479,
      "propose new hierarchical attention model": 92854,
      "natural language processing nlp task": 76463,
      "experiments language modeling machine translation": 40972,
      "models achieved stateoftheart results various": 72697,
      "play crucial role natural language": 87692,
      "crucial role natural language processing": 25165,
      "bert bidirectional encoder representations transformers": 13081,
      "conversational machine comprehension conversational machine": 23015,
      "machine comprehension conversational machine comprehension": 65749,
      "techniques natural language processing nlp": 119938,
      "recurrent neural network rnn models": 97219,
      "machine reading comprehension machine reading": 65855,
      "reading comprehension machine reading comprehension": 95989,
      "model terms bleu entity f1": 72157,
      "terms bleu entity f1 scores": 120286,
      "phrasebased statistical machine translation pbsmt": 87395,
      "dataset demonstrate effectiveness proposed method": 26855,
      "udpipe 20 bestperforming systems conll": 126541,
      "20 bestperforming systems conll 2018": 492,
      "bestperforming systems conll 2018 shared": 13481,
      "systems conll 2018 shared task": 116805,
      "monolingual data demonstrated helpful improving": 74569,
      "data demonstrated helpful improving translation": 25826,
      "demonstrated helpful improving translation quality": 28919,
      "improve translation quality experimental results": 52576,
      "translation quality experimental results chineseenglish": 125173,
      "chineseenglish germanenglish machine translation tasks": 16856,
      "conduct experiments text generation tasks": 20860,
      "text generation tasks abstractive summarization": 121014,
      "prediction auxiliary task multitask learning": 89036,
      "task natural language processing artificial": 118440,
      "stateoftheart models named entity recognition": 112741,
      "models named entity recognition ner": 73602,
      "extensive experiments proposed method performs": 41921,
      "experiments proposed method performs better": 41076,
      "proposed method performs better stateoftheart": 93394,
      "method performs better stateoftheart baselines": 69051,
      "neural machine translation machine translation": 78022,
      "based neural machine translation nmt": 11891,
      "vector space models word meaning": 131377,
      "neural language models recurrent neural": 77953,
      "language models recurrent neural networks": 59653,
      "empirical results benchmark datasets demonstrate": 36180,
      "language understanding pretrained language models": 60255,
      "pretrained language models achieved great": 90063,
      "tasks including named entity recognition": 119183,
      "chain conditional random fields crfs": 15974,
      "massively multilingual neural machine translation": 67345,
      "multilingual neural machine translation recently": 75311,
      "neural machine translation recently proposed": 78093,
      "multilingual neural machine translation nmt": 75310,
      "establish new state art datasets": 38588,
      "state art machine translation mt": 112446,
      "neural machine translation model training": 78026,
      "latest developments natural language processing": 61655,
      "developments natural language processing nlp": 30446,
      "neural machine translation models rely": 78033,
      "natural language understanding tasks natural": 76586,
      "language understanding tasks natural language": 60283,
      "understanding tasks natural language inference": 126988,
      "results demonstrate effectiveness proposed method": 101679,
      "named entity recognition ner tasks": 76095,
      "information extraction tasks named entity": 54608,
      "extraction tasks named entity recognition": 42513,
      "named entity recognition relation extraction": 76110,
      "named entity recognition ner models": 76088,
      "recent work shown promising results": 96587,
      "pretrained language models bert devlin": 90065,
      "language models bert devlin et": 59516,
      "models bert devlin et al": 72835,
      "models natural language processing nlp": 73611,
      "language model neural machine translation": 59357,
      "model neural machine translation neural": 71570,
      "text generation tasks machine translation": 121017,
      "generation tasks machine translation text": 47671,
      "tasks machine translation text summarization": 119282,
      "trained using maximum likelihood estimation": 123330,
      "dataset publicly available research community": 27136,
      "state art natural language processing": 112457,
      "natural language inference nli models": 76339,
      "natural language generation nlg tasks": 76315,
      "deep neural network acoustic model": 28356,
      "convolutional neural network cnn followed": 23185,
      "memory recurrent neural network lstm": 68375,
      "morphological tagging named entity recognition": 74746,
      "recently pretrained language models achieved": 96733,
      "broad range natural language processing": 14679,
      "range natural language processing tasks": 95587,
      "including partofspeech tagging named entity": 53349,
      "named entity recognition ner speech": 76092,
      "entity recognition ner speech pos": 38048,
      "recognition ner speech pos tagging": 96941,
      "conditional random field crf output": 20781,
      "random field crf output layer": 95482,
      "propose neural machine translation nmt": 92821,
      "achieved various natural language processing": 2724,
      "natural language processing tasks using": 76503,
      "automatic speech recognition asr task": 10663,
      "underlying question study adversarial learning": 126696,
      "question study adversarial learning used": 95224,
      "train models higher level abstraction": 122963,
      "models higher level abstraction order": 73330,
      "higher level abstraction order increase": 50189,
      "level abstraction order increase robustness": 63415,
      "abstraction order increase robustness lexical": 1794,
      "order increase robustness lexical stylistic": 82347,
      "strategy evaluated french corpus encyclopedic": 113512,
      "adversarial learning increases models generalization": 4137,
      "learning increases models generalization capabilities": 62649,
      "language named entity recognition ner": 59702,
      "natural language generation natural language": 76307,
      "transformer network vaswani et al": 124358,
      "network vaswani et al 2017": 77473,
      "transfer learning natural language processing": 124115,
      "natural language processing paper present": 76470,
      "achieve stateoftheart performance natural language": 2562,
      "stateoftheart performance natural language processing": 112835,
      "natural language processing tasks text": 76502,
      "107 phenomena organized 14 categories": 198,
      "model achieves performance comparable stateoftheart": 70582,
      "representations experimental results approach achieves": 99647,
      "named entity recognition task named": 76118,
      "formulate machine reading comprehension mrc": 45276,
      "machine reading comprehension mrc task": 65861,
      "additionally query encodes informative prior": 3630,
      "query encodes informative prior knowledge": 94959,
      "encodes informative prior knowledge strategy": 36673,
      "informative prior knowledge strategy facilitates": 55149,
      "prior knowledge strategy facilitates process": 90714,
      "knowledge strategy facilitates process entity": 58190,
      "strategy facilitates process entity extraction": 113517,
      "facilitates process entity extraction leading": 42806,
      "process entity extraction leading better": 91476,
      "entity extraction leading better performances": 37937,
      "datasets experimental results demonstrate effectiveness": 27470,
      "pretrained word embeddings characterlevel word": 90218,
      "sequence generation tasks machine translation": 106954,
      "neural machine translation models generate": 78030,
      "phrasebased statistical machine translation model": 87394,
      "experimental results demonstrate effectiveness approach": 40586,
      "classification tasks natural language inference": 17451,
      "tasks natural language inference task": 119329,
      "neural machine translation nmt paper": 78055,
      "success neural machine translation nmt": 115108,
      "language model pretrained language models": 59377,
      "dependency parsing named entity recognition": 29186,
      "named entity recognition ner existing": 76084,
      "approach significantly outperforms existing stateoftheart": 7909,
      "significantly outperforms existing stateoftheart methods": 109016,
      "focuses applying endtoend dialog technologies": 44890,
      "paper describes task definition provided": 83854,
      "describes task definition provided datasets": 29441,
      "evaluation setup track summarize results": 39392,
      "setup track summarize results submitted": 107862,
      "track summarize results submitted systems": 122739,
      "summarize results submitted systems highlight": 115616,
      "results submitted systems highlight overall": 102223,
      "submitted systems highlight overall trends": 114778,
      "systems highlight overall trends stateoftheart": 116934,
      "highlight overall trends stateoftheart technologies": 50270,
      "overall trends stateoftheart technologies tasks": 83268,
      "recurrent neural network conduct extensive": 97204,
      "proposed framework outperforms stateoftheart methods": 93295,
      "outperforms stateoftheart methods large margin": 83008,
      "representation learning natural language processing": 99302,
      "achieved great success natural language": 2634,
      "translation tasks model outperforms strong": 125358,
      "proposed model significantly outperforms stateoftheart": 93483,
      "word embeddings substantially successful capturing": 133218,
      "embeddings substantially successful capturing semantic": 35962,
      "substantially successful capturing semantic relations": 114916,
      "successful capturing semantic relations words": 115156,
      "deep learning models relation extraction": 28293,
      "representations yielded significant improvements nlp": 100004,
      "yielded significant improvements nlp tasks": 135370,
      "replacing static word embeddings contextualized": 98952,
      "static word embeddings contextualized word": 113070,
      "word embeddings contextualized word representations": 133071,
      "wide range natural language processing": 132509,
      "range natural language processing nlp": 95586,
      "achieves new stateoftheart results datasets": 2826,
      "sequence labeling tasks experimental results": 106987,
      "tagging named entity recognition tasks": 117420,
      "named entity recognition named entity": 76076,
      "bidirectional encoder representation transformers bert": 13912,
      "appropriate responses yielding substantive gains": 8434,
      "responses yielding substantive gains bleu": 101307,
      "yielding substantive gains bleu scores": 135393,
      "recent neural machine translation nmt": 96481,
      "downstream natural language processing nlp": 34008,
      "transformer vaswani et al 2017": 124397,
      "vaswani et al 2017 based": 131290,
      "challenge natural language processing nlp": 16063,
      "natural language processing nlp community": 76442,
      "crucial component taskoriented dialog systems": 25138,
      "systems natural language generation nlg": 117013,
      "natural language generation nlg module": 76313,
      "neural machine translation existing neural": 78001,
      "machine translation existing neural machine": 65981,
      "translation existing neural machine translation": 124807,
      "nlp tasks like sentiment analysis": 79735,
      "natural language processing nlp perspective": 76457,
      "recent advances deep learning led": 96399,
      "advances deep learning led significant": 3983,
      "7th dialog technology challenges dstc7": 1283,
      "various natural language processing applications": 131143,
      "models based deep neural networks": 72807,
      "proposed approach significantly improves performance": 93216,
      "machine learning support vector machine": 65828,
      "learning support vector machine logistic": 63078,
      "support vector machine logistic regression": 116004,
      "deep learning convolutional neural network": 28266,
      "convolutional neural network recurrent neural": 23194,
      "neural network recurrent neural network": 78382,
      "language models pretrained large corpora": 59630,
      "including natural language inference paraphrase": 53335,
      "generation natural language generation nlg": 47500,
      "natural language generation nlg models": 76312,
      "using recurrent neural network rnn": 130092,
      "synthetic data neural machine translation": 116622,
      "synthetic parallel data noisy generated": 116640,
      "parallel data noisy generated imperfect": 84645,
      "demonstrate proposed approach significantly outperforms": 28836,
      "monolingual data neural machine translation": 74578,
      "language pairs experimental results demonstrate": 59758,
      "models experimental results demonstrate proposed": 73200,
      "named entity recognition ner fundamental": 76085,
      "natural language processing systems recent": 76488,
      "current stateoftheart neural machine translation": 25364,
      "metalearning knowledge graphbased dialogue systems": 68554,
      "achieved considerable success natural language": 2616,
      "considerable success natural language processing": 21263,
      "unsupervised neural machine translation unsupervised": 127684,
      "neural machine translation unsupervised neural": 78127,
      "machine translation unsupervised neural machine": 66288,
      "unsupervised neural machine translation unmt": 127683,
      "neural machine translation unmt recently": 78125,
      "unmt recently achieved remarkable results": 127472,
      "pretrained language models elmo bert": 90075,
      "connectionist temporal classification ctc loss": 21144,
      "train neural machine translation nmt": 122982,
      "variational autoencoder vae based approaches": 130916,
      "model achieves significant improvements compared": 70592,
      "significant improvements compared strong baselines": 108791,
      "tackle named entity recognition ner": 117298,
      "benchmark datasets demonstrate method outperforms": 12778,
      "approach significantly consistently improves translation": 7900,
      "extensive experiments benchmark datasets method": 41898,
      "networks recurrent neural networks rnns": 77730,
      "bidirectional encoder representations transformers bert": 13918,
      "paper propose novel data augmentation": 84292,
      "automatic generation parallel data iterative": 10548,
      "generation parallel data iterative backtranslation": 47535,
      "transferability outperforming prior best model": 124218,
      "metric significantly higher correlation human": 69904,
      "significantly higher correlation human judgments": 108922,
      "unstructured text existing methods adopt": 127586,
      "pretrained language model pretrained language": 90056,
      "model pretrained language models plms": 71776,
      "natural language processing tasks given": 76494,
      "measuring progress natural language understanding": 67926,
      "prior work controllable text generation": 90746,
      "neural machine translation nmt suffers": 78063,
      "syntactic generalization neural language models": 116413,
      "empirical results model outperforms previous": 36189,
      "results model outperforms previous stateoftheart": 101945,
      "language inference nli task determining": 59184,
      "recurrent neural networks rnns long": 97244,
      "natural language processing nlp including": 76448,
      "learning named entity recognition ner": 62824,
      "pretrained language representation models bidirectional": 90110,
      "language representation models bidirectional encoder": 60045,
      "representation models bidirectional encoder representations": 99337,
      "models bidirectional encoder representations transformers": 72853,
      "language processing tasks paper present": 59964,
      "functional distributional semantics functional distributional": 46059,
      "distributional semantics functional distributional semantics": 32728,
      "semantics functional distributional semantics provides": 105421,
      "models widely used natural language": 74342,
      "tasks machine translation question answering": 119280,
      "documentlevel neural machine translation nmt": 33157,
      "classification using support vector machine": 17495,
      "stateoftheart methods automatic human evaluations": 112707,
      "neural machine translation paper propose": 78073,
      "machine learning models support vector": 65812,
      "downstream natural language processing tasks": 34009,
      "endtoend method based supervised learning": 36924,
      "accuracy endtoend method point corresponding": 2148,
      "endtoend method point corresponding method": 36927,
      "method point corresponding method applied": 69056,
      "text classification tasks sentiment analysis": 120779,
      "validity domain text explored paper": 130752,
      "bidirectional gated recurrent units bigru": 13926,
      "conducted extensive experiments benchmark datasets": 20927,
      "sentiment analysis social media data": 106659,
      "offensive language identification social media": 81549,
      "pretrained language models downstream tasks": 90073,
      "research excellence framework ref 2014": 100495,
      "graph convolutional network relation extraction": 48784,
      "parser achieves new stateoftheart performance": 84915,
      "word segmentation pos tagging named": 133505,
      "proposed framework significantly outperforms stateoftheart": 93300,
      "explicitly modeling mutual interaction relation": 41386,
      "bidirectional encoder representation transformer bert": 13910,
      "semeval 2020 task commonsense validation": 105513,
      "2020 task commonsense validation explanation": 736,
      "tasks semantic role labeling srl": 119484,
      "performs substantially better prior work": 87033,
      "provided domain experts use targetopinion": 93966,
      "domain experts use targetopinion word": 33531,
      "experts use targetopinion word pairs": 41252,
      "use targetopinion word pairs supervision": 128317,
      "word pairs extracted using dependency": 133395,
      "pairs extracted using dependency parsers": 83547,
      "extracted using dependency parsers simple": 42193,
      "using dependency parsers simple rules": 129558,
      "dependency parsers simple rules objective": 29168,
      "predict opinion word given target": 88913,
      "opinion word given target word": 82110,
      "word given target word ultimate": 133303,
      "given target word ultimate goal": 48149,
      "target word ultimate goal learn": 117751,
      "word ultimate goal learn sentiment": 133616,
      "introducing latent variable sentiment polarity": 56644,
      "latent variable sentiment polarity objective": 61634,
      "variable sentiment polarity objective function": 130850,
      "sentiment polarity objective function inject": 106777,
      "polarity objective function inject sentiment": 87920,
      "supervised method hundreds labels aspect": 115802,
      "chinese clinical named entity recognition": 16748,
      "clinical named entity recognition clinical": 17776,
      "named entity recognition clinical named": 76060,
      "entity recognition clinical named entity": 38004,
      "recognition clinical named entity recognition": 96837,
      "clinical named entity recognition cner": 17777,
      "named entity recognition cner aims": 76062,
      "computational results ccks2017 task benchmark": 20429,
      "results ccks2017 task benchmark dataset": 101566,
      "language natural language processing nlp": 59707,
      "model achieve better local optimum": 70536,
      "model achieves new stateoftheart results": 70579,
      "approach achieves new stateoftheart performance": 7311,
      "tasks natural language inference nli": 119328,
      "neural machine translation nmt work": 78068,
      "partofspeech pos tagging named entity": 85538,
      "pos tagging named entity recognition": 88234,
      "tagging named entity recognition ner": 117418,
      "named entity recognition ner dependency": 76081,
      "entity recognition ner dependency parsing": 38034,
      "ubiquitous natural language processing models": 126513,
      "languages paper present new dataset": 60780,
      "propose novel data augmentation method": 92910,
      "language processing nlp tasks despite": 59920,
      "pretrained bidirectional encoder representations transformers": 89997,
      "chinese word segmentation partofspeech tagging": 16839,
      "word segmentation partofspeech tagging named": 133502,
      "segmentation partofspeech tagging named entity": 104616,
      "applications question answering text summarization": 6999,
      "graph graph neural networks gnns": 48819,
      "stateoftheart models natural language processing": 112744,
      "come high computational cost work": 18767,
      "high computational cost work explore": 50047,
      "computational cost work explore simple": 20372,
      "data neural machine translation systems": 26167,
      "language models recent work shown": 59648,
      "approach significantly outperforms stateoftheart baselines": 7914,
      "datasets indicate model significantly outperform": 27524,
      "indicate model significantly outperform stateoftheart": 53844,
      "aspectcategory sentiment analysis acsa aims": 9364,
      "sentiment analysis acsa aims predict": 106564,
      "analysis acsa aims predict sentiment": 5488,
      "acsa aims predict sentiment polarities": 3059,
      "experimental results public datasets demonstrate": 40687,
      "results public datasets demonstrate effectiveness": 102099,
      "tasks natural language processing involve": 119331,
      "establishing new state art task": 38619,
      "datasets demonstrate effectiveness proposed method": 27403,
      "results models achieve competitive results": 101956,
      "recent work natural language processing": 96579,
      "natural language processing nlp focused": 76447,
      "learning approaches natural language processing": 62362,
      "language understanding dialogue state tracking": 60217,
      "major tasks natural language processing": 66590,
      "natural language processing computational linguistics": 76412,
      "research attention natural language processing": 100421,
      "verify effectiveness proposed method conduct": 131585,
      "effectiveness proposed method conduct extensive": 34934,
      "proposed method conduct extensive experiments": 93357,
      "deep neural networks widely used": 28387,
      "augmentation natural language understanding data": 10291,
      "language understanding intent detection slot": 60229,
      "understanding intent detection slot filling": 126866,
      "slot filling dialogue state tracking": 110042,
      "benchmarks indicate model significantly outperform": 12911,
      "using largescale pretrained language models": 129796,
      "abstract meaning representation amr parsing": 1772,
      "method outperforms previous methods large": 69019,
      "outperforms previous methods large margin": 82938,
      "previous stateoftheart methods large margin": 90474,
      "mechanism neural machine translation neural": 68020,
      "using pretrained language models bert": 130048,
      "research area natural language processing": 100414,
      "stateoftheart natural language processing nlp": 112771,
      "natural language processing tasks inspired": 76497,
      "code data models publicly available": 18081,
      "gaussian process preference learning gppl": 46552,
      "neural network based approaches process": 78256,
      "network based approaches process formation": 77157,
      "based approaches process formation splitting": 11519,
      "neural machine translation nmt approaches": 78044,
      "large amounts highquality training data": 61012,
      "model achieves better performance stateoftheart": 70562,
      "simulated multiple reference training smrt": 109642,
      "visual question answering vqa tasks": 131805,
      "language processing tasks including text": 59958,
      "number interesting findings help humans": 80902,
      "interesting findings help humans better": 56076,
      "findings help humans better analyze": 44299,
      "help humans better analyze understand": 49735,
      "humans better analyze understand improve": 51072,
      "models based observations propose new": 72811,
      "language generation nlg critical component": 59084,
      "neural machine translation nmt aims": 78043,
      "require large amounts annotated data": 100157,
      "comprehension machine reading comprehension mrc": 20194,
      "machine answer questions based given": 65743,
      "answer questions based given context": 6566,
      "natural language inference nli fundamental": 76337,
      "neural network model joint partofspeech": 78332,
      "network model joint partofspeech pos": 77331,
      "model joint partofspeech pos tagging": 71382,
      "joint partofspeech pos tagging dependency": 57309,
      "language model pretraining proven useful": 59380,
      "large pretrained language models bert": 61219,
      "pretrained language models plms like": 90093,
      "language models plms like bert": 59623,
      "paper propose task relation classification": 84334,
      "language models pretrained large unannotated": 59631,
      "widely used neural machine translation": 132582,
      "used neural machine translation nmt": 128656,
      "important role natural language processing": 52245,
      "role natural language processing tasks": 103208,
      "brings substantial improvements strong baselines": 14656,
      "advances natural language processing nlp": 4001,
      "long short term memory network": 65097,
      "lowresource languages neural machine translation": 65523,
      "relation extraction predict relation type": 98011,
      "extraction predict relation type entities": 42440,
      "predict relation type entities mentioned": 88926,
      "bert devlin et al 2019": 13099,
      "natural language inference nli tasks": 76341,
      "demonstrate effectiveness universality proposed approach": 28726,
      "bleu points respectively source code": 14312,
      "recent success language model pretraining": 96536,
      "delivers new stateoftheart results various": 28627,
      "new stateoftheart results various crosslingual": 79175,
      "language understanding tasks text classification": 60286,
      "universal conceptual cognitive annotation ucca": 127279,
      "conceptual cognitive annotation ucca abend": 20660,
      "cognitive annotation ucca abend rappoport": 18266,
      "annotation ucca abend rappoport 2013": 6402,
      "shared task covid19 fake news": 108033,
      "task covid19 fake news detection": 118033,
      "covid19 fake news detection english": 24565,
      "natural language inference question answering": 76347,
      "covering text classification sequence labeling": 24550,
      "entity recognition ner tasks natural": 38052,
      "integral natural language processing systems": 55743,
      "natural language processing nlp provides": 76460,
      "language models based recurrent neural": 59511,
      "tagging named entity recognition text": 117421,
      "named entity recognition text classification": 76122,
      "high resource languages like english": 50133,
      "convolutional neural network cnn conditional": 23183,
      "neural network cnn conditional random": 78282,
      "network cnn conditional random field": 77192,
      "cnn conditional random field crf": 17999,
      "machine translation documentlevel machine translation": 65958,
      "encoder representations transformers bert model": 36571,
      "xlmr conneau et al 2020": 135203,
      "transformerbased language models achieved state": 124417,
      "language models achieved state art": 59500,
      "labeled data required training new": 58442,
      "data required training new domain": 26348,
      "deep learning based model uses": 28257,
      "extensive experiments public datasets experimental": 41927,
      "experiments public datasets experimental results": 41095,
      "public datasets experimental results demonstrate": 94254,
      "results demonstrate proposed model outperforms": 101705,
      "word error rate wer reduction": 133272,
      "hindienglish code mixed tweets using": 50483,
      "code mixed tweets using bilingual": 18114,
      "mixed tweets using bilingual word": 70406,
      "tweets using bilingual word embeddings": 126072,
      "using bilingual word embeddings derived": 129414,
      "bilingual word embeddings derived fasttext": 14073,
      "word embeddings derived fasttext word2vec": 133080,
      "embeddings derived fasttext word2vec approaches": 35644,
      "various deep learning models including": 131078,
      "deep learning models including cnns": 28288,
      "learning models including cnns lstms": 62783,
      "models including cnns lstms bidirectional": 73381,
      "including cnns lstms bidirectional lstms": 53274,
      "cnns lstms bidirectional lstms attention": 18041,
      "topic models latent dirichlet allocation": 122553,
      "models latent dirichlet allocation lda": 73471,
      "neural machine translation models learn": 78031,
      "tasks morphological probing pos tagging": 119311,
      "morphological probing pos tagging ner": 74723,
      "masked language modeling mlm task": 67294,
      "machine translation nmt systems translate": 66123,
      "identifying protected health information phi": 51616,
      "experimental results proposed model achieves": 40681,
      "artificial intelligence ai machine learning": 9249,
      "intelligence ai machine learning ml": 55825,
      "bidirectional encoder representations transformers achieved": 13917,
      "used natural language processing applications": 128649,
      "natural language processing applications general": 76394,
      "propose progressive selfsupervised attention learning": 93006,
      "progressive selfsupervised attention learning approach": 92194,
      "activemisleading influence correctincorrect prediction instance": 3129,
      "augment conventional training objective regularization": 10254,
      "conventional training objective regularization term": 22910,
      "automatic speech recognition asr models": 10660,
      "aspect category sentiment analysis acsa": 9325,
      "modern standard arabic msa dialectal": 74419,
      "standard arabic msa dialectal arabic": 112191,
      "multilingual training neural machine translation": 75389,
      "training neural machine translation systems": 123737,
      "multilingual neural machine translation mnmt": 75309,
      "machine reading comprehension natural language": 65863,
      "proposed model achieves stateoftheart performance": 93434,
      "semeval2021 task toxic spans detection": 105562,
      "significantly outperforms multilingual bert mbert": 109020,
      "text generation pretrained language models": 121000,
      "yield performance improvements especially sequencelevel": 135350,
      "performance improvements especially sequencelevel metrics": 86449,
      "neural networks work propose new": 78610,
      "experimental results benchmark datasets demonstrate": 40564,
      "results benchmark datasets demonstrate effectiveness": 101543,
      "recurrent neural networks rnns transformers": 97246,
      "deep convolutional neural network cnn": 28226,
      "pretrained language models pretrained language": 90096,
      "language models pretrained language models": 59628,
      "model learns disentangle semantics syntax": 71445,
      "learns disentangle semantics syntax sentence": 63207,
      "intent classification slot filling tasks": 55899,
      "semeval2021 task multilingual crosslingual wordincontext": 105559,
      "task multilingual crosslingual wordincontext disambiguation": 118418,
      "multilingual crosslingual wordincontext disambiguation mclwic": 75233,
      "named entity recognition ner partofspeech": 76090,
      "large pretrained language models plms": 61221,
      "queries masked sentences paris capital": 94928,
      "masked sentences paris capital mask": 67301,
      "sentences paris capital mask used": 106439,
      "paris capital mask used probes": 84868,
      "model achieves stateoftheart rouge scores": 70607,
      "ae": 4204,
      "suspected": 116220,
      "hood": 50624,
      "exams": 39900,
      "speechdriven": 111853,
      "trecstyle": 125571,
      "interrogative": 56299,
      "rectify": 97151,
      "specialist": 111379,
      "top1": 122481,
      "nonnegative": 80249,
      "mctest": 67586,
      "lstmrnn": 65695,
      "maxpooling": 67557,
      "x1": 135186,
      "entertainment": 37696,
      "leaning": 61981,
      "510": 1092,
      "brnn": 14664,
      "selectionbased": 104851,
      "718": 1222,
      "200000": 540,
      "hermann": 49842,
      "endeavor": 36848,
      "ham": 49306,
      "casual": 15682,
      "yahoo": 135228,
      "graders": 48593,
      "fora": 45010,
      "barely": 11430,
      "paperno": 84496,
      "quiz": 95408,
      "engineer": 37038,
      "jiang": 57236,
      "skim": 109977,
      "expedite": 40406,
      "identifier": 51465,
      "832": 1310,
      "openness": 82003,
      "pasupat": 85660,
      "wikitablequestions": 132695,
      "387": 985,
      "neelakantan": 76904,
      "url": 127850,
      "multiround": 75771,
      "suppression": 116048,
      "enthusiasts": 37698,
      "evidences": 39682,
      "nonadversarial": 80150,
      "newsqa": 79414,
      "466": 1046,
      "conjunctions": 21070,
      "mult": 75024,
      "policygradient": 87975,
      "ex": 39708,
      "684": 1194,
      "dev": 30165,
      "yang": 135230,
      "426": 1022,
      "kernelbased": 57534,
      "closes": 17879,
      "conductor": 20965,
      "intensively": 55889,
      "regulating": 97777,
      "713": 1221,
      "567": 1123,
      "preselected": 89345,
      "truncation": 125869,
      "trivially": 125832,
      "samplespecific": 103598,
      "richardson": 102801,
      "overflow": 83321,
      "zhihu": 135543,
      "repetitions": 98916,
      "legibility": 63319,
      "syntacticbased": 116522,
      "cyber": 25479,
      "indexingbased": 53811,
      "silverstandard": 109069,
      "agricultural": 4414,
      "unusually": 127772,
      "traceable": 122718,
      "questionfocused": 95262,
      "multifactor": 75105,
      "relieves": 98667,
      "16m": 393,
      "clarifications": 17023,
      "exam": 39729,
      "infobox": 54330,
      "authoritative": 10374,
      "nips": 79494,
      "costefficient": 24381,
      "inductively": 54053,
      "piqa": 87570,
      "narrating": 76164,
      "narration": 76165,
      "stocks": 113367,
      "180k": 424,
      "wordcharacter": 133704,
      "downsampled": 33983,
      "repurposed": 100096,
      "weaving": 132218,
      "clarifying": 17027,
      "askubuntu": 9311,
      "lexicalize": 63850,
      "modulate": 74472,
      "heilman": 49679,
      "smith": 110259,
      "topranking": 122692,
      "nlvr": 79817,
      "compete": 19604,
      "traverse": 125546,
      "30000": 911,
      "unanswerable": 126589,
      "answerable": 6595,
      "abstain": 1759,
      "subwordaugmented": 115043,
      "generatorevaluator": 47785,
      "answerability": 6594,
      "verifier": 131575,
      "742": 1239,
      "28th": 868,
      "xie": 135191,
      "confuse": 21048,
      "ubiquitously": 126514,
      "707": 1217,
      "repurposing": 100098,
      "i2b2": 51294,
      "duplicates": 34272,
      "semanticrich": 105392,
      "openbookqa": 81952,
      "querylike": 95007,
      "lime": 64151,
      "subdivided": 114662,
      "wikianswers": 132626,
      "113k": 238,
      "machinereading": 66328,
      "20000": 539,
      "master": 67348,
      "impart": 51912,
      "nonextractive": 80204,
      "forth": 45299,
      "rocstories": 103147,
      "openset": 82013,
      "emphposterior": 36146,
      "inroads": 55504,
      "csa": 25186,
      "task11": 118864,
      "addsent": 3820,
      "ui": 126545,
      "fragility": 45370,
      "neglects": 76974,
      "solvers": 110634,
      "activated": 3097,
      "multistyle": 75804,
      "rqa": 103331,
      "argumentative": 8968,
      "argumentatively": 8974,
      "bidirectionally": 13985,
      "6444": 1177,
      "aggravated": 4342,
      "countering": 24428,
      "cove": 24479,
      "todate": 122234,
      "brittleness": 14663,
      "interrelationships": 56297,
      "turker": 125955,
      "onehop": 81673,
      "twohop": 126144,
      "hops": 50653,
      "2hop": 883,
      "hypothetical": 51293,
      "rounds": 103312,
      "minmax": 70269,
      "min": 70142,
      "quizbowl": 95409,
      "hearing": 49642,
      "exhibition": 40010,
      "vibrant": 131646,
      "coordinating": 23263,
      "236": 796,
      "conll04": 21092,
      "resume": 102353,
      "singlestep": 109842,
      "exaction": 39723,
      "singledocument": 109823,
      "accumulate": 2051,
      "wikihop": 132635,
      "curiosity": 25251,
      "unprompted": 127494,
      "nonfactoid": 80205,
      "bm25": 14386,
      "squash": 112084,
      "frodo": 45898,
      "1978": 453,
      "datasparse": 27807,
      "01": 5,
      "indexable": 53807,
      "indowordnet": 53999,
      "bypassing": 14994,
      "270k": 857,
      "twoagent": 126134,
      "multispan": 75794,
      "799": 1277,
      "24k": 809,
      "890": 1353,
      "pathology": 85685,
      "charts": 16650,
      "rankers": 95655,
      "consumes": 21725,
      "5001000": 1086,
      "coin": 18340,
      "compulsory": 20341,
      "commonsenseqa": 19014,
      "159": 371,
      "attentionoverattention": 10103,
      "superhuman": 115675,
      "incredibly": 53718,
      "70000": 1214,
      "reasoner": 96219,
      "biobert": 14150,
      "hazards": 49573,
      "postprocess": 88505,
      "qabased": 94526,
      "typeaware": 126232,
      "supervisions": 115933,
      "deeplift": 28463,
      "subtraction": 115007,
      "iterates": 57117,
      "802": 1293,
      "singlehop": 109828,
      "multimodule": 75466,
      "ccks2019": 15867,
      "2233": 780,
      "automobile": 10851,
      "coordinators": 23267,
      "qc": 94528,
      "hgn": 49889,
      "875": 1342,
      "whyquestions": 132484,
      "patientspecific": 85709,
      "questiondriven": 95260,
      "anchors": 6088,
      "anchoring": 6087,
      "wechat": 132310,
      "dirty": 31926,
      "misunderstandings": 70353,
      "worsens": 135070,
      "regulation": 97778,
      "judicial": 57453,
      "lawyer": 61692,
      "easytohard": 34492,
      "furthering": 46140,
      "551": 1119,
      "prlm": 90790,
      "83k": 1316,
      "deterministically": 30163,
      "incoming": 53414,
      "caches": 15010,
      "sim": 109070,
      "budgets": 14735,
      "prerequisites": 89343,
      "longtext": 65209,
      "discards": 32009,
      "markert": 67236,
      "bashi": 12511,
      "logicguided": 65028,
      "robertabased": 103004,
      "oversensitivity": 83355,
      "pseudotraining": 94198,
      "crossmodal": 25060,
      "semanticlevel": 105391,
      "centric": 15917,
      "latently": 61641,
      "743": 1240,
      "579": 1129,
      "shortanswer": 108273,
      "613": 1161,
      "kth": 58355,
      "unanimously": 126582,
      "resourceefficient": 100894,
      "exactmatch": 39726,
      "520": 1097,
      "knearestneighbor": 57720,
      "miami": 70031,
      "excels": 39922,
      "longrunning": 65185,
      "pregnant": 89259,
      "babies": 11280,
      "rerankers": 100364,
      "multigrained": 75116,
      "metafeatures": 68538,
      "quote": 95415,
      "situating": 109885,
      "questionspecific": 95379,
      "underwent": 127039,
      "unstudied": 127590,
      "datacentric": 26652,
      "746": 1243,
      "revolves": 102669,
      "streamline": 113557,
      "pathbased": 85679,
      "regulate": 97775,
      "knowledgegraphbased": 58273,
      "decisive": 27970,
      "comve": 20531,
      "piaf": 87483,
      "understands": 127021,
      "detectable": 29823,
      "largevolume": 61548,
      "ocean": 81524,
      "greener": 49081,
      "weaklylabeled": 132201,
      "betterformed": 13775,
      "incurs": 53754,
      "sent": 105759,
      "pn": 87775,
      "tydiqa": 126181,
      "631": 1169,
      "4413": 1028,
      "startend": 112406,
      "23000": 793,
      "174": 407,
      "nba": 76687,
      "timelinebased": 122159,
      "531": 1102,
      "md": 67587,
      "mis": 70286,
      "sat": 103646,
      "19k": 471,
      "nationality": 76216,
      "ethnicity": 38753,
      "religion": 98668,
      "629": 1166,
      "52": 1096,
      "nutshell": 81036,
      "milestone": 70086,
      "delaying": 28590,
      "prompting": 92326,
      "reconsider": 97106,
      "absorbs": 1757,
      "multirole": 75770,
      "debating": 27896,
      "mcq": 67585,
      "chime": 16734,
      "blanc": 14261,
      "automl": 10850,
      "briefs": 14625,
      "selfexplainable": 104919,
      "13k": 308,
      "311": 923,
      "dpps": 34076,
      "determinantal": 30112,
      "urges": 127849,
      "crossed": 24907,
      "healthrelated": 49635,
      "medicalrelated": 68231,
      "ct": 25189,
      "postulates": 88530,
      "encounter": 36742,
      "reformulated": 97634,
      "departing": 29035,
      "outbound": 82611,
      "reevaluate": 97486,
      "knowledgeintensive": 58279,
      "invented": 56684,
      "presuppositions": 89969,
      "dietary": 30926,
      "allergies": 5027,
      "nutrition": 81035,
      "negations": 76920,
      "597": 1137,
      "aristotle": 9004,
      "priming": 90661,
      "refrain": 97642,
      "taxonomybased": 119670,
      "src": 112086,
      "searched": 104368,
      "ri": 102723,
      "nonbert": 80165,
      "intense": 55876,
      "selfteaching": 104953,
      "defenses": 28475,
      "defense": 28473,
      "testbeds": 120559,
      "defect": 28468,
      "onestop": 81724,
      "tenth": 120200,
      "respondent": 101183,
      "sufficiency": 115268,
      "720": 1225,
      "811": 1298,
      "377": 976,
      "falsenegative": 43046,
      "visionlanguage": 131774,
      "shortform": 108307,
      "humanauthored": 50999,
      "groundings": 49125,
      "clevr": 17745,
      "eqa": 38219,
      "deconfounding": 28164,
      "necessitating": 76766,
      "matching question": 67426,
      "question sentence": 95218,
      "extraction answer": 42262,
      "extraction ae": 42257,
      "document directly": 32991,
      "answer given": 6533,
      "user question": 129030,
      "entire documents": 37705,
      "generated knowledge": 47088,
      "retrieves relevant": 102468,
      "constructed combining": 21655,
      "resolution procedure": 100782,
      "answers queries": 6742,
      "big issue": 13993,
      "answers natural": 6735,
      "kinds questions": 57690,
      "ranked according": 95642,
      "provide evaluation": 93815,
      "results trec": 102288,
      "task set": 118703,
      "comprehension exams": 20181,
      "multiple answer": 75492,
      "analyze common": 5955,
      "common approaches": 18855,
      "term overlap": 120236,
      "sentence identification": 105898,
      "analyzing strengths": 6053,
      "results quantify": 102102,
      "transcription asr": 123995,
      "candidates target": 15147,
      "method robust": 69110,
      "trecstyle evaluation": 125572,
      "evaluation workshop": 39449,
      "method means": 68957,
      "means experiments": 67761,
      "answering natural": 6651,
      "behavior person": 12667,
      "answer yes": 6593,
      "place time": 87610,
      "identification algorithms": 51361,
      "recent technological": 96549,
      "known quite": 58319,
      "processing goal": 91674,
      "goal ai": 48327,
      "try answer": 125894,
      "user goal": 128991,
      "following sections": 44986,
      "new level": 78992,
      "questions present": 95343,
      "present opendomain": 89631,
      "learns lowdimensional": 63219,
      "lowdimensional embeddings": 65410,
      "used score": 128737,
      "pairs questions": 83623,
      "benchmark literature": 12836,
      "literature deep": 64750,
      "approach solving": 7925,
      "require feature": 100136,
      "engineering does": 37046,
      "making model": 66852,
      "languages experimental": 60569,
      "task employ": 118139,
      "model calculate": 70797,
      "extracting features": 42212,
      "algorithm train": 4809,
      "demonstrate question": 28856,
      "top1 accuracy": 122482,
      "algorithm works": 4818,
      "works significantly": 134978,
      "traditional method": 122835,
      "aspects semantic": 9406,
      "parsing question": 85205,
      "compositionality existing": 20143,
      "aspect paper": 9340,
      "domain results": 33629,
      "combinatorial explosion": 18605,
      "space logical": 111016,
      "forms propose": 45258,
      "improvements natural": 52878,
      "shown remarkable": 108520,
      "investigate recently": 56807,
      "consist different": 21348,
      "reasoning previous": 96292,
      "understanding semantics": 126953,
      "models inspired": 73405,
      "perform knowledge": 86019,
      "knowledge encoding": 57893,
      "inference achieve": 54110,
      "nearperfect accuracy": 76740,
      "reasoning path": 96289,
      "answering paper": 6658,
      "generate answers": 46903,
      "corpus questionanswer": 23959,
      "pairs associated": 83491,
      "varying degrees": 131257,
      "study measure": 114435,
      "measure effects": 67796,
      "sparse nonnegative": 111231,
      "nonnegative matrix": 80250,
      "automatically induce": 10787,
      "document various": 33107,
      "data instance": 26044,
      "questions new": 95335,
      "related resources": 97894,
      "resources facilitate": 100976,
      "knowledge reasoning": 58134,
      "4th grade": 1062,
      "grade science": 48587,
      "science exams": 103975,
      "collection curated": 18467,
      "facts tables": 42923,
      "resources useful": 101063,
      "type identification": 126204,
      "identification lexical": 51395,
      "primary challenges": 90647,
      "nlp machine": 79631,
      "systems ability": 116706,
      "understand text": 126780,
      "content work": 21978,
      "engineering approaches": 37045,
      "approaches come": 8100,
      "employs attention": 36327,
      "phrases key": 87435,
      "key sentences": 57597,
      "sentences key": 106364,
      "relevant answering": 98523,
      "approaches big": 8088,
      "network large": 77300,
      "recently cnn": 96670,
      "cnn daily": 18000,
      "daily mail": 25511,
      "size datasets": 109920,
      "deeplearning techniques": 28460,
      "techniques currently": 119858,
      "attention directly": 9820,
      "makes model": 66795,
      "models sets": 74014,
      "art evaluated": 9050,
      "evaluated datasets": 38968,
      "major goal": 66562,
      "goal natural": 48368,
      "partly limited": 85516,
      "focused mainly": 44864,
      "answer variety": 6591,
      "opposed using": 82149,
      "sentence fragments": 105879,
      "networks operate": 77682,
      "text trained": 121370,
      "designed help": 29597,
      "cope limited": 23275,
      "model sets": 71993,
      "15 absolute": 340,
      "structural decomposition": 113763,
      "semantic trees": 105338,
      "relevance score": 98517,
      "learned statistical": 62261,
      "modeling final": 72432,
      "documents containing": 33210,
      "modeling relational": 72527,
      "pairs convolutional": 83500,
      "propose convolutional": 92603,
      "representation question": 99391,
      "additional parameters": 3554,
      "interactions questions": 55998,
      "results clearly": 101580,
      "information allows": 54373,
      "approach state": 7931,
      "metric learning": 69883,
      "works using": 134989,
      "based multimodal": 11866,
      "performance surpassing": 86777,
      "map mrr": 67109,
      "dataset simple": 27204,
      "reasoning abilities": 96222,
      "scale news": 103742,
      "semantics help": 105424,
      "help machine": 49747,
      "efficiency training": 35049,
      "bases provide": 12507,
      "rich sources": 102790,
      "base propose": 11472,
      "questions knowledge": 95323,
      "embeddings proposed": 35889,
      "realworld text": 96184,
      "research goal": 100525,
      "based estimated": 11690,
      "experiments demonstrating": 40898,
      "benchmarks outperforming": 12929,
      "keyvalue memory": 57630,
      "networks directly": 77569,
      "documents able": 33173,
      "kbs suffer": 57512,
      "wikipedia documents": 132653,
      "analysis tool": 5901,
      "attentive convolutional": 10107,
      "fact form": 42824,
      "subject predicate": 114681,
      "predicate object": 88852,
      "task simple": 118718,
      "mention question": 68426,
      "characterlevel convolutional": 16559,
      "network charcnn": 77183,
      "makes main": 66791,
      "stateoftheart entity": 112657,
      "question representation": 95211,
      "representation effectively": 99218,
      "experiments sets": 41128,
      "task need": 118446,
      "simple questions": 109498,
      "factoid question": 42850,
      "attempts apply": 9762,
      "apply complex": 7167,
      "accuracy benchmark": 2106,
      "benchmark sets": 12858,
      "formulates task": 45290,
      "detecting entities": 29833,
      "question relation": 95210,
      "network solve": 77438,
      "yields substantial": 135445,
      "based complex": 11586,
      "efficiency modularity": 35034,
      "x1 entertainment": 135187,
      "entertainment platform": 37697,
      "millions users": 70127,
      "stanford question": 112377,
      "dataset squad": 27215,
      "text corresponding": 120839,
      "understand types": 126784,
      "trees build": 125684,
      "improvement simple": 52760,
      "simple baseline": 109373,
      "20 human": 505,
      "indicating dataset": 53870,
      "dataset presents": 27104,
      "answering recently": 6688,
      "recently bidirectional": 96667,
      "generate representation": 46999,
      "supervision loss": 115894,
      "brnn model": 14665,
      "problem answering": 90934,
      "problem requires": 91206,
      "present parsing": 89641,
      "use background": 127916,
      "programs execute": 92136,
      "diverse challenging": 32797,
      "reduce word": 97368,
      "domains systems": 33867,
      "systems compared": 116796,
      "providing strong": 94137,
      "results future": 101803,
      "novel largescale": 80618,
      "largescale natural": 61481,
      "publiclyavailable dataset": 94336,
      "million instances": 70100,
      "instances task": 55636,
      "corresponding wikipedia": 24313,
      "articles task": 9221,
      "task contains": 118017,
      "classification extraction": 17208,
      "stateoftheart dnnbased": 112644,
      "answering models": 6647,
      "model wordlevel": 72327,
      "model mechanism": 71510,
      "mechanism copy": 67965,
      "multiple choice": 75510,
      "choice reading": 16902,
      "comprehension problems": 20218,
      "problems constructed": 91310,
      "newswire corpus": 79419,
      "features contrast": 43426,
      "contrast cnn": 22689,
      "mail datasets": 66391,
      "hermann et": 49843,
      "question formation": 95159,
      "events used": 39609,
      "form question": 45115,
      "question second": 95216,
      "person named": 87100,
      "dataset challenge": 26781,
      "task community": 117981,
      "spoken content": 111965,
      "comprehension test": 20238,
      "presents attractive": 89821,
      "information plain": 54843,
      "highly attractive": 50293,
      "understand spoken": 126777,
      "summarize key": 115605,
      "information humans": 54659,
      "english learners": 37191,
      "task achieving": 117838,
      "wordlevel attention": 133729,
      "improved machine": 52614,
      "content proposed": 21927,
      "treestructured sequential": 125731,
      "framework question": 45666,
      "especially question": 38495,
      "corpus selection": 23984,
      "model sentences": 71977,
      "integrate character": 55746,
      "training compared": 123387,
      "different benchmarks": 31028,
      "selection datasets": 104777,
      "datasets exhibit": 27460,
      "minimal supervision": 70178,
      "supervision common": 115875,
      "model question": 71837,
      "related question": 97891,
      "lexical models": 63785,
      "argue better": 8910,
      "better approach": 13511,
      "relevant way": 98600,
      "steps generate": 113319,
      "causeeffect pairs": 15835,
      "patterns second": 85779,
      "second train": 104461,
      "embeddings data": 35629,
      "contexts context": 22384,
      "finally extend": 44187,
      "extend stateoftheart": 41798,
      "embeddings evaluate": 35699,
      "yahoo answers": 135229,
      "generation knowledge": 47449,
      "questions related": 95353,
      "propose generating": 92702,
      "fluent natural": 44713,
      "questions kb": 95322,
      "significantly reduces": 109047,
      "question set": 95219,
      "generated applying": 47048,
      "handcrafted templates": 49357,
      "relevance evaluated": 98508,
      "answering english": 6625,
      "arabic paper": 8542,
      "studies impact": 114238,
      "features applied": 43362,
      "applied learning": 7088,
      "questions community": 95287,
      "released semeval2016": 98495,
      "task targeted": 118776,
      "bagofwords features": 11374,
      "features bow": 43384,
      "syntactic tree": 116495,
      "tree kernels": 125598,
      "features best": 43379,
      "knowledge structural": 58191,
      "reranking task": 100379,
      "model paraphrase": 71707,
      "baseline given": 12230,
      "accurate model": 2343,
      "case arabic": 15576,
      "task time": 118792,
      "applied tree": 7136,
      "tasks obtained": 119347,
      "obtained second": 81409,
      "subtasks english": 114967,
      "datasets test": 27755,
      "like recent": 64085,
      "paperno et": 84497,
      "broader context": 14699,
      "problem apply": 90936,
      "choose word": 16927,
      "100 instances": 161,
      "resolution external": 100763,
      "knowledge needed": 58080,
      "knowledge questions": 58131,
      "novel problem": 80691,
      "evaluate knowledge": 38845,
      "knowledge specific": 58182,
      "approach approach": 7362,
      "approach selects": 7873,
      "question desired": 95147,
      "finally approach": 44148,
      "structured query": 114035,
      "question key": 95171,
      "human users": 50985,
      "historical data": 50518,
      "suitable features": 115399,
      "variable lengths": 130841,
      "neural rc": 78643,
      "focused predicting": 44872,
      "encodes document": 36668,
      "mechanism acquire": 67942,
      "representations document": 99605,
      "followed generation": 44948,
      "module propose": 74507,
      "answer experimental": 6528,
      "squad dataset": 112076,
      "intents expressed": 55923,
      "answering focused": 6630,
      "conversation humans": 22957,
      "qa setting": 94520,
      "tables wikipedia": 117271,
      "systems face": 116901,
      "major problems": 66579,
      "evaluated dataset": 38967,
      "question corresponding": 95142,
      "corresponding entries": 24286,
      "conclude proposing": 20716,
      "answering reading": 6684,
      "given evidence": 48031,
      "central problem": 15904,
      "typically focused": 126431,
      "arbitrary strings": 8592,
      "task presenting": 118552,
      "spans evidence": 111200,
      "representations significantly": 99891,
      "improves best": 52952,
      "wang jiang": 132010,
      "baseline 50": 12176,
      "query given": 94966,
      "requires modeling": 100300,
      "modeling complex": 72398,
      "interactions context": 55982,
      "attention focus": 9840,
      "small portion": 110187,
      "context summarize": 22285,
      "uses bidirectional": 129198,
      "summarization experimental": 115505,
      "document identify": 33021,
      "fast model": 43144,
      "model selecting": 71959,
      "relevant sentences": 98578,
      "selection latent": 104794,
      "performance challenging": 86200,
      "ibm watson": 51303,
      "understand large": 126755,
      "pairs large": 83571,
      "significantly limits": 108973,
      "training recently": 123805,
      "burden manual": 14969,
      "tried use": 125776,
      "authors knowledge": 10383,
      "attempt use": 9755,
      "use largescale": 128119,
      "insights lessons": 55539,
      "vectors neural": 131454,
      "recurrent entity": 97175,
      "entity networks": 37978,
      "longterm memory": 65208,
      "state world": 112541,
      "neural turing": 78725,
      "turing machine": 125952,
      "fixed size": 44605,
      "size memory": 109934,
      "obtains competitive": 81461,
      "performance reading": 86655,
      "matching machine": 67410,
      "ability current": 1502,
      "techniques newly": 119944,
      "newly released": 79278,
      "beginning ending": 12636,
      "lstm aggregate": 65590,
      "aggregate information": 4348,
      "using paragraph": 129999,
      "generation technique": 47675,
      "91 accuracy": 1383,
      "models experimented": 73202,
      "accuracy remaining": 2261,
      "model improvements": 71314,
      "question analysis": 95014,
      "systems carry": 116771,
      "analysis uses": 5929,
      "uses techniques": 129291,
      "question focus": 95158,
      "extraction components": 42285,
      "analysis detect": 5572,
      "important elements": 52154,
      "identify type": 51569,
      "answer types": 6586,
      "types elements": 126274,
      "accurate detection": 2330,
      "using mean": 129871,
      "mean reciprocal": 67603,
      "answering transfer": 6718,
      "task question": 118594,
      "significantly benefit": 108881,
      "results visual": 102321,
      "analysis similar": 5856,
      "task iterative": 118319,
      "attention multiple": 9939,
      "information needs": 54794,
      "intelligent agent": 55841,
      "proper way": 92428,
      "user profile": 129023,
      "relevant answers": 98525,
      "based artificial": 11524,
      "dataset assessing": 26746,
      "assessing performance": 9516,
      "language support": 60135,
      "learns solve": 63237,
      "simultaneously tasks": 109695,
      "proposed semeval2016": 93545,
      "exploit previous": 41436,
      "learns encode": 63209,
      "questions comments": 95285,
      "challenge test": 16110,
      "produces higher": 91997,
      "convergence rates": 22924,
      "use manual": 128136,
      "approaches state": 8350,
      "explore deep": 41535,
      "approach weakly": 8014,
      "following work": 44990,
      "liang 2015": 63956,
      "character convolutional": 16423,
      "interpretation question": 56273,
      "wikitablequestions dataset": 132696,
      "score task": 104132,
      "neelakantan et": 76905,
      "learning reasoning": 62941,
      "reasoning understanding": 96318,
      "correctly answering": 24172,
      "little human": 64809,
      "trees machine": 125698,
      "utilizes word": 130564,
      "embeddings explicitly": 35712,
      "explicitly taking": 41395,
      "advantage structured": 4062,
      "trees paper": 125701,
      "algorithm framework": 4742,
      "utilize structured": 130530,
      "information encode": 54522,
      "performance algorithms": 86132,
      "comprehension evaluate": 20180,
      "model squad": 72078,
      "accurately identify": 2383,
      "extract answers": 42066,
      "syntactically coherent": 116505,
      "embeddings representation": 35911,
      "larger impact": 61368,
      "impact architectural": 51857,
      "systematically explore": 116695,
      "qa paper": 94512,
      "problems closely": 91305,
      "modelling questions": 72618,
      "framework introduce": 45583,
      "help encode": 49715,
      "adaptation models": 3234,
      "models stanford": 74087,
      "baseline question": 12290,
      "wikipedia structured": 132677,
      "kbs work": 57513,
      "twostep approach": 126168,
      "handcrafted model": 49353,
      "relevant articles": 98527,
      "dataset reducing": 27146,
      "questions paper": 95339,
      "article task": 9164,
      "challenges document": 16149,
      "finding relevant": 44283,
      "trained detect": 123114,
      "combination effective": 18560,
      "effective complete": 34648,
      "automatic question": 10634,
      "sentence related": 106022,
      "questions work": 95376,
      "questions natural": 95332,
      "reads input": 96044,
      "fed decoder": 43800,
      "generate answer": 46902,
      "conduct preliminary": 20886,
      "produce fluent": 91891,
      "augmented context": 10323,
      "article generate": 9133,
      "approach built": 7406,
      "pairs pair": 83598,
      "believe valuable": 12705,
      "test baseline": 120422,
      "selection deep": 104778,
      "question machine": 95177,
      "task models": 118402,
      "model reflect": 71880,
      "second pass": 104439,
      "novel information": 80601,
      "information fusion": 54627,
      "flow model": 44685,
      "novel layer": 80619,
      "representation intermediate": 99268,
      "bidaf model": 13897,
      "model multihop": 71535,
      "attention structure": 10018,
      "matches surpasses": 67390,
      "surpasses performance": 116107,
      "systems question": 117092,
      "answering knowledge": 6643,
      "universal schema": 127329,
      "methods good": 69525,
      "kb unstructured": 57504,
      "support reasoning": 115987,
      "reasoning union": 96320,
      "combination text": 18589,
      "pairs evaluation": 83535,
      "propose recurrent": 93011,
      "standard maximum": 112256,
      "training finetune": 123636,
      "techniques maximize": 119930,
      "generation means": 47468,
      "means improve": 67765,
      "attention deficiency": 9813,
      "new optimization": 79068,
      "optimization approach": 82188,
      "learning introduced": 62661,
      "extend standard": 41797,
      "experiments stanford": 41152,
      "trivia enthusiasts": 125827,
      "comparison recently": 19569,
      "introduced largescale": 56579,
      "complex compositional": 19803,
      "syntactic lexical": 116429,
      "questions corresponding": 95293,
      "cross sentence": 24866,
      "close human": 17818,
      "challenging testbed": 16339,
      "significant future": 108755,
      "future study": 46315,
      "dual tasks": 34242,
      "given answer": 47983,
      "useful evidence": 128884,
      "leverages probabilistic": 63656,
      "guide training": 49249,
      "learning qa": 62930,
      "network components": 77201,
      "involved models": 56884,
      "framework improves": 45568,
      "tasks improved": 119171,
      "generation machine": 47465,
      "dataset unlike": 27252,
      "dataset aims": 26728,
      "defines task": 28510,
      "results specifically": 102199,
      "features question": 43681,
      "networks single": 77772,
      "comprehension propose": 20219,
      "additional task": 3580,
      "extraction multiple": 42397,
      "networks extracted": 77600,
      "methods crosslanguage": 69413,
      "networks application": 77506,
      "crosslanguage adaptation": 24920,
      "trained input": 123162,
      "given labeled": 48052,
      "language particular": 59804,
      "use adversarial": 127889,
      "input languages": 55354,
      "results sizable": 102192,
      "networks transfer": 77799,
      "documents domain": 33222,
      "domain use": 33688,
      "indomain models": 53970,
      "combined global": 18656,
      "focused augmenting": 44848,
      "networks simple": 77771,
      "similarity question": 109289,
      "embeddings time": 35981,
      "using separate": 130153,
      "language involves": 59226,
      "multiple pieces": 75640,
      "pieces evidence": 87497,
      "datasets semantic": 27696,
      "document paper": 33049,
      "models comparing": 72938,
      "queries web": 94945,
      "designed focus": 29593,
      "compositional language": 20115,
      "35 f1": 962,
      "stateoftheart analysis": 112578,
      "requires building": 100247,
      "capable providing": 15215,
      "recognizes entities": 97066,
      "network exploiting": 77250,
      "daily news": 25512,
      "designing systems": 29650,
      "human does": 50794,
      "embedded using": 35364,
      "score pair": 104101,
      "best answer": 13304,
      "chat data": 16654,
      "sample selection": 103562,
      "learning question": 62931,
      "domain domain": 33512,
      "domain useful": 33690,
      "gathering new": 46529,
      "method question": 69089,
      "effectiveness transfer": 34968,
      "performance corpora": 86259,
      "addition using": 3485,
      "task unified": 118819,
      "model follows": 71201,
      "decoder framework": 28011,
      "framework encoder": 45518,
      "query input": 94967,
      "attentionbased long": 10075,
      "copy coverage": 23286,
      "coverage mechanisms": 24515,
      "policygradient reinforcement": 87976,
      "baselines generative": 12405,
      "rapid progress": 95722,
      "progress question": 92177,
      "extractive qa": 42555,
      "end word": 36837,
      "space search": 111056,
      "search makes": 104327,
      "model globally": 71255,
      "models considered": 72964,
      "hierarchical gated": 49962,
      "tensor network": 120189,
      "yang et": 135231,
      "answers question": 6743,
      "popular recently": 88122,
      "attracted lot": 10156,
      "chinese reading": 16812,
      "community proposed": 19087,
      "data humanannotated": 26008,
      "set dataset": 107413,
      "questions different": 95300,
      "particular explore": 85413,
      "crosslanguage word": 24921,
      "degrade performance": 28558,
      "compares original": 19498,
      "closes performance": 17881,
      "gap respect": 46478,
      "respect monolingual": 101092,
      "use represent": 128244,
      "represent arabic": 99099,
      "network overall": 77379,
      "shows better": 108557,
      "task wide": 118843,
      "order consider": 82300,
      "novel group": 80589,
      "set propose": 107546,
      "representation respect": 99396,
      "traditional cnns": 122807,
      "cnns proposed": 18044,
      "intensively studied": 55890,
      "model research": 71916,
      "models consists": 72969,
      "consists multiple": 21484,
      "fusion layers": 46239,
      "dotproduct attention": 33957,
      "simultaneously encoding": 109674,
      "dynamic changes": 34296,
      "produce globally": 91893,
      "achieve score": 2530,
      "years researchers": 135296,
      "success machine": 115093,
      "means text": 67770,
      "completely change": 19779,
      "yield different": 135335,
      "answers paper": 6739,
      "optimization model": 82197,
      "model answer": 70667,
      "output multiple": 83096,
      "final answer": 44096,
      "baselines challenging": 12367,
      "learning reading": 62940,
      "multiple evidence": 75560,
      "documents answer": 33182,
      "scale entire": 103718,
      "selecting single": 104759,
      "focusing specific": 44929,
      "making trivially": 66869,
      "trivially parallelizable": 125833,
      "approach scale": 7866,
      "answer candidate": 6518,
      "candidate document": 15102,
      "empirically approach": 36216,
      "architectures empirical": 8803,
      "attention uses": 10030,
      "uses reinforcement": 129271,
      "dynamically control": 34344,
      "improves fixed": 52983,
      "types particularly": 126335,
      "stateoftheart datasets": 112629,
      "datasets neural": 27594,
      "supervised language": 115762,
      "processing requires": 91785,
      "detection question": 30030,
      "question type": 95231,
      "knowledge leads": 58044,
      "steps compared": 113317,
      "effective small": 34746,
      "performance transfer": 86810,
      "learning fast": 62578,
      "natural fit": 76256,
      "models latency": 73467,
      "texts present": 121582,
      "alternative recurrent": 5241,
      "dilated convolutional": 31737,
      "time achieving": 121978,
      "fusion network": 46245,
      "attention process": 9980,
      "questions answer": 95270,
      "learning input": 62656,
      "input sample": 55419,
      "attention strategy": 10017,
      "strategy apply": 113504,
      "result reported": 101400,
      "mrc dataset": 74924,
      "variety types": 131030,
      "produce attention": 91875,
      "mrc models": 74928,
      "largescale opendomain": 61489,
      "mrc datasets": 74925,
      "dataset far": 26928,
      "make improvements": 66685,
      "encourage exploration": 36753,
      "successful tasks": 115172,
      "wellstudied paper": 132435,
      "investigate transferability": 56819,
      "dataset target": 27231,
      "stateoftheart target": 112998,
      "target datasets": 117598,
      "learning helpful": 62629,
      "available deep": 10979,
      "communities paper": 19062,
      "framework identifying": 45561,
      "network combines": 77200,
      "user feature": 128983,
      "representations question": 99836,
      "work method": 134632,
      "embedding feature": 35406,
      "domains proposed": 33837,
      "framework trained": 45718,
      "dataset stack": 27216,
      "stack overflow": 112120,
      "best baseline": 13309,
      "generative approach": 47716,
      "frame problem": 45385,
      "problem generative": 91068,
      "task encoder": 118142,
      "encoder network": 36545,
      "models relationship": 73910,
      "encoding vector": 36734,
      "affect overall": 4221,
      "coverage vector": 24525,
      "vector model": 131329,
      "examples improved": 39845,
      "graphs neural": 48965,
      "problem question": 91192,
      "problem entity": 91031,
      "linking relation": 64670,
      "simple strong": 109523,
      "lstms grus": 65711,
      "results gains": 101805,
      "used reinforcement": 128723,
      "unique feature": 127186,
      "feature use": 43323,
      "layer neural": 61731,
      "training simple": 123865,
      "achieves results": 2848,
      "content attracted": 21858,
      "attracted substantial": 10162,
      "evaluate importance": 38842,
      "rich contextualized": 102737,
      "model allowing": 70658,
      "model choose": 70823,
      "contextindependent word": 22371,
      "dataset syntactic": 27228,
      "text contribute": 120821,
      "educational activities": 34571,
      "applications goal": 6933,
      "documents set": 33291,
      "sentences domain": 106287,
      "refer source": 97503,
      "sentences slightly": 106498,
      "knowledge simple": 58174,
      "contrary existing": 22680,
      "transform source": 124256,
      "applying series": 7267,
      "evaluated domain": 38971,
      "using series": 130163,
      "pipeline separately": 87554,
      "endtoend proposed": 36954,
      "analyses corpora": 5447,
      "types answer": 126249,
      "categories corpora": 15733,
      "corpora analyzed": 23408,
      "method creation": 68751,
      "using entire": 129638,
      "presented analysis": 89776,
      "suggests better": 115370,
      "selection introduce": 104789,
      "dedicated study": 28182,
      "authors different": 10379,
      "different backgrounds": 31023,
      "service dataset": 107321,
      "questions challenging": 95282,
      "seeking information": 104522,
      "resources suitable": 101048,
      "aim gap": 4457,
      "challenges computational": 16141,
      "models gives": 73301,
      "gives reasonable": 48206,
      "answering challenging": 6608,
      "reasoning multiple": 96277,
      "fact triples": 42836,
      "triples knowledge": 125814,
      "network employs": 77237,
      "dynamically decides": 34346,
      "predicts relation": 89218,
      "relation corresponds": 97963,
      "predictions reasoning": 89185,
      "reasoning analysis": 96227,
      "predicting final": 88986,
      "attentive recurrent": 10125,
      "semantic gap": 105062,
      "gap sentence": 46479,
      "minimize gap": 70195,
      "models augmenting": 72787,
      "network solving": 77440,
      "solving lexical": 110643,
      "answering introduce": 6640,
      "introduce tokenlevel": 56554,
      "space introduce": 111009,
      "optimization training": 82216,
      "datasets outperforming": 27606,
      "tasks best": 118966,
      "networks random": 77721,
      "duplicate question": 34268,
      "question pairs": 95193,
      "pairs determining": 83515,
      "given questions": 48099,
      "highly used": 50352,
      "forest adaboost": 45048,
      "got best": 48532,
      "result using": 101419,
      "forest classifier": 45050,
      "quora question": 95411,
      "task takes": 118774,
      "answer span": 6581,
      "interaction scenarios": 55967,
      "develop generative": 30201,
      "generative extractive": 47726,
      "extractive approaches": 42548,
      "generates words": 47193,
      "extractive approach": 42547,
      "designed measure": 29609,
      "relevance question": 98515,
      "multifactor attention": 75106,
      "answering neural": 6654,
      "focus capturing": 44742,
      "distributed multiple": 32614,
      "crucial achieving": 25129,
      "aggregation mechanism": 4362,
      "words question": 134152,
      "improvements best": 52815,
      "best prior": 13421,
      "model relation": 71886,
      "role knowledge": 103180,
      "methods follow": 69516,
      "question candidate": 95135,
      "max average": 67495,
      "fixeddimensional vectors": 44614,
      "similar traditional": 109161,
      "sequences vectors": 107145,
      "vectors instead": 131444,
      "relation learned": 98048,
      "compared convolutional": 19352,
      "finally performing": 44214,
      "lowlevel representations": 65481,
      "fixeddimensional vector": 44613,
      "vector comparison": 131300,
      "results relation": 102122,
      "accuracy demonstrating": 2135,
      "sequential question": 107246,
      "pairs knowledge": 83567,
      "humans typically": 51114,
      "questions significant": 95366,
      "need study": 76862,
      "scenarios faced": 103849,
      "involving tasks": 56920,
      "task complex": 117993,
      "sequential qa": 107245,
      "contain simple": 21759,
      "quantitative comparative": 94863,
      "parse complex": 84881,
      "resolve coreferences": 100800,
      "utterances iii": 130647,
      "queries finally": 94921,
      "models clearly": 72909,
      "object types": 81056,
      "architecture paired": 8721,
      "copy action": 23284,
      "questions benchmark": 95281,
      "computer aided": 20477,
      "diagnosis clinical": 30504,
      "clinical medicine": 17771,
      "problem field": 91057,
      "nlp work": 79791,
      "questionanswering task": 95258,
      "medicine using": 68238,
      "knowledge largescale": 58039,
      "largescale document": 61427,
      "modular endtoend": 74466,
      "attention architecture": 9794,
      "ability simultaneously": 1570,
      "achieved large": 2651,
      "large increase": 61113,
      "additionally develop": 3603,
      "series novel": 107277,
      "paper development": 83868,
      "promising accuracy": 92263,
      "primary objective": 90653,
      "objective natural": 81099,
      "nlp text": 79782,
      "achieving objective": 2966,
      "chat bots": 16652,
      "designed better": 29584,
      "better emulate": 13569,
      "words data": 133897,
      "data generate": 25974,
      "features popular": 43659,
      "network interpret": 77287,
      "subtitles automatic": 114992,
      "manual inspection": 67004,
      "promising outcomes": 92282,
      "experiment work": 40522,
      "work fact": 134526,
      "forums community": 45315,
      "cqa forums": 24582,
      "new dimension": 78883,
      "dimension context": 31742,
      "checking veracity": 16699,
      "problem create": 90987,
      "baseline unsupervised": 12329,
      "question retrieval": 95214,
      "unsupervised framework": 127640,
      "matching network": 67421,
      "matching questions": 67427,
      "information initial": 54688,
      "surface matching": 116066,
      "word question": 133426,
      "cqa datasets": 24581,
      "using script": 130133,
      "introduce large": 56444,
      "texts intended": 121532,
      "intended used": 55873,
      "require commonsense": 100119,
      "knowledge specifically": 58183,
      "results substantial": 102225,
      "forms basis": 45232,
      "knowledge organized": 58092,
      "model commonsense": 70848,
      "got stateoftheart": 48534,
      "calculating attention": 15027,
      "got place": 48533,
      "place final": 87603,
      "neural information": 77930,
      "requires participants": 100307,
      "models conventional": 72985,
      "conventional information": 22879,
      "competition won": 19629,
      "representative task": 100021,
      "proven quite": 93728,
      "quite useful": 95407,
      "mechanisms called": 68057,
      "called bidirectional": 15050,
      "suggest new": 115322,
      "simpler attention": 109553,
      "particularly attention": 85470,
      "dataset clinical": 26785,
      "case reports": 15606,
      "dataset uses": 27261,
      "dataset observe": 27060,
      "performance 20": 86103,
      "20 f1": 503,
      "f1 best": 42649,
      "machine readers": 65848,
      "skills required": 109975,
      "knowledge object": 58088,
      "reasoning difficult": 96245,
      "spoken squad": 112020,
      "task speech": 118732,
      "impact machine": 51877,
      "questionanswering systems": 95257,
      "humans interact": 51082,
      "questions obtain": 95336,
      "context largely": 22161,
      "domainspecific applications": 33887,
      "leverage metadata": 63603,
      "retrieval ii": 102396,
      "performance demonstrated": 86283,
      "actual use": 3151,
      "pairs annotated": 83488,
      "presents promising": 89891,
      "promising path": 92284,
      "answering new": 6657,
      "formalize new": 45197,
      "document encoder": 33000,
      "addresses key": 3798,
      "document discourse": 32992,
      "phrases document": 87418,
      "experiment baseline": 40451,
      "achieve reasonable": 2520,
      "multistep inference": 75802,
      "inference strategies": 54228,
      "results given": 101814,
      "given inputs": 48046,
      "model maintains": 71490,
      "iteratively refines": 57147,
      "inference snli": 54218,
      "multigenre natural": 75111,
      "inference multinli": 54170,
      "multinli dataset": 75468,
      "challenges neural": 16181,
      "movie plots": 74898,
      "different authors": 31019,
      "workers create": 134890,
      "create questions": 24637,
      "different set": 31421,
      "set workers": 107641,
      "created different": 24664,
      "underlying story": 126702,
      "little lexical": 64819,
      "perform complex": 85970,
      "performance f1": 86374,
      "dataset opens": 27070,
      "opens interesting": 82008,
      "interesting research": 56089,
      "research avenues": 100427,
      "studying language": 114557,
      "dynamic nature": 34318,
      "generation critical": 47348,
      "instances based": 55620,
      "studied compared": 114153,
      "results hybrid": 101835,
      "hybrid wordcharacter": 51199,
      "wordcharacter model": 133705,
      "separate character": 106846,
      "components number": 20039,
      "answer problem": 6556,
      "model adversarial": 70643,
      "important source": 52253,
      "present adversarial": 89359,
      "using reinforce": 130100,
      "ngrams different": 79462,
      "need model": 76836,
      "intersentence interactions": 56309,
      "multichoice question": 75047,
      "task makes": 118373,
      "learnt representations": 63254,
      "directly text": 31908,
      "text focus": 120956,
      "networks goal": 77615,
      "multiple conditions": 75523,
      "dataset rajpurkar": 27140,
      "requires machine": 100294,
      "model enables": 71067,
      "predict final": 88886,
      "margin achieves": 67185,
      "performance english": 86340,
      "task ranking": 118601,
      "information good": 54644,
      "useful study": 128935,
      "people routinely": 85889,
      "posts better": 88514,
      "offer assistance": 81561,
      "500 samples": 1081,
      "samples dataset": 103583,
      "judgments demonstrate": 57443,
      "baselines learning": 12423,
      "task substantially": 118753,
      "substantially different": 114885,
      "generation requires": 47601,
      "relevant topics": 98596,
      "question natural": 95185,
      "address key": 3709,
      "making natural": 66854,
      "used modulate": 128641,
      "based clinical": 11576,
      "literature available": 64746,
      "medical practitioners": 68216,
      "lack efficient": 58705,
      "necessary information": 76756,
      "implementation machine": 51946,
      "mining algorithms": 70222,
      "unified medical": 127115,
      "medical language": 68211,
      "questions medical": 95329,
      "documents addition": 33175,
      "employs supervised": 36334,
      "algorithms classification": 4835,
      "documents identifying": 33239,
      "effective domainspecific": 34667,
      "domainspecific heuristics": 33902,
      "cover content": 24482,
      "du et": 34223,
      "knowledge introduced": 58026,
      "generation significantly": 47623,
      "create corpus": 24607,
      "training candidate": 123379,
      "developed reading": 30296,
      "independent manner": 53773,
      "problem worse": 91289,
      "question paper": 95194,
      "learning result": 62968,
      "result approach": 101360,
      "approach improved": 7622,
      "model integrates": 71352,
      "combines knowledge": 18688,
      "stated text": 112544,
      "text relevant": 121238,
      "results strong": 102215,
      "making strong": 66865,
      "knowledge explicitly": 57916,
      "robust question": 103062,
      "context documents": 22069,
      "documents neural": 33258,
      "effective models": 34713,
      "modeling interactions": 72450,
      "shown models": 108494,
      "adversarial inputs": 4127,
      "sentences inspired": 106357,
      "observation propose": 81170,
      "sentences feed": 106316,
      "model overall": 71685,
      "significant reductions": 108856,
      "reductions training": 97472,
      "inference times": 54241,
      "squad newsqa": 112078,
      "furthermore experimental": 46171,
      "years lot": 135269,
      "shown outstanding": 108502,
      "outstanding results": 83205,
      "cost grows": 24354,
      "grows quadratically": 49187,
      "quadratically number": 94544,
      "solve issues": 110599,
      "issues introduce": 57053,
      "memory storage": 68383,
      "module model": 74498,
      "reducing computational": 97409,
      "quadratic linear": 94541,
      "simple ensemble": 109428,
      "paper seek": 84416,
      "questions provide": 95345,
      "provide suggestions": 93934,
      "suggestions effective": 115366,
      "effective question": 34734,
      "questions collected": 95284,
      "collected reddit": 18434,
      "enables development": 36382,
      "new convolutional": 78841,
      "efficacy model": 35012,
      "model comparing": 70857,
      "multihop inference": 75138,
      "modeled graph": 72357,
      "challenging models": 16276,
      "major barrier": 66538,
      "current inference": 25286,
      "empirically characterize": 36218,
      "difficulty building": 31694,
      "sentences connected": 106256,
      "including study": 53384,
      "study guides": 114396,
      "quality low": 94707,
      "maximize likelihood": 67511,
      "traditional chinese": 122805,
      "dataset aimed": 26727,
      "build baseline": 14746,
      "rouge bleu": 103289,
      "focus lexical": 44781,
      "reference answers": 97506,
      "used specific": 128773,
      "entity lists": 37966,
      "metrics better": 69929,
      "better correlate": 13552,
      "overlap human": 83330,
      "proves effectiveness": 93744,
      "know dont": 57730,
      "dont know": 33946,
      "unanswerable questions": 126590,
      "document tend": 33093,
      "identify address": 51474,
      "address weaknesses": 3778,
      "combines existing": 18684,
      "look similar": 65216,
      "questions possible": 95342,
      "answering squad": 6702,
      "involves matching": 56898,
      "pair paper": 83446,
      "paper formalize": 83958,
      "formalize problem": 45198,
      "offer promising": 81571,
      "structural representation": 113791,
      "networks question": 77719,
      "networks nns": 77679,
      "propose inject": 92723,
      "predicting labels": 88989,
      "corpus question": 23957,
      "subwordaugmented embedding": 115044,
      "comprehension representation": 20227,
      "learning foundation": 62589,
      "foundation machine": 45346,
      "level representations": 63498,
      "minimal linguistic": 70166,
      "embedding previous": 35472,
      "models actually": 72711,
      "subword character": 115016,
      "augment word": 10270,
      "handle rare": 49402,
      "words effectively": 133929,
      "challenging ai": 16223,
      "ai task": 4427,
      "gated neural": 46511,
      "extra labels": 42061,
      "obtains substantial": 81480,
      "model baselines": 70752,
      "terms multiple": 120349,
      "investigated problem": 56841,
      "thousand questions": 121916,
      "knowledge largest": 58040,
      "units involved": 127251,
      "qa training": 94524,
      "examples improve": 39844,
      "query document": 94955,
      "recently deeplearning": 96677,
      "similarity representation": 109297,
      "grammatically incorrect": 48734,
      "learn words": 62195,
      "demonstrate superior": 28877,
      "visualization attention": 131825,
      "query terms": 94991,
      "terms deep": 120303,
      "integration strategies": 55816,
      "characterlevel representation": 16582,
      "especially rare": 38498,
      "public benchmarks": 94238,
      "networkbased approaches": 77482,
      "directly optimizes": 31892,
      "generator sequencetosequence": 47779,
      "generated generator": 47079,
      "model evaluates": 71112,
      "question based": 95131,
      "reward functions": 102673,
      "conformity generated": 21040,
      "testing evaluation": 120593,
      "predict additional": 88871,
      "cases fail": 15644,
      "extract candidate": 42072,
      "dataset achieves": 26721,
      "f1 test": 42697,
      "context aggregation": 22004,
      "task measuring": 118383,
      "years paper": 135287,
      "training heterogeneous": 123649,
      "aggregate context": 4346,
      "simple aggregation": 109356,
      "network final": 77257,
      "insufficient labeled": 55721,
      "mechanism automatically": 67950,
      "generate training": 47035,
      "xie et": 135192,
      "documents processed": 33274,
      "network order": 77376,
      "understood especially": 127025,
      "documents used": 33308,
      "used prior": 128699,
      "adaptive approach": 3329,
      "multiple benchmark": 75504,
      "datasets context": 27382,
      "despite current": 29682,
      "numerous models": 81022,
      "approaches vulnerable": 8401,
      "vulnerable adversarial": 131973,
      "attacks paper": 9715,
      "model single": 72053,
      "vanilla knowledge": 130814,
      "ensemble experiments": 37595,
      "04 f1": 25,
      "set compared": 107394,
      "networks success": 77784,
      "propose machine": 92755,
      "limitations model": 64180,
      "generate adversarial": 46900,
      "confuse model": 21049,
      "furthermore assess": 46147,
      "assess generalizability": 9479,
      "generalizability model": 46751,
      "differences human": 30959,
      "evaluating question": 39090,
      "models respect": 73945,
      "able reason": 1680,
      "models memory": 73561,
      "memory augmentation": 68288,
      "fail tasks": 42973,
      "world models": 135037,
      "random sentences": 95512,
      "questions understanding": 95372,
      "problem involves": 91094,
      "able perform": 1670,
      "report accuracy": 98974,
      "comprehension learning": 20188,
      "learning attend": 62373,
      "answering remains": 6693,
      "requires models": 100303,
      "information reasoning": 54898,
      "work typically": 134866,
      "typically formulates": 126433,
      "entailment problem": 37668,
      "evidence retrieved": 39667,
      "directly related": 31902,
      "especially complex": 38442,
      "identifies important": 51469,
      "words predict": 134131,
      "predict answer": 88873,
      "reasoning challenge": 96232,
      "dataset better": 26769,
      "better metric": 13628,
      "performance nlg": 86572,
      "systems metrics": 117001,
      "documents knowledge": 33244,
      "metrics suitable": 70010,
      "suitable task": 115409,
      "work current": 134451,
      "correlate significantly": 24201,
      "developed work": 30323,
      "electronic medical": 35254,
      "methodology generating": 69262,
      "largescale qa": 61500,
      "leveraging existing": 63676,
      "community shared": 19099,
      "explore learning": 41554,
      "training baseline": 123370,
      "question detection": 95148,
      "manually tedious": 67078,
      "tedious costly": 120055,
      "costly automatic": 24383,
      "data questions": 26312,
      "solution use": 110568,
      "multiple pairs": 75633,
      "pairs domains": 83521,
      "domains generating": 33787,
      "works aim": 134921,
      "words short": 134214,
      "short phrases": 108229,
      "goal generate": 48353,
      "dynamic attention": 34295,
      "tackle task": 117310,
      "step generate": 113267,
      "generate readable": 46995,
      "irrelevant sentences": 56951,
      "sentences sentences": 106490,
      "answering present": 6661,
      "new kind": 78969,
      "understanding subject": 126977,
      "elementary level": 35275,
      "application novel": 6873,
      "novel situations": 80731,
      "language expressed": 59035,
      "qa methods": 94505,
      "experiments designed": 40900,
      "datasets existing": 27462,
      "approach hinges": 7607,
      "transformation model": 124268,
      "derive new": 29340,
      "nli examples": 79542,
      "systems prove": 117086,
      "users used": 129180,
      "according knowledge": 1984,
      "knowledge world": 58241,
      "datasets typically": 27770,
      "reasoning mrc": 96275,
      "annotation dataset": 6297,
      "knowledge extracted": 57928,
      "external kb": 42005,
      "mrc model": 74927,
      "lay remaining": 61694,
      "question interpretation": 95169,
      "limiting utility": 64310,
      "evaluation tool": 39425,
      "nl question": 79509,
      "using public": 130065,
      "public benchmark": 94236,
      "datasets suggest": 27738,
      "performance qa": 86643,
      "improving systems": 53174,
      "style approaches": 114568,
      "forums given": 45316,
      "given new": 48068,
      "new question": 79095,
      "related existing": 97862,
      "use auxiliary": 127913,
      "good answers": 48460,
      "setting performing": 107769,
      "results trained": 102278,
      "embeddings crf": 35624,
      "tasks improves": 119173,
      "metrics showing": 70009,
      "contains components": 21819,
      "span detector": 111121,
      "components jointly": 20028,
      "paraphrase clusters": 84804,
      "users interested": 129139,
      "community qa": 19089,
      "qa platform": 94514,
      "platform typically": 87659,
      "process constructing": 91444,
      "measures taken": 67899,
      "making effective": 66838,
      "demonstrating dataset": 28970,
      "multihop question": 75140,
      "answering existing": 6627,
      "datasets fail": 27483,
      "train qa": 123001,
      "provide explanations": 93821,
      "dataset 113k": 26707,
      "require finding": 100138,
      "preexisting knowledge": 89233,
      "sentencelevel supporting": 106175,
      "systems reason": 117099,
      "explainable predictions": 41268,
      "datasets question": 27657,
      "provide complementary": 93782,
      "coverage aspects": 24497,
      "datasets improved": 27512,
      "improved baseline": 52595,
      "similarity models": 109272,
      "pretraining encourage": 90257,
      "longterm context": 65198,
      "methods attention": 69324,
      "tasks systems": 119542,
      "entity features": 37940,
      "particularly difficult": 85477,
      "text builds": 120692,
      "states generally": 113044,
      "evolving knowledge": 39704,
      "text demonstrate": 120865,
      "graphs help": 48957,
      "dataset composed": 26811,
      "real questions": 96075,
      "experts annotated": 41238,
      "need comprehensive": 76788,
      "dataset release": 27154,
      "performance thanks": 86802,
      "useful datasets": 128873,
      "truly understanding": 125864,
      "question asked": 95130,
      "table question": 117257,
      "answer query": 6558,
      "using answers": 129339,
      "question table": 95225,
      "uses multiple": 129249,
      "interpretability answers": 56219,
      "answers model": 6732,
      "information train": 55050,
      "performance interpretability": 86468,
      "models big": 72854,
      "systems tend": 117191,
      "tend learn": 120153,
      "span source": 111133,
      "source document": 110745,
      "instead seeking": 55681,
      "relevant passages": 98570,
      "reason important": 96196,
      "important pieces": 52207,
      "document order": 33046,
      "relational understanding": 98101,
      "model master": 71504,
      "models overview": 73695,
      "tasks evaluated": 119104,
      "need extracted": 76808,
      "selected set": 104733,
      "machine commonsense": 65745,
      "systems fall": 116902,
      "fall far": 43027,
      "far human": 43083,
      "challenge future": 16036,
      "research bridge": 100435,
      "comprehension record": 20226,
      "general reading": 46703,
      "improve comprehension": 52355,
      "levels especially": 63532,
      "science given": 103977,
      "resources just": 100992,
      "model fixed": 71194,
      "instances propose": 55632,
      "order input": 82349,
      "text embedding": 120901,
      "proposed strategies": 93557,
      "result achieved": 101358,
      "strategies finetune": 113464,
      "leading absolute": 61883,
      "understanding selfsupervised": 126951,
      "comprehension work": 20243,
      "algorithm solving": 4801,
      "problems requires": 91369,
      "scientific terms": 104023,
      "issue learning": 57004,
      "problems introduce": 91327,
      "process annotations": 91421,
      "annotations experimental": 6431,
      "knowledge multimodal": 58069,
      "process effective": 91468,
      "models integrating": 73411,
      "marginal loglikelihood": 67207,
      "distillation learning": 32509,
      "methods explicit": 69493,
      "structured queries": 114033,
      "queries directly": 94917,
      "way users": 132140,
      "users express": 129117,
      "look case": 65213,
      "queries entities": 94920,
      "combines distributed": 18682,
      "query answering": 94948,
      "simplify model": 109613,
      "relations wikidata": 98280,
      "necessary training": 76763,
      "relationships natural": 98325,
      "provides tools": 94089,
      "challenges present": 16195,
      "techniques begin": 119843,
      "begin address": 12631,
      "parsing challenges": 85082,
      "task built": 117949,
      "outperforms offtheshelf": 82928,
      "zeroshot capability": 135491,
      "handle new": 49395,
      "requiring additional": 100335,
      "scaling new": 103777,
      "model reading": 71856,
      "fully extract": 45961,
      "representations furthermore": 99663,
      "various attention": 131045,
      "results propose": 102060,
      "convolutional operation": 23211,
      "different size": 31430,
      "datasets deep": 27393,
      "deep cascade": 28210,
      "cascade model": 15569,
      "effectiveness comes": 34876,
      "functions extractive": 46074,
      "components candidate": 20020,
      "document selection": 33072,
      "irrelevant documents": 56947,
      "filtered simple": 44078,
      "texts better": 121465,
      "methods largescale": 69581,
      "describes novel": 29420,
      "comprehension style": 20233,
      "answering aims": 6600,
      "given narrative": 48064,
      "method attention": 68649,
      "specifically encode": 111545,
      "introduces hierarchical": 56614,
      "datasets validate": 27784,
      "writing paper": 135092,
      "achieves position": 2834,
      "extracting answers": 42201,
      "language actively": 58811,
      "studied existing": 114159,
      "provides model": 94051,
      "model explanation": 71138,
      "global normalization": 48254,
      "normalization multiple": 80340,
      "recent evidence": 96461,
      "task baseline": 117925,
      "reciprocal rank": 96796,
      "methods solve": 69765,
      "work neglects": 134650,
      "information tasks": 55032,
      "motivated following": 74850,
      "improved learning": 52612,
      "scheme utilizes": 103943,
      "method performance": 69043,
      "attention scheme": 9998,
      "processing especially": 91667,
      "answer paper": 6552,
      "combines information": 18687,
      "models relevant": 73916,
      "combined using": 18673,
      "network predict": 77390,
      "comprehension coreference": 20172,
      "context traditional": 22301,
      "context extract": 22102,
      "bert contextual": 13091,
      "model empirical": 71057,
      "art result": 9085,
      "model 16": 70498,
      "f1 ensemble": 42654,
      "deeper level": 28446,
      "level instead": 63462,
      "method multiple": 68977,
      "accurate information": 2337,
      "fusion experimental": 46236,
      "study tackles": 114531,
      "key characteristics": 57550,
      "focused extracting": 44856,
      "model instead": 71350,
      "second previous": 104444,
      "studies built": 114190,
      "built specific": 14941,
      "specific model": 111464,
      "difficulty acquiring": 31692,
      "style experiments": 114576,
      "task qa": 118591,
      "nlg task": 79528,
      "summary task": 115647,
      "attention academic": 9790,
      "academic research": 1862,
      "research industry": 100536,
      "studies assume": 114182,
      "perceive human": 85909,
      "raise questions": 95457,
      "attempt understand": 9754,
      "understand true": 126782,
      "true meaning": 125852,
      "qa approach": 94493,
      "work human": 134554,
      "spans model": 111206,
      "recently works": 96781,
      "problem reading": 91195,
      "task directly": 118101,
      "performances models": 86884,
      "exist given": 40023,
      "given paragraph": 48075,
      "end position": 36820,
      "level answer": 63421,
      "span level": 111125,
      "problems experiments": 91322,
      "study participants": 114465,
      "human upper": 50983,
      "time demonstrate": 122012,
      "designed human": 29599,
      "chinese learners": 16781,
      "approach investigate": 7649,
      "incorporating dialogue": 53529,
      "structure different": 113835,
      "rulebased neural": 103371,
      "dataset effectiveness": 26883,
      "improving question": 53153,
      "areas science": 8900,
      "science require": 103982,
      "require broad": 100115,
      "corpus work": 24069,
      "effective methods": 34709,
      "corpus relevant": 23966,
      "relevant text": 98594,
      "potentially ambiguous": 88602,
      "ambiguous concepts": 5290,
      "research second": 100621,
      "additional indomain": 3525,
      "instances experiments": 55626,
      "experiments challenging": 40819,
      "accuracy 81": 2074,
      "employing additional": 36313,
      "performance degrades": 86281,
      "exhibit higher": 40000,
      "level difficulty": 63435,
      "data studies": 26512,
      "predictions existing": 89167,
      "existing mrc": 40214,
      "explain support": 41261,
      "directly extracted": 31877,
      "reference documents": 97514,
      "lack ground": 58712,
      "sentence labels": 105917,
      "apply distant": 7171,
      "generate imperfect": 46955,
      "use train": 128335,
      "sentence extractor": 105870,
      "labels apply": 58579,
      "apply recently": 7210,
      "extracted evidence": 42154,
      "evaluate endtoend": 38833,
      "reference document": 97513,
      "expression diversity": 41735,
      "words explicitly": 133951,
      "based cnn": 11579,
      "maintaining strong": 66526,
      "information matching": 54751,
      "advantages better": 4069,
      "distill rich": 32500,
      "challenges using": 16220,
      "examine performance": 39753,
      "rewards agent": 102682,
      "drastically improve": 34102,
      "improve precision": 52497,
      "questions answering": 95272,
      "employing supervised": 36323,
      "training question": 123803,
      "humans provide": 51104,
      "educational purposes": 34574,
      "models sufficient": 74130,
      "vocabulary paper": 131891,
      "employing variety": 36324,
      "design multitask": 29549,
      "labeling strategy": 58544,
      "identify question": 51539,
      "question word": 95239,
      "accurate boundaries": 2327,
      "generation furthermore": 47414,
      "range features": 95569,
      "features prediction": 43663,
      "clue word": 17914,
      "tree representation": 125610,
      "approaches question": 8303,
      "prediction test": 89141,
      "model asked": 70688,
      "sequence causing": 106921,
      "number authors": 80844,
      "authors proposed": 10385,
      "using discriminator": 129586,
      "directly training": 31910,
      "used rewards": 128734,
      "model simply": 72047,
      "report introduce": 99010,
      "retrieval accuracy": 102379,
      "reasoning paragraphs": 96288,
      "seen rapid": 104538,
      "english reading": 37258,
      "understanding content": 126817,
      "necessary prior": 76758,
      "datasets apply": 27312,
      "literature dataset": 64749,
      "accuracy metric": 2208,
      "reasoning achieve": 96225,
      "f1 bidirectional": 42650,
      "different question": 31367,
      "existing embeddingbased": 40123,
      "types relation": 126347,
      "directly model": 31889,
      "kb novel": 57502,
      "webquestions benchmark": 132288,
      "benchmark method": 12839,
      "handcrafted semantic": 49355,
      "method offers": 68998,
      "corpora knowledge": 23509,
      "challenge human": 16039,
      "fully understand": 45974,
      "scale knowledge": 103726,
      "corpora example": 23477,
      "learn templates": 62164,
      "learning endtoend": 62546,
      "endtoend knowledge": 36909,
      "approach knowledge": 7657,
      "users questions": 129163,
      "questions appropriate": 95275,
      "multiple instance": 75581,
      "problem noisy": 91147,
      "selection different": 104781,
      "instance weighting": 55611,
      "entity accuracy": 37903,
      "rougel score": 103307,
      "score stateoftheart": 104126,
      "question article": 95128,
      "article existing": 9127,
      "mimicking humans": 70139,
      "better identify": 13603,
      "identify subtle": 51562,
      "mechanical turker": 67937,
      "dataset question": 27138,
      "paper dataset": 83792,
      "web tables": 132261,
      "asked questions": 9304,
      "dataset novel": 27059,
      "paired table": 83475,
      "sql query": 112070,
      "classification conduct": 17152,
      "experiments measure": 40996,
      "task extract": 118193,
      "parts work": 85594,
      "generate sql": 47019,
      "studied recent": 114170,
      "published models": 94361,
      "provides dataset": 94017,
      "process data": 91452,
      "preparation model": 89295,
      "easier making": 34417,
      "making neural": 66855,
      "challenge develop": 16020,
      "computationally faster": 20455,
      "faster model": 43184,
      "applied train": 7132,
      "entity entity": 37931,
      "questions including": 95317,
      "given textual": 48158,
      "network gan": 77261,
      "model discriminator": 71008,
      "function models": 46034,
      "question evaluate": 95154,
      "evaluate datasets": 38817,
      "model ablations": 70507,
      "model eliminating": 71051,
      "choice questions": 16901,
      "question pair": 95192,
      "humans perform": 51099,
      "just focus": 57462,
      "option selection": 82247,
      "selection use": 104844,
      "use combination": 127942,
      "selection specifically": 104828,
      "light new": 63997,
      "information ignore": 54662,
      "process repeated": 91559,
      "multiple times": 75729,
      "portions passage": 88187,
      "multiple rounds": 75675,
      "representation finally": 99241,
      "model 13": 70496,
      "types dataset": 126270,
      "clinical trial": 17794,
      "values study": 130806,
      "articles specifically": 9219,
      "trained massive": 123195,
      "structured clinical": 113994,
      "article based": 9112,
      "values extracted": 130793,
      "answering benchmarks": 6606,
      "variant turing": 130868,
      "intelligence paper": 55835,
      "setting research": 107777,
      "research model": 100557,
      "model calibration": 70799,
      "learning challenges": 62423,
      "incorporates knowledge": 53507,
      "dataset developing": 26871,
      "developing model": 30348,
      "spanning years": 111192,
      "matching paper": 67423,
      "language mathematical": 59276,
      "particles quantum": 85375,
      "extraction pretrained": 42443,
      "layers question": 61803,
      "classification layers": 17250,
      "leads superior": 61973,
      "knowledge challenging": 57824,
      "relevant given": 98552,
      "knowledge linguistic": 58055,
      "realworld problems": 96174,
      "gap best": 46443,
      "require prior": 100189,
      "study effects": 114366,
      "platform study": 87658,
      "study leverage": 114425,
      "graph multihop": 48834,
      "gradually builds": 48617,
      "iterative process": 57133,
      "extraction module": 42395,
      "bert graph": 13138,
      "millions documents": 70115,
      "best competitor": 13321,
      "entityrelation extraction": 38147,
      "task multiturn": 118428,
      "problem extraction": 91051,
      "key advantages": 57539,
      "modeling entity": 72423,
      "conll04 datasets": 21093,
      "datasets increasing": 27520,
      "extensively recent": 41967,
      "supporting evidence": 116029,
      "fusion layer": 46238,
      "mentioned given": 68437,
      "dynamically built": 34340,
      "finds relevant": 44319,
      "supporting entities": 116028,
      "given documents": 48024,
      "documents evaluate": 33225,
      "dataset requiring": 27160,
      "reasoning heterogeneous": 96253,
      "types nodes": 126329,
      "candidates documents": 15133,
      "evaluated blind": 38958,
      "blind test": 14353,
      "paradigm natural": 84542,
      "bert proposed": 13212,
      "glue leaderboard": 48314,
      "bert applied": 13064,
      "different nuances": 31307,
      "modeling relevance": 72531,
      "length answer": 63352,
      "answer different": 6524,
      "bert answer": 13063,
      "demonstrating success": 28982,
      "learning multihop": 62810,
      "textual sources": 121747,
      "study focuses": 114386,
      "focuses task": 44915,
      "qa requires": 94518,
      "query focused": 94962,
      "extraction uses": 42530,
      "method extracts": 68835,
      "consider dependency": 21196,
      "sentences cover": 106267,
      "cover important": 24485,
      "textual database": 121686,
      "larger variety": 61387,
      "higher levels": 50190,
      "nlp human": 79621,
      "terms learning": 120344,
      "input modalities": 55367,
      "potential directions": 88546,
      "corpus endtoend": 23772,
      "new collection": 78831,
      "results challenging": 101568,
      "particular script": 85448,
      "challenging humans": 16259,
      "effectiveness range": 34939,
      "best method": 13365,
      "significant gap": 108760,
      "dialogue using": 30812,
      "model successful": 72106,
      "limit number": 64155,
      "uses bert": 129195,
      "bert encode": 13112,
      "paragraph independently": 84567,
      "conditioned question": 20805,
      "context method": 22178,
      "encoded bert": 36470,
      "21 f1": 760,
      "effects number": 34999,
      "number types": 80990,
      "gold answer": 48431,
      "history given": 50557,
      "supervised open": 115819,
      "evidence candidates": 39638,
      "fundamentally different": 46133,
      "task evaluate": 118155,
      "answer new": 6548,
      "hard obtain": 49473,
      "using motivation": 129901,
      "language hindi": 59123,
      "dataset hindi": 26965,
      "sentences generating": 106330,
      "converts input": 23107,
      "related specific": 97898,
      "datasets general": 27493,
      "general specific": 46716,
      "input pipelined": 55393,
      "conditional neural": 20768,
      "model extensively": 71156,
      "experiments report": 41115,
      "report strong": 99050,
      "conventional paradigm": 22897,
      "content limited": 21899,
      "contrast work": 22720,
      "extends conventional": 41824,
      "neural qa": 78638,
      "leverages different": 63644,
      "design allows": 29513,
      "estimation outperforms": 38658,
      "systems significant": 117143,
      "corpus dynamic": 23766,
      "provides effective": 94021,
      "novel powerful": 80686,
      "challenging baseline": 16227,
      "paper considers": 83784,
      "task multiple": 118423,
      "upstream components": 127829,
      "help downstream": 49714,
      "pipelined approaches": 87563,
      "context candidate": 22022,
      "pipelined baseline": 87564,
      "baseline achieves": 12180,
      "reasoning present": 96291,
      "challenging highly": 16257,
      "monolingual spanish": 74618,
      "crosslingual english": 24950,
      "english experiments": 37134,
      "lag human": 58779,
      "benchmark future": 12827,
      "consistency introduce": 21361,
      "resulting corpora": 101435,
      "corpora obtain": 23545,
      "model extractive": 71163,
      "suitable realtime": 115406,
      "realtime usage": 96142,
      "input query": 55409,
      "query paper": 94974,
      "allows reach": 5182,
      "targets particular": 117791,
      "encoding effectively": 36690,
      "captures syntactic": 15457,
      "optimization strategies": 82211,
      "reduced computational": 97370,
      "endtoend inference": 36906,
      "propose data": 92611,
      "paired corresponding": 83471,
      "way construct": 132068,
      "data question": 26310,
      "performs consistently": 87004,
      "baseline use": 12330,
      "means data": 67759,
      "19 absolute": 433,
      "bertbase model": 13265,
      "bertlarge model": 13281,
      "ai challenge": 4420,
      "passages web": 85618,
      "web question": 132246,
      "uses bilstm": 129202,
      "additionally use": 3635,
      "different embeddings": 31124,
      "achieved mean": 2653,
      "present rulebased": 89683,
      "annotated multiple": 6210,
      "multiple annotators": 75491,
      "scale evaluation": 103719,
      "help various": 49789,
      "various semantic": 131188,
      "fed neural": 43805,
      "years research": 135295,
      "remains lack": 98801,
      "lack comprehensive": 58690,
      "survey summarizing": 116199,
      "summarizing existing": 115623,
      "trends motivated": 125747,
      "specifically thorough": 111597,
      "thorough review": 121893,
      "review research": 102569,
      "field covering": 43938,
      "approaches new": 8257,
      "emerging areas": 36065,
      "corresponding challenges": 24277,
      "achieved far": 2625,
      "open issues": 81913,
      "aims teach": 4577,
      "text like": 121090,
      "challenging direction": 16244,
      "aspects corpus": 9378,
      "corpus techniques": 24037,
      "specific characteristics": 111417,
      "compared main": 19381,
      "benchmarks like": 12915,
      "surpassing human": 116115,
      "huge corpus": 50719,
      "developing scalable": 30359,
      "largescale sentencelevel": 61508,
      "models establish": 73166,
      "models classical": 72905,
      "release evaluation": 98452,
      "evaluation code": 39147,
      "dataset social": 27209,
      "rely realtime": 98732,
      "knowledge previous": 58120,
      "like news": 64078,
      "news wikipedia": 79400,
      "limited performance": 64263,
      "results point": 102038,
      "point need": 87807,
      "need improved": 76820,
      "long form": 65074,
      "documents help": 33238,
      "abstractive model": 1803,
      "modeling strong": 72552,
      "strong extractive": 113674,
      "extractive baseline": 42549,
      "model far": 71173,
      "improvement modeling": 52726,
      "current ai": 25256,
      "network extensive": 77251,
      "settings propose": 107832,
      "novel grammarbased": 80584,
      "trained reinforcement": 123260,
      "human data": 50791,
      "systems downstream": 116860,
      "inputs machine": 55484,
      "translation optical": 125070,
      "speech investigate": 111701,
      "model forced": 71202,
      "create train": 24646,
      "noisy sentences": 80118,
      "evaluate human": 38840,
      "incorporating relation": 53555,
      "relation knowledge": 98045,
      "advantage external": 4042,
      "learning traditional": 63111,
      "knowledge mentioned": 58062,
      "ability leverage": 1538,
      "leverage external": 63587,
      "reasoning specifically": 96312,
      "design auxiliary": 29520,
      "type words": 126231,
      "document candidate": 32957,
      "story test": 113420,
      "information required": 54924,
      "simultaneously achieving": 109670,
      "graphtosequence model": 48983,
      "answer previous": 6554,
      "text ii": 121040,
      "solely rely": 110534,
      "issues like": 57056,
      "information address": 54367,
      "rl based": 102888,
      "based encoder": 11673,
      "mixed objective": 70400,
      "generation syntactically": 47650,
      "text introduce": 121061,
      "effective deep": 34659,
      "passage word": 85612,
      "levels model": 63553,
      "stateoftheart scores": 112941,
      "margin standard": 67201,
      "comprehension requires": 20231,
      "achieved human": 2642,
      "settings performance": 107830,
      "applied realistic": 7112,
      "realistic scenarios": 96103,
      "involve various": 56878,
      "types multiple": 126323,
      "abilities required": 1480,
      "required paper": 100222,
      "producing multiple": 92022,
      "coreferential reasoning": 23384,
      "coreference present": 23356,
      "new crowdsourced": 78850,
      "resolving coreference": 100816,
      "paragraphs wikipedia": 84580,
      "phenomena challenging": 87221,
      "challenging hard": 16255,
      "lexical cues": 63752,
      "reasoning deal": 96242,
      "stateoftheart reading": 112879,
      "worse humans": 135063,
      "estimated human": 38640,
      "text ability": 120622,
      "new situation": 79138,
      "situation order": 109889,
      "facilitate progress": 42787,
      "reading present": 96029,
      "analyze challenges": 5953,
      "f1 human": 42656,
      "text structuring": 121334,
      "model clinical": 70835,
      "task clinical": 117967,
      "research traditional": 100649,
      "models pipeline": 73745,
      "lack dataset": 58699,
      "unify different": 127148,
      "different specific": 31446,
      "tasks make": 119286,
      "make dataset": 66648,
      "aims introduce": 4543,
      "features clinical": 43399,
      "reports collected": 99081,
      "performance specific": 86729,
      "competes favorably": 19614,
      "favorably strong": 43219,
      "applying general": 7247,
      "annotated natural": 6214,
      "variety stateoftheart": 131017,
      "comprehension demonstrate": 20177,
      "search similar": 104348,
      "retrieval paper": 102417,
      "short sentence": 108232,
      "sentence ranking": 106014,
      "ranking task": 95690,
      "task retrievalbased": 118657,
      "retrieve similar": 102455,
      "query question": 94976,
      "rank sentences": 95628,
      "design specific": 29566,
      "bert training": 13241,
      "data semantic": 26390,
      "build tree": 14815,
      "based kmeans": 11786,
      "predicting time": 89017,
      "sentence query": 106012,
      "results methods": 101929,
      "ranking accuracy": 95661,
      "accuracy adversarial": 2097,
      "adaptation machine": 3229,
      "domain large": 33568,
      "data unlabeled": 26594,
      "adaptation framework": 3224,
      "ii domain": 51706,
      "predict domain": 88882,
      "encoder jointly": 36536,
      "generalizable different": 46753,
      "combined pretrained": 18663,
      "answering evaluation": 6626,
      "order resolve": 82399,
      "approach english": 7529,
      "qa present": 94515,
      "singletask models": 109847,
      "trained auxiliary": 123073,
      "resolution datasets": 100758,
      "datasets clearly": 27346,
      "clearly outperforming": 17741,
      "phrase ellipsis": 87344,
      "question previous": 95203,
      "representation ignore": 99252,
      "modeling relationship": 72529,
      "best information": 13342,
      "question work": 95240,
      "multichoice questions": 75049,
      "supporting sentences": 116032,
      "using logical": 129827,
      "model appropriate": 70677,
      "data attention": 25625,
      "question used": 95236,
      "used resolve": 128730,
      "generator model": 47775,
      "generation hierarchical": 47427,
      "model boosts": 70783,
      "propose automatically": 92568,
      "model guide": 71269,
      "generation deal": 47355,
      "accuracy question": 2252,
      "prediction finally": 89063,
      "used systems": 128801,
      "compute similarity": 20471,
      "improvement field": 52711,
      "proposed select": 93543,
      "select relevant": 104706,
      "model ranks": 71847,
      "distinguish positive": 32572,
      "model indicates": 71334,
      "indicates strong": 53866,
      "using contextsensitive": 129481,
      "sentences help": 106337,
      "requires reading": 100313,
      "understanding long": 126879,
      "passages text": 85617,
      "interactive interface": 56019,
      "user select": 129040,
      "group similar": 49142,
      "filtering techniques": 44089,
      "comprehension ability": 20163,
      "study work": 114553,
      "systems developing": 116847,
      "people context": 85871,
      "tasks beneficial": 118962,
      "levels machine": 63552,
      "showing different": 108402,
      "identifying basic": 51582,
      "performing inference": 86953,
      "inference understanding": 54244,
      "answering commonsense": 6609,
      "knowledge generate": 57949,
      "available question": 11088,
      "vocabulary knowledge": 131879,
      "knowledge process": 58124,
      "utilize symbolic": 130532,
      "symbolic knowledge": 116315,
      "knowledge useful": 58227,
      "exploit external": 41414,
      "answer empirical": 6525,
      "quality models": 94721,
      "knowledge existing": 57910,
      "effectiveness leveraging": 34901,
      "knowledge bert": 57816,
      "surpass human": 116091,
      "bert achieved": 13056,
      "knowledge like": 58053,
      "achieve human": 2485,
      "require common": 100118,
      "knowledge achieve": 57739,
      "level experiments": 63444,
      "knowledge included": 58010,
      "included knowledge": 53247,
      "entities finegrained": 37787,
      "types neural": 126327,
      "encode additional": 36420,
      "questions evaluate": 95305,
      "knowledge enriched": 57900,
      "score squad": 104125,
      "dataset respectively": 27163,
      "order successfully": 82413,
      "replicate experiments": 98959,
      "experiments paper": 41051,
      "shown yield": 108542,
      "yield stateoftheart": 135356,
      "superhuman performance": 115676,
      "performance contrary": 86256,
      "contrary prior": 22685,
      "results evaluate": 101767,
      "learning bert": 62405,
      "machine understand": 66315,
      "dataset extractive": 26923,
      "retrieve supporting": 102456,
      "corpus existing": 23786,
      "work opendomain": 134663,
      "techniques retrieve": 119978,
      "topic second": 122570,
      "resolve issue": 100801,
      "trained weakly": 123334,
      "14 point": 316,
      "f1 using": 42700,
      "embeddings result": 35916,
      "evidence needed": 39654,
      "process challenging": 91435,
      "f1 pretrained": 42670,
      "biomedical question": 14203,
      "largely attributed": 61333,
      "pretrained general": 90036,
      "biobert pretrained": 14151,
      "biomedical language": 14192,
      "structure various": 113981,
      "models biobert": 72857,
      "guided approach": 49253,
      "textual question": 121726,
      "answering requires": 6694,
      "sentences focus": 106322,
      "model retrieve": 71928,
      "provided knowledge": 93972,
      "knowledge answer": 57753,
      "choice based": 16886,
      "partial knowledge": 85291,
      "study content": 114340,
      "understanding conversational": 126819,
      "processing field": 91672,
      "recently collected": 96672,
      "collected datasets": 18416,
      "work achieved": 134350,
      "questions design": 95298,
      "design different": 29532,
      "indicate potential": 53846,
      "light models": 63995,
      "learn datasets": 62016,
      "benefit future": 12970,
      "future progress": 46290,
      "project propose": 92223,
      "multitask adversarial": 75808,
      "26 f1": 840,
      "labeling machine": 58507,
      "useful studying": 128936,
      "particular phenomena": 85439,
      "task correctly": 118029,
      "benchmarks automatic": 12886,
      "crucial importance": 25143,
      "importance training": 52088,
      "tiny fraction": 122212,
      "increases efficiency": 53643,
      "models prevents": 73793,
      "contiguous spans": 22591,
      "spans input": 111204,
      "avoid need": 11238,
      "datasets include": 27515,
      "simple architecture": 109365,
      "output model": 83094,
      "challenges opportunities": 16187,
      "text books": 120690,
      "corpus subset": 24025,
      "pretraining improve": 90265,
      "significantly time": 109054,
      "need novel": 76840,
      "research needed": 100567,
      "including commonsense": 53275,
      "neural seq2seq": 78668,
      "seq2seq based": 106888,
      "close words": 17839,
      "methods substantially": 69774,
      "improve relevance": 52520,
      "relevance generated": 98510,
      "problem conversational": 90984,
      "base recent": 11480,
      "task subtasks": 118755,
      "following issues": 44971,
      "affect downstream": 4215,
      "downstream ones": 34019,
      "propose innovative": 92724,
      "designed resolve": 29621,
      "alleviates effect": 5059,
      "entities proposed": 37849,
      "stateoftheart work": 113031,
      "success recently": 115127,
      "tackling challenging": 117323,
      "networks pretrained": 77703,
      "finetuned large": 44432,
      "models run": 73970,
      "similar human": 109094,
      "human reasoning": 50951,
      "input paragraph": 55388,
      "reasoning addition": 96226,
      "addition subtraction": 3478,
      "humans reading": 51107,
      "comprehension considered": 20167,
      "considered existing": 21291,
      "models considering": 72965,
      "information missing": 54766,
      "neural retrieval": 78656,
      "models golden": 73306,
      "language search": 60069,
      "available context": 10964,
      "scale efficiently": 103716,
      "reasoning maintaining": 96268,
      "maintaining interpretability": 66521,
      "bert unsupervised": 13245,
      "dl models": 32924,
      "models succeeded": 74119,
      "accuracy various": 2316,
      "approach similar": 7918,
      "fever dataset": 43882,
      "bert classifier": 13089,
      "label based": 58384,
      "based correctly": 11614,
      "popular choice": 88084,
      "choice neural": 16892,
      "architectures machine": 8821,
      "information relatively": 54916,
      "dramatically affect": 34090,
      "models multidocument": 73588,
      "require information": 100149,
      "results current": 101652,
      "input sequencetosequence": 55434,
      "methods tfidf": 69803,
      "base query": 11474,
      "search information": 104323,
      "standard sequencetosequence": 112304,
      "tasks long": 119273,
      "adapting models": 3312,
      "domain finetuning": 33536,
      "problem deep": 90997,
      "framework domain": 45504,
      "manner models": 66953,
      "features apply": 43363,
      "task 2019": 117823,
      "evaluating generalization": 39057,
      "teams submitted": 119734,
      "systems explored": 116899,
      "data sampling": 26368,
      "absolute points": 1745,
      "initial baseline": 55208,
      "baseline based": 12193,
      "time leading": 122043,
      "datasets squad": 27729,
      "texts key": 121535,
      "dataset offer": 27067,
      "help address": 49694,
      "annotations present": 6453,
      "approach mitigate": 7711,
      "problems remain": 91367,
      "remain unsolved": 98775,
      "using multisource": 129923,
      "potential solution": 88583,
      "preliminary step": 89280,
      "question generator": 95164,
      "train rnnbased": 123008,
      "single document": 109727,
      "distributions document": 32759,
      "strategy significantly": 113540,
      "existing baseline": 40075,
      "using automated": 129363,
      "approaches investigate": 8198,
      "investigate bert": 56725,
      "training train": 123933,
      "train bert": 122903,
      "observe finetuned": 81194,
      "prediction instead": 89071,
      "information solve": 54990,
      "retrieval semantic": 102429,
      "topic entity": 122513,
      "sparql query": 111219,
      "text possibly": 121186,
      "grammar vocabulary": 48665,
      "vocabulary question": 131895,
      "word classifier": 132954,
      "module predicts": 74503,
      "second module": 104431,
      "information collected": 54421,
      "collected multiple": 18427,
      "derive answer": 29337,
      "assess ability": 9470,
      "ability various": 1578,
      "reasoning chain": 96231,
      "models surprisingly": 74138,
      "surprisingly using": 116151,
      "approaches limited": 8220,
      "13 absolute": 279,
      "improvement bert": 52685,
      "readers including": 95958,
      "existence potential": 40037,
      "necessity develop": 76769,
      "investigate number": 56792,
      "demonstrate strategy": 28873,
      "short answers": 108202,
      "dev set": 30166,
      "problem demands": 91002,
      "providing supporting": 94142,
      "effective interpretable": 34694,
      "problem filters": 91058,
      "information achieved": 54358,
      "trained novel": 123223,
      "learningtorank loss": 63190,
      "model optimized": 71617,
      "prediction attentionbased": 89032,
      "systems leaderboard": 116974,
      "popular widely": 88143,
      "requires research": 100317,
      "research systems": 100634,
      "consist parts": 21350,
      "methods presented": 69680,
      "dataset gap": 26948,
      "gap language": 46460,
      "models employed": 73142,
      "employed implementing": 36300,
      "language empirical": 58999,
      "applying model": 7256,
      "bert enhance": 13116,
      "effect language": 34597,
      "component results": 19999,
      "demonstrate language": 28766,
      "requirement natural": 100231,
      "knowledgeenhanced graph": 58269,
      "reasoning indispensable": 96255,
      "attention current": 9810,
      "capture entities": 15297,
      "relational facts": 98088,
      "build entity": 14764,
      "baselines methods": 12428,
      "analysis illustrates": 5648,
      "new specialized": 79144,
      "domains current": 33754,
      "effective answering": 34627,
      "techniques problem": 119954,
      "manual domain": 66987,
      "domain demonstrate": 33501,
      "adapting dnn": 3307,
      "generation focuses": 47411,
      "aspect multiple": 9339,
      "documents model": 33252,
      "using targeted": 130267,
      "set generate": 107451,
      "questions cover": 95294,
      "cover larger": 24486,
      "challenge introduce": 16043,
      "introduce contrastive": 56396,
      "given positive": 48083,
      "sets documents": 107665,
      "negative set": 76955,
      "propose multisource": 92805,
      "framework includes": 45570,
      "effective auxiliary": 34636,
      "auxiliary objective": 10881,
      "evaluation source": 39403,
      "classification qc": 17360,
      "selective attention": 104854,
      "sentences sentence": 106489,
      "model conducted": 70881,
      "demonstrate overall": 28812,
      "overall effectiveness": 83229,
      "fluent relevant": 44715,
      "network hgn": 77273,
      "nodes different": 80033,
      "given hierarchical": 48040,
      "node representations": 80021,
      "representations updated": 99958,
      "graph edges": 48796,
      "heterogeneous nodes": 49865,
      "unified graph": 127111,
      "qa approaches": 94494,
      "technique finetuning": 119788,
      "specifically transfer": 111603,
      "dataset enable": 26887,
      "transfer step": 124187,
      "approach establishes": 7536,
      "establishes state": 38612,
      "art wellknown": 9103,
      "wellknown benchmarks": 132416,
      "highest scores": 50236,
      "optimal hyperparameters": 82162,
      "adaptation step": 3268,
      "enables effective": 36384,
      "use noisy": 128177,
      "noisy datasets": 80100,
      "finetuning finally": 44460,
      "specific datasets": 111428,
      "subject different": 114676,
      "answering whyquestions": 6724,
      "text materials": 121106,
      "materials methods": 67445,
      "methods bidirectional": 69348,
      "varying data": 131255,
      "partial match": 85292,
      "clinical language": 17770,
      "suggested model": 115343,
      "model did": 70992,
      "did really": 30921,
      "perform deep": 85978,
      "sophisticated solutions": 110675,
      "clinical information": 17768,
      "questions recent": 95351,
      "witnessed great": 132752,
      "works focusing": 134950,
      "novel knowledge": 80610,
      "consists modules": 21483,
      "texts knowledge": 121536,
      "higher precision": 50193,
      "matching strategy": 67434,
      "strategy enables": 113508,
      "different realworld": 31373,
      "tasks realworld": 119437,
      "studies significant": 114282,
      "confirm effectiveness": 21013,
      "datasets approaches": 27319,
      "sentences higher": 106342,
      "supervised sentence": 115833,
      "method assessing": 68647,
      "features associated": 43369,
      "scores baseline": 104155,
      "datasets need": 27593,
      "task rewriting": 118660,
      "question wellformed": 95238,
      "annotations subset": 6467,
      "subset dataset": 114829,
      "dataset quality": 27137,
      "improves average": 52948,
      "average 45": 11177,
      "models constructed": 72974,
      "summary generation": 115640,
      "users solve": 129172,
      "novel joint": 80607,
      "essential information": 38552,
      "alleviate noise": 5042,
      "pairs addition": 83483,
      "reference summaries": 97535,
      "answer summarization": 6584,
      "selection text": 104841,
      "applicability resourcepoor": 6821,
      "summaries learning": 115446,
      "learning retrieve": 62970,
      "retriever model": 102466,
      "model trains": 72223,
      "retrieved documents": 102459,
      "robustness method": 103109,
      "14 points": 317,
      "points unsupervised": 87881,
      "adaptation reading": 3256,
      "linguistic intelligence": 64496,
      "outdomain datasets": 82621,
      "training hypothesize": 123652,
      "discrepancy caused": 32155,
      "caused lack": 15832,
      "lack language": 58723,
      "task allows": 117872,
      "domain unlabeled": 33686,
      "problem provide": 91189,
      "models learns": 73494,
      "lm indomain": 64864,
      "models retain": 73957,
      "domain lm": 33577,
      "domain evaluated": 33521,
      "outperformed model": 82784,
      "model yielded": 72335,
      "agent paper": 4309,
      "agent architecture": 4296,
      "architecture specific": 8751,
      "users provide": 129159,
      "positive feedback": 88321,
      "feedback study": 43837,
      "dataset legal": 27000,
      "methods retrieving": 69734,
      "baselines help": 12407,
      "used current": 128468,
      "current search": 25332,
      "using vietnamese": 130361,
      "exploration data": 41498,
      "paraphrases generated": 84841,
      "generated backtranslation": 47053,
      "backtranslation simple": 11341,
      "achieved second": 2691,
      "efforts automatically": 35188,
      "generating complex": 47204,
      "question datasets": 95145,
      "separate scenarios": 106860,
      "scenarios using": 103874,
      "outperform base": 82688,
      "variant using": 130870,
      "demonstrates importance": 28946,
      "description analysis": 29453,
      "analysis baseline": 5511,
      "simple paraphrase": 109491,
      "typing entity": 126473,
      "availability datasets": 10905,
      "datasets comprehensive": 27367,
      "reports performance": 99087,
      "performance seven": 86709,
      "understanding wide": 127011,
      "test bed": 120423,
      "suitable datasets": 115396,
      "testing models": 120597,
      "knowledge opendomain": 58090,
      "underlying knowledge": 126681,
      "actually learning": 3154,
      "knowledge trained": 58209,
      "stateoftheart qa": 112877,
      "reasoning fundamental": 96250,
      "fundamental complex": 46088,
      "introduce methodology": 56457,
      "allowing systematic": 5119,
      "systematic control": 116667,
      "evaluation automatically": 39128,
      "carefully control": 15518,
      "types structural": 126363,
      "slight increase": 110018,
      "introduced models": 56585,
      "leave room": 63263,
      "improvement assessed": 52681,
      "including approaches": 53261,
      "focus research": 44811,
      "generation single": 47624,
      "single multidocument": 109767,
      "paper ends": 83894,
      "cited papers": 16995,
      "web building": 132224,
      "database schema": 26647,
      "field key": 43959,
      "key concept": 57559,
      "help corpus": 49706,
      "query templates": 94989,
      "data small": 26461,
      "train novel": 122992,
      "domains restaurants": 33852,
      "schema additional": 103893,
      "additional manual": 3547,
      "effort needed": 35177,
      "needed create": 76872,
      "schema furthermore": 103897,
      "hotel domain": 50668,
      "comparable google": 19144,
      "multiplechoice machine": 75751,
      "teaching reading": 119714,
      "utilizes semantic": 130559,
      "text compare": 120803,
      "higher best": 50165,
      "big gap": 13992,
      "indicates significant": 53865,
      "process solving": 91570,
      "mrc problem": 74929,
      "evaluated benchmark": 38956,
      "model reach": 71849,
      "reach new": 95896,
      "entails ability": 37686,
      "showing quality": 108428,
      "showing used": 108434,
      "alleviate annotation": 5029,
      "learning contextualized": 62460,
      "contextualized document": 22541,
      "present contextual": 89418,
      "contextual discourse": 22454,
      "representation efficient": 99219,
      "documents approach": 33184,
      "text medical": 121110,
      "encode position": 36443,
      "clinical entities": 17767,
      "use continuous": 127964,
      "representations resolve": 99856,
      "approximate nearest": 8456,
      "level apply": 63422,
      "patients medical": 85708,
      "medical professionals": 68217,
      "data wikipedia": 26627,
      "generalized model": 46830,
      "able adapt": 1606,
      "additional finetuning": 3518,
      "followup questions": 44996,
      "instantiation framework": 55648,
      "train pointergenerator": 122996,
      "predict followup": 88888,
      "network applied": 77136,
      "challenge text": 16112,
      "prominent performance": 92252,
      "bert stateoftheart": 13235,
      "typically follow": 126432,
      "follow retrieve": 44941,
      "module bert": 74480,
      "efficiency problem": 35038,
      "encoding framework": 36694,
      "framework dual": 45506,
      "bert encodes": 13114,
      "approaches opendomain": 8265,
      "extracted documents": 42151,
      "natural order": 76613,
      "use huge": 128089,
      "encoder achieve": 36491,
      "annotation strategies": 6376,
      "objective develop": 81069,
      "performance conduct": 86249,
      "demonstrate experimentally": 28740,
      "experimentally training": 40739,
      "performance help": 86425,
      "generalizing unseen": 46845,
      "datasets second": 27695,
      "subset samples": 114835,
      "effectiveness demonstrated": 34881,
      "based domain": 11661,
      "clinical setting": 17788,
      "subset 12": 114826,
      "complete dataset": 19763,
      "budgets limited": 14736,
      "framework evaluation": 45530,
      "gain popularity": 46349,
      "noticeable performance": 80399,
      "issues raised": 57070,
      "used establish": 128521,
      "establish performance": 38591,
      "data design": 25831,
      "draw comparisons": 34111,
      "features contribute": 43427,
      "expected answers": 40389,
      "measuring similarities": 67932,
      "scenario challenging": 103821,
      "improve deep": 52370,
      "leverage domain": 63585,
      "pair evaluate": 83430,
      "chinese qa": 16811,
      "psychological perspective": 94212,
      "terms end": 120318,
      "provides theoretical": 94087,
      "future datasets": 46266,
      "capability model": 15183,
      "model constructing": 70902,
      "explanation task": 41288,
      "design reinforced": 29561,
      "models surpass": 74136,
      "limited capacity": 64218,
      "questions inspired": 95318,
      "aims generating": 4534,
      "employ multitask": 36270,
      "learning auxiliary": 62380,
      "fact prediction": 42830,
      "rl framework": 102890,
      "approach experiments": 7550,
      "rouge human": 103294,
      "quality coverage": 94622,
      "learning recover": 62948,
      "propose cooperative": 92605,
      "approach deal": 7472,
      "supervision evaluation": 115883,
      "created benchmarks": 24657,
      "benchmarks based": 12888,
      "chains experimental": 15983,
      "approach graph": 7593,
      "trend nlp": 125741,
      "areas paper": 8898,
      "adjacency matrix": 3847,
      "visualized analysis": 131838,
      "entire graph": 37707,
      "called sentence": 15063,
      "clozestyle machine": 17899,
      "evaluate difficulty": 38823,
      "task add": 117843,
      "correct ones": 24112,
      "contains 100k": 21814,
      "narrative stories": 76176,
      "underperforms human": 126716,
      "models reached": 73858,
      "models generalizability": 73271,
      "outofdomain examples": 82654,
      "data ability": 25550,
      "variations single": 130940,
      "methods following": 69518,
      "evaluate task": 38933,
      "datasets shared": 27704,
      "answering relies": 6692,
      "contexts traditional": 22431,
      "method work": 69230,
      "using dense": 129551,
      "framework evaluated": 45528,
      "broad applicability": 14667,
      "tutoring systems": 126003,
      "methods long": 69602,
      "models increasing": 73390,
      "task future": 118229,
      "generation graph": 47426,
      "focus simple": 44817,
      "kg subgraph": 57662,
      "information kg": 54707,
      "apply bidirectional": 7164,
      "enhance rnn": 37489,
      "rnn decoder": 102913,
      "task mean": 118380,
      "strong pipeline": 113700,
      "intuitively useful": 56673,
      "graphbased reasoning": 48910,
      "complexity results": 19936,
      "simple pipeline": 109492,
      "use bert": 127921,
      "bert identify": 13140,
      "context standard": 22280,
      "standard bert": 112209,
      "sentences strong": 106505,
      "value complex": 130775,
      "complex techniques": 19889,
      "assume gold": 9646,
      "mention information": 68421,
      "paper cast": 83758,
      "given anaphor": 47982,
      "answering framework": 6632,
      "pretrained dataset": 90020,
      "bridging corpora": 14607,
      "markert et": 67237,
      "augment labeled": 10259,
      "regularizer train": 97769,
      "achieves large": 2811,
      "methods variety": 69841,
      "robertabased models": 103005,
      "15 datasets": 343,
      "learn effectively": 62038,
      "using pretraining": 130053,
      "generalize language": 46810,
      "objective propose": 81108,
      "better aligns": 13506,
      "input better": 55302,
      "evaluated text": 39016,
      "models result": 73951,
      "summaries generated": 115441,
      "concise informative": 20702,
      "far robust": 43099,
      "comprehensively verify": 20290,
      "verify robustness": 131593,
      "realworld chinese": 96154,
      "presents challenges": 89831,
      "behavior existing": 12655,
      "models challenge": 72892,
      "codes publicly": 18212,
      "improved unsupervised": 52652,
      "common approach": 18854,
      "scarce costly": 103789,
      "largescale humanlabeled": 61446,
      "pseudotraining data": 94199,
      "applying simple": 7269,
      "sentence original": 105963,
      "original context": 82508,
      "sentence improves": 105902,
      "learn complex": 62002,
      "relationships training": 98337,
      "aim build": 4449,
      "language provided": 60004,
      "explored work": 41640,
      "88 96": 1344,
      "result produced": 101394,
      "different random": 31369,
      "random seeds": 95510,
      "method represent": 69103,
      "documents apply": 33183,
      "entities mention": 37817,
      "allows direct": 5141,
      "encoders knowledge": 36651,
      "pretraining selfsupervised": 90318,
      "selfsupervised masked": 104944,
      "objective words": 81127,
      "improvements 16": 52792,
      "31 f1": 920,
      "f1 comparable": 42651,
      "gains indomain": 46396,
      "dynamic contexts": 34298,
      "environment neural": 38190,
      "impressive accuracy": 52317,
      "knowledge generally": 57948,
      "interpretable work": 56258,
      "automatically built": 10729,
      "bases develop": 12495,
      "knowledge improves": 58007,
      "based network": 11884,
      "multimodal input": 75434,
      "forms context": 45238,
      "alignment multimodal": 4975,
      "dataset multimodal": 27042,
      "graphs generating": 48954,
      "facilitate reasoning": 42788,
      "reasoning propose": 96298,
      "framework constructs": 45471,
      "graph input": 48822,
      "encodes semantic": 36676,
      "perform joint": 86018,
      "model greatly": 71267,
      "collected public": 18433,
      "public school": 94272,
      "school english": 103965,
      "discourselevel context": 32110,
      "experiments significant": 41140,
      "encouraging future": 36777,
      "suffers challenges": 115249,
      "challenges answer": 16127,
      "decision support": 27944,
      "makes attempt": 66757,
      "knowledge enhance": 57898,
      "enhance representation": 37487,
      "graph extracted": 48810,
      "knowledge original": 58093,
      "representation key": 99271,
      "great improvements": 49007,
      "focused generating": 44858,
      "generation largely": 47457,
      "acquire new": 3034,
      "given shared": 48124,
      "space potential": 111043,
      "good candidates": 48467,
      "learning optimize": 62863,
      "propose combined": 92585,
      "generated baseline": 47054,
      "evaluated metrics": 38990,
      "utilizing set": 130581,
      "manner natural": 66954,
      "knowledge single": 58175,
      "single human": 109740,
      "furthermore ensemble": 46166,
      "tasks achieving": 118917,
      "sentence position": 105992,
      "position bias": 88282,
      "answering extractive": 6629,
      "predict start": 88942,
      "simplicity effectiveness": 109568,
      "positional cues": 88298,
      "bias popular": 13831,
      "layer bert": 61701,
      "effective reducing": 34735,
      "trained biased": 123080,
      "structured annotations": 113991,
      "annotations reading": 6459,
      "framework introduces": 45584,
      "textual span": 121748,
      "framework implemented": 45565,
      "new highquality": 78947,
      "comprehension english": 20179,
      "including error": 53293,
      "error distributions": 38305,
      "ability experiments": 1518,
      "provides alternative": 93997,
      "alternative test": 5246,
      "substantially higher": 114891,
      "performance progressively": 86624,
      "rely information": 98713,
      "thanks efficiency": 121770,
      "simply use": 109628,
      "achieved pretraining": 2669,
      "pretraining effective": 90256,
      "stored index": 113389,
      "expensive requires": 40431,
      "method pretrain": 69069,
      "encoders perform": 36660,
      "scoring document": 104225,
      "bagofwords models": 11378,
      "analysis establish": 5602,
      "document length": 33031,
      "support precise": 115984,
      "events described": 39573,
      "phenomena systems": 87246,
      "querying temporal": 95006,
      "exactmatch score": 39727,
      "performance clinical": 86207,
      "clinical domain": 17766,
      "domain expertise": 33527,
      "templates existing": 120085,
      "quantitative experiments": 94869,
      "experiments surprising": 41162,
      "entire dataset": 37703,
      "ability utilize": 1577,
      "knn search": 57727,
      "encountered training": 36745,
      "large datastore": 61076,
      "embedded text": 35363,
      "contributions follows": 22788,
      "large margins": 61146,
      "identifies correct": 51468,
      "easily handle": 34455,
      "berts training": 13288,
      "common situation": 18927,
      "capabilities artificial": 15166,
      "situations training": 109903,
      "set created": 107407,
      "scores evaluation": 104171,
      "challenging nature": 16283,
      "nature task": 76670,
      "generation transformer": 47685,
      "corresponding input": 24290,
      "problem rely": 91203,
      "complexity rely": 19934,
      "auxiliary data": 10874,
      "data unavailable": 26591,
      "additional taskspecific": 3581,
      "finetuned gpt2": 44428,
      "095 meteor": 112,
      "meteor points": 68571,
      "varying model": 131265,
      "order verify": 82428,
      "pairs dialogues": 83517,
      "sites using": 109879,
      "method crowdsourcing": 68755,
      "welldefined information": 132400,
      "possible build": 88386,
      "build high": 14772,
      "ample room": 5371,
      "text address": 120630,
      "new sota": 79140,
      "improvement downstream": 52702,
      "large transformerbased": 61305,
      "use applications": 127895,
      "applications requiring": 7008,
      "classification large": 17248,
      "investigated approaches": 56834,
      "approaches reduce": 8313,
      "reduce model": 97339,
      "adapt transformerbased": 3197,
      "model shared": 72001,
      "pairs unsupervised": 83657,
      "availability largescale": 10921,
      "improve unsupervised": 52580,
      "syntactically divergent": 116510,
      "wikipedia automatically": 132642,
      "extract appropriate": 42067,
      "bert access": 13051,
      "access manually": 1916,
      "models effectiveness": 73119,
      "extraction neural": 42408,
      "typically consist": 126419,
      "reference text": 97538,
      "answers extracted": 6729,
      "importance evidence": 52057,
      "present selftraining": 89684,
      "labels extra": 58604,
      "extra supervision": 42063,
      "demonstrate improvement": 28762,
      "code obtained": 18124,
      "document modeling": 33039,
      "modeling graph": 72439,
      "answer entities": 6526,
      "dependencies address": 29063,
      "comprehension framework": 20183,
      "framework focuses": 45548,
      "networks obtain": 77680,
      "levels representations": 63561,
      "learned simultaneously": 62258,
      "respectively way": 101177,
      "systems long": 116982,
      "addressed question": 3788,
      "models separately": 74006,
      "separately using": 106879,
      "modeling problems": 72516,
      "websites using": 132307,
      "using recently": 130086,
      "transformers investigate": 124450,
      "trained separate": 123270,
      "novel contextaware": 80514,
      "generate ranked": 46993,
      "spans tokens": 111212,
      "tokens given": 122312,
      "document approach": 32952,
      "adapt stateoftheart": 3193,
      "bertbased methods": 13274,
      "collection speech": 18490,
      "associated news": 9604,
      "performance bertbased": 86178,
      "task outperform": 118489,
      "relative ranking": 98368,
      "tasks confirm": 119002,
      "best bert": 13312,
      "highquality conversational": 50373,
      "nlg approach": 79512,
      "fluent grammatical": 44706,
      "responses maintaining": 101281,
      "augmentation generate": 10277,
      "develop syntactic": 30239,
      "2019 human": 698,
      "conversational responses": 23026,
      "wikidata paper": 132633,
      "complexity english": 19910,
      "sparql queries": 111218,
      "containing entities": 21788,
      "labels dataset": 58587,
      "domains covering": 33752,
      "relations added": 98104,
      "previously unstudied": 90631,
      "domains added": 33726,
      "training making": 123700,
      "challenging propose": 16300,
      "lexical variety": 63841,
      "use distant": 128006,
      "express relation": 41705,
      "unseen domain": 127521,
      "reading goal": 96021,
      "base text": 11485,
      "text require": 121256,
      "reasoning paper": 96283,
      "strategy utilizing": 113545,
      "interpretable visualizing": 56255,
      "structurally similar": 113800,
      "need consolidate": 76789,
      "research literature": 100549,
      "approach tackling": 7954,
      "models highlighting": 73333,
      "future researchers": 46311,
      "performance room": 86689,
      "systems document": 116856,
      "independently considering": 53785,
      "relationship documents": 98290,
      "relevant document": 98543,
      "question documents": 95151,
      "recently paper": 96726,
      "pathbased approaches": 85680,
      "approaches better": 8087,
      "contains new": 21833,
      "question information": 95168,
      "accuracy previously": 2246,
      "surpasses human": 116106,
      "constructed corpus": 21656,
      "reader read": 95955,
      "queries searching": 94939,
      "way combine": 132065,
      "single answer": 109706,
      "demonstrate achieves": 28655,
      "superior accuracy": 115678,
      "reasoning skills": 96310,
      "sentiment perform": 106768,
      "mining work": 70268,
      "released dataset": 98479,
      "subjective opinions": 114700,
      "domains answer": 33730,
      "task primary": 118563,
      "investigate inner": 56771,
      "representations corresponding": 99577,
      "observation holds": 81164,
      "objective subjective": 81117,
      "correct compared": 24097,
      "predictions results": 89189,
      "implications downstream": 51990,
      "know neural": 57734,
      "corpora necessary": 23536,
      "health domain": 49618,
      "applications little": 6958,
      "research domain": 100479,
      "corpus vietnamese": 24063,
      "corpus comprises": 23707,
      "crowdworkers create": 25127,
      "based collection": 11582,
      "process creating": 91448,
      "creating corpus": 24696,
      "word matching": 133358,
      "types machine": 126315,
      "baseline performances": 12284,
      "machine model": 65843,
      "achieves exact": 2782,
      "corpus significant": 23997,
      "corpus indicate": 23838,
      "indicate improvements": 53838,
      "research purpose": 100600,
      "scale chinese": 103705,
      "opinion based": 82083,
      "finally abstractive": 44141,
      "support evidence": 115969,
      "inference logical": 54162,
      "bert achieve": 13052,
      "performance indicating": 86461,
      "codes datasets": 18200,
      "task scarcity": 118663,
      "issues recently": 57071,
      "relationships sentence": 98331,
      "general qa": 46700,
      "focus applying": 44735,
      "dataset obtains": 27063,
      "obtains better": 81457,
      "challenge bioasq": 16002,
      "phase present": 87213,
      "present sequential": 89690,
      "tasks finetuned": 119134,
      "reconstruction strategy": 97123,
      "systems subtasks": 117168,
      "input reconstruction": 55412,
      "construct input": 21633,
      "input prompt": 55406,
      "systems approaches": 116735,
      "964 accuracy": 1434,
      "release annotation": 98434,
      "collection effort": 18473,
      "effort data": 35167,
      "obtained preliminary": 81393,
      "limit development": 64154,
      "poor diversity": 88055,
      "scale end": 103717,
      "question provide": 95207,
      "provide corresponding": 93796,
      "datasets suggests": 27739,
      "diagnostic dataset": 30510,
      "testing multiple": 120598,
      "thorough evaluation": 121884,
      "datasets obtained": 27601,
      "processing makes": 91709,
      "based texts": 12116,
      "problems word": 91382,
      "different vietnamese": 31557,
      "constructing virtual": 21682,
      "providing finegrained": 94116,
      "building reliable": 14881,
      "framework automatically": 45433,
      "require external": 100135,
      "tuples unstructured": 125950,
      "context semantics": 22257,
      "structure allows": 113805,
      "performance retrieval": 86683,
      "studies reading": 114270,
      "understanding entities": 126835,
      "entities relationships": 37862,
      "focus entities": 44758,
      "datasets exist": 27461,
      "domain covid19": 33494,
      "achieved tremendous": 2716,
      "largevolume traffic": 61549,
      "networks online": 77681,
      "additional cost": 3505,
      "accuracy drop": 2142,
      "aim extract": 4455,
      "represent relations": 99129,
      "context instead": 22148,
      "messages extracted": 68509,
      "occur context": 81497,
      "context implicitly": 22128,
      "represent relation": 99127,
      "finetuning strategy": 44518,
      "weaklylabeled data": 132202,
      "knowledge student": 58195,
      "extraction generative": 42349,
      "generative machine": 47733,
      "crucial model": 25150,
      "suffer generating": 115222,
      "redundant words": 97480,
      "thorough experiments": 121887,
      "demonstrate novel": 28810,
      "approach alleviate": 7343,
      "iterative document": 57129,
      "reranking existing": 100370,
      "reasoning make": 96269,
      "number relevant": 80953,
      "extraction address": 42255,
      "process improve": 91506,
      "improve retrieval": 52527,
      "framework method": 45611,
      "usually employ": 130417,
      "text modern": 121129,
      "selection step": 104830,
      "issue investigate": 57003,
      "investigate best": 56726,
      "approach groups": 7594,
      "candidates retrievalbased": 15145,
      "question examine": 95155,
      "demonstrate ensemble": 28737,
      "module represents": 74508,
      "models faster": 73241,
      "generalizes crossdomain": 46836,
      "large complex": 61057,
      "complex documents": 19809,
      "size given": 109925,
      "produce relevant": 91919,
      "retrieval propose": 102423,
      "time unsupervised": 122132,
      "explore suitability": 41592,
      "methods biomedical": 69351,
      "biobert scibert": 14154,
      "representations biomedical": 99523,
      "entities context": 37760,
      "correct entity": 24098,
      "denoising task": 29002,
      "task enables": 118141,
      "learn good": 62057,
      "good representations": 48497,
      "abundant unlabeled": 1839,
      "unlabeled biomedical": 127373,
      "text helps": 121034,
      "requiring model": 100351,
      "experiments pretraining": 41064,
      "retrieval large": 102402,
      "early fusion": 34390,
      "train accurate": 122897,
      "crossattention model": 24869,
      "examples neural": 39859,
      "data significantly": 26450,
      "given various": 48170,
      "10 years": 156,
      "years mainly": 135271,
      "studies stateoftheart": 114284,
      "summary used": 115650,
      "coverage mechanism": 24514,
      "respectively performed": 101156,
      "context explaining": 22097,
      "extraction essential": 42326,
      "comprehension existing": 20182,
      "given corresponding": 48009,
      "probability vectors": 90872,
      "matrix matrix": 67478,
      "address computational": 3667,
      "competitive models": 19656,
      "backbone proposed": 11288,
      "approach consistent": 7446,
      "improvements datasets": 52839,
      "vietnamese dataset": 131672,
      "lack benchmark": 58687,
      "datasets vietnamese": 27792,
      "articles wikipedia": 9229,
      "experiments stateoftheart": 41153,
      "chinese experimental": 16767,
      "result substantial": 101410,
      "collected chinese": 18405,
      "require understanding": 100204,
      "accuracy 531": 2066,
      "question data": 95143,
      "tasks treat": 119577,
      "highquality diverse": 50380,
      "utilizes transformer": 130562,
      "continuous embedding": 22617,
      "discrete space": 32176,
      "additional question": 3563,
      "data comprehensive": 25759,
      "learning generating": 62613,
      "comprehension given": 20184,
      "given article": 47985,
      "main points": 66450,
      "present learning": 89535,
      "scheme generate": 103927,
      "syntactic tags": 116490,
      "tags particular": 117481,
      "scheme called": 103915,
      "sequences training": 107143,
      "interrogative sentence": 56300,
      "sentence generate": 105886,
      "tests large": 120608,
      "97 accuracy": 1439,
      "variety settings": 131015,
      "type question": 126221,
      "tries gap": 125779,
      "range models": 95581,
      "pragmatic strategies": 88758,
      "remains unexplored": 98829,
      "bias estimates": 13803,
      "studies use": 114294,
      "important classes": 52122,
      "ethnicity religion": 38754,
      "models broad": 72862,
      "size context": 109916,
      "choice question": 16899,
      "comprehension main": 20195,
      "extract evidence": 42080,
      "evidence given": 39646,
      "rely human": 98709,
      "efforts address": 35187,
      "model relationships": 71891,
      "sentences filter": 106317,
      "reduction human": 97449,
      "graph reasoning": 48860,
      "deals challenge": 27874,
      "challenge learning": 16047,
      "reasoning multihop": 96276,
      "qa propose": 94516,
      "proposed graph": 93305,
      "documentlevel graph": 33143,
      "sentence question": 106013,
      "sentence contain": 105805,
      "incorporating argument": 53517,
      "types argument": 126252,
      "argument phrases": 8949,
      "graph encoder": 48801,
      "setting benchmark": 107734,
      "benchmark compared": 12743,
      "shallow heuristics": 107919,
      "limit ability": 64153,
      "learn correct": 62012,
      "behavior identify": 12661,
      "examples benchmark": 39812,
      "behavior experiments": 12657,
      "worse overall": 135066,
      "performance adding": 86124,
      "provide concrete": 93791,
      "given database": 48012,
      "covers different": 24555,
      "uses automatic": 129194,
      "automatic paraphrasing": 10621,
      "attribute different": 10179,
      "speech uses": 111842,
      "sentences apply": 106213,
      "tested natural": 120579,
      "lower model": 65442,
      "language annotations": 58831,
      "paraphrase data": 84807,
      "collected crowdworkers": 18413,
      "overnight dataset": 83350,
      "methods generally": 69521,
      "rely external": 98694,
      "resources endtoend": 100968,
      "model comprises": 70874,
      "need separate": 76858,
      "model generator": 71247,
      "lm using": 64873,
      "models outperforming": 73689,
      "process current": 91450,
      "set perform": 107526,
      "users judge": 129141,
      "usefulness model": 128960,
      "model practical": 71755,
      "practical setting": 88713,
      "making promising": 66861,
      "answer wrong": 6592,
      "underexplored area": 126661,
      "paper question": 84394,
      "different answer": 30994,
      "approaches introduce": 8196,
      "simple framework": 109436,
      "enables automated": 36376,
      "little variation": 64834,
      "question benefit": 95132,
      "automatically use": 10839,
      "use methodology": 128144,
      "models really": 73866,
      "era big": 38264,
      "domain gained": 33537,
      "gained wide": 46373,
      "transparency fairness": 125534,
      "understand particular": 126764,
      "fail address": 42949,
      "generating explanations": 47219,
      "various stages": 131203,
      "approach supervised": 7949,
      "measure effectiveness": 67795,
      "user survey": 129047,
      "study significant": 114519,
      "dimensions human": 31767,
      "interaction community": 55940,
      "layers transformerbased": 61813,
      "wikipedia key": 132660,
      "high complexity": 50042,
      "filter corpus": 44070,
      "solution consists": 110551,
      "consists applying": 21468,
      "applying generic": 7248,
      "significant speedup": 108868,
      "cases better": 15638,
      "field education": 43950,
      "large impact": 61105,
      "topic generating": 122524,
      "area work": 8887,
      "work train": 134856,
      "showed statistically": 108391,
      "queries users": 94942,
      "capabilities facilitate": 15169,
      "questions focus": 95307,
      "focus main": 44785,
      "document topics": 33098,
      "model outofdomain": 71624,
      "lack gold": 58711,
      "framework design": 45488,
      "model ability": 70504,
      "generate accurate": 46898,
      "framework shows": 45681,
      "shows interesting": 108590,
      "span selection": 111131,
      "using distantly": 129593,
      "trained positive": 123240,
      "confidence predictions": 20994,
      "annotations perform": 6452,
      "false positive": 43042,
      "accuracy natural": 2212,
      "reduces search": 97397,
      "question community": 95138,
      "evidence indicates": 39649,
      "originally intended": 82568,
      "approximate search": 8461,
      "method enhance": 68807,
      "rate endtoend": 95792,
      "efforts natural": 35201,
      "event influence": 39522,
      "fundamental understanding": 46130,
      "generate event": 46938,
      "methods event": 69486,
      "rouge points": 103298,
      "furthermore event": 46168,
      "require background": 100113,
      "knowledge multihop": 58067,
      "retrieval recent": 102426,
      "recent pretrained": 96489,
      "document datasets": 32985,
      "outperform human": 82711,
      "human annotator": 50749,
      "contrast paper": 22700,
      "encoder models": 36543,
      "used retrieve": 128733,
      "domainspecific data": 33895,
      "tasks clear": 118978,
      "clear strategies": 17731,
      "different strategies": 31452,
      "performance par": 86600,
      "text develop": 120871,
      "develop unified": 30247,
      "employ single": 36281,
      "subtasks retrieving": 114979,
      "iterative fashion": 57130,
      "like knowledge": 64054,
      "combining existing": 18724,
      "wikipedia corpora": 132647,
      "corpora versions": 23617,
      "benchmarks new": 12926,
      "benchmark make": 12838,
      "make new": 66702,
      "benchmark available": 12733,
      "real scenario": 96076,
      "query work": 94994,
      "crossdomain texttosql": 24906,
      "improvement task": 52774,
      "given questionanswer": 48098,
      "method combination": 68700,
      "dataset produce": 27110,
      "experiments evaluations": 40942,
      "evaluations human": 39465,
      "expense quality": 40409,
      "novel applications": 80467,
      "pretraining knowledge": 90267,
      "strategies significant": 113488,
      "clinical texts": 17792,
      "model retraining": 71927,
      "models requiring": 73940,
      "requiring manual": 100348,
      "annotations order": 6448,
      "role achieving": 103154,
      "hierarchical memory": 49973,
      "review question": 102564,
      "introducing auxiliary": 56627,
      "memory module": 68350,
      "components context": 20022,
      "memory working": 68396,
      "syntactically wellformed": 116520,
      "dataset comprehensive": 26812,
      "evaluation reasoning": 39353,
      "provide complete": 93784,
      "examples existing": 39830,
      "datasets require": 27675,
      "information benefits": 54395,
      "rules create": 103389,
      "demonstrate dataset": 28696,
      "predicting correct": 88979,
      "answer text": 6585,
      "important number": 52199,
      "based main": 11824,
      "learning manner": 62706,
      "attention method": 9928,
      "learns context": 63200,
      "experiments reading": 41103,
      "number answer": 80838,
      "conduct experiment": 20844,
      "models zeroshot": 74370,
      "aims select": 4572,
      "question task": 95226,
      "highest confidence": 50224,
      "parameters experimental": 84746,
      "generating fact": 47220,
      "number active": 80834,
      "contemporary media": 21849,
      "good intentions": 48481,
      "claim detection": 17008,
      "providing information": 94117,
      "generates set": 47174,
      "train components": 122911,
      "increases accuracy": 53636,
      "reduce time": 97362,
      "explanation generation": 41282,
      "various benchmarks": 131053,
      "models reallife": 73864,
      "humans paper": 51096,
      "achieve consistent": 2451,
      "traditional attention": 122796,
      "incomplete information": 53422,
      "information reading": 54896,
      "humans read": 51106,
      "rc tasks": 95875,
      "performance identifying": 86435,
      "lack sufficient": 58749,
      "answer missing": 6543,
      "increasing difficulty": 53670,
      "difficulty task": 31709,
      "work various": 134877,
      "datasets construct": 27378,
      "construct baseline": 21619,
      "f1 task": 42696,
      "code baseline": 18075,
      "time user": 122134,
      "set ranking": 107555,
      "based determinantal": 11641,
      "determinantal point": 30113,
      "point processes": 87811,
      "processes dpps": 91605,
      "built dataset": 14917,
      "dataset demonstrated": 26860,
      "demonstrated proposed": 28928,
      "detection duplicate": 29928,
      "leverage annotations": 63576,
      "work leverage": 134610,
      "neural representations": 78651,
      "representations study": 99909,
      "method aggregate": 68624,
      "crossdomain scenarios": 24898,
      "widespread form": 132615,
      "form standard": 45127,
      "quite challenging": 95397,
      "knowledge currently": 57849,
      "applies attention": 7150,
      "control level": 22810,
      "stateoftheart chinese": 112607,
      "chinese medical": 16785,
      "medical question": 68218,
      "matching methods": 67411,
      "focus semantic": 44813,
      "ignoring semantic": 51693,
      "semantic association": 104995,
      "models tackle": 74152,
      "better adapt": 13501,
      "advantages different": 4075,
      "structures propose": 114098,
      "network extract": 77253,
      "different structure": 31457,
      "representation experiments": 99234,
      "stages neural": 112161,
      "health care": 49616,
      "questions online": 95337,
      "numerous data": 81016,
      "automated machine": 10449,
      "perspective information": 87163,
      "thinking skills": 121867,
      "evaluated paper": 38999,
      "framework understand": 45726,
      "thought process": 121910,
      "human thinking": 50974,
      "benefits adopting": 13004,
      "biases cause": 13865,
      "cause errors": 15824,
      "provides set": 94073,
      "component automatic": 19969,
      "issues concerning": 57041,
      "unnatural language": 127476,
      "mainly obtained": 66492,
      "main strategies": 66466,
      "long contexts": 65066,
      "answer key": 6539,
      "utilized enhance": 130540,
      "baselines addition": 12351,
      "work established": 134493,
      "research future": 100522,
      "work make": 134626,
      "pipeline building": 87531,
      "reduce search": 97356,
      "text respectively": 121261,
      "improve computational": 52356,
      "scalability challenge": 103682,
      "processing documents": 91666,
      "heavily depend": 49650,
      "work time": 134853,
      "time learn": 122044,
      "learn dense": 62018,
      "learn phrase": 62111,
      "learning reduce": 62949,
      "matches performance": 67387,
      "model easy": 71038,
      "study opendomain": 114461,
      "unstructured semistructured": 127581,
      "semistructured knowledge": 105588,
      "text applies": 120653,
      "11 points": 225,
      "unified knowledge": 127112,
      "combine heterogeneous": 18626,
      "advancing stateoftheart": 4033,
      "26 points": 842,
      "memory efficient": 68311,
      "baseline open": 12279,
      "important improvements": 52169,
      "tasks effective": 119079,
      "study memory": 114437,
      "filtering evaluate": 44082,
      "text wikipedia": 121410,
      "link entity": 64634,
      "context corresponding": 22043,
      "representing unstructured": 100067,
      "nli identify": 79544,
      "relations choice": 98121,
      "knowledgeintensive tasks": 58280,
      "tasks opendomain": 119351,
      "degrades considerably": 28563,
      "question neural": 95190,
      "model universal": 72245,
      "perform robustly": 86058,
      "tasks closely": 118979,
      "art multiple": 9064,
      "multiple benchmarks": 75506,
      "proposed provides": 93530,
      "systems discuss": 116855,
      "proposed implemented": 93311,
      "framework breaks": 45445,
      "breaks problem": 14582,
      "problem steps": 91249,
      "report progress": 99028,
      "modest gains": 74438,
      "reached human": 95906,
      "pretraining scheme": 90316,
      "mask set": 67283,
      "replaced special": 98936,
      "representation later": 99277,
      "used finetuning": 128557,
      "obtains surprisingly": 81481,
      "fail consider": 42954,
      "ignore crucial": 51665,
      "utilize rich": 130527,
      "unified way": 127134,
      "additional constraints": 3499,
      "validate idea": 130723,
      "idea create": 51321,
      "framework equipped": 45524,
      "equipped novel": 38242,
      "stateoftheart extractive": 112660,
      "need expensive": 76802,
      "used realworld": 128716,
      "paradigm paper": 84546,
      "possibility transfer": 88371,
      "search task": 104355,
      "fashion model": 43120,
      "previous stateofart": 90468,
      "f1score results": 42718,
      "originally trained": 82572,
      "systems future": 116914,
      "key limitation": 57586,
      "limitation current": 64160,
      "inferred using": 54265,
      "using strategy": 130228,
      "fundamental challenge": 46087,
      "crowdsourcing workers": 25125,
      "careful control": 15510,
      "examples consisting": 39816,
      "cover wide": 24493,
      "strategies empirically": 113457,
      "singlehop reasoning": 109829,
      "complex queries": 19864,
      "semantic feature": 105049,
      "feature interaction": 43290,
      "generally captures": 46851,
      "fails capture": 42981,
      "completion strategies": 19795,
      "simple ones": 109489,
      "ones condition": 81680,
      "additional annotations": 3491,
      "represent word": 99153,
      "document extract": 33006,
      "sentences related": 106470,
      "systems quickly": 117093,
      "providing accurate": 94099,
      "answer machine": 6540,
      "document generate": 33015,
      "generate final": 46942,
      "document tackle": 33092,
      "task pretraining": 118557,
      "methods codes": 69373,
      "incremental knowledge": 53729,
      "base developed": 11451,
      "approaches assume": 8067,
      "static knowledge": 113065,
      "directly apply": 31863,
      "forgetting problem": 45064,
      "new incremental": 78953,
      "learning capacity": 62419,
      "problem taking": 91259,
      "task stateoftheart": 118740,
      "performance extractive": 86372,
      "requiring models": 100353,
      "text production": 121217,
      "confidence estimates": 20985,
      "improve systems": 52559,
      "measured area": 67832,
      "area curve": 8869,
      "curve auc": 25430,
      "auc scores": 10219,
      "models gradients": 73311,
      "prediction particularly": 89101,
      "question matching": 95181,
      "taxonomy paper": 119668,
      "technique semantic": 119813,
      "obtained deep": 81357,
      "knowledge taxonomy": 58203,
      "contents web": 21994,
      "requires understand": 100328,
      "requires certain": 100249,
      "certain structural": 15955,
      "structural understanding": 113798,
      "information visual": 55089,
      "baselines publicly": 12448,
      "datasets survey": 27743,
      "datasets representations": 27672,
      "tables web": 117270,
      "interpret standard": 56216,
      "article describing": 9123,
      "work aim": 134363,
      "aim improve": 4461,
      "information surrounding": 55022,
      "surrounding text": 116157,
      "obtaining significant": 81448,
      "achieve low": 2495,
      "models score": 73978,
      "individually ignoring": 53944,
      "document potential": 33051,
      "expensive models": 40425,
      "designed machine": 29607,
      "approach efficiently": 7516,
      "context improves": 22131,
      "minimal impact": 70163,
      "ranking formulate": 95667,
      "optimization objectives": 82201,
      "scheme experimental": 103924,
      "improvements hold": 52859,
      "improvements choice": 52823,
      "question introduce": 95170,
      "set 20": 107341,
      "number benchmarks": 80848,
      "useful machine": 128905,
      "question collect": 95137,
      "use incomplete": 128092,
      "use generated": 128067,
      "instances improve": 55627,
      "2018 stateoftheart": 686,
      "baselines demonstrating": 12385,
      "examples question": 39869,
      "vanilla models": 130815,
      "prediction combined": 89041,
      "defense techniques": 28474,
      "calling question": 15072,
      "strong adversarial": 113618,
      "retrieving information": 102471,
      "challenging deal": 16243,
      "paper point": 84071,
      "architecture models": 8696,
      "data considers": 25776,
      "published stateoftheart": 94373,
      "survey approaches": 116161,
      "approaches challenges": 8093,
      "challenges automatic": 16137,
      "applications effectively": 6921,
      "effectively perceive": 34837,
      "understand complex": 126740,
      "question entailment": 95153,
      "approaches survey": 8366,
      "potential future": 88556,
      "directions explore": 31839,
      "set evaluation": 107434,
      "systems deployed": 116838,
      "voice assistants": 131924,
      "questions languages": 95325,
      "identifying correct": 51593,
      "components pipeline": 20043,
      "sources error": 110886,
      "models conclude": 72951,
      "conclude substantial": 20718,
      "substantial room": 114877,
      "realworld use": 96188,
      "hope findings": 50626,
      "systems actually": 116718,
      "questionanswer qa": 95250,
      "pairs critical": 83508,
      "advancing research": 4030,
      "areas like": 8893,
      "documents requires": 33284,
      "mining appropriate": 70228,
      "conventional pipeline": 22898,
      "approach specifically": 7928,
      "trained deployed": 123113,
      "solve complex": 110592,
      "questions quality": 95346,
      "effective approaches": 34632,
      "approaches adapted": 8047,
      "model roberta": 71941,
      "representations influence": 99692,
      "representations contextaware": 99565,
      "attention computation": 9806,
      "present information": 89515,
      "items text": 57113,
      "keyword searches": 57639,
      "significant challenging": 108738,
      "model determines": 70983,
      "implicitly explicitly": 52029,
      "propose multistep": 92806,
      "continuously integrate": 22652,
      "achieves remarkable": 2845,
      "analysis verifies": 5934,
      "experience work": 40443,
      "leverage recent": 63618,
      "work summarization": 134833,
      "evaluate factuality": 38837,
      "context transformer": 22305,
      "provide flexibility": 93831,
      "handling problem": 49418,
      "graph framework": 48813,
      "sequence actions": 106914,
      "executed knowledge": 39971,
      "types compared": 126261,
      "improvement 11": 52664,
      "generating reading": 47252,
      "languages languagespecific": 60673,
      "corpus required": 23970,
      "architectures method": 8824,
      "method surpasses": 69173,
      "literature shows": 64773,
      "evaluation complementary": 39153,
      "complementary evidence": 19744,
      "evidence identification": 39648,
      "problem aims": 90925,
      "end proposes": 36829,
      "learns vector": 63242,
      "diversity selected": 32894,
      "set addition": 107350,
      "qa domain": 94504,
      "domain open": 33602,
      "tackle opendomain": 117302,
      "tables time": 117269,
      "designed handle": 29596,
      "rely retrieval": 98736,
      "rendering training": 98889,
      "problem exacerbated": 91039,
      "rely lexical": 98716,
      "training allows": 123356,
      "annotations test": 6472,
      "points achieves": 87846,
      "just right": 57470,
      "predicted best": 88957,
      "groundtruth answer": 49129,
      "case existing": 15584,
      "question propose": 95206,
      "introduce semantic": 56532,
      "improvements code": 52824,
      "available link": 11034,
      "utilizing relation": 130578,
      "entities effective": 37772,
      "models solely": 74065,
      "rely fixed": 98696,
      "information subgraphs": 55012,
      "relation representation": 98066,
      "information additionally": 54366,
      "ability experimental": 1516,
      "contain abundant": 21732,
      "issues demonstrate": 57043,
      "setting requires": 107776,
      "requires generating": 100273,
      "freeform text": 45777,
      "humangenerated explanations": 51023,
      "involving entities": 56914,
      "relations provide": 98243,
      "networks approach": 77509,
      "correlations entity": 24255,
      "recognition module": 96911,
      "ranks relevant": 95705,
      "types specifically": 126361,
      "improves f1score": 52982,
      "increase f1score": 53600,
      "models compositional": 72944,
      "structured models": 114018,
      "problems provide": 91364,
      "provide interpretable": 93862,
      "intermediate outputs": 56136,
      "confidence model": 20991,
      "models challenging": 72896,
      "intermediate tasks": 56149,
      "correctly work": 24183,
      "provide stronger": 93931,
      "share internal": 107953,
      "encourage consistency": 36750,
      "external supervision": 42045,
      "signal provided": 108692,
      "individual training": 53937,
      "examples apply": 39808,
      "improve compositional": 52354,
      "using templates": 130278,
      "best place": 13408,
      "tasks unclear": 119580,
      "unclear models": 126635,
      "learn study": 62160,
      "different subdomains": 31461,
      "shows learn": 108593,
      "broad notion": 14673,
      "notion entities": 80407,
      "performance highly": 86430,
      "specific entities": 111437,
      "entities observed": 37828,
      "set true": 107618,
      "questions study": 95367,
      "results difficulty": 101728,
      "making claims": 66830,
      "models world": 74363,
      "linguistic competence": 64442,
      "based performance": 11925,
      "specific benchmark": 111410,
      "benchmark problems": 12849,
      "foster future": 45336,
      "reasoning unstructured": 96321,
      "chain consists": 15975,
      "text assuming": 120665,
      "captures implicit": 15445,
      "attention present": 9975,
      "neurosymbolic approach": 78752,
      "2019 learning": 703,
      "clevr dataset": 17746,
      "performance visual": 86849,
      "global knowledge": 48244,
      "ambiguity humans": 5281,
      "contexts form": 22394,
      "useful context": 128870,
      "context inspired": 22147,
      "taking difference": 117546,
      "useful generate": 128889,
      "access knowledge": 1912,
      "different questions": 31368,
      "specialized particular": 111394,
      "data overcome": 26203,
      "literature address": 64742,
      "central research": 15905,
      "suitable qa": 115405,
      "challenge showing": 16102,
      "1000 examples": 174,
      "benefit interpretability": 12974,
      "knowledge compared": 57831,
      "highly constrained": 50305,
      "model comparable": 70851,
      "requires precise": 100309,
      "precise reasoning": 88782,
      "reasoning entity": 96247,
      "relations represented": 98248,
      "terms labels": 120342,
      "predicting sequential": 89012,
      "relation path": 98062,
      "graph features": 48811,
      "lacks interpretability": 58776,
      "transparent model": 125536,
      "text relations": 121236,
      "relations unified": 98274,
      "steps step": 113334,
      "results sequence": 102159,
      "extraction setting": 42486,
      "setting natural": 107759,
      "following general": 44968,
      "article investigate": 9139,
      "biomedical domains": 14177,
      "domains discover": 33764,
      "tagging approach": 117373,
      "variable number": 130846,
      "requiring postprocessing": 100354,
      "question major": 95178,
      "major qa": 66583,
      "reveals little": 102518,
      "accuracy lower": 2201,
      "people based": 85869,
      "using syntactic information": 130256,
      "information question answering": 54893,
      "written natural language": 135139,
      "generated knowledge base": 47089,
      "answers natural language": 6736,
      "analyzing strengths limitations": 6054,
      "trecstyle evaluation workshop": 125573,
      "effectiveness method means": 34907,
      "method means experiments": 68958,
      "answering natural language": 6652,
      "recent technological advances": 96550,
      "language processing goal": 59866,
      "features model learns": 43610,
      "answer given question": 6534,
      "approach solving task": 7926,
      "external resources approach": 42039,
      "does require feature": 33392,
      "require feature engineering": 100137,
      "feature engineering does": 43274,
      "wide range domains": 132501,
      "languages experimental results": 60570,
      "experimental results standard": 40703,
      "results standard benchmark": 102201,
      "standard benchmark dataset": 112204,
      "works significantly better": 134979,
      "compositional semantic parsing": 20131,
      "important aspects semantic": 52106,
      "semantic parsing question": 105159,
      "parsing question answering": 85206,
      "space logical forms": 111017,
      "logical forms propose": 65009,
      "shown remarkable progress": 108521,
      "paper investigate recently": 84021,
      "investigate recently proposed": 56809,
      "reasoning previous work": 96293,
      "question answering paper": 95069,
      "trained corpus questionanswer": 123102,
      "empirical study shows": 36205,
      "demonstrates proposed model": 28958,
      "proposed model outperform": 93455,
      "sparse nonnegative matrix": 111232,
      "4th grade science": 1063,
      "grade science exams": 48588,
      "processing nlp machine": 91742,
      "content work investigate": 21979,
      "based feature engineering": 11715,
      "feature engineering approaches": 43273,
      "manually designed features": 67057,
      "relevant answering question": 98524,
      "cnn daily mail": 18001,
      "model uses attention": 72263,
      "ensemble models sets": 37611,
      "models sets new": 74015,
      "state art evaluated": 112440,
      "major goal natural": 66563,
      "goal natural language": 48369,
      "model sets new": 71994,
      "syntactic semantic trees": 116478,
      "final model gives": 44110,
      "pairs convolutional neural": 83501,
      "propose convolutional neural": 92604,
      "neural networks learning": 78503,
      "approach state art": 7932,
      "based deep convolutional": 11628,
      "deep convolutional network": 28222,
      "neural networks based": 78443,
      "boost performance various": 14436,
      "problem previous works": 91172,
      "used deep learning": 128478,
      "semantic representation sentence": 105231,
      "novel endtoend learning": 80557,
      "proposed model demonstrates": 93442,
      "surpassing previous stateoftheart": 116118,
      "previous stateoftheart systems": 90484,
      "large scale news": 61256,
      "largescale knowledge bases": 61449,
      "answer questions like": 6568,
      "knowledge bases provide": 57811,
      "knowledge base propose": 57782,
      "text evaluate models": 120922,
      "benchmarks outperforming previous": 12930,
      "previous neural models": 90437,
      "models significant margin": 74036,
      "keyvalue memory networks": 57631,
      "able answer questions": 1608,
      "stateoftheart results existing": 112901,
      "attentive convolutional neural": 10108,
      "subject predicate object": 114682,
      "characterlevel convolutional neural": 16560,
      "neural network charcnn": 78274,
      "makes main contributions": 66792,
      "outperforms stateoftheart entity": 83004,
      "simple recurrent neural": 109502,
      "factoid question answering": 42851,
      "neural network solve": 78397,
      "network solve problem": 77439,
      "yields substantial improvements": 135446,
      "networks based complex": 77518,
      "based complex architectures": 11587,
      "x1 entertainment platform": 135188,
      "stanford question answering": 112378,
      "answering dataset squad": 6619,
      "future research dataset": 46296,
      "question answering recently": 95095,
      "recently bidirectional recurrent": 96668,
      "answering qa tasks": 6676,
      "models extract information": 73222,
      "select correct answer": 104695,
      "use background knowledge": 127917,
      "datasets various domains": 27788,
      "providing strong baseline": 94138,
      "baseline results future": 12302,
      "largescale language understanding": 61458,
      "largescale natural language": 61482,
      "structured knowledge base": 114007,
      "question answering models": 95061,
      "word character sequences": 132944,
      "multiple choice reading": 75514,
      "choice reading comprehension": 16903,
      "hermann et al": 49844,
      "information plain text": 54844,
      "new task machine": 79201,
      "native language english": 76224,
      "native languages english": 76228,
      "framework question answering": 45667,
      "especially question answering": 38496,
      "integrate character embedding": 55747,
      "model question answering": 71838,
      "argue better approach": 8911,
      "data using taskspecific": 26611,
      "task using data": 118831,
      "improves performance tasks": 53022,
      "task best model": 117934,
      "outperforming strong baseline": 82834,
      "generation knowledge base": 47450,
      "significantly reduces human": 109048,
      "question answering english": 95042,
      "english arabic paper": 37070,
      "impact different types": 51869,
      "questions community question": 95288,
      "released semeval2016 task": 98496,
      "task community question": 117982,
      "robust noisy data": 103057,
      "context language modeling": 22158,
      "paperno et al": 84498,
      "neural networks models": 78516,
      "problem automatically generating": 90948,
      "propose endtoend approach": 92650,
      "question key challenge": 95172,
      "comprehension paper proposes": 20211,
      "model able extract": 70515,
      "neural rc models": 78644,
      "attention mechanism acquire": 9879,
      "answer experimental results": 6529,
      "question answering focused": 95047,
      "questionanswer pairs total": 95249,
      "question answering reading": 95091,
      "reading comprehension task": 96014,
      "best published results": 13425,
      "answering query given": 6678,
      "requires modeling complex": 100301,
      "modeling complex interactions": 72399,
      "context different levels": 22060,
      "experimental evaluations model": 40536,
      "evaluations model achieves": 39470,
      "long documents present": 65073,
      "improving performance stateoftheart": 53139,
      "learning experiments demonstrate": 62566,
      "using automatically generated": 129373,
      "automatically generated dataset": 10778,
      "use dataset train": 127980,
      "hidden state vectors": 49910,
      "recurrent entity networks": 97176,
      "neural turing machine": 78726,
      "fixed size memory": 44606,
      "obtains competitive performance": 81462,
      "endtoend deep learning": 36884,
      "models based dataset": 72803,
      "based dataset propose": 11623,
      "using bidirectional lstms": 129402,
      "carry detailed analysis": 15551,
      "classifier using features": 17593,
      "different types elements": 31527,
      "question answering transfer": 95121,
      "transfer learning large": 124098,
      "task question answering": 118595,
      "transfer learning models": 124107,
      "learning models trained": 62795,
      "transfer learning procedure": 124124,
      "intelligent agent able": 55842,
      "based artificial neural": 11525,
      "performance model tasks": 86535,
      "multitask learning deep": 75830,
      "developed deep neural": 30268,
      "achieving better results": 2935,
      "multiple tasks results": 75725,
      "challenge test set": 16111,
      "methods paper explore": 69664,
      "explore deep learning": 41536,
      "approach weakly supervised": 8015,
      "learned vector space": 62282,
      "character convolutional neural": 16424,
      "scoring function used": 104228,
      "best single model": 13453,
      "single model achieves": 109759,
      "neelakantan et al": 76906,
      "introduce new paradigm": 56492,
      "incremental learning approach": 53731,
      "reinforcement learning experimental": 97807,
      "taking advantage structured": 117543,
      "dependency trees paper": 29246,
      "utilize structured information": 130531,
      "stateoftheart neural attention": 112774,
      "reading comprehension tasks": 96015,
      "use pretrained word": 128210,
      "models stanford question": 74088,
      "baseline question answering": 12291,
      "based attention model": 11532,
      "model neural model": 71572,
      "dataset reducing error": 27147,
      "results demonstrate importance": 101686,
      "highly competitive respect": 50301,
      "multitask learning using": 75873,
      "challenging task neural": 16327,
      "traditional methods mainly": 122838,
      "questions work propose": 95377,
      "work propose apply": 134712,
      "questions natural language": 95333,
      "fed decoder generate": 43801,
      "conduct preliminary study": 20887,
      "preliminary study neural": 89282,
      "results method produce": 101925,
      "largescale dataset called": 61421,
      "human evaluation test": 50840,
      "selection deep learning": 104779,
      "task models need": 118403,
      "context vector representation": 22320,
      "attention mechanism applied": 9882,
      "raw text knowledge": 95860,
      "text knowledge base": 121071,
      "propose recurrent neural": 93012,
      "recurrent neural model": 97191,
      "neural model generates": 78149,
      "standard maximum likelihood": 112257,
      "model using policy": 72283,
      "stateoftheart results model": 112909,
      "reading comprehension present": 96003,
      "recently introduced largescale": 96698,
      "close human performance": 17819,
      "guide training process": 49250,
      "based sequencetosequence learning": 12029,
      "model based recurrent": 70741,
      "conduct experiments datasets": 20850,
      "neural networks single": 78579,
      "reading comprehension propose": 96004,
      "training data language": 123481,
      "use adversarial training": 127890,
      "adversarial training neural": 4182,
      "results sizable improvements": 102193,
      "networks transfer learning": 77800,
      "propose new attention": 92837,
      "new attention mechanism": 78805,
      "current stateoftheart results": 25365,
      "multiple pieces evidence": 75641,
      "document paper propose": 33050,
      "compositional language model": 20116,
      "language model obtains": 59360,
      "attention mechanism model": 9905,
      "similarity score pair": 109300,
      "learning question answering": 62932,
      "transfer learning deep": 124089,
      "method question answering": 69090,
      "effectiveness transfer learning": 34969,
      "transfer learning results": 124129,
      "framework encoder takes": 45519,
      "attentionbased long short": 10076,
      "memory lstm model": 68335,
      "policygradient reinforcement learning": 87977,
      "exposure bias major": 41694,
      "automatically generated questions": 10779,
      "rapid progress question": 95723,
      "progress question answering": 92178,
      "text existing neural": 120930,
      "approaches make use": 8229,
      "bidirectional attention mechanisms": 13900,
      "beam search makes": 12606,
      "single model performance": 109762,
      "named entities knowledge": 76031,
      "method improves performance": 68890,
      "hierarchical gated recurrent": 49963,
      "gated recurrent neural": 46513,
      "recurrent neural tensor": 97256,
      "neural tensor network": 78703,
      "yang et al": 135232,
      "model capture context": 70807,
      "recently attracted lot": 96663,
      "attracted lot attention": 10157,
      "largescale training data": 61515,
      "test set dataset": 120503,
      "parallel corpus available": 84614,
      "available training machine": 11133,
      "translation bilingual dictionaries": 124676,
      "crosslanguage word embeddings": 24922,
      "word embeddings observe": 133171,
      "closes performance gap": 17882,
      "neural network use": 78413,
      "use parallel corpus": 128191,
      "shows better performance": 108558,
      "compared neural network": 19398,
      "classification important task": 17227,
      "task wide applications": 118844,
      "propose novel group": 92926,
      "significantly outperform strong": 108993,
      "nlp tasks machine": 79738,
      "dotproduct attention function": 33958,
      "problem adapting neural": 90918,
      "data experiments demonstrate": 25926,
      "able achieve score": 1597,
      "paper consider problem": 83783,
      "recent years researchers": 96632,
      "achieved significant success": 2697,
      "success machine reading": 115094,
      "neural network consists": 78294,
      "work different approach": 134476,
      "making trivially parallelizable": 66870,
      "challenging task requires": 16333,
      "empirically investigate performance": 36235,
      "uses reinforcement learning": 129272,
      "achieve results competitive": 2525,
      "language processing requires": 59941,
      "solving task paper": 110651,
      "use neural network": 128168,
      "approach transfer knowledge": 7981,
      "question type classification": 95232,
      "conduct empirical evaluation": 20838,
      "performance transfer learning": 86811,
      "reading comprehension models": 95991,
      "recurrent neural nets": 97193,
      "achieve results comparable": 2524,
      "results comparable state": 101593,
      "presents novel neural": 89885,
      "dynamic fusion network": 34309,
      "achieve best result": 2421,
      "best result reported": 13434,
      "detailed empirical analysis": 29769,
      "comprehension mrc dataset": 20200,
      "current stateoftheart baseline": 25352,
      "unsupervised transfer learning": 127743,
      "transfer learning question": 124126,
      "extensive experiments investigate": 41911,
      "neural network combines": 78289,
      "work method does": 134633,
      "languages different domains": 60501,
      "domains proposed framework": 33838,
      "dataset stack overflow": 27217,
      "experiments framework outperforms": 40950,
      "outperforms best baseline": 82856,
      "knowledge graphs neural": 57988,
      "graphs neural networks": 48966,
      "problem question answering": 91193,
      "entity linking relation": 37961,
      "techniques proposed literature": 119960,
      "used reinforcement learning": 128724,
      "layer neural network": 61734,
      "neural network training": 78410,
      "achieves results competitive": 2850,
      "requires deeper understanding": 100263,
      "endtoend proposed approach": 36955,
      "customer service dataset": 25445,
      "question answering challenging": 95028,
      "base paper present": 11466,
      "present novel model": 89616,
      "experiments model yields": 41020,
      "yields stateoftheart results": 135442,
      "predicting final answer": 88987,
      "handcrafted features paper": 49347,
      "features paper propose": 43647,
      "question answering introduce": 95056,
      "training proposed model": 123802,
      "stateoftheart performance task": 112848,
      "outperforming current stateoftheart": 82807,
      "challenging task given": 16320,
      "random forest adaboost": 95491,
      "random forest classifier": 95493,
      "quora question pairs": 95412,
      "information extraction present": 54594,
      "remedy introduce new": 98859,
      "new dataset called": 78860,
      "multifactor attention network": 75107,
      "question answering neural": 95065,
      "answering neural network": 6655,
      "network models recently": 77353,
      "models recently proposed": 73886,
      "significant improvements best": 108787,
      "prior stateoftheart results": 90734,
      "traditional deep learning": 122811,
      "max average pooling": 67496,
      "paper propose learn": 84257,
      "accuracy demonstrating effectiveness": 2136,
      "sequential question answering": 107247,
      "generation knowledge graphs": 47451,
      "new stateoftheart zeroshot": 79181,
      "clinical medicine using": 17772,
      "based lstm networks": 11818,
      "integrate information multiple": 55753,
      "information multiple documents": 54777,
      "present challenges development": 89405,
      "natural language primary": 76386,
      "objective natural language": 81100,
      "processing nlp text": 91762,
      "natural language used": 76596,
      "automatic evaluation shows": 10540,
      "training data manual": 123491,
      "novel unsupervised framework": 80766,
      "obtains comparable performance": 81460,
      "introduce large dataset": 56445,
      "commonsense knowledge dataset": 18993,
      "questions require commonsense": 95359,
      "require commonsense knowledge": 100120,
      "commonsense knowledge paper": 18999,
      "predictions experimental results": 89169,
      "information retrieval models": 54940,
      "using supervised machine": 130240,
      "model achieved best": 70545,
      "best performance systems": 13395,
      "attention mechanisms called": 9922,
      "provides better results": 94006,
      "proposed models achieve": 93491,
      "models achieve superior": 72676,
      "new dataset machine": 78867,
      "human machine readers": 50912,
      "spoken content difficult": 111966,
      "spoken content paper": 111967,
      "web search engines": 132253,
      "transfer learning order": 124118,
      "question answering new": 95068,
      "models new task": 73640,
      "language inference snli": 59197,
      "inference snli dataset": 54219,
      "multigenre natural language": 75112,
      "language inference multinli": 59175,
      "rc models achieved": 95873,
      "traditional nlp techniques": 122856,
      "address challenges presented": 3663,
      "performance f1 score": 86375,
      "largescale dataset containing": 61422,
      "results hybrid wordcharacter": 101836,
      "hybrid wordcharacter model": 51200,
      "model adversarial training": 70644,
      "important source information": 52254,
      "increasingly popular research": 53708,
      "present adversarial training": 89360,
      "levels granularity evaluate": 63537,
      "datasets achieves stateoftheart": 27296,
      "proposed relation extraction": 93536,
      "specific training set": 111509,
      "multichoice question answering": 75048,
      "dataset rajpurkar et": 27141,
      "outperforms baseline large": 82842,
      "large margin achieves": 61139,
      "stateoftheart performance english": 112823,
      "network model task": 77338,
      "expert human judgments": 41223,
      "human judgments demonstrate": 50883,
      "outperform stateoftheart baselines": 82752,
      "stateoftheart baselines generate": 112595,
      "recent developments field": 96454,
      "paper present implementation": 84107,
      "unified medical language": 127116,
      "shows effectiveness approach": 108572,
      "novel gating mechanism": 80580,
      "du et al": 34224,
      "models outperform current": 73676,
      "stateoftheart performance significantly": 112845,
      "commonsense knowledge introduce": 18998,
      "commonsense knowledge encoded": 18995,
      "features prior work": 43669,
      "explicitly stated text": 41394,
      "results strong baseline": 102216,
      "knowledge explicitly model": 57917,
      "robust question answering": 103063,
      "comparable better stateoftheart": 19132,
      "furthermore experimental results": 46172,
      "shown outstanding results": 108503,
      "computational cost grows": 20367,
      "cost grows quadratically": 24355,
      "grows quadratically number": 49188,
      "reducing computational complexity": 97410,
      "computational models learn": 20413,
      "provide suggestions effective": 93935,
      "new convolutional neural": 78842,
      "demonstrate efficacy model": 28729,
      "model comparing stateoftheart": 70858,
      "knowledge graphs built": 57981,
      "baseline model achieves": 12254,
      "metrics better correlate": 69930,
      "know dont know": 57731,
      "challenging task involves": 16322,
      "dataset demonstrate approach": 26850,
      "nlp machine learning": 79632,
      "empirical study based": 36199,
      "language learning platform": 59245,
      "syntactic parsing information": 116445,
      "neural networks nns": 78526,
      "using tree kernels": 130321,
      "approach learn accurate": 7676,
      "comprehension representation learning": 20228,
      "representation learning foundation": 99290,
      "learning foundation machine": 62590,
      "foundation machine reading": 45347,
      "use word character": 128360,
      "minimal linguistic unit": 70167,
      "word embedding previous": 133029,
      "solution paper propose": 110562,
      "performance generalization ability": 86413,
      "challenging ai task": 16224,
      "ai task requires": 4428,
      "work propose hybrid": 134721,
      "propose hybrid neural": 92714,
      "hybrid neural model": 51188,
      "small training dataset": 110219,
      "substantial performance gains": 114873,
      "baselines terms multiple": 12478,
      "terms multiple evaluation": 120350,
      "multiple evaluation metrics": 75558,
      "best knowledge largest": 13352,
      "consistent improvements models": 21387,
      "possible achieve comparable": 88376,
      "propose novel ranking": 92958,
      "uses attention mechanism": 129193,
      "terms deep learning": 120304,
      "improve word representations": 52586,
      "especially rare words": 38499,
      "baseline model significantly": 12259,
      "challenging task nlp": 16328,
      "neural networkbased approaches": 78422,
      "task work attempt": 118855,
      "generator sequencetosequence model": 47780,
      "contrast previous works": 22706,
      "shows approach significantly": 108552,
      "f1 test set": 42698,
      "recent years paper": 96628,
      "paper design novel": 83860,
      "context information different": 22140,
      "information different perspectives": 54489,
      "insufficient labeled data": 55722,
      "labeled data propose": 58439,
      "mechanism automatically generate": 67951,
      "xie et al": 135193,
      "results method achieves": 101917,
      "method achieves new": 68603,
      "neural network order": 78366,
      "experimental results showing": 40700,
      "multiple benchmark datasets": 75505,
      "vulnerable adversarial attacks": 131974,
      "adversarial attacks paper": 4106,
      "leveraging knowledge distillation": 63683,
      "neural networks success": 78586,
      "generate adversarial examples": 46901,
      "assess generalizability model": 9480,
      "evaluating question answering": 39091,
      "neural models memory": 78182,
      "question natural language": 95186,
      "language processing pipeline": 59930,
      "understanding introduce new": 126869,
      "question answering remains": 95099,
      "work typically formulates": 134867,
      "typically formulates task": 126434,
      "systems automatically generate": 116749,
      "documents knowledge graphs": 33245,
      "contain relevant information": 21755,
      "developed work publicly": 30324,
      "corpus question answering": 23958,
      "electronic medical records": 35255,
      "nlp tasks demonstrate": 79712,
      "training baseline models": 123371,
      "duplicate question detection": 34269,
      "use domain adaptation": 128015,
      "words short phrases": 134215,
      "framework outperforms strong": 45646,
      "question answering present": 95072,
      "datasets existing datasets": 27463,
      "new method automatically": 79007,
      "successfully applied variety": 115178,
      "knowledge based machine": 57796,
      "paper build new": 83756,
      "new dataset consisting": 78863,
      "lay remaining challenges": 61695,
      "existing evaluation methods": 40133,
      "public benchmark datasets": 94237,
      "benchmark datasets suggest": 12805,
      "facilitate research field": 42791,
      "ensure high quality": 37639,
      "making effective use": 66839,
      "present extensive analysis": 89488,
      "multihop question answering": 75141,
      "question answering existing": 95044,
      "systems perform complex": 117058,
      "new dataset 113k": 78857,
      "question answering squad": 95105,
      "outperform previous state": 82727,
      "text using machine": 121397,
      "comprehension mrc model": 20201,
      "results model competitive": 101938,
      "english paper introduce": 37233,
      "extract span source": 42124,
      "dataset finally present": 26934,
      "far human performance": 43084,
      "research bridge gap": 100436,
      "reading comprehension record": 96011,
      "limited computational resources": 64221,
      "number training instances": 80987,
      "accuracy previous best": 2245,
      "pretrained model finetuned": 90131,
      "leading absolute improvement": 61884,
      "novel algorithm solving": 80458,
      "based graph convolutional": 11747,
      "annotations experimental results": 6432,
      "outperforms prior stateoftheart": 82965,
      "bring significant performance": 14638,
      "representations effective way": 99614,
      "structured queries directly": 114034,
      "tasks work present": 119611,
      "train models large": 122964,
      "relationships natural language": 98326,
      "parsing task mapping": 85258,
      "techniques begin address": 119844,
      "semantic parsing models": 105152,
      "requiring additional training": 100336,
      "address problem develop": 3734,
      "develop novel deep": 30222,
      "extraction machine reading": 42378,
      "paper describes novel": 83836,
      "reading comprehension style": 96012,
      "question answering aims": 95020,
      "answering aims answer": 6601,
      "embeddings better capture": 35577,
      "datasets validate effectiveness": 27785,
      "work introduces novel": 134588,
      "introduces novel method": 56623,
      "mean reciprocal rank": 67604,
      "external knowledge knowledge": 42013,
      "propose novel multitask": 92948,
      "proposed method performance": 93391,
      "language processing especially": 59862,
      "answering deep learning": 6622,
      "learning models natural": 62785,
      "neural network predict": 78373,
      "model outperforms best": 71639,
      "comprehension coreference resolution": 20173,
      "extract relevant information": 42108,
      "novel method integrate": 80637,
      "model empirical results": 71058,
      "results effectiveness model": 101745,
      "state art result": 112469,
      "respectively experimental results": 101137,
      "official test set": 81607,
      "previous studies built": 90490,
      "style experiments model": 114577,
      "understand true meaning": 126783,
      "chinese data sets": 16753,
      "conventional text classification": 22906,
      "problem reading comprehension": 91196,
      "datasets significantly outperforms": 27713,
      "user study participants": 129045,
      "model based bert": 70723,
      "human upper bound": 50984,
      "requires complex reasoning": 100251,
      "time demonstrate effectiveness": 122013,
      "model evaluate model": 71106,
      "experimental result proposed": 40545,
      "improving question answering": 53154,
      "external knowledge focus": 42011,
      "comparison previous stateoftheart": 19564,
      "absolute gains accuracy": 1732,
      "models paper focus": 73700,
      "paper focus extracting": 83948,
      "lack ground truth": 58713,
      "apply distant supervision": 7172,
      "apply recently proposed": 7211,
      "short text matching": 108259,
      "based cnn model": 11580,
      "outperform stateoftheart matching": 82756,
      "models strong baselines": 74103,
      "learning based question": 62398,
      "used prior work": 128700,
      "question generation models": 95161,
      "vocabulary paper propose": 131892,
      "dependency tree representation": 29234,
      "previous stateoftheart neural": 90481,
      "advances machine translation": 3996,
      "test time model": 120546,
      "novel approach using": 80493,
      "perform human evaluation": 86009,
      "seen rapid progress": 104539,
      "popular datasets task": 88089,
      "semantic parsing literature": 105150,
      "new model combines": 79029,
      "memory networks question": 68360,
      "networks question answering": 77720,
      "language questions knowledge": 60020,
      "methods knowledge base": 69568,
      "entity types relation": 38121,
      "based methods use": 11841,
      "approach knowledge base": 7658,
      "experimental results human": 40620,
      "existing methods significantly": 40195,
      "language question corresponding": 60018,
      "classification conduct experiments": 17153,
      "studied recent years": 114171,
      "data preparation model": 26261,
      "understanding tasks bert": 126984,
      "used knowledge distillation": 128599,
      "relation extraction framework": 97994,
      "proposed framework enables": 93289,
      "stateoftheart models achieves": 112725,
      "models achieves competitive": 72700,
      "present approach generating": 89379,
      "adversarial network gan": 4151,
      "multiple choice questions": 75513,
      "evaluate model recently": 38867,
      "model trained massive": 72200,
      "question answering benchmarks": 95026,
      "variant turing test": 130869,
      "machine learning challenges": 65785,
      "achieves comparable performances": 2762,
      "cnn rnn baselines": 18023,
      "answering text classification": 6712,
      "extraction pretrained language": 42444,
      "classification regression models": 17373,
      "approach leads superior": 7674,
      "significant performance gap": 108841,
      "performance gap best": 86408,
      "require prior knowledge": 100190,
      "great challenges existing": 48997,
      "answering paper propose": 6660,
      "extraction entities relations": 42324,
      "best models able": 13378,
      "sota results datasets": 110696,
      "studied extensively recent": 114161,
      "extensively recent years": 41968,
      "existing approaches focus": 40052,
      "documents paper propose": 33265,
      "requires reasoning multiple": 100315,
      "answer paper propose": 6553,
      "different types nodes": 31536,
      "types nodes edges": 126330,
      "evaluated blind test": 38959,
      "blind test set": 14354,
      "ensemble model achieves": 37603,
      "models recently pretrained": 73885,
      "paradigm natural language": 84543,
      "performance finetuning bert": 86388,
      "results popular datasets": 102042,
      "pretrained models task": 90154,
      "extractive summarization models": 42567,
      "task challenging humans": 117957,
      "massive pretrained language": 67335,
      "access large monolingual": 1915,
      "proposed approach using": 93220,
      "approach using different": 8006,
      "new question answering": 79096,
      "conditional neural language": 20769,
      "systems significant margin": 117144,
      "prior work shown": 90763,
      "network trained endtoend": 77459,
      "challenging highly specialized": 16258,
      "lag human performance": 58780,
      "introduce novel method": 56514,
      "corpora obtain significant": 23546,
      "obtain significant improvements": 81322,
      "synthetic data generation": 116618,
      "query paper introduce": 94975,
      "reduced computational cost": 97371,
      "propose data augmentation": 92612,
      "training data question": 123521,
      "performs consistently better": 87005,
      "means data augmentation": 67760,
      "web question answering": 132247,
      "various semantic syntactic": 131189,
      "improve generation quality": 52392,
      "methods able generate": 69273,
      "remains lack comprehensive": 98802,
      "lack comprehensive survey": 58691,
      "comprehensive survey summarizing": 20276,
      "survey summarizing existing": 116200,
      "summarizing existing approaches": 115624,
      "existing approaches recent": 40058,
      "approaches recent trends": 8307,
      "recent trends motivated": 96556,
      "trends motivated work": 125748,
      "motivated work presented": 74867,
      "work presented article": 134691,
      "specifically thorough review": 111598,
      "thorough review research": 121894,
      "review research field": 102570,
      "research field covering": 100508,
      "field covering different": 43939,
      "covering different aspects": 24535,
      "different aspects including": 31012,
      "surpassing human performance": 116116,
      "dataset social media": 27210,
      "media increasingly popular": 68121,
      "text like news": 121091,
      "finetuned bert model": 44420,
      "performance large margin": 86488,
      "results point need": 102039,
      "form question answering": 45116,
      "introduce largescale corpus": 56447,
      "abstractive model trained": 1804,
      "language modeling strong": 59469,
      "strong extractive baseline": 113675,
      "current ai systems": 25257,
      "network extensive experiments": 77252,
      "settings propose novel": 107833,
      "trained reinforcement learning": 123261,
      "reinforcement learning able": 97792,
      "machine translation optical": 66132,
      "translation optical character": 125071,
      "recognition speech recognition": 97007,
      "leverage external knowledge": 63588,
      "model commonsense reasoning": 70849,
      "method achieves superior": 68611,
      "answer previous works": 6555,
      "fail fully exploit": 42960,
      "address limitations paper": 3715,
      "paper propose reinforcement": 84312,
      "propose reinforcement learning": 93016,
      "learning rl based": 62974,
      "new stateoftheart scores": 79177,
      "achieved human parity": 2644,
      "reasoning abilities required": 96223,
      "coreference resolution current": 23363,
      "present new crowdsourced": 89582,
      "new crowdsourced dataset": 78851,
      "strong baseline model": 113633,
      "stateoftheart reading comprehension": 112880,
      "models perform significantly": 73728,
      "significantly worse humans": 109057,
      "evaluate performance stateoftheart": 38890,
      "models usually suffer": 74311,
      "features clinical named": 43400,
      "information pretrained language": 54865,
      "strong baseline models": 113634,
      "annotated natural language": 6215,
      "variety stateoftheart methods": 131018,
      "reading comprehension demonstrate": 95981,
      "sentence ranking task": 106015,
      "model solve problem": 72066,
      "experimental results methods": 40644,
      "methods outperform strong": 69658,
      "outperform strong baseline": 82763,
      "domain adaptation machine": 33437,
      "labeled data unlabeled": 58449,
      "domain adaptation framework": 33434,
      "encoder jointly trained": 36537,
      "trained using adversarial": 123325,
      "question answering evaluation": 95043,
      "answering qa present": 6670,
      "coreference resolution datasets": 23364,
      "verb phrase ellipsis": 131523,
      "previous approaches usually": 90386,
      "question work propose": 95241,
      "graphs using graph": 48976,
      "pointer network model": 87826,
      "multitask learning language": 75846,
      "learning language modeling": 62672,
      "language modeling question": 59459,
      "hierarchical multitask learning": 49984,
      "representation input sequence": 99267,
      "achieve significant improvement": 2536,
      "significant improvement accuracy": 108766,
      "widely used systems": 132592,
      "neural networks used": 78600,
      "distinguish positive negative": 32573,
      "models based sequencetosequence": 72816,
      "reading comprehension ability": 95974,
      "question answering commonsense": 95029,
      "different existing work": 31138,
      "new neural model": 79054,
      "empirical study public": 36204,
      "knowledge plays important": 58106,
      "generation model able": 47478,
      "model able outperform": 70520,
      "yield stateoftheart results": 135358,
      "requires machine understand": 100295,
      "issue introduce new": 57001,
      "trained weakly supervised": 123335,
      "answering qa requires": 6673,
      "language model biomedical": 59302,
      "biomedical question answering": 14204,
      "pretrained general domain": 90037,
      "general domain corpora": 46652,
      "biobert pretrained biomedical": 14152,
      "pretrained biomedical language": 90001,
      "bioasq challenge task": 14149,
      "outperformed previous stateoftheart": 82788,
      "stateoftheart models biobert": 112727,
      "textual question answering": 121727,
      "question answering requires": 95100,
      "information multiple sentences": 54779,
      "develop novel approach": 30221,
      "jointly training model": 57396,
      "approaches empirical study": 8140,
      "language processing field": 59865,
      "recent work achieved": 96567,
      "different training settings": 31508,
      "models learn datasets": 73480,
      "test set data": 120502,
      "role labeling machine": 103185,
      "labeling machine translation": 58508,
      "world applications paper": 135020,
      "tasks demonstrates effectiveness": 119045,
      "task sequence tagging": 118700,
      "sequence tagging problem": 107082,
      "including commonsense knowledge": 53276,
      "commonsense knowledge generating": 18996,
      "paper propose methods": 84264,
      "propose methods address": 92778,
      "analyses demonstrate proposed": 5450,
      "proposed methods substantially": 93418,
      "methods substantially improve": 69775,
      "knowledge base recent": 57788,
      "proposed framework improves": 93290,
      "challenging nlp tasks": 16286,
      "neural networks pretrained": 78538,
      "data paper investigate": 26212,
      "numerical reasoning numerical": 81009,
      "reasoning numerical reasoning": 96281,
      "existing machine reading": 40166,
      "dataset outperforming existing": 27078,
      "outperforms best previously": 82857,
      "learning dl models": 62515,
      "nlp tasks require": 79763,
      "tasks require information": 119460,
      "input sequence models": 55431,
      "adapting models new": 3313,
      "models new domain": 73638,
      "problem deep learning": 90998,
      "deep learning paper": 28305,
      "answering qa task": 6675,
      "shared task 2019": 108019,
      "performance compared baseline": 86229,
      "shared task evaluating": 108045,
      "present results machine": 89670,
      "evaluating generalization capabilities": 39058,
      "teams submitted systems": 119736,
      "best achieved average": 13302,
      "achieved average f1": 2595,
      "model proposed approach": 71814,
      "proposed approach improves": 93198,
      "approach improves current": 7627,
      "current stateoftheart language": 25354,
      "using trained model": 130301,
      "outperforms existing baseline": 82888,
      "existing baseline models": 40076,
      "evaluated using automated": 39019,
      "using automated metrics": 129364,
      "achieved impressive performance": 2646,
      "leading significant performance": 61908,
      "answering knowledge base": 6644,
      "knowledge base task": 57790,
      "task open domain": 118479,
      "domain knowledge based": 33556,
      "obtained f1 score": 81371,
      "conduct systematic analysis": 20898,
      "13 absolute improvement": 280,
      "answering natural questions": 6653,
      "improves f1 score": 52981,
      "outperforms previous sota": 82943,
      "challenging problem demands": 16296,
      "multiple information sources": 75577,
      "multitask learning objective": 75858,
      "data set proposed": 26414,
      "compared existing systems": 19371,
      "bert language model": 13150,
      "model recent years": 71862,
      "popular widely used": 88144,
      "uses bert language": 129196,
      "model bert enhance": 70765,
      "bert enhance performance": 13117,
      "requirement natural language": 100232,
      "relational facts knowledge": 98089,
      "data question answering": 26311,
      "systems using deep": 117224,
      "specialized domains paper": 111390,
      "learning techniques work": 63099,
      "training data new": 123501,
      "learning based methods": 62390,
      "performance computational complexity": 86248,
      "learning based approaches": 62387,
      "based approaches used": 11520,
      "paper propose study": 84327,
      "demonstrate overall effectiveness": 28813,
      "able generate fluent": 1646,
      "answering paper present": 6659,
      "graph network hgn": 48838,
      "pretrained contextual encoders": 90005,
      "tasks specifically transfer": 119515,
      "task finetuning large": 118219,
      "large highquality dataset": 61104,
      "model target domain": 72141,
      "establishes state art": 38613,
      "state art wellknown": 112484,
      "deep learning language": 28275,
      "learning language model": 62671,
      "clinical text materials": 17790,
      "text materials methods": 121107,
      "bert models trained": 13182,
      "different training data": 31506,
      "clinical information extraction": 17769,
      "improve performance stateoftheart": 52487,
      "obtain new stateoftheart": 81305,
      "largescale dataset task": 61423,
      "users solve problems": 129173,
      "novel joint learning": 80608,
      "construct new largescale": 21644,
      "summarization experimental results": 115506,
      "joint learning method": 57286,
      "text summarization tasks": 121353,
      "datasets used training": 27780,
      "training data source": 123543,
      "source domain unlabeled": 110749,
      "model domain adaptation": 71026,
      "stateoftheart models achieve": 112723,
      "language best knowledge": 58860,
      "natural language experiments": 76296,
      "exploration data augmentation": 41499,
      "shared task investigate": 108056,
      "negative sampling technique": 76952,
      "additional training examples": 3589,
      "phenomena natural language": 87235,
      "entity typing entity": 38128,
      "understanding wide variety": 127012,
      "models actually learning": 72713,
      "stateoftheart qa models": 112878,
      "allowing systematic control": 5120,
      "using pretrained embeddings": 130044,
      "datasets evaluation metrics": 27458,
      "prohibitively expensive paper": 92203,
      "based bert pretrained": 11548,
      "bert pretrained model": 13206,
      "additional manual effort": 3548,
      "performance comparable google": 86222,
      "multiplechoice machine reading": 75752,
      "benchmark datasets task": 12806,
      "utilizes semantic similarity": 130560,
      "compare performance proposed": 19276,
      "best baseline model": 13310,
      "measure human performance": 67800,
      "model reach new": 71850,
      "reach new stateoftheart": 95897,
      "dual encoder architecture": 34235,
      "approximate nearest neighbor": 8457,
      "sentence level apply": 105925,
      "endtoend training data": 36987,
      "challenge text generation": 16113,
      "approaches opendomain question": 8266,
      "approaches task use": 8370,
      "neural models bert": 78167,
      "case study based": 15615,
      "using various metrics": 130358,
      "various metrics including": 131133,
      "experiments approach outperforms": 40781,
      "approach outperforms existing": 7770,
      "strong baselines approach": 113641,
      "inspired recent works": 55583,
      "generation aims generating": 47303,
      "employ multitask learning": 36271,
      "multitask learning auxiliary": 75826,
      "learning rl framework": 62976,
      "weakly supervised signals": 132198,
      "approach deal problem": 7473,
      "distant supervision evaluation": 32469,
      "chains experimental results": 15984,
      "use pretrained models": 128208,
      "clozestyle machine reading": 17900,
      "dataset contains 100k": 26829,
      "underperforms human performance": 126717,
      "question answering relies": 95098,
      "sentences propose novel": 106457,
      "propose novel evaluation": 92919,
      "empirically demonstrate performance": 36227,
      "methods outperform existing": 69655,
      "enhance rnn decoder": 37490,
      "set selected sentences": 107574,
      "model tackle problem": 72134,
      "question answering framework": 95049,
      "furthermore propose novel": 46206,
      "novel method generate": 80634,
      "method generate large": 68857,
      "dataset achieves new": 26722,
      "markert et al": 67238,
      "approach achieves large": 7307,
      "achieves large improvements": 2812,
      "large improvements previous": 61110,
      "text generation recent": 121004,
      "pretraining finetuning approaches": 90259,
      "method better aligns": 68678,
      "better language models": 13616,
      "models target task": 74157,
      "evaluated text generation": 39017,
      "terms automatic metrics": 120274,
      "metrics human evaluators": 69969,
      "behavior existing models": 12656,
      "models challenge test": 72893,
      "codes publicly available": 18213,
      "scarce costly obtain": 103790,
      "model learn complex": 71428,
      "model f1 score": 71166,
      "f1 score squad": 42688,
      "proposed models achieved": 93492,
      "ensemble model using": 37608,
      "different random seeds": 31371,
      "encyclopedic knowledge present": 36789,
      "knowledge present method": 58114,
      "background knowledge multiple": 11307,
      "selfsupervised masked language": 104945,
      "various types knowledge": 131231,
      "interpretable work propose": 56259,
      "propose graphbased approach": 92708,
      "train model endtoend": 122957,
      "types knowledge improves": 126304,
      "reasoning propose novel": 96299,
      "improves performance questions": 53016,
      "stateoftheart performance code": 112813,
      "task requires model": 118643,
      "experiments significant performance": 41142,
      "correct answer candidate": 24091,
      "external knowledge encoded": 42010,
      "generated baseline model": 47055,
      "understanding paper introduce": 126914,
      "new model called": 79028,
      "achieving best performance": 2932,
      "question answering extractive": 95046,
      "models trained predict": 74222,
      "new annotation framework": 78776,
      "alternative test set": 5247,
      "propose simple neural": 93065,
      "paper provide indepth": 84377,
      "improve language model": 52401,
      "models human performance": 73344,
      "challenging nature task": 16284,
      "generation task model": 47661,
      "task model trained": 118398,
      "increase model complexity": 53605,
      "natural human speech": 76259,
      "introduced new set": 56592,
      "goal work build": 48401,
      "build high quality": 14773,
      "high quality conversational": 50112,
      "indomain training data": 53986,
      "data good results": 25985,
      "ample room improvement": 5372,
      "text address task": 120631,
      "address task explaining": 3771,
      "multitask learning problem": 75862,
      "large transformerbased language": 61306,
      "models shown effective": 74026,
      "previous works investigated": 90572,
      "shown great success": 108479,
      "availability largescale datasets": 10922,
      "effectiveness neural models": 34921,
      "neural models recent": 78194,
      "recent research works": 96509,
      "access manually annotated": 1917,
      "data approach outperforms": 25615,
      "previous unsupervised approaches": 90514,
      "extraction neural models": 42409,
      "present selftraining method": 89685,
      "labels extra supervision": 58605,
      "source code obtained": 110720,
      "dependencies address issue": 29064,
      "reading comprehension framework": 95983,
      "levels representations learned": 63562,
      "experiments approach significantly": 40782,
      "deep bidirectional transformer": 28207,
      "transfer learning qa": 124125,
      "bert model finetuning": 13170,
      "generate ranked list": 46994,
      "given source document": 48133,
      "collection speech transcripts": 18491,
      "associated news articles": 9605,
      "data augmentation generate": 25633,
      "generate training data": 47036,
      "al 2019 human": 4642,
      "2019 human evaluation": 699,
      "training data cover": 123447,
      "data cover small": 25802,
      "method automatically generates": 68657,
      "use distant supervision": 128007,
      "given knowledge base": 48050,
      "knowledge base text": 57791,
      "reasoning paper present": 96284,
      "present new framework": 89586,
      "answer questions related": 6569,
      "enhances overall performance": 37532,
      "recently paper propose": 96727,
      "novel approach tackle": 80489,
      "dataset approach achieves": 26741,
      "recently released dataset": 96750,
      "implications downstream applications": 51991,
      "labeled data new": 58438,
      "process creating corpus": 91449,
      "experiments using different": 41193,
      "different types machine": 31533,
      "f1score test set": 42721,
      "test set corpus": 120501,
      "encourage research community": 36764,
      "large scale chinese": 61243,
      "qa models perform": 94509,
      "learning method significantly": 62729,
      "approaches achieve significant": 8041,
      "achieve significant performance": 2541,
      "official test sets": 81608,
      "annotation tool developed": 6394,
      "methods achieve promising": 69282,
      "achieve promising results": 2517,
      "requires research efforts": 100318,
      "conduct thorough evaluation": 20902,
      "evaluation existing models": 39209,
      "codes datasets obtained": 18201,
      "network models vietnamese": 77360,
      "word representation important": 133436,
      "like english chinese": 64040,
      "paper conduct experiments": 83777,
      "impact word representation": 51903,
      "embeddings bert model": 35573,
      "requires human intervention": 100275,
      "framework automatically construct": 45434,
      "does require external": 33391,
      "domain covid19 pandemic": 33495,
      "achieved tremendous success": 2717,
      "model generate large": 71234,
      "results demonstrate method": 101687,
      "generally perform poorly": 46873,
      "thorough experiments demonstrate": 121888,
      "demonstrate novel approach": 28811,
      "datasets including natural": 27518,
      "retrieving relevant documents": 102473,
      "sentence selection step": 106061,
      "address issue investigate": 3690,
      "investigate best practices": 56727,
      "text processing techniques": 121214,
      "faster inference time": 43180,
      "methods biomedical text": 69352,
      "enables model learn": 36393,
      "unlabeled biomedical text": 127374,
      "requiring model predict": 100352,
      "performance outperforms previous": 86590,
      "augmentation neural models": 10295,
      "matching questions answers": 67428,
      "bert based classifier": 13071,
      "training examples neural": 123624,
      "data significantly outperforms": 26451,
      "textual data research": 121685,
      "using sequencetosequence approach": 130160,
      "vectors paper propose": 131461,
      "address computational cost": 3668,
      "proposed approach consistent": 93187,
      "million people speak": 70104,
      "lack benchmark datasets": 58688,
      "lowresource language vietnamese": 65506,
      "model performance dataset": 71725,
      "code data paper": 18082,
      "continuous embedding space": 22618,
      "reading comprehension given": 95984,
      "main points document": 66451,
      "sequence sequence vectors": 107068,
      "training data form": 123466,
      "partofspeech tagging namedentity": 85558,
      "trained small dataset": 123283,
      "challenging task automatic": 16317,
      "despite recent progress": 29722,
      "models trained existing": 74204,
      "trained existing datasets": 123142,
      "highlevel semantic discourse": 50250,
      "based gpt2 model": 11744,
      "multiple choice question": 75511,
      "choice question answering": 16900,
      "challenging task machine": 16323,
      "context existing work": 22088,
      "work deals challenge": 134455,
      "deals challenge learning": 27875,
      "challenge learning reasoning": 16048,
      "answering qa propose": 6671,
      "network based semantic": 77165,
      "based semantic structure": 12012,
      "proposed approach shows": 93211,
      "approach shows competitive": 7893,
      "compared recent stateoftheart": 19431,
      "relation extraction models": 98001,
      "training data introduce": 123479,
      "extraction models trained": 42394,
      "experiments stateoftheart models": 41154,
      "data training examples": 26573,
      "different parts speech": 31323,
      "model new regularization": 71582,
      "benchmark data set": 12750,
      "automatic approach generating": 10493,
      "able generate semantically": 1649,
      "showed statistically significant": 108392,
      "trained natural questions": 123215,
      "model produces better": 71799,
      "comprehension propose novel": 20220,
      "propose novel bidirectional": 92905,
      "using distantly supervised": 129594,
      "negative examples training": 76927,
      "make model robust": 66699,
      "efforts natural language": 35202,
      "leverage pretrained language": 63614,
      "research evaluation methods": 100490,
      "automated evaluation metrics": 10440,
      "questions require background": 95357,
      "require background knowledge": 100114,
      "background knowledge multihop": 11305,
      "knowledge multihop reasoning": 58068,
      "recent pretrained language": 96490,
      "automatic question answering": 10635,
      "semantic similarity lexical": 105282,
      "par best performing": 84518,
      "benchmarks new benchmark": 12927,
      "models achieved significant": 72690,
      "achieved significant results": 2696,
      "score test set": 104134,
      "domain knowledge bases": 33557,
      "evaluations human judges": 39466,
      "paper explore novel": 83922,
      "neural qa models": 78639,
      "clinical texts different": 17793,
      "simple effective framework": 109409,
      "hierarchical memory network": 49974,
      "structured unstructured data": 114050,
      "predicting correct answer": 88980,
      "answer question given": 6560,
      "parameters experimental results": 84747,
      "tasks model achieves": 119299,
      "stateoftheart results single": 112919,
      "generating fact checking": 47221,
      "model generates set": 71240,
      "models reallife applications": 73865,
      "humans paper propose": 51098,
      "traditional attention mechanism": 122797,
      "human evaluations proposed": 50850,
      "construct baseline model": 21620,
      "based determinantal point": 11642,
      "determinantal point processes": 30114,
      "point processes dpps": 87812,
      "demonstrated proposed method": 28929,
      "method outperformed baseline": 69005,
      "outperformed baseline methods": 82777,
      "detection duplicate question": 29929,
      "source target domain": 110832,
      "domains paper propose": 33828,
      "existing models achieves": 40204,
      "new stateoftheart chinese": 79156,
      "deep learning method": 28279,
      "models tackle problem": 74153,
      "neural network structures": 78400,
      "learn semantic features": 62137,
      "stateoftheart models chinese": 112728,
      "models chinese medical": 72901,
      "paper propose innovative": 84253,
      "results stateoftheart performance": 102209,
      "make data code": 66642,
      "reduce search space": 97357,
      "learn dense representations": 62019,
      "negative sampling methods": 76949,
      "structured unstructured knowledge": 114051,
      "extraction paper propose": 42424,
      "methods large margins": 69579,
      "propose method leverage": 92774,
      "inference nli identify": 54179,
      "learning model learns": 62763,
      "model approach outperforms": 70675,
      "training data abundant": 123418,
      "state art multiple": 112453,
      "shortcomings current models": 108282,
      "presenting novel framework": 89810,
      "reached human parity": 95907,
      "training examples available": 123622,
      "model select correct": 71958,
      "representation later used": 99278,
      "surprisingly good results": 116139,
      "results multiple benchmarks": 101970,
      "propose novel problem": 92957,
      "framework equipped novel": 45525,
      "paper explore possibility": 83923,
      "rich semantic representations": 102782,
      "cover wide range": 24494,
      "qa paper present": 94513,
      "significant improvement previous": 108774,
      "knowledge base developed": 57768,
      "existing approaches assume": 40049,
      "catastrophic forgetting problem": 15713,
      "task stateoftheart models": 118741,
      "closing gap human": 17886,
      "area curve auc": 8870,
      "obtained deep learning": 81358,
      "approach achieve stateoftheart": 7297,
      "dataset empirical analysis": 26886,
      "empirical analysis shows": 36151,
      "effective deep learning": 34660,
      "paper introduce task": 84004,
      "task requires understand": 118647,
      "baselines publicly available": 12449,
      "resource researchers interested": 100874,
      "work aim improve": 134364,
      "text present effective": 121193,
      "obtaining significant improvements": 81449,
      "models designed machine": 73051,
      "incorporate contextual information": 53460,
      "sentences source document": 106500,
      "answers given question": 6731,
      "scheme experimental results": 103925,
      "investigate question collect": 56805,
      "adversarial examples question": 4121,
      "examples question answering": 39870,
      "models work present": 74360,
      "existing works consider": 40341,
      "understand complex biomedical": 126741,
      "potential future directions": 88557,
      "future directions explore": 46271,
      "challenge set evaluation": 16100,
      "deployed real world": 29309,
      "conclude substantial room": 20719,
      "like machine reading": 64059,
      "task conduct comprehensive": 118001,
      "datasets squad newsqa": 27730,
      "translation paper introduces": 125089,
      "simple effective approaches": 109405,
      "achieves remarkable improvement": 2846,
      "user experience work": 128982,
      "semantic parsing approaches": 105144,
      "novel framework named": 80576,
      "largescale knowledge graph": 61451,
      "knowledge graph evaluate": 57967,
      "observe performance improvements": 81207,
      "method provides strong": 69084,
      "terms human evaluation": 120336,
      "learns vector representations": 63243,
      "text corpora wikipedia": 120833,
      "f1 points achieves": 42669,
      "data available link": 25677,
      "ability experimental results": 1517,
      "widely used datasets": 132568,
      "correlations entity types": 24256,
      "entity recognition module": 38024,
      "learning models challenging": 62777,
      "empirically demonstrate proposed": 36228,
      "large neural language": 61167,
      "nlp tasks unclear": 79775,
      "models actually learn": 72712,
      "entities observed training": 37829,
      "observed training set": 81237,
      "models world knowledge": 74364,
      "foster future research": 45337,
      "reasoning unstructured text": 96322,
      "stateoftheart systems using": 112993,
      "al 2019 learning": 4644,
      "model able learn": 70518,
      "systems fall short": 116903,
      "overcome limitation propose": 83291,
      "carry extensive experiments": 15554,
      "experiments datasets demonstrate": 40868,
      "surpasses stateoftheart models": 116113,
      "stateoftheart models large": 112736,
      "results sequence tagging": 102161,
      "general biomedical domains": 46637,
      "sequence tagging approach": 107076,
      "freely available download": 45782,
      "datasets better represent": 27337,
      "answers natural language questions": 6737,
      "effectiveness method means experiments": 34908,
      "natural language processing goal": 76420,
      "does require feature engineering": 33393,
      "experimental results standard benchmark": 40704,
      "task method outperforms previous": 118389,
      "semantic parsing question answering": 105160,
      "paper investigate recently proposed": 84022,
      "4th grade science exams": 1064,
      "language processing nlp machine": 59906,
      "ensemble models sets new": 37612,
      "major goal natural language": 66564,
      "goal natural language processing": 48370,
      "model sets new state": 71995,
      "pairs convolutional neural networks": 83502,
      "convolutional neural networks paper": 23202,
      "convolutional neural network based": 23179,
      "attentive convolutional neural network": 10109,
      "characterlevel convolutional neural network": 16561,
      "convolutional neural network charcnn": 23181,
      "deep neural networks achieve": 28367,
      "neural network solve problem": 78398,
      "neural networks based complex": 78444,
      "networks based complex architectures": 77519,
      "stanford question answering dataset": 112379,
      "question answering dataset squad": 95036,
      "question answering qa tasks": 95087,
      "multiple choice reading comprehension": 75515,
      "hermann et al 2015": 49845,
      "propose new task machine": 92886,
      "competitive performance compared stateoftheart": 19661,
      "performance compared stateoftheart methods": 86240,
      "significantly outperforming strong baseline": 109001,
      "questions community question answering": 95289,
      "task community question answering": 117983,
      "paperno et al 2016": 84499,
      "natural language understanding recent": 76579,
      "requires modeling complex interactions": 100302,
      "context different levels granularity": 22061,
      "experimental evaluations model achieves": 40537,
      "evaluations model achieves stateoftheart": 39471,
      "improving performance stateoftheart models": 53140,
      "task question answering qa": 118596,
      "multitask learning deep neural": 75831,
      "learning deep neural networks": 62490,
      "developed deep neural network": 30269,
      "explore deep learning techniques": 41537,
      "character convolutional neural networks": 16425,
      "reinforcement learning experimental results": 97808,
      "stateoftheart neural attention model": 112775,
      "dataset experimental results demonstrate": 26911,
      "use pretrained word embeddings": 128211,
      "models stanford question answering": 74089,
      "raw text knowledge base": 95861,
      "standard maximum likelihood training": 112258,
      "model using policy gradient": 72284,
      "model based recurrent neural": 70742,
      "method outperforms stateoftheart methods": 69033,
      "propose new attention mechanism": 92838,
      "term memory lstm model": 120232,
      "rapid progress question answering": 95724,
      "method improves performance models": 68891,
      "recently attracted lot attention": 96664,
      "available training machine translation": 11134,
      "machine translation bilingual dictionaries": 65915,
      "significantly outperform strong baselines": 108994,
      "success machine reading comprehension": 115095,
      "achieve results competitive stateoftheart": 2526,
      "results comparable state art": 101594,
      "paper presents novel neural": 84194,
      "machine reading comprehension dataset": 65851,
      "reading comprehension mrc dataset": 95993,
      "transfer learning question answering": 124127,
      "conduct extensive experiments investigate": 20874,
      "knowledge graphs neural networks": 57989,
      "reasoning machine reading comprehension": 96267,
      "knowledge base paper present": 57776,
      "paper present novel model": 84126,
      "handcrafted features paper propose": 49348,
      "open information extraction present": 81910,
      "introduce new dataset called": 56481,
      "question answering neural network": 95066,
      "answering neural network models": 6656,
      "neural network models recently": 78351,
      "base question answering kbqa": 11477,
      "model sets new stateoftheart": 71997,
      "objective natural language processing": 81101,
      "language processing nlp text": 59926,
      "sequencetosequence neural network model": 107190,
      "propose novel unsupervised framework": 92975,
      "questions require commonsense knowledge": 95360,
      "paper present neural network": 84117,
      "using supervised machine learning": 130241,
      "supervised machine learning model": 115793,
      "model achieved best performance": 70546,
      "machine reading comprehension present": 65866,
      "spoken content paper propose": 111968,
      "achieves stateoftheart results benchmarks": 2902,
      "natural language inference snli": 76352,
      "language inference snli dataset": 59198,
      "multigenre natural language inference": 75113,
      "natural language inference multinli": 76332,
      "proposed relation extraction task": 93537,
      "dataset rajpurkar et al": 27142,
      "experimental results method outperforms": 40640,
      "outperforms baseline large margin": 82843,
      "achieves stateoftheart performance english": 2885,
      "neural network model task": 78339,
      "supervised machine learning algorithms": 115790,
      "external commonsense knowledge introduce": 41996,
      "computational cost grows quadratically": 20368,
      "new convolutional neural network": 78843,
      "machine reading comprehension task": 65867,
      "challenging natural language understanding": 16281,
      "dataset demonstrate approach achieves": 26851,
      "comprehension representation learning foundation": 20229,
      "representation learning foundation machine": 99291,
      "learning foundation machine reading": 62591,
      "foundation machine reading comprehension": 45348,
      "significantly outperform stateoftheart baselines": 108988,
      "challenging ai task requires": 16225,
      "baselines terms multiple evaluation": 12479,
      "terms multiple evaluation metrics": 120351,
      "address problem propose novel": 3745,
      "context information different perspectives": 22141,
      "insufficient labeled data propose": 55723,
      "xie et al 2017": 135194,
      "method achieves new stateoftheart": 68604,
      "vulnerable adversarial attacks paper": 131975,
      "recurrent neural networks success": 97250,
      "recurrent neural networks generate": 97231,
      "natural language processing pipeline": 76471,
      "remains challenging task requires": 98793,
      "work typically formulates task": 134868,
      "developed work publicly available": 30325,
      "propose new method automatically": 92863,
      "machine reading comprehension models": 65857,
      "information recent work shown": 54903,
      "outperform previous state art": 82728,
      "reading comprehension mrc model": 95994,
      "based graph convolutional networks": 11748,
      "results model significantly outperforms": 101950,
      "semantic parsing task mapping": 105172,
      "requiring additional training data": 100337,
      "extraction machine reading comprehension": 42379,
      "proposed method outperforms previous": 93385,
      "method outperforms previous stateoftheart": 69022,
      "question answering aims answer": 95021,
      "answering aims answer questions": 6602,
      "validate effectiveness proposed method": 130719,
      "propose novel multitask learning": 92949,
      "realworld datasets demonstrate effectiveness": 96160,
      "natural language processing especially": 76417,
      "question answering deep learning": 95039,
      "deep learning models natural": 28289,
      "learning models natural language": 62786,
      "model achieves stateoftheart accuracy": 70597,
      "experiments model achieves stateoftheart": 41013,
      "experimental result proposed model": 40546,
      "proposed model achieves new": 93428,
      "models significantly outperform stateoftheart": 74047,
      "significantly outperform stateoftheart matching": 108991,
      "learning based question answering": 62399,
      "outperforms previous stateoftheart neural": 82951,
      "memory networks question answering": 68361,
      "natural language questions knowledge": 76522,
      "method significantly outperforms existing": 69147,
      "model outperforms existing methods": 71648,
      "outperforms existing methods significantly": 82893,
      "natural language question corresponding": 76520,
      "language understanding tasks bert": 60281,
      "models achieves competitive performance": 72701,
      "generative adversarial network gan": 47712,
      "current state art model": 25343,
      "question answering text classification": 95115,
      "significant performance gap best": 108842,
      "question answering paper propose": 95071,
      "significantly outperforms previous best": 109022,
      "outperforms previous best models": 82935,
      "studied extensively recent years": 114162,
      "years existing approaches focus": 135263,
      "different types nodes edges": 31537,
      "evaluated blind test set": 38960,
      "ensemble model achieves stateoftheart": 37604,
      "paradigm natural language processing": 84544,
      "massive pretrained language models": 67336,
      "large number training instances": 61193,
      "conditional neural language model": 20770,
      "web question answering deep": 132248,
      "remains lack comprehensive survey": 98803,
      "lack comprehensive survey summarizing": 58692,
      "comprehensive survey summarizing existing": 20277,
      "survey summarizing existing approaches": 116201,
      "summarizing existing approaches recent": 115625,
      "existing approaches recent trends": 40059,
      "approaches recent trends motivated": 8308,
      "recent trends motivated work": 96557,
      "trends motivated work presented": 125749,
      "motivated work presented article": 74868,
      "specifically thorough review research": 111599,
      "thorough review research field": 121895,
      "review research field covering": 102571,
      "research field covering different": 100509,
      "field covering different aspects": 43940,
      "covering different aspects including": 24536,
      "social media increasingly popular": 110387,
      "machine translation optical character": 66133,
      "translation optical character recognition": 125072,
      "based pretrained language model": 11946,
      "effectiveness proposed method achieves": 34931,
      "method achieves superior performance": 68612,
      "paper propose reinforcement learning": 84313,
      "propose reinforcement learning rl": 93017,
      "reinforcement learning rl based": 97824,
      "present new crowdsourced dataset": 89583,
      "features clinical named entity": 43401,
      "information pretrained language model": 54866,
      "question answering qa present": 95081,
      "graph neural network model": 48844,
      "deep neural networks used": 28385,
      "using convolutional neural networks": 129500,
      "propose new neural model": 92876,
      "knowledge plays important role": 58107,
      "knowledge paper propose method": 58098,
      "pretrained language models like": 90083,
      "stateoftheart results benchmark datasets": 112894,
      "question answering qa requires": 95084,
      "pretrained language model biomedical": 90050,
      "pretrained language models language": 90080,
      "language models language models": 59574,
      "biobert pretrained biomedical language": 14153,
      "outperformed previous stateoftheart models": 82789,
      "previous stateoftheart models biobert": 90479,
      "substantially outperforms previous approaches": 114909,
      "natural language processing field": 76419,
      "challenging task requires understanding": 16334,
      "semantic role labeling machine": 105249,
      "role labeling machine translation": 103186,
      "real world applications paper": 96090,
      "task sequence tagging problem": 118701,
      "numerical reasoning numerical reasoning": 81010,
      "existing machine reading comprehension": 40167,
      "outperforms best previously published": 82858,
      "deep learning dl models": 28269,
      "adapting models new domain": 3314,
      "question answering qa task": 95086,
      "model outperform stateoftheart methods": 71627,
      "improves current stateoftheart language": 52970,
      "evaluated using automated metrics": 39020,
      "models achieved impressive performance": 72683,
      "paper conduct systematic analysis": 83780,
      "uses bert language model": 129197,
      "language model bert enhance": 59299,
      "model bert enhance performance": 70766,
      "requirement natural language processing": 100233,
      "deep neural networks paper": 28377,
      "using deep neural network": 129547,
      "machine learning based methods": 65783,
      "question answering paper present": 95070,
      "models natural language tasks": 73613,
      "clinical text materials methods": 17791,
      "unsupervised domain adaptation language": 127629,
      "data source domain unlabeled": 26473,
      "paper present comprehensive survey": 84087,
      "novel neural network based": 80672,
      "multiplechoice machine reading comprehension": 75753,
      "compare performance proposed model": 19277,
      "model reach new stateoftheart": 71851,
      "approximate nearest neighbor search": 8458,
      "performance compared stateoftheart approaches": 86238,
      "approaches opendomain question answering": 8267,
      "reinforcement learning rl framework": 97826,
      "clozestyle machine reading comprehension": 17901,
      "automatic human evaluation results": 10554,
      "human evaluation results demonstrate": 50831,
      "evaluation results demonstrate model": 39368,
      "results demonstrate model achieves": 101691,
      "markert et al 2012": 67239,
      "approach achieves large improvements": 7308,
      "terms automatic metrics human": 120275,
      "automatic metrics human evaluators": 10606,
      "pretrained multilingual bert model": 90162,
      "encyclopedic knowledge present method": 36790,
      "manner experimental results benchmark": 66948,
      "language understanding paper introduce": 60250,
      "introduce new model called": 56489,
      "paper provide indepth analysis": 84378,
      "pretrained language models better": 90067,
      "improve language model performance": 52402,
      "large transformerbased language models": 61307,
      "relatively little attention paid": 98399,
      "data approach outperforms previous": 25616,
      "extraction neural models achieved": 42410,
      "neural models achieved great": 78163,
      "tasks experimental results demonstrate": 119120,
      "dependencies address issue present": 29065,
      "machine reading comprehension framework": 65852,
      "experiments approach significantly outperforms": 40783,
      "approach significantly outperforms previous": 7910,
      "et al 2019 human": 38730,
      "al 2019 human evaluation": 4643,
      "neural language models bert": 77941,
      "recently paper propose novel": 96728,
      "approach achieves stateoftheart accuracy": 7323,
      "transfer learning method significantly": 124104,
      "methods achieve promising results": 69283,
      "neural network models vietnamese": 78358,
      "understanding natural language texts": 126899,
      "languages like english chinese": 60694,
      "deep neural networks achieved": 28368,
      "experimental results demonstrate method": 40592,
      "results demonstrate method outperforms": 101689,
      "statistically significant improvement accuracy": 113185,
      "unsupervised representation learning methods": 127705,
      "introduce new pretraining task": 56494,
      "data augmentation neural models": 25642,
      "partofspeech tagging namedentity recognition": 85559,
      "models trained existing datasets": 74205,
      "multiple choice question answering": 75512,
      "work deals challenge learning": 134456,
      "deals challenge learning reasoning": 27876,
      "question answering qa propose": 95082,
      "proposed approach shows competitive": 93212,
      "approach shows competitive performance": 7894,
      "synthetic training data propose": 116651,
      "transformerbased language models bert": 124418,
      "language models bert shown": 59518,
      "reading comprehension propose novel": 96005,
      "leverage pretrained language models": 63615,
      "pretrained language models generate": 90077,
      "questions require background knowledge": 95358,
      "background knowledge multihop reasoning": 11306,
      "recent pretrained language models": 96491,
      "f1 score test set": 42690,
      "presents novel approach automatic": 89882,
      "propose simple effective framework": 93056,
      "new stateoftheart results single": 79171,
      "based determinantal point processes": 11643,
      "determinantal point processes dpps": 30115,
      "demonstrated proposed method outperformed": 28930,
      "proposed method outperformed baseline": 93382,
      "method outperformed baseline methods": 69006,
      "demonstrate model significantly outperforms": 28799,
      "outperforms existing models achieves": 82895,
      "sentence representation learning models": 106031,
      "dataset model significantly outperforms": 27035,
      "make data code publicly": 66645,
      "code publicly available research": 18142,
      "choice reading comprehension task": 16904,
      "language inference nli identify": 59181,
      "datasets experimental results method": 27471,
      "results method outperforms stateoftheart": 101923,
      "method outperforms stateoftheart baselines": 69032,
      "development deep learning techniques": 30382,
      "deep learning based question": 28261,
      "approach achieve stateoftheart results": 7298,
      "adversarial examples question answering": 4122,
      "automatic question answering qa": 10636,
      "like machine reading comprehension": 64060,
      "task conduct comprehensive experiments": 118002,
      "propose novel framework named": 92922,
      "code data available link": 18078,
      "propose novel neural model": 92953,
      "empirically demonstrate proposed approach": 36229,
      "large neural language models": 61168,
      "et al 2019 learning": 38732,
      "stateoftheart models large margin": 112737,
      "challenges natural language processing nlp": 16180,
      "natural language processing nlp machine": 76452,
      "major goal natural language processing": 66565,
      "model sets new state art": 71996,
      "neural networks based complex architectures": 78445,
      "stanford question answering dataset squad": 112380,
      "competitive performance compared stateoftheart methods": 19662,
      "experimental evaluations model achieves stateoftheart": 40538,
      "evaluations model achieves stateoftheart results": 39472,
      "multitask learning deep neural networks": 75832,
      "models stanford question answering dataset": 74090,
      "model based recurrent neural network": 70743,
      "paper propose novel neural network": 84306,
      "question answering neural network models": 95067,
      "knowledge base question answering kbqa": 57786,
      "natural language processing nlp text": 76467,
      "stanford natural language inference snli": 112373,
      "natural language inference snli dataset": 76353,
      "multigenre natural language inference multinli": 75114,
      "dataset rajpurkar et al 2016": 27143,
      "machine reading comprehension mrc dataset": 65859,
      "challenging natural language understanding task": 16282,
      "comprehension representation learning foundation machine": 20230,
      "representation learning foundation machine reading": 99292,
      "learning foundation machine reading comprehension": 62592,
      "baselines terms multiple evaluation metrics": 12480,
      "method achieves new stateoftheart performance": 68605,
      "use deep neural networks dnns": 127989,
      "machine reading comprehension mrc model": 65860,
      "experimental results model significantly outperforms": 40653,
      "proposed method outperforms previous stateoftheart": 93386,
      "question answering aims answer questions": 95022,
      "realworld datasets demonstrate effectiveness proposed": 96161,
      "deep learning models natural language": 28290,
      "experiments model achieves stateoftheart performance": 41014,
      "important natural language processing nlp": 52190,
      "proposed model achieves new stateoftheart": 93429,
      "models significantly outperform stateoftheart matching": 74048,
      "model outperforms existing methods significantly": 71649,
      "natural language understanding tasks bert": 76585,
      "significantly outperforms previous best models": 109023,
      "recent years existing approaches focus": 96614,
      "remains lack comprehensive survey summarizing": 98804,
      "lack comprehensive survey summarizing existing": 58693,
      "comprehensive survey summarizing existing approaches": 20278,
      "survey summarizing existing approaches recent": 116202,
      "summarizing existing approaches recent trends": 115626,
      "existing approaches recent trends motivated": 40060,
      "approaches recent trends motivated work": 8309,
      "recent trends motivated work presented": 96558,
      "trends motivated work presented article": 125750,
      "specifically thorough review research field": 111600,
      "thorough review research field covering": 121896,
      "review research field covering different": 102572,
      "research field covering different aspects": 100510,
      "field covering different aspects including": 43941,
      "machine translation optical character recognition": 66134,
      "demonstrate effectiveness proposed method achieves": 28721,
      "paper propose reinforcement learning rl": 84314,
      "paper propose new neural model": 84284,
      "pretrained language models language models": 90081,
      "semantic role labeling machine translation": 105250,
      "existing machine reading comprehension models": 40168,
      "proposed model outperforms previous stateoftheart": 93468,
      "model outperforms previous stateoftheart methods": 71658,
      "achieves new stateoftheart results task": 2828,
      "language model bert enhance performance": 59300,
      "using deep neural networks paper": 129550,
      "using deep neural network dnn": 129548,
      "model significantly outperforms strong baselines": 72039,
      "automatic human evaluation results demonstrate": 10555,
      "endtoend manner experimental results benchmark": 36919,
      "neural models achieved great success": 78164,
      "success machine reading comprehension mrc": 115096,
      "dependencies address issue present novel": 29066,
      "experiments approach significantly outperforms previous": 40784,
      "et al 2019 human evaluation": 38731,
      "experimental results demonstrate method outperforms": 40594,
      "work deals challenge learning reasoning": 134457,
      "proposed approach shows competitive performance": 93213,
      "achieves new stateoftheart results single": 2827,
      "proposed method outperformed baseline methods": 93383,
      "significantly outperforms existing stateoftheart models": 109017,
      "make data code publicly available": 66646,
      "task natural language inference nli": 118437,
      "natural language inference nli identify": 76338,
      "experimental results method outperforms stateoftheart": 40641,
      "proposed model achieves significant improvement": 93431,
      "datasets demonstrate effectiveness proposed approach": 27402,
      "leftmost": 63303,
      "brills": 14626,
      "prosodybased": 93653,
      "gleaned": 48223,
      "handlabeling": 49372,
      "metalearners": 68542,
      "celex": 15884,
      "invertible": 56707,
      "rescaling": 100386,
      "normalize": 80354,
      "dictated": 30853,
      "blindly": 14355,
      "unaffected": 126575,
      "stationary": 113072,
      "crossmedia": 25059,
      "963": 1432,
      "systmes": 117245,
      "reconnaissance": 97105,
      "parole": 84876,
      "mother": 74829,
      "postedition": 88477,
      "exacerbates": 39711,
      "benign": 13044,
      "deepening": 28440,
      "suppress": 116046,
      "receiver": 96378,
      "departure": 29041,
      "gmm": 48321,
      "triphones": 125807,
      "gaussians": 46555,
      "mel": 68265,
      "mfcc": 70025,
      "unfolded": 127073,
      "underestimated": 126655,
      "featurerich": 43345,
      "scrfs": 104248,
      "hinge": 50495,
      "std": 113221,
      "summing": 115661,
      "reestimated": 97485,
      "relabeled": 97840,
      "humanlabelled": 51048,
      "slt": 110107,
      "wholeword": 132478,
      "categorised": 15760,
      "pns": 87776,
      "oovs": 81876,
      "globalized": 48283,
      "inexpensively": 54086,
      "mgb": 70027,
      "normalised": 80331,
      "noiseaware": 80076,
      "ivector": 57154,
      "anatomy": 6078,
      "lvcsr": 65727,
      "dataintensive": 26690,
      "directory": 31920,
      "13000": 293,
      "deployable": 29304,
      "appeal": 6789,
      "frontend": 45899,
      "ascii": 9284,
      "trainingdata": 123957,
      "framelevel": 45392,
      "ctcbased": 25197,
      "verbatim": 131542,
      "1400": 324,
      "statelevel": 112545,
      "881": 1345,
      "successively": 115201,
      "hlstm": 50579,
      "ob": 81041,
      "classifierbased": 17594,
      "arbitrarylength": 8594,
      "voicing": 131931,
      "noisychannel": 80134,
      "bn": 14388,
      "characterised": 16512,
      "htm": 50697,
      "distortions": 32595,
      "scrf": 104247,
      "652": 1181,
      "synchronization": 116336,
      "bayesrisk": 12589,
      "resourcelimited": 100904,
      "singlelabel": 109830,
      "fst": 45917,
      "monotonous": 74648,
      "voices": 131930,
      "largevocabulary": 61543,
      "colossal": 18536,
      "caller": 15069,
      "diarization": 30847,
      "pashto": 85602,
      "summer": 115656,
      "camp": 15083,
      "vgg": 131635,
      "oracles": 82258,
      "multiplications": 75761,
      "nonlinearity": 80228,
      "nsfc": 80810,
      "librispeech": 63966,
      "mat": 67353,
      "multitarget": 75805,
      "kws": 58367,
      "invocabulary": 56873,
      "ams": 5404,
      "cts": 25199,
      "transcribers": 123990,
      "backchannel": 11294,
      "goodbad": 48512,
      "frenchtoenglish": 45831,
      "sped": 111639,
      "broadcaster": 14691,
      "maximized": 67514,
      "affine": 4254,
      "canonicalize": 15159,
      "recentlyreleased": 96783,
      "lexiconfree": 63921,
      "randomlyinitialized": 95534,
      "ers": 38418,
      "abx": 1848,
      "seventeen": 107875,
      "wavenet": 132050,
      "spectrogram": 111631,
      "453": 1037,
      "situ": 109881,
      "outperformance": 82774,
      "a2p": 1457,
      "psd": 94176,
      "consults": 21716,
      "cca": 15856,
      "smallsize": 110251,
      "heterogenous": 49869,
      "withincorpus": 132746,
      "crosscorpus": 24873,
      "zoneout": 135551,
      "chose": 16937,
      "bestpublished": 13482,
      "programmes": 92125,
      "064": 48,
      "lowerresource": 65458,
      "1438": 329,
      "fcnn": 43229,
      "wfst": 132465,
      "overheads": 83327,
      "hamper": 49308,
      "densenets": 29028,
      "schedules": 103888,
      "15m": 372,
      "vocal": 131921,
      "trainingtesting": 123959,
      "202": 720,
      "logographic": 65049,
      "logographs": 65052,
      "unimodal": 127162,
      "tacotron2": 117328,
      "425": 1021,
      "memad": 68268,
      "clm": 17804,
      "lfmmi": 63947,
      "singlechannel": 109820,
      "rehearsed": 97781,
      "gauging": 46540,
      "semanticsyntactic": 105484,
      "2dimensional": 882,
      "sensory": 105758,
      "scientist": 104029,
      "yuan": 135464,
      "tips": 122216,
      "decent": 27919,
      "voiceenabled": 131929,
      "cd": 15868,
      "undersampled": 126731,
      "wordpiecebased": 133794,
      "916": 1390,
      "clips": 17801,
      "pm": 87770,
      "triphone": 125806,
      "utters": 130681,
      "nonshared": 80290,
      "243": 804,
      "underconstrained": 126653,
      "internally": 56181,
      "2k": 884,
      "iu": 57150,
      "slides": 110011,
      "imperfection": 51922,
      "obstruct": 81251,
      "megabytes": 68263,
      "reversed": 102528,
      "sinhalese": 109867,
      "lend": 63347,
      "blockwise": 14366,
      "decentralized": 27921,
      "singlegpu": 109826,
      "replay": 98954,
      "pa": 83400,
      "adda": 3354,
      "547": 1107,
      "wellused": 132445,
      "assembles": 9460,
      "pioneer": 87518,
      "jit": 57238,
      "schools": 103967,
      "northern": 80371,
      "aged": 4288,
      "11000": 234,
      "audiovisual": 10247,
      "turnaround": 125981,
      "polyphone": 88021,
      "hinting": 50504,
      "inevitable": 54080,
      "committing": 18851,
      "allosaurus": 5072,
      "pronounce": 92345,
      "250000": 824,
      "smallsized": 110252,
      "sigmorphon": 108669,
      "executes": 39972,
      "vad": 130696,
      "promptly": 92327,
      "disparities": 32372,
      "924": 1399,
      "mit": 70356,
      "hts": 50699,
      "manytomany": 67100,
      "clir": 17802,
      "lithuanian": 64784,
      "layernorm": 61758,
      "crossmodality": 25064,
      "mbart": 67561,
      "overloaded": 83342,
      "perword": 87204,
      "speechtext": 111855,
      "auditory": 10249,
      "eventual": 39614,
      "extinct": 42054,
      "enlarged": 37549,
      "zenodo": 135469,
      "texthypothesis": 121438,
      "stand": 112179,
      "819": 1302,
      "tedx": 120057,
      "aphasia": 6775,
      "hallmark": 49299,
      "stroke": 113614,
      "836": 1313,
      "895": 1357,
      "asrs": 9456,
      "59": 1133,
      "downgrade": 33976,
      "317": 924,
      "tibetoburman": 121955,
      "det": 29748,
      "administered": 3883,
      "subnetwork": 114787,
      "abnormal": 1706,
      "contributor": 22797,
      "licensed": 63970,
      "mock": 70472,
      "finitestate method": 44545,
      "segmenting words": 104668,
      "set handcrafted": 107456,
      "accuracy best": 2108,
      "best using": 13468,
      "speech sentences": 111799,
      "speech audio": 111650,
      "extraction topic": 42519,
      "prosody information": 93650,
      "using decision": 129537,
      "prosodic model": 93647,
      "better wordbased": 13771,
      "recognized words": 97061,
      "prosodic events": 93642,
      "features highly": 43540,
      "highly informative": 50331,
      "accuracy automatic": 2104,
      "consists generating": 21477,
      "representation turn": 99455,
      "turn based": 125970,
      "based actual": 11498,
      "speech recordings": 111793,
      "possible approaches": 88381,
      "entropy modeling": 38170,
      "classifiers stacked": 17635,
      "accuracy obtaining": 2223,
      "single classifiers": 109714,
      "error reductions": 38343,
      "level accuracy": 63416,
      "integrating prosodic": 55800,
      "automatic topic": 10699,
      "segmentation present": 104621,
      "present probabilistic": 89647,
      "lexical prosodic": 63793,
      "prosodic information": 93644,
      "obtained speech": 81414,
      "lexicon learning": 63892,
      "words transcribed": 134278,
      "empirical tests": 36211,
      "recently task": 96771,
      "boundaries based": 14514,
      "model described": 70971,
      "tasks presented": 119393,
      "speakers paper": 111323,
      "dynamic manner": 34314,
      "speech representations": 111795,
      "representations largely": 99717,
      "information spoken": 55003,
      "retrieval terms": 102434,
      "terms vocabulary": 120405,
      "collection complete": 18462,
      "documents relevant": 33282,
      "results associated": 101524,
      "domains spoken": 33862,
      "accuracy experiments": 2159,
      "using target": 130266,
      "queries propose": 94935,
      "nonlinear function": 80224,
      "function form": 46018,
      "form scale": 45120,
      "determined previously": 30138,
      "previously encountered": 90595,
      "form time": 45135,
      "describes test": 29442,
      "task performed": 118530,
      "used collection": 128430,
      "size speech": 109945,
      "text queries": 121225,
      "formulate queries": 45282,
      "associated target": 9609,
      "extracts audio": 42577,
      "vocabulary continuous": 131870,
      "obtained human": 81377,
      "queries produce": 94934,
      "generated automatic": 47051,
      "paper improve": 83979,
      "uses speech": 129281,
      "models hmm": 73337,
      "iterative procedure": 57132,
      "hmm models": 50582,
      "models defined": 73029,
      "automated method": 10453,
      "based detection": 11640,
      "work developed": 134472,
      "tested techniques": 120584,
      "models retraining": 73958,
      "models representing": 73930,
      "spoken native": 112013,
      "models adapted": 72715,
      "joint use": 57330,
      "best combination": 13319,
      "la parole": 58378,
      "non native": 80146,
      "models integration": 73412,
      "propose combine": 92584,
      "error detection": 38302,
      "composing words": 20086,
      "words average": 133833,
      "transfer generation": 124056,
      "present architecture": 89383,
      "quality used": 94820,
      "directly output": 31894,
      "online offline": 81787,
      "automatically changed": 10731,
      "survey automatic": 116163,
      "advances past": 4010,
      "past 60": 85631,
      "60 years": 1148,
      "technological perspective": 120021,
      "variations context": 130934,
      "requires careful": 100248,
      "speech classes": 111653,
      "speech representation": 111794,
      "speech classifiers": 111656,
      "problems existing": 91319,
      "techniques solve": 119986,
      "authors hope": 10382,
      "work shall": 134797,
      "known methods": 58310,
      "model hmm": 71288,
      "shown speech": 108526,
      "algorithm extended": 4734,
      "performance despite": 86289,
      "analogous behavior": 5414,
      "previously understood": 90627,
      "understood does": 127024,
      "does appear": 33329,
      "second fitting": 104409,
      "phenomenon result": 87258,
      "standard methodology": 112260,
      "modify standard": 74457,
      "based mean": 11833,
      "illustrate performance": 51741,
      "presented approach": 89777,
      "decoding incorporating": 28098,
      "improvement achieved": 52680,
      "accurate automatic": 2325,
      "applied automatic": 7047,
      "fact key": 42826,
      "data fact": 25941,
      "better able": 13497,
      "model choice": 70822,
      "contemporary research": 21852,
      "preliminary exploration": 89272,
      "aimed understanding": 4488,
      "uses standard": 129282,
      "data demonstrates": 25827,
      "statistical dependency": 113092,
      "using simulation": 130180,
      "dependency data": 29138,
      "text independent": 121049,
      "speaker recognition": 111296,
      "using source": 130199,
      "used representation": 128726,
      "captured using": 15440,
      "speech contains": 111662,
      "effectively captured": 34798,
      "continuous stream": 22639,
      "distinct words": 32550,
      "statistical relationships": 113157,
      "languages deep": 60485,
      "english outperforms": 37231,
      "word accuracy": 132902,
      "different scoring": 31400,
      "combinations methods": 18602,
      "article instead": 9136,
      "earlier algorithms": 34376,
      "algorithms performance": 4874,
      "combining method": 18734,
      "method methods": 68962,
      "recognition large": 96902,
      "increasing training": 53686,
      "model estimated": 71100,
      "data depending": 25828,
      "availability training": 10926,
      "rate 10": 95771,
      "10 relative": 147,
      "scale distributed": 103714,
      "acoustic modeling": 3018,
      "size measured": 109933,
      "measured number": 67838,
      "model approximately": 70678,
      "accomplished using": 1955,
      "recognition experiments": 96880,
      "rescoring framework": 100390,
      "proves effective": 93743,
      "hours training": 50684,
      "recognition confidence": 96843,
      "models ranging": 73851,
      "ranging size": 95619,
      "estimated using": 38641,
      "relative reductions": 98374,
      "information respectively": 54925,
      "lexical access": 63725,
      "speech understanding": 111841,
      "encoding lexical": 36706,
      "determining intended": 30149,
      "sequence phonemes": 107028,
      "problem model": 91129,
      "word partofspeech": 133401,
      "read speech": 95940,
      "using equivalence": 129641,
      "equivalence classes": 38246,
      "recognition process": 96972,
      "segmentation paper": 104607,
      "detect words": 29822,
      "location word": 64969,
      "compare algorithm": 19220,
      "algorithm previous": 4787,
      "corpus basic": 23679,
      "report effort": 98989,
      "issues challenges": 57039,
      "aware language": 11260,
      "ongoing process": 81737,
      "task local": 118360,
      "method comprises": 68719,
      "proposed requires": 93538,
      "identification automatic": 51366,
      "development proper": 30414,
      "lexicon usually": 63913,
      "based work": 12171,
      "names given": 76143,
      "database propose": 26645,
      "propose algorithms": 92540,
      "function results": 46042,
      "independent language": 53772,
      "performance better": 86182,
      "spoken speech": 112018,
      "ability classify": 1494,
      "speech based": 111652,
      "based style": 12081,
      "american british": 5308,
      "real speech": 96078,
      "metric performs": 69893,
      "different human": 31173,
      "accuracy general": 2171,
      "deployed service": 29311,
      "interface human": 56100,
      "addition language": 3446,
      "recognition application": 96806,
      "engine separate": 37034,
      "technique automatic": 119769,
      "recognition involves": 96898,
      "vector classification": 131299,
      "step question": 113299,
      "improved classification": 52599,
      "using subset": 130234,
      "used classification": 128422,
      "model algorithm": 70651,
      "classification accuracies": 17099,
      "telephone speech": 120066,
      "techniques beneficial": 119845,
      "number outputs": 80931,
      "outputs trained": 83186,
      "data joint": 26057,
      "model lastly": 71418,
      "techniques result": 119975,
      "result 80": 101356,
      "set 23": 107342,
      "23 relative": 792,
      "probabilities words": 90855,
      "underestimated language": 126656,
      "method involving": 68923,
      "words produce": 134141,
      "phone recognition": 87274,
      "fields scrfs": 44005,
      "structured support": 114045,
      "success speech": 115130,
      "decoding models": 28109,
      "slow decoding": 110101,
      "expensive features": 40420,
      "features segment": 43713,
      "highorder features": 50358,
      "approximate decoding": 8453,
      "decoding beam": 28077,
      "hinge loss": 50496,
      "structured svms": 114047,
      "suitable learning": 115402,
      "gives good": 48194,
      "instead consider": 55656,
      "consider approach": 21188,
      "boundary features": 14520,
      "paper analyzed": 83734,
      "recognizer using": 97063,
      "distribution information": 32658,
      "confusion matrices": 21053,
      "document ranking": 33059,
      "term detection": 120208,
      "algorithm improve": 4752,
      "detection std": 30064,
      "algorithm calculates": 4701,
      "improve reliability": 52521,
      "english respectively": 37264,
      "method confidence": 68726,
      "applicable languages": 6825,
      "using cascaded": 129431,
      "optimization techniques": 82214,
      "huge quantities": 50725,
      "hard acquire": 49449,
      "given raw": 48100,
      "raw speech": 95857,
      "data linguistic": 26093,
      "structure includes": 113878,
      "patterns patterns": 85771,
      "learned unlabelled": 62274,
      "linguistic lexical": 64505,
      "labels respectively": 58636,
      "structure learned": 113895,
      "layer layer": 61727,
      "news including": 79347,
      "including task": 53393,
      "compared parallel": 19405,
      "test using": 120549,
      "detection spoken": 30060,
      "corpus different": 23758,
      "states model": 113047,
      "model number": 71596,
      "number distinct": 80874,
      "distinct models": 32540,
      "jointly capture": 57335,
      "caused different": 15830,
      "outperformed unsupervised": 82795,
      "considering context": 21320,
      "approach enhancing": 7531,
      "patterns form": 85745,
      "space multiple": 111030,
      "space shown": 111062,
      "jointly capturing": 57336,
      "capturing characteristics": 15460,
      "corpus representing": 23969,
      "good improvements": 48479,
      "improvements observed": 52887,
      "acoustic modelling": 3019,
      "modelling speech": 72623,
      "highly domain": 50318,
      "fact widely": 42838,
      "specific work": 111520,
      "paper concentrates": 83770,
      "method perform": 69042,
      "assumed exist": 9652,
      "data audio": 25626,
      "data covering": 25803,
      "based domains": 11662,
      "domains shown": 33856,
      "relative compared": 98346,
      "training 10": 123342,
      "10 compared": 126,
      "international workshop": 56188,
      "2014 paper": 578,
      "network systems": 77448,
      "systems adaptation": 116719,
      "techniques machine": 119926,
      "achieved phrasebased": 2665,
      "gives bleu": 48187,
      "englishtofrench englishtogerman": 37441,
      "translation gives": 124841,
      "phonetic units": 87301,
      "space resulting": 111054,
      "space compare": 110981,
      "old new": 81643,
      "speech segments": 111798,
      "requires stronger": 100324,
      "yield large": 135345,
      "task latent": 118342,
      "broadcast media": 14686,
      "domains diverse": 33765,
      "multigenre broadcast": 75109,
      "mixture latent": 70421,
      "better robust": 13711,
      "method latent": 68932,
      "used adapt": 128377,
      "dnns using": 32937,
      "using unique": 130337,
      "binary code": 14125,
      "shows training": 108641,
      "13 relative": 288,
      "document words": 33110,
      "approach input": 7639,
      "layer learns": 61728,
      "learns assign": 63194,
      "assign importance": 9543,
      "words ability": 133799,
      "convergence training": 22927,
      "translation slt": 125266,
      "increasingly globalized": 53698,
      "globalized world": 48284,
      "world social": 135049,
      "social economic": 110336,
      "research activities": 100396,
      "language real": 60023,
      "news shows": 79364,
      "stateoftheart acoustic": 112575,
      "modeling asr": 72380,
      "factored translation": 42866,
      "comparable bilingual": 19133,
      "corpora preparation": 23557,
      "task mt": 118413,
      "recognition english": 96866,
      "english mandarin": 37205,
      "replaces entire": 98945,
      "networks endtoend": 77587,
      "techniques resulting": 119976,
      "efficiency experiments": 35024,
      "architectures algorithms": 8780,
      "datasets finally": 27484,
      "finally using": 44246,
      "deployed online": 29307,
      "mgb challenge": 70028,
      "advancing state": 4031,
      "recognition speaker": 97003,
      "investigated work": 56847,
      "unreliable data": 127507,
      "shows acoustic": 108546,
      "highly variable": 50354,
      "segmentation followed": 104589,
      "set consisting": 107399,
      "embedding generated": 35411,
      "resource management": 100863,
      "noiseaware training": 80077,
      "ivector framework": 57155,
      "framework multitask": 45619,
      "years numerous": 135285,
      "model traditional": 72177,
      "far attention": 43070,
      "conjunction traditional": 21069,
      "networks potential": 77699,
      "dictionaries language": 30861,
      "similar problem": 109124,
      "problem faced": 91052,
      "categorical linguistic": 15723,
      "unsupervised bayesian": 127607,
      "approach potential": 7802,
      "word segment": 133488,
      "arbitrary length": 8586,
      "performing segmentation": 86968,
      "20 error": 502,
      "10 absolute": 120,
      "using ivector": 129769,
      "based distances": 11648,
      "hours speech": 50683,
      "speaker independent": 111290,
      "observed proposed": 81229,
      "modeling use": 72582,
      "polyglot language": 88010,
      "sequences different": 107119,
      "language predicted": 59818,
      "perplexity qualitative": 87068,
      "applications make": 6964,
      "representations higher": 99676,
      "quality learned": 94704,
      "approach spoken": 7929,
      "low resourced": 65396,
      "languages automatic": 60416,
      "linguistically meaningful": 64591,
      "based automatically": 11538,
      "dataset improvement": 26972,
      "lower best": 65421,
      "proposes efficient": 93599,
      "datadriven technique": 26680,
      "build better": 14747,
      "regular dictionaries": 97723,
      "temporal modelling": 120112,
      "modelling approach": 72596,
      "dependencies better": 29067,
      "novel incremental": 80598,
      "combines features": 18685,
      "selection criteria": 104776,
      "relies small": 98661,
      "orthographic word": 82600,
      "speech critical": 111666,
      "predict speech": 88940,
      "specifically try": 111604,
      "learn dynamics": 62035,
      "prediction error": 89055,
      "dataset improvements": 26974,
      "similar methods": 109112,
      "networks deep": 77560,
      "advanced stateoftheart": 3958,
      "years compared": 135246,
      "usually larger": 130433,
      "previously investigated": 90603,
      "accuracy smaller": 2278,
      "smaller number": 110240,
      "parameters compared": 84740,
      "paper push": 84392,
      "technique known": 119797,
      "teacherstudent training": 119705,
      "cumbersome model": 25236,
      "ami meeting": 5316,
      "meeting speech": 68256,
      "corpus technique": 24036,
      "technique significantly": 119814,
      "flexible feature": 44659,
      "techniques possible": 119951,
      "reducing decoding": 97412,
      "pruning training": 94173,
      "recognition distant": 96864,
      "recognition challenge": 96827,
      "challenge particularly": 16080,
      "caused large": 15833,
      "order cope": 82303,
      "realworld situations": 96180,
      "modeling including": 72447,
      "time 60": 121977,
      "texttospeech synthesis": 121661,
      "seen lot": 104534,
      "challenges systems": 16214,
      "systems assume": 116739,
      "computer interaction": 20482,
      "variations spelling": 130941,
      "spelling word": 111917,
      "synthesize speech": 116605,
      "approach methods": 7710,
      "methods convert": 69409,
      "learn deep": 62017,
      "hindi tamil": 50472,
      "native scripts": 76232,
      "released public": 98491,
      "nonparametric approach": 80271,
      "modelling statistical": 72626,
      "trained output": 123230,
      "form gaussian": 45089,
      "synthesis approach": 116594,
      "prediction consistent": 89045,
      "method competitive": 68716,
      "second used": 104463,
      "used transform": 128829,
      "evaluate commonly": 38807,
      "published performance": 94367,
      "performance similar": 86714,
      "similar systems": 109153,
      "technology automatic": 120037,
      "addresses automatic": 3793,
      "large feature": 61089,
      "translation built": 124682,
      "features remain": 43698,
      "feedback user": 43839,
      "wordlike units": 133761,
      "transcribed audio": 123983,
      "written form": 135127,
      "text translations": 121378,
      "audio information": 10234,
      "translations test": 125499,
      "corpus spanish": 24007,
      "precision results": 88803,
      "range configurations": 95550,
      "methods joint": 69566,
      "based endtoend": 11679,
      "learns mapping": 63221,
      "conditional independence": 20757,
      "independence assumptions": 53759,
      "results noisy": 101997,
      "model flexible": 71195,
      "cases lack": 15650,
      "fast convergence": 43130,
      "convergence using": 22928,
      "issue experiment": 56996,
      "baselines showing": 12464,
      "baseline present": 12286,
      "80 hours": 1290,
      "constructed deep": 21660,
      "appropriate number": 8425,
      "convolutional layers": 23159,
      "pooling operations": 88047,
      "feature maps": 43296,
      "dimensions input": 31769,
      "adding convolutional": 3372,
      "feature map": 43294,
      "deep cnn": 28211,
      "features ivector": 43574,
      "substantial word": 114879,
      "rate reductions": 95803,
      "standard cnn": 112212,
      "likelihood score": 64126,
      "score combination": 104060,
      "joint decoding": 57266,
      "decoding deep": 28084,
      "shown success": 108533,
      "success endtoend": 115079,
      "used shallow": 128757,
      "build deep": 14761,
      "convolutional structures": 23218,
      "structure feature": 113865,
      "dnn training": 32933,
      "recently attempts": 96659,
      "process deep": 91454,
      "initial alignment": 55207,
      "training criterion": 123413,
      "final phase": 44117,
      "approach iterative": 7652,
      "iterative retraining": 57135,
      "better endtoend": 13571,
      "shown competitive": 108453,
      "existing endtoend": 40126,
      "gain better": 46336,
      "evaluation document": 39190,
      "speech systems": 111814,
      "data alleviate": 25591,
      "joint rnn": 57318,
      "effective algorithm": 34623,
      "modules trained": 74530,
      "outputs considered": 83163,
      "overall model": 83240,
      "results tested": 102266,
      "word segments": 133511,
      "tasks speech": 119516,
      "corresponding word": 24314,
      "relatively unexplored": 98419,
      "unexplored paper": 127061,
      "training losses": 123692,
      "loss word": 65306,
      "present analyses": 89366,
      "analyses learned": 5458,
      "using audio": 129360,
      "audio samples": 10236,
      "samples train": 103594,
      "speech english": 111683,
      "lstm gated": 65621,
      "unit gru": 127205,
      "propagation neural": 92393,
      "network hypothesis": 77276,
      "sequence features": 106941,
      "network multiview": 77364,
      "representations arbitrarylength": 99505,
      "words make": 134045,
      "idea map": 51332,
      "map acoustic": 67104,
      "examples word": 39898,
      "embeddings jointly": 35768,
      "models multiview": 73597,
      "verification paper": 131571,
      "adaptation using": 3278,
      "utterances particular": 130661,
      "speaker models": 111295,
      "likelihood ml": 64121,
      "identification step": 51446,
      "using gaussian": 129692,
      "approaches build": 8090,
      "endtoend architecture": 36872,
      "tasks online": 119349,
      "endtoend recurrent": 36956,
      "classification loss": 17256,
      "allow model": 5084,
      "architecture does": 8646,
      "aligned training": 4925,
      "allows deploy": 5139,
      "translation does": 124766,
      "learning decoding": 62485,
      "collection methodology": 18479,
      "spoken arabic": 111963,
      "devoted collection": 30475,
      "translation successful": 125300,
      "speakers directly": 111314,
      "systems comparison": 116797,
      "describes methods": 29414,
      "using measures": 129873,
      "manner place": 66956,
      "criteria focus": 24793,
      "acquire probabilistic": 3035,
      "noisychannel model": 80135,
      "innovations improve": 55283,
      "respectively empirical": 101133,
      "zero resource": 135480,
      "acoustic feature": 3002,
      "analysis lda": 5672,
      "crosslingual generalization": 24958,
      "downstream speech": 34027,
      "adaptation automatic": 3210,
      "propose exploit": 92668,
      "hypotheses perform": 51257,
      "unsupervised adaptation": 127597,
      "run experiments": 103461,
      "sets released": 107704,
      "characterised different": 16513,
      "realistic conditions": 96101,
      "adaptation performed": 3247,
      "performed data": 86917,
      "selected according": 104721,
      "predictions allow": 89158,
      "closely approximate": 17850,
      "better strong": 13730,
      "translating speech": 124606,
      "text lowresource": 121095,
      "scenarios automatic": 103841,
      "form audio": 45069,
      "present problem": 89649,
      "low recall": 65383,
      "architectures achieve": 8775,
      "robust automatic": 103021,
      "provide deep": 93798,
      "features additionally": 43358,
      "additionally identify": 3609,
      "low medium": 65373,
      "medium high": 68245,
      "consistently outperformed": 21427,
      "properties measure": 92465,
      "consistent observed": 21394,
      "benefit proposed": 12987,
      "alignments language": 5015,
      "transcriptions investigate": 124007,
      "valuable training": 130773,
      "technique produces": 119808,
      "models define": 73028,
      "sequence probability": 107037,
      "label paper": 58404,
      "paper train": 84475,
      "rnn encoder": 102914,
      "encoder used": 36585,
      "objective improves": 81088,
      "used pretrain": 128693,
      "encoder improves": 36532,
      "convergence rate": 22923,
      "rate learning": 95795,
      "recognition deep": 96852,
      "significant accuracy": 108718,
      "layers deep": 61767,
      "signals simultaneously": 108706,
      "mutually complementary": 75988,
      "proposed combination": 93241,
      "wer reductions": 132452,
      "sets respectively": 107706,
      "average model": 11201,
      "update local": 127794,
      "tasks accurate": 118914,
      "accurate recognition": 2355,
      "conversational corpus": 22998,
      "range human": 95570,
      "issues human": 57051,
      "set human": 107459,
      "performance measurements": 86527,
      "performance considerably": 86253,
      "community significantly": 19100,
      "significantly harder": 108915,
      "second lstm": 104426,
      "lstm trained": 65670,
      "lstm rnns": 65664,
      "effective network": 34720,
      "network sequential": 77430,
      "recognition deeper": 96854,
      "train deeper": 122923,
      "deeper network": 28449,
      "network introduce": 77289,
      "layerwise training": 61816,
      "methods deeper": 69428,
      "models competitive": 72940,
      "models layers": 73474,
      "model recognition": 71865,
      "reduces relative": 97395,
      "minimum bayesrisk": 70203,
      "dataset outperform": 27076,
      "asr produce": 9445,
      "develop standard": 30234,
      "unavailable paper": 126600,
      "investigate alternative": 56714,
      "training asr": 123364,
      "spoken document": 111982,
      "provides competitive": 94012,
      "similarity functions": 109236,
      "variations significant": 130939,
      "similarity function": 109235,
      "errors propose": 38403,
      "based binary": 11559,
      "task methods": 118390,
      "voice conversion": 131925,
      "adversarial networks": 4152,
      "networks building": 77529,
      "building voice": 14900,
      "highly valuable": 50353,
      "focus explaining": 44763,
      "observations latent": 81178,
      "objective building": 81067,
      "building speech": 14888,
      "results corroborate": 101646,
      "data demonstrate": 25822,
      "automatic measurement": 10588,
      "network works": 77479,
      "level input": 63461,
      "containing single": 21807,
      "output pair": 83103,
      "yields higher": 135413,
      "generalizes new": 46840,
      "prediction algorithm": 89027,
      "mobile keyboard": 70468,
      "finitestate transducer": 44548,
      "transducer fst": 124019,
      "latency constraints": 61565,
      "extend framework": 41784,
      "word predictions": 133417,
      "framework short": 45680,
      "implementation details": 51941,
      "decoder enables": 28005,
      "support advanced": 115954,
      "similarity embeddings": 109226,
      "identification verification": 51457,
      "experiments distinct": 40916,
      "adapting model": 3311,
      "reducing bias": 97407,
      "object recognition": 81053,
      "improved neural": 52619,
      "strategies increase": 113470,
      "broadly applied": 14705,
      "conventional neural": 22892,
      "model decoder": 70953,
      "decoding long": 28101,
      "rnns using": 102983,
      "application speech": 6884,
      "words low": 134041,
      "problem addressed": 90922,
      "improving acoustic": 53070,
      "output decoder": 83065,
      "methods literature": 69601,
      "connection words": 21137,
      "words recognised": 134160,
      "relatedness computed": 97917,
      "novelty paper": 80791,
      "web automatic": 132223,
      "recently encoderdecoder": 96684,
      "shown impressive": 108482,
      "alignments source": 5019,
      "mechanisms used": 68073,
      "used today": 128814,
      "sequence generated": 106947,
      "generated encoder": 47076,
      "nature tasks": 76671,
      "explored experimental": 41625,
      "attention achieve": 9791,
      "improvements reduce": 52904,
      "used standard": 128776,
      "architecture deep": 8642,
      "augmenting neural": 10343,
      "different voices": 31560,
      "model starting": 72083,
      "based similar": 12042,
      "improvement demonstrate": 52701,
      "demonstrate technique": 28887,
      "multispeaker tts": 75796,
      "recognition conversational": 96845,
      "largevocabulary continuous": 61544,
      "recognition lvcsr": 96904,
      "thousand hours": 121914,
      "hours audio": 50681,
      "collection labeled": 18476,
      "evaluating results": 39092,
      "results updated": 102296,
      "transcribed utterances": 123988,
      "speech task": 111819,
      "develop specific": 30233,
      "specific tools": 111504,
      "automatic semiautomatic": 10650,
      "data tools": 26563,
      "speaker diarization": 111285,
      "released open": 98488,
      "open license": 81914,
      "discusses current": 32308,
      "using heterogeneous": 129729,
      "heterogeneous corpora": 49851,
      "summer camp": 115657,
      "camp applied": 15084,
      "language exploration": 59033,
      "good proxy": 48494,
      "language exhibiting": 59021,
      "little benefit": 64798,
      "useful order": 128914,
      "order benefit": 82285,
      "benefit multiple": 12981,
      "multiple different": 75540,
      "event recognition": 39533,
      "detecting classifying": 29828,
      "events words": 39612,
      "specifically pitch": 111573,
      "pitch accents": 87573,
      "accents phrase": 1883,
      "context adding": 22000,
      "setup proposed": 107854,
      "yields strong": 135443,
      "present stateoftheart": 89717,
      "endtoend automatic": 36876,
      "asr model": 9436,
      "cnn based": 17989,
      "attentionbased decoder": 10064,
      "combine ctc": 18620,
      "ctc predictions": 25195,
      "prior systems": 90737,
      "beats traditional": 12625,
      "framework speech": 45694,
      "paper automatically": 83751,
      "automatically obtaining": 10806,
      "sequence acoustic": 106913,
      "aspect problem": 9343,
      "tend good": 120152,
      "thousand words": 121918,
      "lexicon terms": 63907,
      "better lexicons": 13621,
      "queries proposed": 94936,
      "distance cosine": 32408,
      "discriminate words": 32195,
      "model followed": 71199,
      "performance embeddings": 86332,
      "trained optimize": 123226,
      "runtime efficiency": 103480,
      "prosodic structure": 93648,
      "networks ability": 77498,
      "learning classifying": 62439,
      "variation paper": 130897,
      "network convnet": 77213,
      "outperforms lstm": 82916,
      "95 classification": 1420,
      "order select": 82401,
      "features information": 43567,
      "highly depends": 50313,
      "hypotheses introduce": 51255,
      "features compensate": 43415,
      "information apply": 54378,
      "carried different": 15535,
      "quality hypotheses": 94681,
      "advanced state": 3956,
      "require significant": 100198,
      "significant amounts": 108723,
      "amounts memory": 5350,
      "reasons challenging": 96327,
      "deploy neural": 29303,
      "platforms paper": 87671,
      "study applications": 114322,
      "results wsj": 102349,
      "networks report": 77732,
      "demonstrate combined": 28689,
      "multilingual research": 75352,
      "research data": 100456,
      "published online": 94364,
      "gap automatic": 46442,
      "based hybrid": 11761,
      "hybrid deep": 51178,
      "dnns trained": 32936,
      "entropy ce": 38161,
      "data step": 26505,
      "step number": 113289,
      "combination data": 18557,
      "learning speech": 63044,
      "discovery multiple": 32145,
      "bottleneck features": 14495,
      "framework framework": 45549,
      "challenge organized": 16072,
      "keyword search": 57638,
      "results presented": 102050,
      "speed single": 111878,
      "low memory": 65375,
      "systems monolingual": 117006,
      "require languagespecific": 100153,
      "data handcrafted": 25995,
      "rules available": 103382,
      "alternative present": 5239,
      "11 improvement": 218,
      "model compact": 70850,
      "relative previous": 98367,
      "approaches comparison": 8102,
      "elegant approach": 35265,
      "vectors output": 131458,
      "output symbols": 83130,
      "loss language": 65280,
      "ngram lms": 79440,
      "search finally": 104316,
      "sequence individual": 106960,
      "word string": 133589,
      "analyze error": 5970,
      "2017 conversational": 630,
      "2016 recent": 616,
      "aware lstm": 11261,
      "adopt twostage": 3899,
      "level followed": 63450,
      "comparing human": 19504,
      "automatic recognition": 10641,
      "accuracy levels": 2198,
      "human transcribers": 50975,
      "raises question": 95462,
      "comparing output": 19510,
      "standard speech": 112306,
      "degree overlap": 28573,
      "attached words": 9692,
      "overlap training": 83331,
      "finally report": 44230,
      "investigate automatic": 56721,
      "modules paper": 74527,
      "errors using": 38417,
      "based joint": 11783,
      "transcription translation": 124001,
      "translation step": 125295,
      "inhouse corpus": 55202,
      "classifier ctc": 17536,
      "general corpus": 46648,
      "directly predict": 31898,
      "increasing computational": 53668,
      "trained subtitles": 123296,
      "facilitate process": 42786,
      "trained available": 123074,
      "corpus spoken": 24014,
      "language compare": 58895,
      "words small": 134225,
      "freely downloaded": 45789,
      "time investigate": 122040,
      "time budget": 121991,
      "location size": 64967,
      "model estimates": 71101,
      "cost model": 24365,
      "training cost": 123410,
      "models ongoing": 73661,
      "removes need": 98881,
      "realistic user": 96108,
      "improvements 15": 52791,
      "relative average": 98342,
      "model updating": 72252,
      "framework capture": 45452,
      "observe large": 81200,
      "released far": 98484,
      "far largest": 43091,
      "including audio": 53263,
      "including transcriptions": 53399,
      "increase depth": 53593,
      "time dimension": 122015,
      "memory cell": 68299,
      "recognition proposed": 96975,
      "reported tasks": 99076,
      "corpus computational": 23709,
      "information world": 55110,
      "systems constructed": 116810,
      "resource conditions": 100832,
      "example tasks": 39797,
      "transcriptions available": 124005,
      "community reproducible": 19097,
      "attentionbased seq2seq": 10098,
      "describes various": 29445,
      "including batch": 53265,
      "residual network": 100741,
      "network finally": 77258,
      "accuracy provides": 2250,
      "investigated paper": 56840,
      "vectors constant": 131417,
      "internal feature": 56165,
      "performance widelyused": 86854,
      "trained sequencetosequence": 123274,
      "order adapt": 82270,
      "adapt models": 3181,
      "keyword systems": 57640,
      "explore application": 41517,
      "application deep": 6839,
      "dilated convolutions": 31739,
      "benchmark best": 12735,
      "network resnet": 77410,
      "networks terms": 77793,
      "model depth": 70970,
      "examine approaches": 39736,
      "approaches keyword": 8203,
      "development future": 30389,
      "classification connectionist": 17154,
      "recognition typically": 97034,
      "instead rnns": 55680,
      "recognition particular": 96959,
      "explore range": 41581,
      "cnnbased models": 18030,
      "models typical": 74254,
      "time decoding": 122010,
      "train decode": 122918,
      "learning free": 62607,
      "corpus shared": 23991,
      "paper designed": 83861,
      "designed novel": 29613,
      "data transcription": 26578,
      "characters paper": 16617,
      "finally observe": 44208,
      "embeddings speech": 35949,
      "representations audio": 99511,
      "contain semantic": 21758,
      "rnn encoderdecoder": 102915,
      "information audio": 54384,
      "segments taken": 104675,
      "directly raw": 31901,
      "images challenging": 51808,
      "challenging expensive": 16253,
      "expensive collect": 40416,
      "languages typically": 60932,
      "recipe language": 96789,
      "labels share": 58642,
      "parameters work": 84797,
      "exploit recent": 41438,
      "create single": 24640,
      "single multilingual": 109768,
      "use universal": 128347,
      "used initialize": 128592,
      "monolingual speech": 74619,
      "systems achieving": 116717,
      "model online": 71608,
      "resources explore": 100975,
      "combined proposed": 18666,
      "learning improvements": 62643,
      "results 19": 101485,
      "19 relative": 436,
      "compared randomlyinitialized": 19428,
      "randomlyinitialized baseline": 95535,
      "corpora human": 23493,
      "information corpora": 54457,
      "corpora multiple": 23533,
      "theoretical framework": 121795,
      "based connectionist": 11595,
      "model emits": 71056,
      "model reduce": 71874,
      "model 30": 70500,
      "challenge 2017": 15987,
      "challenge aimed": 15992,
      "word units": 133618,
      "languages adapt": 60387,
      "features evaluation": 43491,
      "models discussed": 73090,
      "tts synthesis": 125918,
      "mean opinion": 67599,
      "opinion score": 82097,
      "score mos": 104094,
      "present ablation": 89358,
      "features demonstrate": 43443,
      "using compact": 129456,
      "representation enables": 99221,
      "text contrast": 120820,
      "contrast using": 22717,
      "using characters": 129438,
      "allows good": 5161,
      "good tradeoff": 48503,
      "set available": 107370,
      "grapheme based": 48918,
      "connected convolutional": 21124,
      "adaptation scheme": 3261,
      "corpus improved": 23833,
      "portion evaluation": 88184,
      "combination systems": 18587,
      "phone sets": 87275,
      "recognition obtained": 96949,
      "respectively best": 101126,
      "reported far": 99062,
      "latest work": 61660,
      "corpus direct": 23759,
      "language aligned": 58824,
      "aligned text": 4923,
      "paper tries": 84479,
      "translations obtain": 125480,
      "subset corpus": 114827,
      "bilingual alignment": 14020,
      "source speech": 110827,
      "text single": 121309,
      "baseline corpus": 12205,
      "future linguistic": 46281,
      "exploring computational": 41660,
      "computational scientific": 20430,
      "subwords words": 115055,
      "language help": 59118,
      "support natural": 115979,
      "speech technologies": 111820,
      "models naturally": 73617,
      "units finally": 127242,
      "decoder task": 28057,
      "translation leads": 124899,
      "input automatic": 55299,
      "languages serve": 60863,
      "universal phone": 127319,
      "natural consider": 76245,
      "available train": 11130,
      "language adapting": 58812,
      "hours transcribed": 50685,
      "needs studied": 76900,
      "particularly stateoftheart": 85496,
      "darpa lorelei": 25544,
      "lorelei program": 65236,
      "evaluating asr": 39036,
      "humanitarian assistance": 51039,
      "approach asr": 7366,
      "rapid adaptation": 95708,
      "obtain significantly": 81323,
      "competing approaches": 19616,
      "lstm blstm": 65606,
      "tts systems": 125919,
      "generated speech": 47124,
      "inference cost": 54130,
      "shown consistent": 108455,
      "generation time": 47680,
      "directly map": 31886,
      "usually required": 130451,
      "techniques utilized": 120012,
      "novel modular": 80651,
      "performing endtoend": 86949,
      "sequence lengths": 107005,
      "loss finally": 65265,
      "modules integrated": 74525,
      "classification attention": 17127,
      "attention modeling": 9933,
      "framework particular": 45650,
      "using time": 130292,
      "improve attention": 52336,
      "extracted network": 42167,
      "applied context": 7051,
      "vectors time": 131477,
      "individual components": 53902,
      "e2e model": 34359,
      "attempt improve": 9746,
      "testing time": 120606,
      "improves baseline": 52949,
      "baseline wordbased": 12339,
      "proposed attention": 93229,
      "improve speech": 52542,
      "example using": 39799,
      "using canonical": 129426,
      "analysis cca": 5524,
      "learned feature": 62218,
      "feature models": 43298,
      "difficult port": 31653,
      "port new": 88174,
      "settings based": 107789,
      "similar proposed": 109126,
      "speech version": 111848,
      "use semantic": 128260,
      "learned word2vec": 62289,
      "aim collect": 4451,
      "language collect": 58884,
      "explore improve": 41551,
      "assistance text": 9572,
      "neural multisource": 78215,
      "shared attention": 107965,
      "baselines reducing": 12451,
      "applications lowresource": 6959,
      "approach cascading": 7413,
      "usually available": 130408,
      "available lowresource": 11037,
      "intermediate transcription": 56150,
      "transcription investigate": 123999,
      "improvements allow": 52804,
      "allow train": 5092,
      "train fewer": 122936,
      "words explore": 133952,
      "explore models": 41562,
      "models selfattention": 73987,
      "modeling discrete": 72420,
      "modeling computational": 72400,
      "quadratically sequence": 94545,
      "length address": 63351,
      "models end": 73151,
      "stress importance": 113580,
      "explicit control": 41311,
      "model approaches": 70676,
      "approaches strong": 8355,
      "based lstms": 11820,
      "heads learn": 49608,
      "achieve unsupervised": 2585,
      "read english": 95930,
      "dataset audio": 26747,
      "utterances represented": 130664,
      "represented vector": 100037,
      "method utilized": 69216,
      "segments text": 104676,
      "transformed embeddings": 124283,
      "framework unsupervised": 45729,
      "audio embeddings": 10230,
      "annotation learning": 6329,
      "audio signals": 10238,
      "data basic": 25691,
      "learn mapping": 62085,
      "study develop": 114355,
      "use teacherstudent": 128319,
      "learning adapt": 62308,
      "cost need": 24367,
      "learning utilizes": 63153,
      "data boost": 25706,
      "model losing": 71476,
      "losing accuracy": 65241,
      "speakers using": 111331,
      "using low": 129838,
      "audio feature": 10231,
      "recognition possible": 96966,
      "features speech": 43741,
      "ngrams approach": 79460,
      "approach followed": 7579,
      "adapted language": 3284,
      "data believe": 25693,
      "believe results": 12703,
      "simple accurate": 109354,
      "partofspeech induction": 85526,
      "pos induction": 88208,
      "information induced": 54684,
      "context focus": 22107,
      "descent sgd": 29373,
      "robust generalized": 103036,
      "new network": 79048,
      "multiple attentions": 75500,
      "integrates outputs": 55786,
      "leading improvement": 61895,
      "conduct experimental": 20845,
      "corpus spontaneous": 24016,
      "japanese experimental": 57188,
      "contexts attentionbased": 22379,
      "prediction unseen": 89148,
      "task prediction": 118546,
      "dedicated task": 28183,
      "prediction approaches": 89030,
      "approaches compared": 8101,
      "new strategy": 79187,
      "particularly focus": 85482,
      "use textual": 128328,
      "features did": 43456,
      "inputs cnns": 55480,
      "twice data": 126081,
      "speaker adaptation": 111281,
      "train acoustic": 122898,
      "development automatic": 30372,
      "outperforms endtoend": 82883,
      "respectively word": 101178,
      "designed make": 29608,
      "make experiments": 66666,
      "effects word": 35008,
      "detection makes": 29987,
      "use acoustic": 127878,
      "stateoftheart vector": 113024,
      "examine effect": 39743,
      "languages openset": 60770,
      "embeddings temporal": 35976,
      "context includes": 22135,
      "pairs original": 83596,
      "segment pairs": 104563,
      "pairs obtain": 83592,
      "matching acoustic": 67393,
      "representations reduce": 99846,
      "reduce runtime": 97355,
      "paper investigated": 84027,
      "regularization techniques": 97757,
      "replicate results": 98961,
      "results help": 101820,
      "training convolutional": 123397,
      "annotated speech": 6237,
      "infeasible use": 54092,
      "matching apply": 67396,
      "cnn classifier": 17995,
      "keywords improving": 57647,
      "area roc": 8883,
      "roc curve": 103139,
      "mandarin cantonese": 66912,
      "fundamental frequency": 46096,
      "tree framework": 125593,
      "framework performs": 45653,
      "forest model": 45051,
      "advantage new": 4054,
      "model according": 70526,
      "retrieval novel": 102416,
      "genetic algorithm": 47818,
      "syllable boundaries": 116302,
      "tested english": 120569,
      "words lowresource": 134042,
      "unstructured audio": 127569,
      "method topic": 69190,
      "using cascade": 129430,
      "modeling translation": 72576,
      "classification instead": 17234,
      "dependencies sequential": 29114,
      "recognition improved": 96892,
      "improved hierarchical": 52608,
      "consistent previous": 21398,
      "recognition specifically": 97004,
      "training subwordlevel": 123890,
      "lowerresource settings": 65459,
      "learning observe": 62858,
      "improves standard": 53045,
      "multi task": 75027,
      "recognition challenging": 96828,
      "abstract target": 1784,
      "words reason": 134157,
      "phoneme based": 87278,
      "tend outperform": 120157,
      "encourage formation": 36754,
      "representations achieve": 99489,
      "achieve performing": 2510,
      "targets different": 117790,
      "architecture exhibits": 8661,
      "exhibits improvements": 40014,
      "decoder language": 28026,
      "models open": 73663,
      "applications research": 7009,
      "research stateoftheart": 100628,
      "models german": 73299,
      "previously best": 90590,
      "achieves word": 2921,
      "topics included": 122630,
      "included training": 53248,
      "variation topic": 130901,
      "text paired": 121160,
      "backtranslation technique": 11342,
      "sequence hidden": 106958,
      "extracted pretrained": 42171,
      "achieve faster": 2461,
      "faster attention": 43162,
      "encoder use": 36584,
      "model speaker": 72073,
      "dependencies unlike": 29122,
      "states large": 113046,
      "unpaired text": 127486,
      "using generated": 129700,
      "generated hidden": 47084,
      "dataset demonstrates": 26861,
      "number unknown": 80993,
      "words need": 134075,
      "need paired": 76842,
      "articulatory features": 9237,
      "speech work": 111850,
      "type speech": 126227,
      "performance high": 86427,
      "high variability": 50144,
      "articulatory information": 9238,
      "network fcnn": 77254,
      "features time": 43760,
      "dutch flemish": 34280,
      "asr using": 9453,
      "training scenarios": 123827,
      "based acoustic": 11497,
      "problem employing": 91024,
      "difficulty getting": 31700,
      "lvcsr based": 65728,
      "models systematically": 74149,
      "efficient sequence": 35114,
      "approaches obtained": 8261,
      "process speech": 91572,
      "sequence labels": 106998,
      "weighted finitestate": 132352,
      "transducer wfst": 124020,
      "computational overheads": 20417,
      "classifiers deep": 17605,
      "text vectors": 121405,
      "carrying semantic": 15562,
      "word2vec unsupervised": 133685,
      "utterance directly": 130595,
      "represented sequence": 100033,
      "applications significantly": 7016,
      "presents latest": 89860,
      "networks demonstrated": 77563,
      "neuralbased models": 78735,
      "results wall": 102322,
      "previous paper": 90442,
      "paper presented": 84147,
      "using cnns": 129446,
      "encode text": 36460,
      "text asr": 120662,
      "model try": 72238,
      "captured deep": 15428,
      "relation different": 97971,
      "clear signal": 17730,
      "utterances according": 130624,
      "metalearning approach": 68546,
      "testing conditions": 120588,
      "conditions example": 20816,
      "update weights": 127800,
      "principled way": 90683,
      "adaptation outperforms": 3243,
      "experiments adapting": 40759,
      "largest opensource": 61536,
      "released baseline": 98475,
      "mmi objective": 70458,
      "dev test": 30167,
      "community hope": 19077,
      "topics like": 122636,
      "like transfer": 64105,
      "learning robust": 62978,
      "given audio": 47988,
      "matched mismatched": 67375,
      "using articulatory": 129346,
      "improves lowresource": 52993,
      "translation st": 125287,
      "pretrain model": 89976,
      "model highresource": 71287,
      "task finetune": 118216,
      "asr data": 9426,
      "shared language": 107986,
      "language audio": 58849,
      "lowresource task": 65570,
      "multimodal neural": 75448,
      "spoken languages": 112011,
      "languages logographic": 60703,
      "logographic origin": 65051,
      "languages encode": 60526,
      "like spanish": 64095,
      "modeling standard": 72550,
      "logographic characters": 65050,
      "multimodal approach": 75414,
      "unimodal multimodal": 127163,
      "multimodal baselines": 75420,
      "inference approaches": 54116,
      "examining trajectories": 39771,
      "using multilayer": 129911,
      "features combination": 43405,
      "classifiers prove": 17629,
      "data lstm": 26104,
      "features engineered": 43483,
      "inspired success": 55584,
      "network neural": 77371,
      "help multihead": 49755,
      "constructed parallel": 21667,
      "improves training": 53055,
      "directly selfattention": 31904,
      "output final": 83076,
      "conducted test": 20951,
      "speed training": 111880,
      "rigorous human": 102855,
      "human quality": 50937,
      "multitask transfer": 75897,
      "children learning": 16725,
      "application scenario": 6878,
      "scenario characterized": 103822,
      "students learning": 114149,
      "discusses techniques": 32314,
      "models starting": 74093,
      "nonnative evaluation": 80244,
      "results resulting": 102135,
      "memad submission": 68271,
      "describes memad": 29410,
      "memad project": 68269,
      "project entry": 92215,
      "task addressing": 117850,
      "english audio": 37073,
      "time systems": 122118,
      "systems start": 117156,
      "systems employ": 116864,
      "leading final": 61892,
      "final systems": 44131,
      "experimented various": 40748,
      "pre postprocessing": 88766,
      "years task": 135308,
      "task adversarial": 117853,
      "novel adversarial": 80455,
      "challenge learn": 16046,
      "able gain": 1641,
      "auxiliary text": 10896,
      "paper recent": 84395,
      "stateoftheart semisupervised": 112946,
      "absolute word": 1749,
      "performance 2017": 86104,
      "especially natural": 38484,
      "speech separation": 111801,
      "bulk research": 14950,
      "constructing highquality": 21675,
      "highquality synthetic": 50408,
      "applications using": 7036,
      "performance demonstrate": 86282,
      "variety data": 130967,
      "generalize multiple": 46812,
      "audio signal": 10237,
      "efficient example": 35074,
      "modeling classifying": 72392,
      "information phonetic": 54840,
      "word syllable": 133599,
      "unlimited number": 127460,
      "better cluster": 13538,
      "corresponding linguistic": 24292,
      "pairs unlabeled": 83656,
      "style training": 114600,
      "set including": 107464,
      "special consideration": 111357,
      "speech previous": 111729,
      "intermediate step": 56145,
      "text significantly": 121294,
      "speech datasets": 111670,
      "exist typically": 40031,
      "necessary evaluation": 76755,
      "evaluation introduce": 39246,
      "handle disfluencies": 49380,
      "endtoend translation": 36989,
      "representations crucial": 99580,
      "human computer": 50780,
      "interaction paper": 55958,
      "cues contextual": 25211,
      "contextual cues": 22450,
      "information focus": 54622,
      "focus model": 44788,
      "morphological transformations": 74752,
      "techniques train": 120000,
      "data asr": 25622,
      "similarity evaluations": 109231,
      "word relations": 133432,
      "enriching word": 37569,
      "using principal": 130054,
      "analysis embedding": 5589,
      "helps resolve": 49829,
      "resolve ambiguities": 100797,
      "pipeline specifically": 87555,
      "endtoend multimodal": 36932,
      "multimodal asr": 75417,
      "systems compare": 116795,
      "s2s models": 103516,
      "improvement 14": 52667,
      "08 absolute": 73,
      "ensemble decoding": 37591,
      "text directly": 120877,
      "characters subwords": 16624,
      "extraction consider": 42288,
      "recognition output": 96951,
      "encoding methods": 36712,
      "multiple hypotheses": 75575,
      "truth labels": 125882,
      "labels evaluate": 58596,
      "achieved 98": 2591,
      "effective speech": 34749,
      "analyze representations": 5999,
      "compare deep": 19235,
      "noise type": 80073,
      "result similar": 101405,
      "framework endtoend": 45520,
      "attentionbased methods": 10080,
      "promising research": 92289,
      "architectures multitask": 8829,
      "training joint": 123668,
      "decoding work": 28148,
      "model heterogeneous": 71280,
      "architectures temporal": 8850,
      "acoustic information": 3007,
      "information demonstrate": 54469,
      "research linguistic": 100547,
      "advances speech": 4019,
      "computational power": 20419,
      "data viable": 26621,
      "provides general": 94037,
      "e2e models": 34360,
      "predict output": 88914,
      "e2e speech": 34363,
      "number evaluation": 80882,
      "evaluation categories": 39143,
      "search decoding": 104307,
      "decoding using": 28147,
      "manner articulation": 66938,
      "networks require": 77736,
      "require priori": 100191,
      "source datasets": 110741,
      "lexical acoustic": 63726,
      "clustering process": 17953,
      "integrate word": 55770,
      "comprehensive way": 20283,
      "boundary information": 14522,
      "diverse handcrafted": 32814,
      "architecture apply": 8616,
      "accuracy observe": 2221,
      "feature generation": 43287,
      "contributions work": 22796,
      "work efficient": 134486,
      "aims making": 4549,
      "making better": 66827,
      "parameters associated": 84737,
      "memory previous": 68367,
      "contribution method": 22772,
      "score significantly": 104122,
      "recognition current": 96849,
      "systems build": 116762,
      "extraction pipelines": 42435,
      "approach trained": 7978,
      "matches current": 67378,
      "speech trained": 111826,
      "trained 12": 123061,
      "12 times": 261,
      "2x faster": 898,
      "faster optimized": 43186,
      "enable fast": 36353,
      "fast iteration": 43141,
      "crucial factor": 25141,
      "model tuning": 72239,
      "query recognition": 94977,
      "limited capabilities": 64217,
      "platform explore": 87653,
      "accurate stateoftheart": 2365,
      "resources evaluation": 100972,
      "dataset representing": 27157,
      "low false": 65363,
      "rate model": 95797,
      "issues exist": 57047,
      "exist current": 40018,
      "means model": 67767,
      "model frequently": 71208,
      "words tagging": 134255,
      "fixed set": 44604,
      "words study": 134244,
      "particular introduce": 85423,
      "word letter": 133338,
      "shared hidden": 107983,
      "achieves relative": 2844,
      "mtl approaches": 75014,
      "issue speaker": 57029,
      "adapts model": 3344,
      "model forcing": 71203,
      "output distribution": 83067,
      "regularization particular": 97753,
      "trees deep": 125686,
      "modeling output": 72499,
      "consider direct": 21199,
      "direct training": 31806,
      "identify issues": 51511,
      "behavior leads": 12663,
      "poor generalization": 88056,
      "trained rest": 123265,
      "model removes": 71901,
      "cases neural": 15659,
      "english performance": 37239,
      "examine impact": 39750,
      "impact choice": 51860,
      "phoneme grapheme": 87279,
      "lists generated": 64719,
      "provides limited": 94047,
      "nbest hypotheses": 76689,
      "search decoder": 104306,
      "decoder introduce": 28025,
      "optimize training": 82227,
      "time inference": 122038,
      "allows combine": 5131,
      "used target": 128802,
      "aligned input": 4916,
      "alignments demonstrate": 5011,
      "recognition jointly": 96899,
      "efforts shown": 35206,
      "networks attentionbased": 77514,
      "model final": 71181,
      "implicitly learning": 52032,
      "level language": 63465,
      "sequence based": 106918,
      "models share": 74019,
      "knowledge languages": 58034,
      "languages enabling": 60525,
      "approaches particularly": 8278,
      "useful low": 128903,
      "performs monolingual": 87009,
      "times fewer": 122178,
      "obtain comparable": 81266,
      "comparable word": 19193,
      "wfst decoding": 132467,
      "lowresource conditions": 65488,
      "paper simulate": 84433,
      "application data": 6838,
      "models hand": 73316,
      "labeling process": 58530,
      "random selection": 95511,
      "simplicity training": 109573,
      "projection matrix": 92235,
      "matrix maps": 67477,
      "softmax output": 110509,
      "use high": 128086,
      "projection layer": 92233,
      "vectors projected": 131466,
      "different projection": 31362,
      "projection matrices": 92234,
      "developed automatically": 30255,
      "automatically grade": 10782,
      "models employing": 73143,
      "employing deep": 36314,
      "original domain": 82513,
      "composed short": 20071,
      "short audio": 108203,
      "audio clips": 10224,
      "resources publicly": 101032,
      "available hope": 11007,
      "hope used": 50643,
      "embeddings improved": 35750,
      "subword systems": 115038,
      "address shortcomings": 3766,
      "idea based": 51316,
      "based treating": 12135,
      "matrix word": 67485,
      "externally trained": 42053,
      "acoustic embeddings": 3001,
      "style control": 114572,
      "speech existing": 111685,
      "style experimental": 114574,
      "convolutional sequencetosequence": 23217,
      "wer librispeech": 132450,
      "rnn baseline": 102908,
      "convolution block": 23137,
      "dramatically reduces": 34097,
      "stable efficient": 112112,
      "effectively integrate": 34825,
      "model coupled": 70936,
      "22 relative": 774,
      "librispeech test": 63967,
      "problem outofvocabulary": 91152,
      "lm perform": 64868,
      "using characterbased": 129435,
      "scenarios especially": 103844,
      "paper adapt": 83698,
      "decoder output": 28040,
      "layer attention": 61700,
      "models findings": 73250,
      "average prediction": 11209,
      "especially unseen": 38522,
      "presents high": 89853,
      "vietnamese speech": 131695,
      "used analyzing": 128387,
      "synthesis models": 116599,
      "males females": 66883,
      "corpus designed": 23750,
      "context speaker": 22274,
      "representations raw": 99839,
      "multilayer convolutional": 75169,
      "available approach": 10947,
      "robust dataefficient": 103028,
      "translation traditionally": 125386,
      "cascaded models": 15572,
      "texts recent": 121590,
      "model stronger": 72090,
      "unrealistic assumption": 127497,
      "data ignoring": 26013,
      "require data": 100126,
      "exploiting data": 41462,
      "modeling source": 72545,
      "auxiliary training": 10897,
      "data effectively": 25877,
      "text attracted": 120667,
      "st models": 112102,
      "model difficult": 70997,
      "model transferring": 72228,
      "possible implement": 88409,
      "model gain": 71218,
      "35 bleu": 960,
      "feedback effective": 43825,
      "speech input": 111698,
      "conversion using": 23090,
      "gan generate": 46429,
      "training mapping": 123701,
      "cycle consistency": 25482,
      "consistency loss": 21362,
      "able successfully": 1692,
      "method baseline": 68671,
      "propose variation": 93150,
      "incorporates alignment": 53502,
      "information computing": 54431,
      "reference hypothesis": 97521,
      "alignment performed": 4980,
      "adjust word": 3876,
      "similar scores": 109140,
      "quality reduce": 94771,
      "introduce variant": 56565,
      "underlying mathematical": 126686,
      "translation applications": 124641,
      "particularly presence": 85490,
      "baseline new": 12274,
      "simply using": 109629,
      "cases worse": 15672,
      "worse performance": 135067,
      "entity labeling": 37948,
      "labeling information": 58504,
      "data similarity": 26453,
      "set representative": 107565,
      "sampled target": 103572,
      "newly published": 79277,
      "published multilingual": 94362,
      "provides data": 94016,
      "languages fact": 60581,
      "languages exploited": 60572,
      "clean dataset": 17701,
      "finnish french": 44557,
      "final corpus": 44100,
      "utterances language": 130652,
      "showcase usefulness": 108363,
      "contextaware translation": 22349,
      "simultaneous interpreting": 109654,
      "simultaneously determine": 109673,
      "fluent translation": 44722,
      "translation simple": 125258,
      "decoding practice": 28124,
      "tradeoff latency": 122777,
      "model presents": 71769,
      "additional relevant": 3564,
      "present test": 89738,
      "capitalization punctuation": 15243,
      "punctuation automatic": 94393,
      "studies automatic": 114186,
      "results reach": 102111,
      "reach human": 95894,
      "output asr": 83049,
      "output semantically": 83117,
      "cause difficulties": 15822,
      "allows build": 5129,
      "decoding parallel": 28120,
      "corpus showed": 23994,
      "crosslingual topic": 25025,
      "speech using": 111843,
      "consider question": 21224,
      "setting small": 107778,
      "small speech": 110209,
      "language paired": 59735,
      "language develop": 58957,
      "translations poor": 125485,
      "correctly classify": 24174,
      "baseline useful": 12332,
      "crisis response": 24786,
      "surprise language": 116123,
      "pretrained acoustic": 89982,
      "corpora arabic": 23412,
      "aggregate outputs": 4349,
      "outputs systems": 83183,
      "voting approach": 131951,
      "weighted voting": 132359,
      "algorithm combining": 4705,
      "consistently reduces": 21438,
      "approaches block": 8089,
      "model update": 72251,
      "achieves linear": 2813,
      "suffers severe": 115265,
      "restricted network": 101342,
      "neighbor nodes": 76986,
      "process apply": 91423,
      "wikitext103 benchmark": 132698,
      "benchmark using": 12872,
      "using gpus": 129712,
      "addition performance": 3459,
      "detection resolution": 30036,
      "extremely simplified": 42610,
      "fast decoding": 43131,
      "model investigated": 71376,
      "used recover": 128718,
      "requires accurate": 100242,
      "lm trained": 64872,
      "trained transcriptions": 123314,
      "information detect": 54477,
      "reducing errors": 97415,
      "especially outofdomain": 38490,
      "scenario investigate": 103825,
      "prediction value": 89152,
      "used convolutional": 128455,
      "different environments": 31130,
      "types attacks": 126253,
      "features included": 43558,
      "training way": 123949,
      "help regularize": 49772,
      "languages popular": 60801,
      "results hint": 101830,
      "increase quality": 53612,
      "translation german": 124839,
      "based german": 11738,
      "50k parallel": 1089,
      "speech low": 111705,
      "sentence alignments": 105769,
      "parallel translation": 84698,
      "score knowledge": 104088,
      "neural document": 77900,
      "document language": 33030,
      "developments deep": 30440,
      "research recently": 100609,
      "retrieval framework": 102394,
      "pioneer study": 87519,
      "framework especially": 45526,
      "especially combined": 38441,
      "layers newly": 61794,
      "experiment number": 40481,
      "number layers": 80907,
      "learning rates": 62938,
      "key findings": 57571,
      "data epoch": 25900,
      "slight improvement": 110017,
      "datasets machine": 27553,
      "critically endangered": 24844,
      "approaches motivated": 8239,
      "dataset parallel": 27089,
      "audio files": 10233,
      "using resources": 130113,
      "github repository": 47978,
      "recognition developed": 96859,
      "assessment second": 9526,
      "recently witnessed": 96778,
      "models creation": 73001,
      "content multiple": 21910,
      "turnaround times": 125982,
      "nmt requires": 79954,
      "highquality large": 50389,
      "corpus comprised": 23706,
      "special symbols": 111370,
      "method annotating": 68636,
      "networks explore": 77597,
      "directly derived": 31872,
      "sequence keywords": 106964,
      "different design": 31091,
      "arbitrary set": 8589,
      "set userdefined": 107628,
      "userdefined keywords": 129066,
      "detection algorithm": 29873,
      "writing introduce": 135091,
      "dictionary containing": 30874,
      "high german": 50076,
      "variation written": 130905,
      "written forms": 135128,
      "uncertainty associated": 126614,
      "german words": 47928,
      "representation major": 99320,
      "transformer trained": 124384,
      "dictionary generation": 30880,
      "studies introduced": 114241,
      "introduced methods": 56583,
      "words encode": 133936,
      "features despite": 43450,
      "spaces learned": 111098,
      "preserve information": 89923,
      "distance words": 32437,
      "various studies": 131208,
      "studies human": 114235,
      "processing lexical": 91695,
      "address open": 3723,
      "date addition": 27822,
      "reported systems": 99075,
      "researchers practitioners": 100704,
      "convert chinese": 23093,
      "sentences chinese": 106237,
      "improving readability": 53155,
      "noisy asr": 80090,
      "tasks maintaining": 119285,
      "maintaining semantic": 66524,
      "address lack": 3710,
      "taskspecific data": 119622,
      "data synthesizing": 26530,
      "models traditional": 74189,
      "systems hope": 116936,
      "work serve": 134795,
      "serve basis": 107291,
      "basis nlp": 12544,
      "scalable multilingual": 103696,
      "experimented training": 40746,
      "languageindependent approach": 60360,
      "technique allows": 119765,
      "sentences models": 106400,
      "18 languages": 418,
      "context endtoend": 22078,
      "current production": 25326,
      "latency training": 61568,
      "complexity models": 19927,
      "compared actual": 19320,
      "unidirectional encoders": 127090,
      "future information": 46280,
      "latency inference": 61566,
      "propose strategies": 93086,
      "strategies training": 113492,
      "training leveraging": 123687,
      "hard alignments": 49450,
      "task studied": 118746,
      "decoder present": 28046,
      "analysis understand": 5922,
      "translation endtoend": 124788,
      "loosely coupled": 65232,
      "discussion main": 32323,
      "approaches stem": 8353,
      "representations speech": 99905,
      "way overcoming": 132114,
      "removing need": 98884,
      "explicit intermediate": 41326,
      "solving issues": 110642,
      "provides unifying": 94091,
      "highlighting tradeoffs": 50281,
      "concrete realizations": 20739,
      "outperforms universal": 83040,
      "explore implications": 41550,
      "endangered minority": 36842,
      "phonological typology": 87316,
      "faces problem": 42753,
      "models boost": 72860,
      "propose solutions": 93075,
      "previously overlooked": 90608,
      "information numerous": 54809,
      "data boosts": 25707,
      "robust unsupervised": 103081,
      "contextual biasing": 22447,
      "incorporate userspecific": 53493,
      "userspecific information": 129187,
      "bias prediction": 13833,
      "prediction common": 89042,
      "common framework": 18881,
      "framework dynamically": 45508,
      "explicit contextual": 41309,
      "bias score": 13843,
      "score word": 104141,
      "corpus bias": 23684,
      "accuracy relevant": 2260,
      "context available": 22015,
      "exhibits high": 40013,
      "learning improves": 62644,
      "based output": 11918,
      "critical accuracy": 24803,
      "improve multitask": 52426,
      "model utilizing": 72291,
      "provide just": 93863,
      "pattern matching": 85717,
      "phonetic encoding": 87291,
      "generic approach": 47791,
      "similar suggestion": 109152,
      "suggestion problem": 115364,
      "learning produce": 62918,
      "way people": 132116,
      "written differently": 135123,
      "learning classifier": 62436,
      "algorithms evaluated": 4848,
      "study including": 114405,
      "phonetic string": 87297,
      "algorithms results": 4880,
      "approach serve": 7885,
      "solving similar": 110649,
      "systems primarily": 117075,
      "primarily evaluated": 90639,
      "output readability": 83113,
      "time work": 122138,
      "scarcity training": 103814,
      "methods endtoend": 69475,
      "length constraints": 63356,
      "experiments limited": 40988,
      "needed training": 76879,
      "major hurdle": 66566,
      "datadriven research": 26678,
      "languages draw": 60518,
      "access data": 1905,
      "greatly facilitate": 49051,
      "scale languages": 103731,
      "languages nontrivial": 60766,
      "discuss caveats": 32239,
      "utility data": 130477,
      "possible research": 88426,
      "series case": 107271,
      "scripts publicly": 104271,
      "noncommercial use": 80171,
      "translation user": 125414,
      "words like": 134036,
      "strategy mt": 113530,
      "output mt": 83095,
      "given enormous": 48027,
      "process translating": 91581,
      "large manual": 61135,
      "methods applying": 69316,
      "cascade approach": 15568,
      "iwslt 2020": 57164,
      "2020 paper": 726,
      "cascaded approach": 15571,
      "approach offline": 7754,
      "provided small": 93983,
      "recognition synthesis": 97018,
      "15 languages": 349,
      "sigmorphon 2020": 108673,
      "task competitive": 117990,
      "words exist": 133944,
      "models 12": 72635,
      "present single": 89711,
      "languages code": 60448,
      "language need": 59710,
      "build textbased": 14812,
      "textbased language": 121425,
      "research case": 100438,
      "proposes convolutional": 93597,
      "autoencoder based": 10399,
      "measuring word": 67934,
      "robustness compared": 103091,
      "input address": 55293,
      "models translate": 74245,
      "applications display": 6918,
      "speech high": 111690,
      "methodology evaluate": 69258,
      "traditional cascaded": 122804,
      "procedure able": 91388,
      "directly optimizing": 31893,
      "model starts": 72084,
      "factor consider": 42856,
      "quality research": 94775,
      "currently lacks": 25408,
      "policy automatically": 87948,
      "used iwslt": 128597,
      "released publication": 98492,
      "translation automatically": 124661,
      "fragments sentences": 45378,
      "model making": 71498,
      "segment context": 104560,
      "context contextaware": 22038,
      "time reading": 122083,
      "reading provide": 96030,
      "joint pretraining": 57313,
      "linguistic content": 64452,
      "content make": 21904,
      "boost models": 14434,
      "representations largescale": 99718,
      "novel semisupervised": 80717,
      "text finetuning": 120954,
      "contextual semantic": 22494,
      "knowledge input": 58019,
      "results summary": 102239,
      "mit license": 70357,
      "detailed error": 29771,
      "preprocessing module": 89327,
      "chinese pinyin": 16803,
      "accuracy 89": 2080,
      "accuracy prior": 2247,
      "translation pipelines": 125121,
      "pipelines work": 87569,
      "focus improving": 44774,
      "segmentation crucial": 104580,
      "systems expect": 116887,
      "performance explore": 86364,
      "models intrinsically": 73424,
      "extrinsically downstream": 42629,
      "retrieval clir": 102385,
      "currently popular": 25411,
      "input directly": 55322,
      "relation underlying": 98083,
      "underlying representations": 126697,
      "information despite": 54476,
      "despite information": 29698,
      "translation direct": 124753,
      "speech inputs": 111699,
      "translations available": 125453,
      "data pretraining": 26273,
      "performance mt": 86545,
      "mt used": 75005,
      "used signal": 128760,
      "general multitask": 46681,
      "framework leverage": 45599,
      "text tasks": 121362,
      "sequencetosequence modeling": 107176,
      "elegant solution": 35266,
      "solution applications": 110545,
      "relies availability": 98644,
      "tasks auxiliary": 118949,
      "sequences reduce": 107137,
      "enhance knowledge": 37472,
      "transfer text": 124196,
      "proven powerful": 93727,
      "count based": 24413,
      "framework work": 45741,
      "wfst decoder": 132466,
      "recognition context": 96844,
      "effectively experimental": 34809,
      "meeting transcription": 68259,
      "improvements obtained": 52888,
      "applying contextual": 7234,
      "hand existing": 49318,
      "depend source": 29051,
      "selfsupervised fashion": 104937,
      "signals including": 108702,
      "pretraining pretraining": 90305,
      "improvement 16": 52670,
      "bleu languages": 14288,
      "translation efficient": 124778,
      "build multilingual": 14788,
      "efficient transfer": 35124,
      "pretrained speech": 90191,
      "finetuning achieve": 44443,
      "finetuning 10": 44442,
      "pretrained parameters": 90176,
      "enables effectively": 36385,
      "models low": 73529,
      "bleu average": 14277,
      "manytomany multilingual": 67101,
      "57 bleu": 1126,
      "average 18": 11174,
      "appealing approach": 6791,
      "selfsupervised speech": 104950,
      "work spoken": 134813,
      "predict intents": 88895,
      "speech address": 111647,
      "built pretrained": 14932,
      "pretrained endtoend": 90034,
      "limited target": 64288,
      "study semisupervised": 114511,
      "asr component": 9425,
      "component supervised": 20001,
      "input perform": 55390,
      "using oracle": 129993,
      "training improved": 123655,
      "address weakness": 3777,
      "propose extensions": 92672,
      "techniques allow": 119831,
      "time resulting": 122092,
      "rate improvement": 95794,
      "provides reliable": 94067,
      "regularize decoder": 97764,
      "additional language": 3535,
      "shallow fusion": 107918,
      "set analyses": 107357,
      "analyses different": 5451,
      "types errors": 126279,
      "information suggesting": 55017,
      "translation aims": 124633,
      "endtoend way": 36991,
      "encoderdecoder structure": 36624,
      "improve endtoend": 52381,
      "corresponding text": 24310,
      "tasks trained": 119569,
      "introduce crossmodal": 56402,
      "corpora shown": 23584,
      "translation augmented": 124656,
      "augmented memory": 10328,
      "memory transformer": 68391,
      "transformer transformerbased": 124386,
      "time spent": 122110,
      "latency quality": 61567,
      "scenarios paper": 103859,
      "partial input": 85290,
      "long continuous": 65067,
      "continuous input": 22619,
      "task hybrid": 118261,
      "representations local": 99742,
      "effective variety": 34774,
      "rely autoregressive": 98677,
      "dependencies generating": 29085,
      "time time": 122124,
      "occurs training": 81523,
      "information reliable": 54918,
      "modeling longterm": 72468,
      "achieve train": 2584,
      "rescoring model": 100392,
      "hypotheses better": 51251,
      "semantic consistency": 105016,
      "investigate types": 56821,
      "progress little": 92159,
      "work combined": 134415,
      "monotonic multihead": 74645,
      "attention endtoend": 9828,
      "text insertion": 121059,
      "article introduce": 9137,
      "method fast": 68838,
      "method additionally": 68618,
      "reducing model": 97428,
      "sizes propose": 109960,
      "computation complexity": 20346,
      "complexity proposed": 19933,
      "lightweight endtoend": 64012,
      "recognition compare": 96842,
      "errors observe": 38393,
      "additionally investigate": 3612,
      "investigate behavior": 56722,
      "data conclude": 25768,
      "better prediction": 13673,
      "pilot experiment": 87508,
      "early stages": 34394,
      "data insufficient": 26047,
      "insufficient train": 55726,
      "isolated words": 56976,
      "searching words": 104373,
      "spoken input": 111986,
      "dataset exists": 26905,
      "dataset report": 27156,
      "report standard": 99046,
      "accuracy possible": 2239,
      "gains obtained": 46402,
      "combine output": 18637,
      "zeroshot manner": 135512,
      "plan make": 87629,
      "need access": 76774,
      "neural transducer": 78714,
      "architecture attentionbased": 8621,
      "proposed achieves": 93167,
      "sequence target": 107085,
      "learning capture": 62420,
      "syntactical information": 116502,
      "results textual": 102273,
      "models word2vec": 74353,
      "demonstrated capability": 28910,
      "work kind": 134598,
      "span embeddings": 111122,
      "low zeroresource": 65401,
      "work methods": 134634,
      "spans words": 111213,
      "languages distinct": 60511,
      "tasks finding": 119132,
      "sequences utterances": 107144,
      "utterances training": 130671,
      "encoder produces": 36556,
      "input decoder": 55317,
      "secondary task": 104471,
      "consecutive utterances": 21166,
      "baseline knowledge": 12234,
      "requiring knowledge": 100342,
      "mt mt": 74975,
      "promising techniques": 92309,
      "distillation paper": 32519,
      "terms translation": 120398,
      "northern southern": 80372,
      "extinct language": 42055,
      "analysis quality": 5791,
      "quality useful": 94821,
      "refined human": 97592,
      "human transcriptions": 50976,
      "results related": 102121,
      "data enlarged": 25896,
      "results line": 101890,
      "using perplexity": 130010,
      "words sharing": 134212,
      "including sentence": 53368,
      "application model": 6865,
      "form words": 45150,
      "search optimal": 104332,
      "mapping problem": 67143,
      "various words": 131239,
      "method tackle": 69174,
      "performing automated": 86944,
      "word unknown": 133619,
      "second scenario": 104452,
      "boundary known": 14523,
      "present using": 89761,
      "model exploring": 71145,
      "endtoend spoken": 36970,
      "alexa siri": 4675,
      "siri google": 109873,
      "nlu component": 79807,
      "smaller faster": 110233,
      "dont advantage": 33945,
      "advantage available": 4037,
      "propose e2e": 92638,
      "facebook dataset": 42737,
      "dataset set": 27193,
      "research applying": 100408,
      "recognition various": 97040,
      "usually pretrained": 130445,
      "english verify": 37332,
      "apply pretrained": 7203,
      "achieve 20": 2407,
      "improvements languages": 52866,
      "achieves gain": 2789,
      "units subword": 127266,
      "systems exist": 116883,
      "data includes": 26032,
      "phoneme inventories": 87280,
      "languages includes": 60633,
      "information wikipedia": 55094,
      "set publicly": 107550,
      "involving lowresource": 56916,
      "producing accurate": 92014,
      "models excel": 73187,
      "layers instead": 61782,
      "dedicated model": 28180,
      "direct st": 31803,
      "language intermediate": 59222,
      "compress input": 20294,
      "pairs englishitalian": 83533,
      "reducing memory": 97426,
      "code enabling": 18100,
      "talks languages": 117563,
      "corpus creation": 23737,
      "provide baselines": 93764,
      "training scratch": 123831,
      "different layer": 31221,
      "consisting 24": 21446,
      "assess usefulness": 9495,
      "corpus best": 23682,
      "used building": 128416,
      "using picture": 130016,
      "tasks key": 119233,
      "like deep": 64032,
      "automated systems": 10470,
      "incorporating deep": 53525,
      "accuracy ranged": 2254,
      "created study": 24686,
      "bilingual pretraining": 14053,
      "suffer limitations": 115229,
      "input modality": 55368,
      "exploit various": 41446,
      "personalized content": 87142,
      "problem endtoend": 91028,
      "degradation general": 28551,
      "models dependent": 73039,
      "debiasing approach": 27899,
      "achieve additional": 2415,
      "rnn layers": 102922,
      "quality does": 94633,
      "furthermore models": 46193,
      "trained efficiently": 123124,
      "encoders learn": 36656,
      "models components": 72943,
      "study endtoend": 114368,
      "translation decoding": 124739,
      "proposed decoding": 93254,
      "tradeoff different": 122776,
      "multidialectal corpus": 75065,
      "german dialect": 47885,
      "dialects used": 30541,
      "used verbal": 128846,
      "major dialects": 66553,
      "standard german": 112245,
      "reference goal": 97520,
      "create make": 24622,
      "datadriven nlp": 26674,
      "conducting experiments": 20962,
      "downgrade performance": 33977,
      "employ data": 36255,
      "compare previous": 19283,
      "error distribution": 38304,
      "effectively improves": 34822,
      "models revolutionized": 73960,
      "revolutionized natural": 102666,
      "transcriptions spoken": 124008,
      "advantage robustness": 4060,
      "10 reduction": 146,
      "multiple automatic": 75502,
      "adaptation scenarios": 3260,
      "ctcbased endtoend": 25198,
      "data having": 25998,
      "using backpropagation": 129378,
      "parameters inference": 84755,
      "production setting": 92061,
      "setting explore": 107748,
      "compression methods": 20312,
      "20 model": 514,
      "times smaller": 122188,
      "degradation word": 28555,
      "varieties models": 130957,
      "pretrained multiple": 90168,
      "promise improving": 92258,
      "using allosaurus": 129331,
      "challenging realworld": 16302,
      "recognition datasets": 96851,
      "recognition finetuning": 96882,
      "varying sizes": 131268,
      "flexible solution": 44663,
      "multiple devices": 75536,
      "computation capacities": 20345,
      "utterance apply": 130590,
      "randomly drops": 95527,
      "different depths": 31090,
      "learning reduces": 62950,
      "reduces model": 97389,
      "aims providing": 4566,
      "resource development": 100837,
      "assistive technologies": 9589,
      "aiming improving": 4491,
      "ability communicate": 1496,
      "addition dataset": 3418,
      "developing effective": 30339,
      "knowledge database": 57851,
      "architecture optimization": 8716,
      "block learns": 14361,
      "score propose": 104111,
      "perspective analyzing": 87159,
      "analyzing attention": 6035,
      "48 absolute": 1050,
      "viterbi decoding": 131851,
      "main contributor": 66414,
      "contributor performance": 22798,
      "constructed based": 21654,
      "videos talks": 131665,
      "used corpus": 128460,
      "systems largescale": 116973,
      "quantities unlabeled": 94898,
      "different complementary": 31048,
      "complementary ways": 19755,
      "ways explore": 132158,
      "26 bleu": 839,
      "pairs simple": 83638,
      "single iteration": 109743,
      "speech audio data": 111651,
      "data information extraction": 26041,
      "model performs par": 71741,
      "automatically recognized words": 10823,
      "model achieves comparable": 70563,
      "features highly informative": 43541,
      "present probabilistic model": 89648,
      "using hidden markov": 129735,
      "broadcast news corpus": 14688,
      "evaluation metrics results": 39295,
      "word boundaries based": 132937,
      "produce language models": 91905,
      "propose method integrate": 92772,
      "paper describes test": 83855,
      "continuous speech recognition": 22637,
      "comparable obtained human": 19153,
      "acoustic language models": 3010,
      "corpus language model": 23853,
      "markov models hmm": 67266,
      "classification paper present": 17323,
      "present automated method": 89387,
      "recognition paper present": 96956,
      "relative word error": 98385,
      "non native speakers": 80147,
      "past 60 years": 85632,
      "60 years research": 1149,
      "hope work shall": 50648,
      "markov model hmm": 67261,
      "performance improvement achieved": 86443,
      "successfully applied automatic": 115176,
      "understanding speech data": 126970,
      "recognition error rates": 96870,
      "algorithm word segmentation": 4815,
      "algorithm designed solve": 4716,
      "large scale language": 61250,
      "increasing training data": 53687,
      "availability training data": 10927,
      "used language model": 128604,
      "ngram language modeling": 79437,
      "data model size": 26135,
      "sequence words corresponding": 107101,
      "using equivalence classes": 129642,
      "segmentation paper introduce": 104608,
      "paper introduce method": 83995,
      "previous approaches unsupervised": 90384,
      "issues challenges faced": 57040,
      "paper report initial": 84401,
      "model hmm based": 71289,
      "experimental results performance": 40666,
      "using novel metric": 129983,
      "spoken speech text": 112019,
      "better performance paper": 13661,
      "feature extraction classification": 43277,
      "large number outputs": 61183,
      "neural networks convolutional": 78461,
      "major challenge automatic": 66544,
      "words paper presents": 134104,
      "multilingual speech recognition": 75375,
      "random fields scrfs": 95489,
      "structured support vector": 114046,
      "demonstrate proposed algorithm": 28832,
      "proposed algorithm achieves": 93173,
      "models trained supervised": 74232,
      "queries using multiple": 94944,
      "detection paper presents": 30016,
      "novel approach enhancing": 80477,
      "models domain specific": 73106,
      "spoken language translation": 111994,
      "2014 paper introduce": 579,
      "neural network systems": 78401,
      "techniques machine translation": 119927,
      "gives bleu score": 48188,
      "word embeddings need": 133168,
      "different word types": 31575,
      "adaptation paper presents": 3245,
      "adaptation deep neural": 3215,
      "speech recognition work": 111789,
      "performance propose new": 86626,
      "experiments conducted set": 40845,
      "new approach input": 78788,
      "models outperform baseline": 73674,
      "faster convergence training": 43170,
      "language translation slt": 60187,
      "increasingly globalized world": 53699,
      "translation spoken language": 125286,
      "stateoftheart acoustic models": 112576,
      "factored translation models": 42867,
      "translation models automatic": 124952,
      "english mandarin chinese": 37206,
      "using technique called": 130274,
      "advancing state art": 4032,
      "recognition propose novel": 96974,
      "novel deep neural": 80540,
      "acoustic features input": 3006,
      "unsupervised bayesian model": 127608,
      "features language modeling": 43584,
      "language modeling use": 59481,
      "case study crosslingual": 15616,
      "representation learning introduce": 99295,
      "polyglot language models": 88011,
      "network models trained": 77356,
      "shared feature representations": 107981,
      "extrinsic evaluation downstream": 42616,
      "applications make use": 6965,
      "data available languages": 25675,
      "slightly lower best": 110027,
      "paper proposes efficient": 84354,
      "novel incremental learning": 80599,
      "learning approach unsupervised": 62353,
      "novel unsupervised algorithm": 80764,
      "sequence prediction models": 107031,
      "model trained predict": 72204,
      "networks deep learning": 77561,
      "models neural network": 73633,
      "network models usually": 77359,
      "deep neural networkbased": 28365,
      "speech recognition challenge": 111744,
      "using maximum posteriori": 129868,
      "statistical model trained": 113132,
      "model trained output": 72203,
      "indicate proposed method": 53850,
      "proposed method competitive": 93354,
      "evaluate commonly used": 38808,
      "previously published performance": 90615,
      "use external language": 128045,
      "speech translation using": 111839,
      "paper addresses automatic": 83713,
      "relatively new task": 98408,
      "large feature set": 61090,
      "translation mt quality": 124982,
      "recent work unsupervised": 96594,
      "data preprocessing methods": 26263,
      "speech recognition using": 111786,
      "multitask learning recently": 75863,
      "attentionbased encoderdecoder framework": 10069,
      "presents novel method": 89884,
      "english words based": 37350,
      "report baseline results": 98980,
      "previous work deep": 90525,
      "adding convolutional layers": 3373,
      "using joint training": 129772,
      "error rate reductions": 38332,
      "deep convolutional networks": 28223,
      "sequencetosequence models shown": 107182,
      "models shown success": 74033,
      "recurrent convolutional structures": 97171,
      "method training process": 69195,
      "paper extend existing": 83937,
      "simple effective algorithm": 109403,
      "propose neural architecture": 92816,
      "loss function designed": 65270,
      "task demonstrate effectiveness": 118058,
      "tasks commonly used": 118988,
      "proposed model superior": 93484,
      "tasks speech recognition": 119517,
      "unexplored paper present": 127062,
      "compare performance using": 19279,
      "memory lstm gated": 68330,
      "lstm gated recurrent": 65622,
      "recurrent unit gru": 97265,
      "neural network multiview": 78359,
      "embeddings recent work": 35901,
      "deep bidirectional lstm": 28205,
      "embeddings improve previous": 35749,
      "present results tasks": 89680,
      "maximum likelihood ml": 67539,
      "model proposed method": 71815,
      "endtoend recurrent neural": 36957,
      "model achieve high": 70538,
      "architecture does require": 8647,
      "target language text": 117656,
      "evaluating automatic speech": 39039,
      "systems comparison human": 116798,
      "target languages work": 117671,
      "process automatically identifying": 91428,
      "acoustic model training": 3017,
      "acoustic feature representations": 3003,
      "representations significantly improved": 99892,
      "discriminant analysis lda": 32191,
      "downstream speech applications": 34028,
      "automatic quality estimation": 10633,
      "paper propose exploit": 84236,
      "data sets released": 26433,
      "method significantly better": 69134,
      "significantly better strong": 108892,
      "aim paper validate": 4468,
      "low medium high": 65374,
      "case study using": 15626,
      "network rnn encoder": 77417,
      "used feature extraction": 128550,
      "recognition deep learning": 96853,
      "achieved significant accuracy": 2694,
      "ctc loss function": 25194,
      "data sets compared": 26420,
      "used neural network": 128657,
      "data sets respectively": 26434,
      "network model learning": 77333,
      "trained proposed method": 123249,
      "speech recognition error": 111748,
      "like speech recognition": 64097,
      "process model trained": 91529,
      "model trained proposed": 72205,
      "use automatic speech": 127909,
      "recognition asr produce": 96815,
      "perform supervised classification": 86082,
      "paper investigate alternative": 84013,
      "based framework learning": 11727,
      "provides competitive performance": 94013,
      "propose methods based": 92779,
      "generative adversarial networks": 47713,
      "speak different languages": 111278,
      "neural network works": 78419,
      "examples results suggest": 39873,
      "yields higher accuracy": 135414,
      "structured prediction algorithm": 114025,
      "finitestate transducer fst": 44549,
      "representation models used": 99343,
      "cosine similarity embeddings": 24340,
      "reduces time consumption": 97401,
      "use knowledge graph": 128106,
      "language model decoder": 59309,
      "monotonic attention mechanism": 74644,
      "speech language processing": 111703,
      "recently encoderdecoder neural": 96685,
      "shown impressive performance": 108483,
      "mechanism allows model": 67945,
      "alignments source target": 5020,
      "explored experimental results": 41626,
      "attention achieve significant": 9792,
      "improvements stateoftheart approaches": 52914,
      "largevocabulary continuous speech": 61545,
      "speech recognition lvcsr": 111758,
      "acoustic language model": 3009,
      "rate wer reductions": 95809,
      "paper discusses current": 83877,
      "using heterogeneous corpora": 129730,
      "summer camp applied": 115658,
      "camp applied language": 15085,
      "applied language exploration": 7082,
      "preliminary experiments reveal": 89270,
      "specifically pitch accents": 111574,
      "pitch accents phrase": 87574,
      "feature representations word": 43304,
      "setup proposed method": 107855,
      "yields strong results": 135444,
      "endtoend automatic speech": 36877,
      "attentionbased encoderdecoder network": 10072,
      "network cnn based": 77189,
      "combine ctc predictions": 18621,
      "language model achieve": 59286,
      "tasks proposed framework": 119414,
      "distance cosine distance": 32409,
      "achieve substantial improvements": 2578,
      "performance runtime efficiency": 86691,
      "neural networks ability": 78434,
      "convolutional network convnet": 23163,
      "95 classification accuracy": 1421,
      "efficient neural network": 35100,
      "advanced state art": 3957,
      "network models require": 77354,
      "models require significant": 73935,
      "require significant amounts": 100199,
      "significant amounts memory": 108724,
      "network based speech": 77166,
      "baseline present data": 12287,
      "neural networks report": 78554,
      "cross entropy ce": 24860,
      "combination data augmentation": 18558,
      "learning framework framework": 62601,
      "results evaluated using": 101769,
      "features work proposes": 43793,
      "work proposes novel": 134755,
      "asr systems proposed": 9447,
      "systems proposed approach": 117085,
      "data handcrafted rules": 25996,
      "beam search finally": 12604,
      "sequencetosequence models used": 107185,
      "models used translate": 74293,
      "human performance task": 50930,
      "performance task using": 86786,
      "standard speech transcription": 112307,
      "high degree overlap": 50058,
      "aim paper investigate": 4467,
      "models trained subtitles": 74231,
      "data size corpus": 26459,
      "relatively large corpus": 98396,
      "corpus spoken language": 24015,
      "model language modeling": 71409,
      "compare various types": 19314,
      "resource language speech": 100857,
      "speech language technologies": 111704,
      "paper deep neural": 83797,
      "internal feature representations": 56166,
      "endtoend trained sequencetosequence": 36984,
      "trained sequencetosequence model": 123275,
      "outofvocabulary words order": 82684,
      "explore application deep": 41518,
      "residual network resnet": 100742,
      "support development future": 115964,
      "temporal classification connectionist": 120094,
      "classification connectionist temporal": 17155,
      "speech recognition typically": 111785,
      "terms training time": 120397,
      "decoding time model": 28142,
      "paper designed novel": 83862,
      "freely available online": 45785,
      "available online learning": 11061,
      "word embeddings speech": 133210,
      "proposed model based": 93435,
      "used word similarity": 128852,
      "achieved competitive results": 2612,
      "model used initialize": 72256,
      "achieving high accuracy": 2952,
      "task proposed method": 118580,
      "compared randomlyinitialized baseline": 19429,
      "corpora human annotation": 23494,
      "based connectionist temporal": 11596,
      "mean opinion score": 67600,
      "opinion score mos": 82098,
      "using characters words": 129439,
      "training data evaluate": 123457,
      "language model able": 59284,
      "able achieve state": 1601,
      "densely connected convolutional": 29026,
      "networks recently introduced": 77724,
      "training corpus improved": 123402,
      "direct speech translation": 31802,
      "translation using source": 125420,
      "using source language": 130200,
      "source parallel corpora": 110803,
      "paper tries gap": 84480,
      "correlated human judgments": 24207,
      "available training time": 11136,
      "endtoend speech translation": 36969,
      "applications paper introduce": 6984,
      "support natural language": 115980,
      "active research field": 3122,
      "intermediate representations provide": 56141,
      "provide useful information": 93948,
      "models improves performance": 73372,
      "leads better performance": 61922,
      "using attention information": 129353,
      "extremely lowresource languages": 42604,
      "universal phone recognition": 127320,
      "using small amounts": 130187,
      "darpa lorelei program": 25545,
      "evaluating asr performance": 39037,
      "recognition paper investigate": 96955,
      "reduce model complexity": 97340,
      "unified model previous": 127119,
      "model previous works": 71787,
      "data usually required": 26613,
      "benefits proposed method": 13020,
      "context vectors using": 22322,
      "improve speech recognition": 52543,
      "using canonical correlation": 129427,
      "correlation analysis cca": 24223,
      "difficult port new": 31654,
      "available speech corpora": 11110,
      "applications lowresource languages": 6960,
      "available lowresource languages": 11038,
      "data models trained": 26141,
      "models recently shown": 73887,
      "quadratically sequence length": 94546,
      "sequence length address": 107004,
      "strong baseline based": 113627,
      "vector representations using": 131360,
      "processing toolkit paper": 91843,
      "new open source": 79065,
      "model size reduced": 72057,
      "identification paper describes": 51416,
      "english speakers using": 37289,
      "features ngram features": 43629,
      "easily adapted language": 34432,
      "provide useful insights": 93949,
      "gradient descent sgd": 48601,
      "obtain competitive performance": 81268,
      "recognition paper presents": 96957,
      "new network architecture": 79049,
      "multihead attention model": 75126,
      "japanese experimental results": 57189,
      "train acoustic models": 122899,
      "endtoend asr systems": 36875,
      "embeddings neural networkbased": 35831,
      "include word embeddings": 53242,
      "results word embeddings": 102342,
      "model neural networks": 71577,
      "baseline results tasks": 12303,
      "word embeddings temporal": 133222,
      "embeddings experiments proposed": 35711,
      "regularization techniques dropout": 97758,
      "recognition task recent": 97026,
      "task recent years": 118611,
      "training convolutional neural": 123398,
      "large unlabelled data": 61318,
      "area roc curve": 8884,
      "random forest model": 95494,
      "spoken document retrieval": 111983,
      "provide large improvements": 93865,
      "speech recognition improved": 111756,
      "speech recognition specifically": 111774,
      "test sets training": 120533,
      "high quality automatic": 50111,
      "model achieves word": 70616,
      "achieves word error": 2922,
      "included training data": 53249,
      "sequence hidden states": 106959,
      "demonstrates proposed method": 28956,
      "speech work investigate": 111851,
      "stateoftheart systems considerably": 112988,
      "neural network fcnn": 78300,
      "stateoftheart performance large": 112826,
      "compared stateoftheart deep": 19456,
      "showed proposed approaches": 108387,
      "consistent significant improvement": 21402,
      "tasks compared previous": 118993,
      "language models best": 59519,
      "weighted finitestate transducer": 132353,
      "finitestate transducer wfst": 44550,
      "classifiers deep learning": 17606,
      "paper presents latest": 84175,
      "results data sets": 101654,
      "results wall street": 102323,
      "outperform models trained": 82717,
      "learned representations deep": 62251,
      "prediction model paper": 89080,
      "speaker adaptation performance": 111282,
      "performance automatic speech": 86159,
      "mismatch training testing": 70308,
      "training testing conditions": 123917,
      "various stateoftheart techniques": 131206,
      "data address problem": 25582,
      "simple approach improve": 109363,
      "languages logographic origin": 60704,
      "languages like spanish": 60696,
      "propose multimodal approach": 92801,
      "related languages proposed": 97878,
      "features help improve": 43535,
      "model capable learning": 70803,
      "efficiency training inference": 35050,
      "transformer network neural": 124353,
      "network neural machine": 77372,
      "help multihead selfattention": 49756,
      "directly selfattention mechanism": 31905,
      "results experiments conducted": 101781,
      "experiments conducted test": 40846,
      "stateoftheart performance outperforms": 112837,
      "multitask transfer learning": 75898,
      "italian german english": 57094,
      "paper describes memad": 83827,
      "describes memad project": 29411,
      "memad project entry": 68270,
      "nmt models systems": 79921,
      "systems experiments indicate": 116893,
      "task adversarial training": 117854,
      "novel adversarial training": 80456,
      "adversarial training approach": 4170,
      "unpaired text data": 127487,
      "approach endtoend training": 7528,
      "absolute word error": 1750,
      "faster decoding speed": 43172,
      "especially natural language": 38485,
      "applications work develop": 7041,
      "training robust models": 123825,
      "models generalize multiple": 73277,
      "data set including": 26409,
      "speech previous work": 111730,
      "significantly improving performance": 108963,
      "human computer interaction": 50781,
      "interaction paper propose": 55959,
      "propose novel word": 92978,
      "tasks like machine": 119256,
      "work present case": 134676,
      "word similarity evaluations": 133562,
      "using principal component": 130055,
      "component analysis embedding": 19966,
      "asr error correction": 9429,
      "training approach improves": 123362,
      "multimodal asr systems": 75418,
      "stateoftheart convolutional neural": 112621,
      "speech recognition output": 111764,
      "knowledge distillation model": 57870,
      "ground truth labels": 49099,
      "effective speech recognition": 34750,
      "analyze representations learned": 6000,
      "network use learned": 77466,
      "based deep cnn": 11627,
      "test set best": 120498,
      "endtoend e2e models": 36896,
      "proposed approach outperform": 93202,
      "beam search decoding": 12603,
      "detection using deep": 30095,
      "neural networks require": 78558,
      "work presents novel": 134698,
      "novel approach leverage": 80482,
      "comprehensive way proposed": 20284,
      "word boundary information": 132939,
      "classification propose neural": 17354,
      "10 absolute accuracy": 121,
      "main contributions work": 66413,
      "making better use": 66828,
      "second contribution method": 104398,
      "score significantly outperforms": 104123,
      "speech recognition current": 111746,
      "recognition current stateoftheart": 96850,
      "paper present alternative": 84078,
      "matches current stateoftheart": 67379,
      "training endtoend neural": 123609,
      "endtoend neural networks": 36948,
      "computational resources training": 20425,
      "annotated speech data": 6238,
      "issues exist current": 57048,
      "learning mtl approaches": 62807,
      "auxiliary task improve": 10888,
      "language models second": 59655,
      "models conduct experiments": 72957,
      "models produce diverse": 73810,
      "beam search decoder": 12602,
      "training time inference": 123925,
      "wordlevel language models": 133742,
      "research efforts shown": 100485,
      "useful low resource": 128904,
      "resource languages paper": 100861,
      "times fewer parameters": 122179,
      "state transducer wfst": 112534,
      "simplicity training efficiency": 109574,
      "softmax output layer": 110510,
      "different projection matrices": 31363,
      "data data augmentation": 25816,
      "resources publicly available": 101033,
      "publicly available hope": 94305,
      "available hope used": 11008,
      "hope used future": 50644,
      "word embeddings improved": 133131,
      "word embedding techniques": 133037,
      "word embeddings test": 133223,
      "embeddings encode words": 35689,
      "style experimental results": 114575,
      "reduces number parameters": 97391,
      "best previously reported": 13420,
      "oov words paper": 81871,
      "models lm perform": 73516,
      "domains performance highly": 33831,
      "speech synthesis models": 111813,
      "multilayer convolutional neural": 75170,
      "classification task experiments": 17431,
      "data available approach": 25672,
      "machine translation corpora": 65932,
      "models attention mechanisms": 72784,
      "training data effectively": 123456,
      "speech translation st": 111836,
      "translates source language": 124583,
      "model transferring knowledge": 72229,
      "model knowledge distillation": 71396,
      "35 bleu points": 961,
      "cycle consistency loss": 25483,
      "language translation applications": 60186,
      "information retrieval paper": 54944,
      "target domain target": 117610,
      "significantly outperforms random": 109027,
      "english finnish french": 37138,
      "achieves promising translation": 2839,
      "outstanding results reach": 83206,
      "results reach human": 102112,
      "reach human parity": 95895,
      "parsing paper propose": 85185,
      "based transformer models": 12131,
      "improving prediction accuracy": 53143,
      "noisy text language": 80126,
      "achieve higher accuracy": 2480,
      "approach paper presents": 7786,
      "language modeling neural": 59447,
      "modeling neural network": 72494,
      "lstm language modeling": 65632,
      "better performance conventional": 13656,
      "oov words words": 81875,
      "models lm trained": 73517,
      "used convolutional neural": 128456,
      "english translation based": 37315,
      "50k parallel sentences": 1090,
      "parallel translation data": 84699,
      "speech translation neural": 111833,
      "language modeling framework": 59432,
      "recent developments deep": 96452,
      "developments deep learning": 30441,
      "datasets machine translation": 27554,
      "parallel corpus containing": 84615,
      "publicly available github": 94304,
      "assessment second language": 9527,
      "second language proficiency": 104422,
      "content multiple languages": 21911,
      "different design choices": 31092,
      "set userdefined keywords": 107629,
      "approach takes advantage": 7958,
      "recent studies introduced": 96523,
      "human speech processing": 50967,
      "processing lexical access": 91696,
      "extended new languages": 41808,
      "sequencetosequence s2s models": 107201,
      "explicit intermediate representations": 41327,
      "models fall short": 73236,
      "problem data scarcity": 90993,
      "method using pretrained": 69212,
      "training data boosts": 123437,
      "data boosts performance": 25708,
      "main language model": 66436,
      "model decoding process": 70955,
      "target language previous": 117649,
      "multitask learning improves": 75843,
      "source language translation": 110786,
      "machine learning classifier": 65787,
      "phonetic string similarity": 87298,
      "scarcity training data": 103815,
      "training data investigate": 123480,
      "experiments limited data": 40989,
      "possible research directions": 88427,
      "series case studies": 107272,
      "scripts publicly available": 104272,
      "publicly available noncommercial": 94313,
      "available noncommercial use": 11059,
      "manual effort required": 66990,
      "speech recognition synthesis": 111780,
      "sigmorphon 2020 shared": 108674,
      "machine translation multilingual": 66075,
      "proposed model able": 93423,
      "translation text speech": 125377,
      "simultaneous translation models": 109666,
      "provide better understanding": 93771,
      "address issue model": 3691,
      "time reading provide": 122084,
      "propose novel semisupervised": 92965,
      "latent space using": 61610,
      "detailed error analysis": 29772,
      "using unsupervised methods": 130346,
      "work focus improving": 134532,
      "evaluate models intrinsically": 38870,
      "crosslingual information retrieval": 24961,
      "information retrieval clir": 54930,
      "general multitask learning": 46682,
      "relies availability large": 98645,
      "text data asr": 120849,
      "task compared baseline": 117986,
      "previous studies focus": 90494,
      "effectively experimental results": 34810,
      "performance improvements obtained": 86452,
      "incorporating contextual information": 53524,
      "conduct experiments different": 20851,
      "improvement 11 bleu": 52665,
      "improvement 16 bleu": 52671,
      "achieves similar accuracy": 2867,
      "efficient finetuning pretrained": 35077,
      "pretrained models present": 90146,
      "efficient transfer learning": 35125,
      "transfer learning pretrained": 124122,
      "large pretrained models": 61222,
      "selfsupervised language models": 104940,
      "models bert finetuned": 72836,
      "training improved neural": 123656,
      "error rate improvement": 38329,
      "strong baselines furthermore": 113646,
      "model multitask training": 71548,
      "additional language model": 3536,
      "different types errors": 31528,
      "better semantic representation": 13718,
      "transformer transformerbased models": 124387,
      "time time step": 122125,
      "acoustic linguistic features": 3013,
      "monotonic multihead attention": 74646,
      "uses recurrent neural": 129270,
      "network predict word": 77391,
      "models high error": 73326,
      "effect data augmentation": 34590,
      "new dataset using": 78870,
      "publicly available resources": 94323,
      "model using augmented": 72274,
      "representation learning paper": 99304,
      "network architecture uses": 77147,
      "unlike existing work": 127435,
      "representation learning capture": 99283,
      "model able predict": 70521,
      "able achieve competitive": 1594,
      "models word2vec fasttext": 74354,
      "word similarity benchmark": 133556,
      "best knowledge existing": 13348,
      "results task using": 102254,
      "providing additional supervision": 94101,
      "knowledge distillation paper": 57871,
      "terms translation quality": 120399,
      "languages study presents": 60903,
      "onetomany mapping problem": 81728,
      "endtoend spoken language": 36971,
      "siri google assistant": 109874,
      "understanding nlu component": 126907,
      "speech text input": 111825,
      "evaluate ability model": 38782,
      "lowresource languages domains": 65513,
      "effectively applied downstream": 34788,
      "resulting data set": 101440,
      "set publicly available": 107551,
      "used variety applications": 128839,
      "involving lowresource languages": 56917,
      "memory lstm layers": 68334,
      "representations work propose": 99999,
      "st models particular": 112103,
      "language pairs englishitalian": 59754,
      "speech recognition translation": 111784,
      "methodology applied languages": 69246,
      "translation performance lowresource": 125115,
      "paper investigate effect": 84015,
      "significantly improves results": 108955,
      "like deep learning": 64033,
      "deep learning research": 28311,
      "significantly better baseline": 108883,
      "tasks existing methods": 119114,
      "address problems propose": 3752,
      "data speech recognition": 26493,
      "representation learning framework": 99293,
      "experiments translation directions": 41187,
      "substantially improve translation": 114895,
      "remains challenging problem": 98791,
      "following previous work": 44978,
      "previous work use": 90561,
      "pair encoding bpe": 83428,
      "achieve comparable results": 2443,
      "use nlp methods": 128176,
      "parallel corpus spoken": 84626,
      "create make available": 24623,
      "improve robustness nlp": 52530,
      "employ data augmentation": 36256,
      "methods solve problem": 69766,
      "previous methods approach": 90426,
      "language models masked": 59598,
      "models masked language": 73554,
      "models revolutionized natural": 73961,
      "revolutionized natural language": 102667,
      "novel approach takes": 80490,
      "language proposed approach": 60001,
      "asr systems trained": 9448,
      "model compression methods": 70873,
      "degradation word error": 28556,
      "particularly lowresource languages": 85488,
      "encoders different depths": 36644,
      "model experiment results": 71126,
      "dataset paper introduces": 27087,
      "systems perform poorly": 117059,
      "translation tasks used": 125367,
      "speech translation paper": 111834,
      "translation paper improve": 125088,
      "large quantities unlabeled": 61231,
      "language pairs simple": 59780,
      "language model different": 59313,
      "code models publicly": 18121,
      "model achieves comparable performance": 70564,
      "using hidden markov models": 129736,
      "speech recognition speech recognition": 111776,
      "hidden markov models hmm": 49902,
      "speech recognition paper present": 111767,
      "relative word error rate": 98386,
      "past 60 years research": 85633,
      "hidden markov model hmm": 49897,
      "large scale language modeling": 61251,
      "training data model size": 123495,
      "markov model hmm based": 67262,
      "recurrent neural networks convolutional": 97229,
      "conditional random fields scrfs": 20787,
      "respectively experimental results demonstrate": 101138,
      "best previously published results": 13419,
      "automatic speech recognition work": 10669,
      "propose neural network models": 92827,
      "spoken language translation slt": 111995,
      "novel deep neural network": 80541,
      "models recurrent neural network": 73894,
      "neural network models trained": 78354,
      "models neural network models": 73634,
      "neural network models usually": 78357,
      "using maximum posteriori map": 129869,
      "paper proposes new approach": 84361,
      "results indicate proposed method": 101859,
      "paper presents novel method": 84193,
      "convolutional recurrent neural network": 23214,
      "recurrent neural network using": 97224,
      "term memory lstm gated": 120230,
      "memory lstm gated recurrent": 68331,
      "lstm gated recurrent unit": 65623,
      "gated recurrent unit gru": 46516,
      "word embeddings recent work": 133197,
      "performance proposed method compared": 86632,
      "endtoend recurrent neural network": 36958,
      "recurrent neural network trained": 97221,
      "evaluating automatic speech recognition": 39040,
      "linear discriminant analysis lda": 64347,
      "features deep learning architectures": 43441,
      "neural network rnn encoder": 78389,
      "widely used neural network": 132583,
      "neural network model learning": 78334,
      "training neural network model": 123740,
      "speech recognition error rates": 111749,
      "use automatic speech recognition": 127910,
      "speech recognition asr produce": 111741,
      "mechanism allows model learn": 67946,
      "propose novel attention mechanism": 92903,
      "largevocabulary continuous speech recognition": 61546,
      "continuous speech recognition lvcsr": 22638,
      "error rate wer reductions": 38337,
      "summer camp applied language": 115659,
      "camp applied language exploration": 15086,
      "endtoend automatic speech recognition": 36878,
      "neural network cnn based": 78280,
      "neural network models require": 78352,
      "require significant amounts memory": 100200,
      "neural network based speech": 78263,
      "paper deep neural network": 83798,
      "endtoend trained sequencetosequence model": 36985,
      "connectionist temporal classification connectionist": 21141,
      "temporal classification connectionist temporal": 120095,
      "classification connectionist temporal classification": 17156,
      "freely available online learning": 45786,
      "based connectionist temporal classification": 11597,
      "mean opinion score mos": 67601,
      "models paper proposes novel": 73708,
      "language model able achieve": 59285,
      "able achieve state art": 1602,
      "support natural language processing": 115981,
      "using language model lm": 129782,
      "using canonical correlation analysis": 129428,
      "canonical correlation analysis cca": 15154,
      "models recently shown promising": 73888,
      "quadratically sequence length address": 94547,
      "nonnative english speakers using": 80243,
      "stochastic gradient descent sgd": 113357,
      "word embeddings experiments proposed": 133110,
      "training convolutional neural network": 123399,
      "deep neural networks different": 28370,
      "model achieves word error": 70617,
      "achieves word error rate": 2923,
      "compared stateoftheart deep learning": 19457,
      "stateoftheart deep learning based": 112633,
      "weighted finitestate transducer wfst": 132354,
      "results wall street journal": 102324,
      "performance automatic speech recognition": 86160,
      "mismatch training testing conditions": 70309,
      "transformer network neural machine": 124354,
      "network neural machine translation": 77373,
      "achieves stateoftheart performance outperforms": 2888,
      "task paper describes memad": 118501,
      "paper describes memad project": 83828,
      "describes memad project entry": 29412,
      "absolute word error rate": 1751,
      "especially natural language processing": 38486,
      "tasks like machine translation": 119257,
      "work present case study": 134677,
      "principal component analysis embedding": 90672,
      "stateoftheart convolutional neural networks": 112622,
      "deep neural networks require": 28378,
      "work presents novel approach": 134699,
      "classification propose neural network": 17355,
      "paper present alternative approach": 84079,
      "multitask learning mtl approaches": 75854,
      "recent research efforts shown": 96508,
      "low resource languages paper": 65390,
      "resource languages paper propose": 100862,
      "weighted finite state transducer": 132350,
      "finite state transducer wfst": 44541,
      "publicly available hope used": 94306,
      "available hope used future": 11009,
      "using word embeddings test": 130381,
      "reduces number parameters model": 97392,
      "outofvocabulary oov words paper": 82676,
      "language models lm perform": 59590,
      "trained large amounts unlabeled": 123176,
      "shown outstanding results reach": 108504,
      "outstanding results reach human": 83207,
      "results reach human parity": 102113,
      "proposed approach outperforms existing": 93206,
      "approach outperforms existing methods": 7771,
      "language modeling neural network": 59448,
      "language models lm trained": 59591,
      "used convolutional neural networks": 128457,
      "recent developments deep learning": 96453,
      "spoken language understanding intent": 111997,
      "easily extended new languages": 34451,
      "method using pretrained language": 69213,
      "training data boosts performance": 123438,
      "publicly available noncommercial use": 94314,
      "sigmorphon 2020 shared task": 108675,
      "neural machine translation multilingual": 78036,
      "language models paper proposes": 59615,
      "analysis paper propose novel": 5741,
      "paper propose novel semisupervised": 84309,
      "crosslingual information retrieval clir": 24962,
      "relies availability large amounts": 98646,
      "availability large amounts training": 10917,
      "task machine translation task": 118368,
      "word error rate improvement": 133269,
      "neural network predict word": 78374,
      "neural network architecture uses": 78249,
      "proposed model significantly improves": 93481,
      "endtoend spoken language understanding": 36972,
      "language understanding nlu component": 60244,
      "training data work propose": 123567,
      "substantially improve translation quality": 114896,
      "byte pair encoding bpe": 14999,
      "pretrained language model generate": 90052,
      "language models masked language": 59599,
      "models masked language models": 73555,
      "models revolutionized natural language": 73962,
      "revolutionized natural language processing": 102668,
      "work propose novel approach": 134734,
      "propose novel approach takes": 92899,
      "based hidden markov models hmm": 11755,
      "hidden markov model hmm based": 49898,
      "short term memory lstm gated": 108251,
      "term memory lstm gated recurrent": 120231,
      "memory lstm gated recurrent unit": 68332,
      "lstm gated recurrent unit gru": 65624,
      "evaluating automatic speech recognition asr": 39041,
      "recurrent neural network rnn encoder": 97216,
      "use automatic speech recognition asr": 127911,
      "automatic speech recognition asr produce": 10661,
      "largevocabulary continuous speech recognition lvcsr": 61547,
      "word error rate wer reductions": 133273,
      "summer camp applied language exploration": 115660,
      "connectionist temporal classification connectionist temporal": 21142,
      "temporal classification connectionist temporal classification": 120096,
      "classification connectionist temporal classification ctc": 17157,
      "using canonical correlation analysis cca": 129429,
      "models recently shown promising results": 73889,
      "relative word error rate reduction": 98387,
      "experimental results demonstrate proposed method": 40601,
      "results demonstrate proposed method outperforms": 101702,
      "model achieves word error rate": 70618,
      "transformer network neural machine translation": 124355,
      "model achieves stateoftheart performance outperforms": 70602,
      "translation task paper describes memad": 125339,
      "task paper describes memad project": 118502,
      "paper describes memad project entry": 83829,
      "absolute word error rate wer": 1752,
      "paper present alternative approach based": 84080,
      "low resource languages paper propose": 65391,
      "weighted finite state transducer wfst": 132351,
      "publicly available hope used future": 94307,
      "performance automatic speech recognition asr": 86161,
      "shown outstanding results reach human": 108505,
      "outstanding results reach human parity": 83208,
      "used convolutional neural networks cnns": 128458,
      "method using pretrained language models": 69214,
      "availability large amounts training data": 10918,
      "natural language understanding nlu component": 76570,
      "language models masked language models": 59600,
      "catching": 15716,
      "paragraphstyle": 84581,
      "condensed": 20749,
      "reflections": 97622,
      "doors": 33952,
      "duc": 34252,
      "summarizers": 115619,
      "documentsummary": 33319,
      "adulteration": 3927,
      "anaphoricity": 6076,
      "timescale": 122192,
      "compositionalities": 20139,
      "ridge": 102823,
      "weed": 132313,
      "inaccurately": 53194,
      "discourages": 32036,
      "encodeattenddecode": 36467,
      "additions": 3638,
      "debatepedia": 27891,
      "universitys": 127356,
      "batches": 12559,
      "ebm": 34497,
      "commenced": 18795,
      "condensing": 20750,
      "absorption": 1758,
      "walks": 131999,
      "contingency": 22594,
      "nice": 79474,
      "illiterate": 51732,
      "resourcebased": 100889,
      "aesop": 4206,
      "informations": 55117,
      "dailymail": 25519,
      "charge": 16637,
      "newsroom": 79415,
      "textrank": 121446,
      "mirrors": 70285,
      "bullet": 14952,
      "graphtograph": 48981,
      "imperfections": 51923,
      "intact": 55732,
      "miss": 70315,
      "summarizations": 115595,
      "programmingbased": 92134,
      "sentenceword": 106553,
      "entice": 37699,
      "transmitting": 125532,
      "verbose": 131549,
      "retention": 102365,
      "polishing": 87987,
      "permissive": 87046,
      "wellperforming": 132427,
      "proliferating": 92245,
      "catering": 15800,
      "marred": 67273,
      "slurs": 110126,
      "promotions": 92323,
      "sentenceparagraph": 106187,
      "slows": 110105,
      "sinusoidal": 109868,
      "11x": 244,
      "measurably": 67775,
      "singletons": 109851,
      "domainadaptive": 33704,
      "superfluous": 115672,
      "blogging": 14371,
      "bwe": 14987,
      "nonmatching": 80238,
      "multicriteria": 75061,
      "270000": 856,
      "2500": 822,
      "rlbased": 102896,
      "diversification": 32861,
      "top5": 122485,
      "residing": 100736,
      "rethink": 102366,
      "modela": 72344,
      "summaryworthy": 115654,
      "rst": 103336,
      "hospitals": 50659,
      "wikiref": 132693,
      "280": 866,
      "000": 2,
      "postags": 88461,
      "selfcritical": 104917,
      "prescriptions": 89344,
      "patents": 85667,
      "ac": 1849,
      "nyt": 81039,
      "855": 1325,
      "793": 1273,
      "791": 1272,
      "enabler": 36371,
      "humaninterpretable": 51034,
      "fullytrained": 45991,
      "creativity": 24734,
      "uniformity": 127144,
      "renowned": 98894,
      "corrector": 24188,
      "exceeded": 39903,
      "emit": 36079,
      "wellunderstood": 132444,
      "wellgrounded": 132414,
      "instantiations": 55649,
      "intermediary": 56129,
      "regulated": 97776,
      "tailoring": 117502,
      "mimiccxr": 70134,
      "semanticdriven": 105390,
      "highscoring": 50436,
      "medicines": 68239,
      "mediqa": 68242,
      "garners": 46495,
      "1800": 423,
      "absorb": 1754,
      "celebrated": 15881,
      "copynet": 23301,
      "wrappers": 135079,
      "reformat": 97632,
      "rcts": 95880,
      "enlist": 37552,
      "synopses": 116356,
      "propositionlevel": 93632,
      "rougebased": 103305,
      "globalaware": 48281,
      "attentionaware": 10056,
      "incomparable": 53415,
      "pitfall": 87575,
      "rotowire": 103284,
      "editorial": 34561,
      "foreground": 45039,
      "outlines": 82638,
      "soliciting": 110539,
      "attaches": 9693,
      "delimiter": 28611,
      "selfattentions": 104906,
      "zhao": 135541,
      "topicguided": 122608,
      "distorting": 32592,
      "cas": 15566,
      "metaevaluation": 68537,
      "niche": 79475,
      "predecessor": 88817,
      "intersentential": 56311,
      "nonredundancy": 80284,
      "fight": 44010,
      "singlehead": 109827,
      "edu": 34566,
      "debaters": 27892,
      "vectorization": 131403,
      "predefining": 88839,
      "questionguided": 95263,
      "uncontrolled": 126645,
      "inflate": 54276,
      "irrelevance": 56944,
      "noncontextualized": 80183,
      "lewis": 63718,
      "taskadaptive": 118868,
      "scorers": 104146,
      "constrains": 21577,
      "ideological": 51645,
      "selectors": 104858,
      "strides": 113591,
      "redefine": 97305,
      "scholar": 103958,
      "contradictions": 22668,
      "textinfilling": 121439,
      "rststyle": 103338,
      "summarization techniques": 115580,
      "evaluation component": 39154,
      "tasks linear": 119269,
      "linear segmentation": 64370,
      "characteristics corpora": 16521,
      "corpora methods": 23528,
      "collection user": 18497,
      "applying natural": 7259,
      "particular document": 85405,
      "task generation": 118242,
      "guidelines corpus": 49262,
      "features implemented": 43549,
      "content planner": 21921,
      "title document": 122221,
      "document conventional": 32982,
      "focuses finding": 44899,
      "pays little": 85816,
      "readers make": 95960,
      "effects content": 34976,
      "common different": 18866,
      "different tendencies": 31483,
      "texts specific": 121618,
      "topics texts": 122658,
      "address order": 3724,
      "models unannotated": 74261,
      "incorporating content": 53521,
      "methods summarizing": 69783,
      "developing automatic": 30335,
      "method compile": 68717,
      "corpus web": 24065,
      "propose summarization": 93095,
      "summarization method": 115532,
      "different viewpoints": 31558,
      "task content": 118018,
      "events despite": 39574,
      "al 2007": 4599,
      "methodology automatic": 69247,
      "summarization sentence": 115564,
      "humans produce": 51103,
      "produce summaries": 91940,
      "possible explanation": 88401,
      "developed extract": 30275,
      "corpora believe": 23427,
      "progress automatic": 92140,
      "development sophisticated": 30426,
      "open doors": 81903,
      "rich data": 102738,
      "algorithms learn": 4863,
      "ability humans": 1528,
      "experiments creating": 40860,
      "automatically model": 10804,
      "model alignment": 70654,
      "extension standard": 41839,
      "word phraselevel": 133407,
      "model document": 71016,
      "overall discourse": 83226,
      "syntactic discourse": 116396,
      "summarization introduce": 115522,
      "graphbased method": 48900,
      "sentences document": 106285,
      "set documents": 107422,
      "consider new": 21218,
      "matrix based": 67472,
      "similarity used": 109327,
      "ranked place": 95648,
      "place task": 87609,
      "duc 2004": 34254,
      "discuss methods": 32268,
      "methods compute": 69389,
      "cases furthermore": 15646,
      "clustering documents": 17943,
      "text book": 120689,
      "challenges seen": 16210,
      "alignment problems": 4982,
      "implicit sensitive": 52018,
      "human capabilities": 50773,
      "generated summary": 47131,
      "conveyed text": 23124,
      "text author": 120673,
      "summary text": 115648,
      "suffer information": 115224,
      "study introduces": 114412,
      "approach reports": 7856,
      "reports new": 99084,
      "techniques extracting": 119887,
      "new summarization": 79189,
      "arabic documents": 8511,
      "documents performed": 33266,
      "experiments applied": 40774,
      "based summarization": 12085,
      "summarization methods": 115533,
      "semantics document": 105411,
      "generation intuitive": 47447,
      "intuitive humans": 56668,
      "humans present": 51101,
      "semantic levels": 105098,
      "levels use": 63567,
      "summarization algorithm": 115477,
      "document graph": 33019,
      "summarize documents": 115603,
      "informative contents": 55132,
      "matching process": 67425,
      "based summary": 12087,
      "lemma form": 63327,
      "evaluation technique": 39418,
      "parameter setting": 84720,
      "highquality summaries": 50406,
      "good overall": 48488,
      "performance produce": 86623,
      "produced different": 91954,
      "different summarization": 31464,
      "discriminative reranking": 32216,
      "useful features": 128886,
      "wordlevel features": 133735,
      "features sentencelevel": 43718,
      "sentencelevel features": 106161,
      "approach abstractive": 7292,
      "encoderdecoder recurrent": 36619,
      "architecture modeling": 8695,
      "words rare": 134154,
      "neural summarization": 78691,
      "words traditional": 134273,
      "sentence features": 105873,
      "develop general": 30199,
      "singledocument summarization": 109824,
      "composed hierarchical": 20066,
      "hierarchical document": 49956,
      "encoder attentionbased": 36495,
      "corpora containing": 23443,
      "containing hundreds": 21792,
      "thousands documentsummary": 121920,
      "documentsummary pairs": 33320,
      "improving information": 53105,
      "accuracy domains": 2141,
      "new sources": 79143,
      "collected approach": 18402,
      "learns select": 63234,
      "select optimal": 104705,
      "optimize reward": 82226,
      "food adulteration": 44999,
      "cases demonstrate": 15640,
      "constraints model": 21604,
      "based rich": 11997,
      "content sentence": 21943,
      "sentence deletion": 105818,
      "pronouns antecedent": 92348,
      "evaluation scientific": 39378,
      "measure similarities": 67824,
      "summaries set": 115457,
      "goldstandard summaries": 48455,
      "different general": 31161,
      "multisentence compression": 75780,
      "microblog summarization": 70047,
      "summarization opinion": 115543,
      "related sentences": 97896,
      "informative grammatical": 55141,
      "time approach": 121984,
      "mapping strategy": 67150,
      "based synonymy": 12100,
      "step identify": 113270,
      "identify best": 51480,
      "summarization framework": 115508,
      "summaries shows": 115458,
      "centrality measures": 15908,
      "rouge score": 103299,
      "content present": 21924,
      "information coverage": 54460,
      "approaches automatic": 8072,
      "task conducted": 118004,
      "generate summaries": 47022,
      "high overall": 50092,
      "quality information": 94690,
      "information diversity": 54503,
      "extraction summarization": 42499,
      "texts controlled": 121484,
      "cases like": 15655,
      "generation story": 47636,
      "amr graphs": 5390,
      "compression techniques": 20326,
      "summarization present": 115548,
      "compression models": 20314,
      "framework integrate": 45578,
      "integrate various": 55769,
      "linguistic motivation": 64512,
      "process deriving": 91456,
      "achieves statistically": 2907,
      "2006 2007": 553,
      "summarization work": 115593,
      "model gated": 71221,
      "implemented encoderdecoder": 51960,
      "encoderdecoder setting": 36622,
      "setting better": 107736,
      "texts proposed": 121585,
      "improve ability": 52330,
      "summarization including": 115519,
      "using set": 130165,
      "consider topical": 21235,
      "documents various": 33312,
      "topics documents": 122625,
      "documents topic": 33302,
      "baselines topic": 12482,
      "stateofart performance": 112556,
      "multidocument abstractive": 75074,
      "informative summaries": 55156,
      "aim developing": 4453,
      "abstractive summarizer": 1819,
      "approach identifies": 7612,
      "important document": 52150,
      "document multidocument": 33041,
      "document aligned": 32948,
      "second generate": 104410,
      "structure finally": 113867,
      "select sentences": 104710,
      "maximizing information": 67521,
      "path information": 85676,
      "quality score": 94784,
      "scores method": 104185,
      "controlling output": 22857,
      "length neural": 63370,
      "success sequence": 115129,
      "crucial applications": 25130,
      "concise summaries": 20703,
      "methods controlling": 69407,
      "learningbased methods": 63180,
      "aims enhance": 4523,
      "summarization aims": 115475,
      "aims model": 4553,
      "model documents": 71018,
      "specific regions": 111483,
      "content input": 21893,
      "content document": 21876,
      "engineering features": 37049,
      "documents long": 33249,
      "approaches suffer": 8360,
      "encoders compute": 36638,
      "problem unknown": 91278,
      "decoding times": 28143,
      "times paper": 122185,
      "mechanism able": 67940,
      "abstract features": 1763,
      "sentencelevel extractive": 106158,
      "conditions architecture": 20815,
      "architecture superior": 8754,
      "ridge regression": 102824,
      "ranking second": 95686,
      "similarity previously": 109286,
      "content important": 21891,
      "important spoken": 52255,
      "original content": 82507,
      "content deep": 21867,
      "error model": 38309,
      "systems news": 117023,
      "response analysis": 101189,
      "specifically define": 111536,
      "context media": 22174,
      "analyze limitations": 5983,
      "limitations future": 64176,
      "ilp formulation": 51756,
      "summaries terms": 115463,
      "set reference": 107559,
      "summaries contain": 115432,
      "corpora demonstrated": 23451,
      "summaries significantly": 115459,
      "simple general": 109439,
      "neural decoder": 77884,
      "generate outputs": 46984,
      "specific properties": 111481,
      "length model": 63369,
      "token generation": 122254,
      "able incorporate": 1656,
      "sources given": 110892,
      "improvements abstractive": 52799,
      "summarization machine": 115527,
      "news training": 79384,
      "summarization approach": 115478,
      "beginning document": 12635,
      "news able": 79282,
      "data growing": 25991,
      "need automatic": 76784,
      "emerged important": 36051,
      "work text": 134851,
      "extraction literature": 42376,
      "domains evaluation": 33775,
      "finally discusses": 44173,
      "discusses briefly": 32305,
      "issues research": 57076,
      "sequencetosequence attentional": 107155,
      "ability produce": 1563,
      "produce novel": 91914,
      "task outperforming": 118491,
      "corpus concept": 23711,
      "structure large": 113892,
      "study variant": 114545,
      "produces summaries": 92012,
      "currently missing": 25410,
      "present newly": 89594,
      "novel crowdsourcing": 80522,
      "elements large": 35280,
      "enable research": 36364,
      "sentences need": 106407,
      "newswire articles": 79418,
      "propose explore": 92669,
      "explore information": 41552,
      "information terms": 55036,
      "summarization propose": 115552,
      "encoding model": 36713,
      "extend sequencetosequence": 41796,
      "framework abstractive": 45405,
      "decoder built": 27998,
      "second level": 104425,
      "driven attention": 34150,
      "summarization abstractive": 115472,
      "generate shorter": 47015,
      "version document": 131605,
      "salient points": 103554,
      "encodeattenddecode paradigm": 36468,
      "systems suffers": 117172,
      "learns focus": 63210,
      "different portions": 31338,
      "diversity based": 32868,
      "aims alleviate": 4499,
      "new querybased": 79094,
      "deep reinforced": 28408,
      "rnnbased encoderdecoder": 102943,
      "models abstractive": 72649,
      "documents summaries": 33298,
      "continuously generated": 22651,
      "rl models": 102892,
      "training standard": 123880,
      "summaries readable": 115454,
      "using abstract": 129314,
      "representation increasing": 99259,
      "generation remains": 47594,
      "pipeline proposed": 87553,
      "generates amr": 47156,
      "input story": 55445,
      "extracts summary": 42581,
      "finally generate": 44189,
      "compared text": 19475,
      "focused use": 44883,
      "runs submitted": 103475,
      "approaches regression": 8314,
      "results challenge": 101567,
      "goal improve": 48357,
      "high semantic": 50138,
      "text represented": 121255,
      "representation produced": 99385,
      "representations keyphrases": 99709,
      "keyphrases furthermore": 57625,
      "methods bring": 69355,
      "improvements state": 52911,
      "coherence automated": 18292,
      "evidencebased medicine": 39680,
      "practice evidencebased": 88731,
      "medicine ebm": 68237,
      "latest research": 61658,
      "various medical": 131130,
      "large medical": 61147,
      "texts development": 121496,
      "techniques task": 119995,
      "serve resource": 107298,
      "volume text": 131942,
      "needs effectively": 76891,
      "useful review": 128927,
      "effect neural": 34603,
      "make initial": 66689,
      "information transferred": 55055,
      "opinion articles": 82082,
      "indicate neural": 53845,
      "combination indomain": 18564,
      "select salient": 104709,
      "style target": 114596,
      "domain extractive": 33534,
      "typical task": 126405,
      "task devoted": 118085,
      "identification informative": 51391,
      "use complex": 127945,
      "nodes represent": 80038,
      "based number": 11903,
      "number shared": 80966,
      "differently previous": 31596,
      "networks including": 77624,
      "symmetry accessibility": 116329,
      "time evaluation": 122025,
      "including based": 53264,
      "random walks": 95520,
      "use discourse": 128004,
      "relation using": 98084,
      "textual genres": 121710,
      "summarization content": 115487,
      "traditional news": 122854,
      "simple greedy": 109441,
      "greedy algorithm": 49068,
      "selecting small": 104760,
      "summary experiments": 115639,
      "summarization spoken": 115569,
      "content based": 21862,
      "powerful approach": 88664,
      "approach sequencetosequence": 7884,
      "include adding": 53219,
      "methods term": 69796,
      "corpora approach": 23411,
      "simplification text": 109595,
      "simplify text": 109615,
      "children nonnative": 16726,
      "brief summary": 14616,
      "simplification aims": 109576,
      "reduce linguistic": 97334,
      "retain original": 102359,
      "original meaning": 82530,
      "recently approaches": 96658,
      "simplified texts": 109606,
      "simplification introduce": 109584,
      "summarization important": 115518,
      "important provide": 52219,
      "provide main": 93870,
      "ideas document": 51348,
      "reconstruction framework": 97118,
      "reconstruction error": 97117,
      "gains compared": 46389,
      "merely relies": 68475,
      "fairly evaluate": 42996,
      "lexical variations": 63840,
      "issue adopt": 56988,
      "rouge capture": 103290,
      "used summaries": 128792,
      "rouge correlate": 103291,
      "summarization learning": 115525,
      "weight distribution": 132325,
      "distribution recently": 32677,
      "noise words": 80074,
      "distribution model": 32667,
      "predict sentence": 88930,
      "way encourage": 132073,
      "information produced": 54875,
      "aims finding": 4530,
      "methods adopted": 69299,
      "task limited": 118355,
      "studies evaluate": 114218,
      "method select": 69115,
      "mds task": 67592,
      "task adopted": 117851,
      "different documents": 31103,
      "proof principle": 92372,
      "methods related": 69715,
      "related textual": 97906,
      "textual models": 121721,
      "learning mechanism": 62714,
      "algorithm learn": 4764,
      "network simple": 77434,
      "features resulting": 43707,
      "obtain encouraging": 81280,
      "entities user": 37891,
      "effective mechanism": 34702,
      "final summaries": 44130,
      "better suit": 13732,
      "suit needs": 115388,
      "input set": 55435,
      "set control": 107405,
      "abstractive systems": 1822,
      "generating summaries": 47268,
      "respect query": 101101,
      "pointergenerator model": 87831,
      "dataset generated": 26953,
      "evaluated measuring": 38987,
      "similarity reference": 109292,
      "summaries results": 115455,
      "similar existing": 109091,
      "summary length": 115645,
      "text suffers": 121341,
      "led stateoftheart": 63287,
      "task abstractive": 117827,
      "trained end": 123127,
      "end method": 36817,
      "address aspects": 3648,
      "incorporate query": 53482,
      "existing abstractive": 40042,
      "setting design": 107740,
      "adapt trained": 3196,
      "text specific": 121320,
      "100 words": 171,
      "words aim": 133812,
      "words design": 133908,
      "adapt target": 3194,
      "target size": 117709,
      "extractive baselines": 42550,
      "model importance": 71303,
      "mainly driven": 66482,
      "notion information": 80408,
      "models importance": 73361,
      "advance understanding": 3935,
      "additionally provide": 3623,
      "summarizing long": 115628,
      "summarization source": 115566,
      "source documents": 110746,
      "documents use": 33307,
      "longer typical": 65166,
      "coherent multisentence": 18321,
      "information reflected": 54908,
      "new phrases": 79076,
      "text recently": 121230,
      "large parts": 61204,
      "text output": 121158,
      "address drawback": 3677,
      "extracted input": 42159,
      "text secondly": 121274,
      "globally optimizes": 48291,
      "objective use": 81125,
      "cnn dailymail": 18002,
      "extractive abstractive": 42546,
      "breaking news": 14579,
      "way collecting": 132064,
      "automatic live": 10580,
      "using wellknown": 130371,
      "corpus encourage": 23770,
      "summarization automatic": 115481,
      "challenging research": 16305,
      "character contains": 16422,
      "existing chinese": 40086,
      "adopt totally": 3898,
      "characterbased wordbased": 16508,
      "representations accurately": 99487,
      "accurately capture": 2376,
      "capture essence": 15298,
      "wordbased characterbased": 133697,
      "characterbased representations": 16499,
      "generates stateoftheart": 47176,
      "text create": 120840,
      "duc 2002": 34253,
      "fixed sized": 44607,
      "generates summary": 47180,
      "baseline abstractive": 12177,
      "points original": 87867,
      "architecture address": 8611,
      "text divided": 120880,
      "divided multiple": 32909,
      "encoder multiple": 36544,
      "automated process": 10460,
      "stateoftheart neuralbased": 112787,
      "addition discuss": 3423,
      "related techniques": 97904,
      "present promising": 89651,
      "relatively short": 98411,
      "documents research": 33285,
      "models datadriven": 73013,
      "applied short": 7116,
      "articles new": 9201,
      "title abstract": 122220,
      "range existing": 95567,
      "methods hardly": 69527,
      "hardly applied": 49492,
      "extend neural": 41789,
      "million articles": 70094,
      "written authors": 135114,
      "styles particular": 114621,
      "abstractive extractive": 1801,
      "extraction strategies": 42495,
      "quantify diversity": 94847,
      "summarization highly": 115515,
      "highly difficult": 50314,
      "difficult problem": 31655,
      "semantically unrelated": 105383,
      "respect source": 101102,
      "propose regularization": 93013,
      "regularization approach": 97745,
      "objective alleviate": 81065,
      "problem addition": 90920,
      "problem existing": 91040,
      "improves semantic": 53042,
      "consistency terms": 21368,
      "report series": 99041,
      "text fail": 120947,
      "fail detect": 42955,
      "weights different": 132375,
      "benchmarks learning": 12913,
      "pretrained vectors": 90210,
      "used tfidf": 128812,
      "similarities semantic": 109187,
      "glove vectors": 48303,
      "algorithms analysis": 4828,
      "proposed tested": 93569,
      "tested possible": 120582,
      "possible improvements": 88410,
      "scores used": 104216,
      "model seq2seq": 71984,
      "content social": 21949,
      "media long": 68125,
      "short written": 108268,
      "shares meaning": 108125,
      "meaning source": 67695,
      "media dataset": 68102,
      "performances benchmark": 86871,
      "inconsistency loss": 53439,
      "loss propose": 65294,
      "paragraph model": 84569,
      "attention words": 10042,
      "sentences likely": 106382,
      "structured selfattentive": 114041,
      "effectiveness representation": 34941,
      "representation traditional": 99443,
      "documents end": 33224,
      "problem incorporating": 91082,
      "knowledge document": 57876,
      "structured selfattention": 114039,
      "embeddings architecture": 35557,
      "enables obtain": 36396,
      "mechanism provides": 68034,
      "wellknown datasets": 132419,
      "summarization online": 115542,
      "online forum": 81773,
      "studies examined": 114221,
      "examined task": 39764,
      "make attempt": 66622,
      "mechanisms build": 68056,
      "investigate new": 56789,
      "training paradigm": 123773,
      "summarization traditionally": 115585,
      "convert human": 23095,
      "content useful": 21969,
      "learning explore": 62569,
      "concise fluent": 20701,
      "fluent informative": 44711,
      "summaries experiments": 115438,
      "standard summarization": 112312,
      "generated daily": 47066,
      "discussion forums": 32321,
      "manner work": 66964,
      "based integer": 11775,
      "feedback corpus": 43823,
      "outperforms range": 82971,
      "range baselines": 95546,
      "timely manner": 122163,
      "responses work": 101303,
      "new methodology": 79015,
      "informative wellformed": 55160,
      "metric work": 69913,
      "points present": 87870,
      "framework draws": 45505,
      "framework source": 45692,
      "set amr": 107355,
      "amr parser": 5391,
      "commonsense representation": 19010,
      "proportion text": 92513,
      "entities original": 37832,
      "entities build": 37752,
      "information linked": 54733,
      "base based": 11445,
      "leverage offtheshelf": 63609,
      "list entities": 64689,
      "representation topic": 99442,
      "current available": 25263,
      "encoding entities": 36692,
      "summarization generating": 115512,
      "generating abstract": 47195,
      "thoroughly investigated": 121904,
      "amr semantic": 5395,
      "language grounded": 59112,
      "realization step": 96117,
      "optimized independently": 82233,
      "independently using": 53791,
      "benchmark summarization": 12861,
      "produced promising": 91964,
      "relations source": 98260,
      "sentence summary": 106095,
      "approach naturally": 7732,
      "naturally combines": 76640,
      "combines source": 18697,
      "source dependency": 110743,
      "effectiveness incorporating": 34896,
      "incorporating sourceside": 53557,
      "sourceside syntactic": 110942,
      "approach compares": 7433,
      "favorably stateoftheart": 43218,
      "modeling copying": 72408,
      "copying source": 23297,
      "named sequential": 76132,
      "explicitly select": 41392,
      "mechanism generation": 67990,
      "learning score": 62985,
      "selection main": 104797,
      "encoder obtain": 36552,
      "obtain representation": 81314,
      "output summary": 83129,
      "given previously": 48089,
      "set products": 107545,
      "faced task": 42750,
      "product domain": 92035,
      "common product": 18912,
      "features set": 43725,
      "features impact": 43548,
      "based seq2seq": 12022,
      "similar humans": 109095,
      "idea propose": 51335,
      "propose seq2seq": 93040,
      "human extracted": 50859,
      "technique extracting": 119787,
      "dataset overall": 27083,
      "make different": 66658,
      "summarize content": 115598,
      "examples include": 39847,
      "different news": 31298,
      "paper overcome": 84063,
      "linear programmingbased": 64363,
      "framework incorporates": 45572,
      "lowrank approximation": 65484,
      "framework effective": 45511,
      "generation provide": 47575,
      "systems process": 117078,
      "readers read": 95962,
      "news items": 79349,
      "compile novel": 19714,
      "architectures task": 8849,
      "approach arabic": 7363,
      "information internet": 54699,
      "makes research": 66803,
      "work try": 134864,
      "schemes calculating": 103948,
      "calculating input": 15029,
      "input matrix": 55361,
      "order syntactic": 82415,
      "relations addition": 98105,
      "informative diverse": 55136,
      "evaluate new": 38876,
      "data consortium": 25783,
      "consortium ldc": 21513,
      "ldc arabic": 61836,
      "2015 dataset": 587,
      "performs comprehensively": 86997,
      "comprehensively better": 20286,
      "methods abstractive": 69275,
      "exploiting various": 41487,
      "extracting important": 42213,
      "seq2seq framework": 106893,
      "generated models": 47094,
      "single documents": 109728,
      "acquired web": 3040,
      "summarization data": 115490,
      "maximal marginal": 67500,
      "marginal relevance": 67211,
      "select representative": 104708,
      "model fuse": 71213,
      "abstractive summary": 1821,
      "data compares": 25756,
      "human assessors": 50755,
      "readable form": 95951,
      "contains important": 21823,
      "facts original": 42920,
      "appear source": 6803,
      "document remains": 33062,
      "parts source": 85584,
      "optimized directly": 82232,
      "encourage generation": 36757,
      "topicaware convolutional": 122603,
      "does favor": 33350,
      "news summary": 79374,
      "realworld largescale": 96170,
      "articles topics": 9222,
      "topics based": 122615,
      "networks demonstrate": 77562,
      "oracle extractive": 82255,
      "stateoftheart abstractive": 112568,
      "progress neural": 92166,
      "models challenged": 72894,
      "challenged lack": 16122,
      "document summary": 33090,
      "generation stage": 47634,
      "guide using": 49251,
      "amr parses": 5392,
      "obtained offtheshelf": 81386,
      "offtheshelf parser": 81628,
      "base sur": 11482,
      "informative content": 55131,
      "reference based": 97508,
      "set evaluations": 107435,
      "quality transcriptions": 94816,
      "sentences remove": 106474,
      "remove need": 98875,
      "paired corpora": 83470,
      "meaning despite": 67630,
      "evaluation grammatical": 39229,
      "meaning semantic": 67685,
      "extractive summary": 42570,
      "weighted word": 132360,
      "word graph": 133305,
      "graph semantic": 48866,
      "pagerank algorithm": 83410,
      "rouge measures": 103295,
      "known learn": 58309,
      "encode information": 36433,
      "decoder produce": 28047,
      "higher rouge": 50204,
      "task approached": 117888,
      "framework deep": 45482,
      "trained global": 123154,
      "implemented neural": 51969,
      "features encoding": 43481,
      "bidirectional beam": 13901,
      "models rnn": 73964,
      "makes models": 66796,
      "tackle aforementioned": 117283,
      "aforementioned issues": 4268,
      "bidirectional encoderdecoder": 13919,
      "architecture encoder": 8653,
      "forward decoder": 45325,
      "encoder backward": 36496,
      "backward decoder": 11349,
      "result experimental": 101371,
      "document summarize": 33089,
      "sentences include": 106351,
      "gradient reinforcement": 48604,
      "approaches addition": 8048,
      "better competing": 13545,
      "sentences appear": 106210,
      "iterative text": 57136,
      "supervised extractive": 115753,
      "article multiple": 9141,
      "order fully": 82326,
      "document model": 33037,
      "machines humans": 66335,
      "points neural": 87866,
      "addressed problem": 3787,
      "did explicitly": 30914,
      "account information": 2027,
      "analyzing information": 6041,
      "information structures": 55010,
      "summaries experimental": 115436,
      "results information": 101863,
      "representation latent": 99276,
      "representation easily": 99217,
      "interpreted humans": 56282,
      "generator input": 47773,
      "input generator": 55346,
      "humanwritten sentences": 51131,
      "sentences taking": 106517,
      "summary input": 115642,
      "dataset indonesian": 26981,
      "indonesian text": 53998,
      "considered challenging": 21287,
      "community challenges": 19068,
      "difficult construct": 31616,
      "notably dataset": 80380,
      "obtained encouraging": 81365,
      "research code": 100443,
      "robust neural": 103051,
      "systems evaluation": 116880,
      "introduced measure": 56582,
      "yields significantly": 135431,
      "temporally sensitive": 120134,
      "summaries compared": 115431,
      "number corpus": 80861,
      "models clear": 72908,
      "features inference": 43564,
      "guarantees scalability": 49220,
      "growing importance": 49173,
      "importance automatic": 52045,
      "translation rapidly": 125194,
      "encoderdecoder networks": 36608,
      "techniques abstractive": 119825,
      "improving state": 53166,
      "generating abstractive": 47196,
      "attention variants": 10034,
      "analyzed context": 6021,
      "based intuition": 11780,
      "generates summaries": 47179,
      "information languages": 54713,
      "languages generate": 60606,
      "evaluation showed": 39395,
      "better stable": 13724,
      "scores languages": 104183,
      "experiments deep": 40872,
      "medical articles": 68195,
      "articles order": 9204,
      "understand content": 126742,
      "selection performed": 104813,
      "performance simpler": 86717,
      "easier create": 34411,
      "domains massive": 33812,
      "massive datasets": 67328,
      "suggest important": 115313,
      "new forms": 78922,
      "needed better": 76871,
      "reddit posts": 97302,
      "forum reddit": 45309,
      "datasets use": 27774,
      "use formal": 128057,
      "documents source": 33294,
      "suffer biases": 115207,
      "sentences usually": 106539,
      "inside text": 55517,
      "memory store": 68384,
      "store information": 113381,
      "dataset highly": 26964,
      "highly abstractive": 50286,
      "salience score": 103543,
      "contrastive training": 22742,
      "salient sentences": 103555,
      "100 training": 169,
      "answering forums": 6631,
      "rich knowledge": 102753,
      "considerably large": 21274,
      "dataset filtering": 26931,
      "build word": 14816,
      "generate dataset": 46922,
      "techniques evaluate": 119882,
      "understand cases": 126739,
      "cases existing": 15643,
      "built entirely": 14920,
      "factual news": 42937,
      "instances introduce": 55628,
      "selfattention model": 104890,
      "employs variational": 36335,
      "vector random": 131338,
      "multimodal attention": 75419,
      "method comes": 68705,
      "framework ignores": 45562,
      "rich internal": 102751,
      "summary using": 115651,
      "produced far": 91955,
      "better content": 13548,
      "predicting long": 88992,
      "summarization baselines": 115484,
      "significant margins": 108826,
      "standards language": 112352,
      "pretraining hierarchical": 90264,
      "longdistance dependencies": 65149,
      "document segmentation": 33071,
      "segmentation sentiment": 104632,
      "analysis effective": 5586,
      "effective usage": 34770,
      "large context": 61059,
      "representations unlabeled": 99952,
      "contextual token": 22503,
      "representations include": 99686,
      "information entire": 54534,
      "summarization demonstrate": 115492,
      "identifies possible": 51470,
      "supervision experimental": 115885,
      "paper employs": 83892,
      "rnnbased classifier": 102942,
      "classifier improve": 17550,
      "75 respectively": 1251,
      "based document": 11660,
      "document class": 32959,
      "context increases": 22137,
      "strategies integrate": 113471,
      "slows inference": 110106,
      "art text": 9097,
      "text version": 121406,
      "results available": 101534,
      "headline corpus": 49595,
      "corpus analyzing": 23659,
      "articles multiple": 9200,
      "upper bounds": 127820,
      "task headline": 118252,
      "article previous": 9149,
      "lengths paper": 63388,
      "confirm validity": 21020,
      "professional editors": 92077,
      "amounts information": 5341,
      "saving time": 103668,
      "time help": 122034,
      "particular automatic": 85384,
      "field known": 43960,
      "different works": 31578,
      "works automatic": 134925,
      "especially recent": 38500,
      "ones present": 81702,
      "present problems": 89650,
      "challenges related": 16201,
      "languages challenges": 60435,
      "summarization questionanswering": 115559,
      "people track": 85898,
      "track salient": 122735,
      "automatically aligning": 10716,
      "guide supervised": 49246,
      "answer important": 6537,
      "informative fluent": 55140,
      "results compare": 101596,
      "strong summarization": 113717,
      "baselines evaluated": 12390,
      "short natural": 108222,
      "descriptions source": 29493,
      "lack suitable": 58752,
      "lack community": 58689,
      "dataset design": 26864,
      "results release": 102123,
      "sentence method": 105948,
      "data guide": 25993,
      "tasks real": 119436,
      "sinusoidal positional": 109869,
      "studies learn": 114247,
      "improve rouge": 52532,
      "method constructs": 68738,
      "model largely": 71416,
      "acts regularizer": 3145,
      "score new": 104098,
      "cost slightly": 24373,
      "publications propose": 94283,
      "produce structured": 91939,
      "combined different": 18652,
      "sample efficient": 103561,
      "unclear best": 126632,
      "efficiency sequencetosequence": 35046,
      "work experimented": 134504,
      "network transformer": 77462,
      "encodes source": 36678,
      "parameters network": 84773,
      "states pretrained": 113053,
      "improves pretrained": 53027,
      "instance achieves": 55592,
      "pretrained encoderdecoder": 90031,
      "source descriptions": 110744,
      "shortage training": 108271,
      "handle paper": 49400,
      "form semantic": 45121,
      "measure performs": 67812,
      "writing summary": 135102,
      "content sentences": 21944,
      "remain poorly": 98772,
      "sentence fusion": 105883,
      "attempts bridge": 9763,
      "framework attempts": 45431,
      "attempts model": 9774,
      "report findings": 99002,
      "learning transformer": 63125,
      "efficiently adapting": 35138,
      "datasets achieving": 27297,
      "finally improvements": 44196,
      "language complicated": 58900,
      "legal documents": 63314,
      "verify quality": 131592,
      "experiments unsupervised": 41189,
      "unsupervised extractive": 127636,
      "annotation method": 6336,
      "coherent concise": 18314,
      "description papers": 29471,
      "created manually": 24674,
      "referenceless evaluation": 97554,
      "approaches manual": 8230,
      "evaluating multiple": 39078,
      "multiple systems": 75718,
      "received lot": 96368,
      "live news": 64850,
      "models aid": 72735,
      "providing suggestions": 94140,
      "dataset abstractive": 26716,
      "model thanks": 72168,
      "dataset additionally": 26726,
      "automatic nlp": 10618,
      "applied existing": 7073,
      "rounds propose": 103313,
      "query user": 94993,
      "efficiently search": 35157,
      "summary results": 115646,
      "easily provide": 34464,
      "generation given": 47424,
      "overwhelming number": 83386,
      "compare baselines": 19227,
      "based correlations": 11615,
      "judgments propose": 57449,
      "given automatic": 47989,
      "generation sentence": 47611,
      "unsupervised summarization": 127727,
      "popularity neural": 88150,
      "languages revisit": 60853,
      "popular graphbased": 88093,
      "capture sentential": 15389,
      "document experimental": 33004,
      "baselines wide": 12490,
      "tasks learn": 119248,
      "learn capture": 61998,
      "capture documentlevel": 15294,
      "fashion experiments": 43117,
      "furthermore pretraining": 46204,
      "approaches create": 8111,
      "single sequence": 109798,
      "existing sequential": 40284,
      "domains automatic": 33734,
      "coverage automatic": 24498,
      "automatically assess": 10721,
      "summaries evaluate": 115435,
      "rely exact": 98691,
      "exact word": 39722,
      "match task": 67370,
      "understanding factual": 126843,
      "factual consistency": 42929,
      "current limitations": 25294,
      "limitations models": 64181,
      "transformerbased framework": 124410,
      "framework generator": 45555,
      "different aspect": 31010,
      "input context": 55311,
      "coherence compared": 18294,
      "based known": 11789,
      "known data": 58295,
      "consider original": 21219,
      "original features": 82524,
      "proposed improved": 93313,
      "combining topic": 18755,
      "original feature": 82523,
      "uses semantic": 129275,
      "quality topic": 94812,
      "better document": 13567,
      "performance single": 86719,
      "learning analyzing": 62332,
      "experiments effect": 40924,
      "seen remarkable": 104540,
      "clear understanding": 17732,
      "improved paper": 52624,
      "seek better": 104515,
      "understand neural": 126763,
      "benefit different": 12963,
      "transferable knowledge": 124221,
      "driving research": 34164,
      "research task": 100635,
      "realworld large": 96169,
      "method postprocessing": 69057,
      "correct information": 24104,
      "information redundancy": 54905,
      "generated noisy": 47101,
      "information summary": 55019,
      "cases parallel": 15662,
      "selects salient": 104864,
      "paraphrases extracted": 84840,
      "extracted sentences": 42180,
      "baselines novel": 12435,
      "journal article": 57406,
      "gives overview": 48202,
      "overview recent": 83371,
      "summarization different": 115497,
      "challenges introduced": 16169,
      "challenge address": 15989,
      "efforts current": 35193,
      "current trends": 25382,
      "incorporate domain": 53465,
      "methods especially": 69478,
      "significant manual": 108823,
      "demonstrate contextualized": 28694,
      "representations extracted": 99651,
      "pretrained deep": 90021,
      "content results": 21937,
      "methods source": 69767,
      "efficiently generate": 35144,
      "task overcome": 118496,
      "report generation": 99003,
      "20 original": 516,
      "documents best": 33194,
      "unsupervised text": 127736,
      "model backtranslation": 70720,
      "backtranslation based": 11333,
      "transfer work": 124211,
      "task perform": 118527,
      "perform backtranslation": 85950,
      "methods outperforms": 69660,
      "summarization fully": 115509,
      "performance recent": 86657,
      "progress benchmark": 92141,
      "critically evaluate": 24845,
      "models highlight": 73332,
      "automatically collected": 10738,
      "contain noise": 21750,
      "important characteristics": 52121,
      "characteristics factual": 16524,
      "models overfit": 73694,
      "understanding scientific": 126949,
      "given information": 48042,
      "summarization module": 115541,
      "module aims": 74478,
      "attention domain": 9824,
      "nature training": 76672,
      "sets shows": 107710,
      "gap different": 46450,
      "influences performance": 54325,
      "shift problem": 108179,
      "methods multidomain": 69627,
      "developments neural": 30447,
      "improvements systems": 52927,
      "remains largely": 98805,
      "largely unexplored": 61348,
      "known bias": 58288,
      "bias news": 13828,
      "importance diversity": 52055,
      "diversity conduct": 32871,
      "analysis biases": 5519,
      "domain different": 33507,
      "meeting minutes": 68254,
      "shows different": 108570,
      "cls task": 17905,
      "task produce": 118568,
      "task steps": 118743,
      "summarization translation": 115586,
      "problem error": 91033,
      "framework refer": 45671,
      "incorporating related": 53554,
      "tasks monolingual": 119307,
      "learning lack": 62669,
      "roundtrip translation": 103315,
      "translation strategy": 125297,
      "strategy acquire": 113503,
      "highquality largescale": 50390,
      "englishtochinese chinesetoenglish": 37437,
      "sets addition": 107645,
      "summaries make": 115449,
      "challenge experiment": 16033,
      "use classification": 127937,
      "approaches reinforcement": 8315,
      "addition conduct": 3411,
      "rouge metrics": 103297,
      "outperforming simpler": 82827,
      "text better": 120687,
      "ratings previous": 95824,
      "approaches human": 8180,
      "function source": 46043,
      "module called": 74481,
      "uses mixture": 129245,
      "mixture experts": 70419,
      "binary masks": 14131,
      "standard encoderdecoder": 112234,
      "given selected": 48112,
      "content source": 21951,
      "nondifferentiable nature": 80193,
      "discrete sampling": 32175,
      "accuracy diversity": 2139,
      "top5 accuracy": 122486,
      "summaries long": 115447,
      "words neural": 134076,
      "extractive step": 42561,
      "tasked generating": 118877,
      "bottleneck principle": 14498,
      "summary information": 115641,
      "mapping information": 67132,
      "modelling objective": 72611,
      "sentence best": 105777,
      "sentence iterative": 105910,
      "iterative algorithm": 57125,
      "sentence conditioned": 105802,
      "supervision approach": 115866,
      "building unsupervised": 14897,
      "method empirical": 68796,
      "outperforms unsupervised": 83041,
      "metrics addition": 69921,
      "including human": 53299,
      "evaluation multiple": 39303,
      "combining global": 18727,
      "incorporating global": 53540,
      "current topic": 25380,
      "papers pubmed": 84511,
      "meteor scores": 68573,
      "benefits method": 13012,
      "patterns learning": 85756,
      "prototype editing": 93684,
      "court judgments": 24472,
      "pairs utilized": 83662,
      "summaries main": 115448,
      "avoid copying": 11231,
      "challenges design": 16146,
      "extract facts": 42082,
      "document editing": 32996,
      "address second": 3762,
      "second challenge": 104396,
      "used estimate": 128522,
      "generator extensive": 47771,
      "control neural": 22814,
      "generation valuable": 47696,
      "sentences specific": 106502,
      "sentence corresponding": 105815,
      "control model": 22813,
      "rl methods": 102891,
      "additionally models": 3615,
      "proposed obtain": 93514,
      "relevant textual": 98595,
      "high volume": 50147,
      "provide details": 93806,
      "produces high": 91995,
      "algorithm alternates": 4693,
      "need complex": 76787,
      "scores manual": 104184,
      "multistage pretraining": 75800,
      "tend achieve": 120139,
      "multiple pretraining": 75645,
      "tasks generic": 119145,
      "various seq2seq": 131192,
      "translation applied": 124642,
      "attention distribution": 9821,
      "distribution introduce": 32660,
      "convolutional seq2seq": 23216,
      "considering quality": 21330,
      "end architecture": 36795,
      "data bias": 25704,
      "paper stock": 84438,
      "datasets influence": 27527,
      "influence generalization": 54311,
      "build connection": 14757,
      "structure design": 113833,
      "dataset example": 26902,
      "understanding characteristics": 126812,
      "analyzing sentence": 6050,
      "resulted higher": 101428,
      "understanding systems": 126981,
      "information taken": 55027,
      "abstractive summarizers": 1820,
      "sentences formed": 106324,
      "news scientific": 79361,
      "properties dataset": 92443,
      "features finally": 43517,
      "important content": 52133,
      "particularly desirable": 85476,
      "sentences better": 106229,
      "potentially lead": 88615,
      "selecting important": 104750,
      "reduces task": 97399,
      "new benchmarks": 78815,
      "baselines generating": 12404,
      "level concept": 63429,
      "network abstractive": 77121,
      "new conceptual": 78837,
      "express concrete": 41698,
      "inspired popular": 55574,
      "presents concept": 89836,
      "network improving": 77281,
      "concepts model": 20629,
      "choice using": 16910,
      "using concept": 129467,
      "set original": 107518,
      "concepts training": 20647,
      "adapts different": 3342,
      "provides statistically": 94080,
      "produced framework": 91956,
      "performing techniques": 86976,
      "sentences form": 106323,
      "modelled using": 72593,
      "using shallow": 130166,
      "informed features": 55169,
      "rise deep": 102864,
      "interesting question": 56086,
      "question extent": 95157,
      "representations remains": 99853,
      "necessary combine": 76752,
      "effective identification": 34688,
      "attention plays": 9970,
      "information avoiding": 54390,
      "vanilla attention": 130811,
      "attention refinement": 9989,
      "supervision attention": 115867,
      "loss optimize": 65289,
      "decoding steps": 28136,
      "metrics assessing": 69924,
      "factually consistent": 42941,
      "modelbased approach": 72350,
      "summary training": 115649,
      "transformations sentences": 124281,
      "documents support": 33299,
      "prediction extract": 89062,
      "models reveals": 73959,
      "inference fact": 54144,
      "recently bert": 96666,
      "bert adopted": 13061,
      "document encoding": 33001,
      "captured bert": 15426,
      "finer granularity": 44398,
      "dependencies discourse": 29078,
      "rst trees": 103337,
      "models optimizing": 73668,
      "reports neural": 99083,
      "models optimized": 73667,
      "reference using": 97544,
      "reward reinforcement": 102674,
      "quality outputs": 94739,
      "summarization corpus": 115488,
      "automatically collect": 10736,
      "develop bertbased": 30180,
      "containing millions": 21801,
      "fraction model": 45365,
      "comparison systems": 19581,
      "classification article": 17125,
      "articles sentence": 9216,
      "annotated news": 6216,
      "preserve original": 89926,
      "sentence syntactic": 106097,
      "abstraction generating": 1790,
      "word introduce": 133325,
      "introduce erroneous": 56419,
      "work twofold": 134865,
      "twofold present": 126142,
      "combines sequential": 18696,
      "sequential decoder": 107222,
      "treebased decoder": 125668,
      "human editors": 50796,
      "remains poorly": 98818,
      "purely extractive": 94415,
      "copying training": 23298,
      "illustrate significance": 51745,
      "mathematical equations": 67455,
      "scientific information": 104007,
      "study explore": 114377,
      "text math": 121108,
      "math equations": 67448,
      "mathematical information": 67457,
      "pointer mechanism": 87823,
      "mathematical representation": 67462,
      "representation augmentation": 99175,
      "math tokens": 67450,
      "tokens source": 122334,
      "mechanism designed": 67967,
      "enrich representation": 37556,
      "sets realworld": 107703,
      "syntaxaware models": 116575,
      "information clear": 54417,
      "approaches standard": 8348,
      "standard rouge": 112299,
      "approaches combination": 8098,
      "information qualitative": 54890,
      "training convergence": 123396,
      "summaries tailored": 115462,
      "recently research": 96752,
      "attempts incorporate": 9771,
      "crucial insights": 25146,
      "certain characteristics": 15933,
      "objective provide": 81109,
      "guidelines appropriate": 49261,
      "incorporate various": 53494,
      "pretraining transformers": 90348,
      "transformers selfsupervised": 124462,
      "propose pretraining": 92997,
      "new selfsupervised": 79120,
      "remaining sentences": 98784,
      "evaluated best": 38957,
      "downstream datasets": 33997,
      "validated results": 130734,
      "feasible solutions": 43243,
      "models hard": 73320,
      "strong encoder": 113671,
      "texts decoder": 121492,
      "propose aggregation": 92536,
      "model review": 71932,
      "history information": 50558,
      "make encoder": 66665,
      "memory capacity": 68297,
      "reinforced selfattention": 97789,
      "novel architectural": 80495,
      "problem based": 90950,
      "solve various": 110626,
      "learning finegrained": 62583,
      "finegrained attention": 44338,
      "token space": 122267,
      "learn dynamic": 62034,
      "distribution tokens": 32684,
      "reduce redundancy": 97354,
      "summaries improve": 115444,
      "pretrained unsupervised": 90207,
      "modeling denoising": 72413,
      "text transform": 121375,
      "transform text": 124259,
      "concise version": 20705,
      "leverage recurrent": 63619,
      "proposed transformer": 93576,
      "corpora resources": 23575,
      "millions unlabeled": 70126,
      "various document": 131083,
      "controllable summarization": 22829,
      "decoder module": 28032,
      "length length": 63363,
      "extracts sequence": 42580,
      "copying words": 23299,
      "lookahead attention": 65220,
      "long sentence": 65088,
      "words sequencetosequence": 134207,
      "models decoder": 73019,
      "decoded words": 27990,
      "sentences decoder": 106271,
      "compressing sentences": 20303,
      "parent child": 84855,
      "capturing important": 15470,
      "google sentence": 48523,
      "performance longer": 86501,
      "furthermore human": 46179,
      "human summaries": 50971,
      "researchers students": 100713,
      "published papers": 94366,
      "relevant work": 98603,
      "characterize human": 16552,
      "domain scientific": 33632,
      "novel large": 80616,
      "number available": 80845,
      "heavy reliance": 49666,
      "reliance handcrafted": 98629,
      "sophisticated models": 110672,
      "new domainspecific": 78895,
      "domainspecific dataset": 33896,
      "datasets domain": 27437,
      "models rapid": 73853,
      "recently models": 96709,
      "known having": 58302,
      "gap actual": 46440,
      "using crowdsourced": 129523,
      "modelgenerated summaries": 72368,
      "results argue": 101522,
      "ability generalize": 1522,
      "method interpretable": 68920,
      "mechanism computes": 67958,
      "computes attention": 20515,
      "semantic kernels": 105089,
      "exhaustive experiments": 39993,
      "encoders general": 36648,
      "performance fullytrained": 86397,
      "encoders task": 36663,
      "random projections": 95505,
      "projections input": 92239,
      "capacity encoder": 15230,
      "general sequence": 46712,
      "trained randomly": 123256,
      "documents text": 33301,
      "humanwritten reference": 51130,
      "framework achieved": 45407,
      "states fed": 113043,
      "output sentence": 83118,
      "usually correspond": 130413,
      "local information": 64926,
      "representation encoder": 99222,
      "decoder decoding": 28001,
      "wmt machine": 132795,
      "extremely popular": 42607,
      "works domain": 134940,
      "domain build": 33475,
      "selection phase": 104814,
      "base architecture": 11443,
      "words summary": 134250,
      "ii limited": 51710,
      "source generating": 110762,
      "structure induction": 113880,
      "framework complements": 45460,
      "complements standard": 19759,
      "implicitly learned": 52031,
      "explicit structures": 41351,
      "dataset improves": 26975,
      "par standard": 84522,
      "knowledge scientific": 58162,
      "scientific research": 104020,
      "research offers": 100574,
      "offers challenging": 81585,
      "environment learn": 38189,
      "creative process": 24729,
      "scientific creativity": 103997,
      "prepared dataset": 89299,
      "using services": 130164,
      "diversity density": 32872,
      "text collections": 120798,
      "texts terms": 121633,
      "terms words": 120408,
      "human intuitions": 50873,
      "metrics highly": 69965,
      "correlated text": 24210,
      "bert inspire": 13145,
      "future applications": 46258,
      "propose factaware": 92674,
      "extract integrate": 42091,
      "integrate factual": 55751,
      "attention design": 9814,
      "generated existing": 47078,
      "combination model": 18569,
      "model extracts": 71164,
      "sequence source": 107072,
      "text experimental": 120933,
      "combination models": 18570,
      "problem unfortunately": 91277,
      "supervised summarization": 115840,
      "text main": 121097,
      "artificially constructed": 9271,
      "sentence reordering": 106027,
      "document generation": 33016,
      "baselines compared": 12372,
      "results demonstrates": 101710,
      "effective automatic": 34635,
      "extraction sentence": 42480,
      "level best": 63425,
      "solution work": 110570,
      "promising alternative": 92264,
      "units based": 127233,
      "tree neural": 125604,
      "comparing sentence": 19517,
      "extraction evaluation": 42327,
      "summarization future": 115510,
      "texts neural": 121561,
      "fixedsize input": 44621,
      "number input": 80897,
      "document content": 32978,
      "input size": 55438,
      "state different": 112490,
      "number tokens": 80982,
      "decoder learns": 28030,
      "render models": 98886,
      "designed identify": 29600,
      "offers natural": 81588,
      "natural form": 76257,
      "promising tool": 92312,
      "text multilingual": 121133,
      "generate short": 47014,
      "outputs combining": 83161,
      "programming method": 92132,
      "graph select": 48865,
      "different keywords": 31187,
      "keywords goal": 57646,
      "goal generating": 48354,
      "led automatic": 63273,
      "advantage fact": 4044,
      "additional sentence": 3570,
      "shows efficacy": 108574,
      "adaptive learning": 3332,
      "datasets modeling": 27576,
      "sourcetarget pairs": 110947,
      "decompose problem": 28153,
      "problem long": 91110,
      "systems instead": 116956,
      "used framework": 128560,
      "shift semantic": 108181,
      "matching framework": 67405,
      "extractors based": 42575,
      "property dataset": 92497,
      "models interpret": 73421,
      "number ways": 80994,
      "black boxes": 14247,
      "provide little": 93869,
      "users given": 129123,
      "captures sentence": 15454,
      "attribute scores": 10183,
      "outperformed stateoftheart": 82791,
      "step extractive": 113265,
      "approaches intuitive": 8197,
      "heterogeneous graphbased": 49858,
      "contains semantic": 21837,
      "structure flexible": 113868,
      "need generate": 76813,
      "control codes": 22801,
      "providing control": 94110,
      "different user": 31544,
      "obtained online": 81388,
      "turkish english": 125960,
      "community report": 19096,
      "existing biases": 40085,
      "motivate use": 74838,
      "documents introduce": 33243,
      "multitarget dataset": 75806,
      "novel annotation": 80463,
      "annotation burden": 6283,
      "baselines automated": 12355,
      "selection clinical": 104773,
      "seq2seq network": 106905,
      "problem clinical": 90963,
      "available clinical": 10959,
      "clinical data": 17763,
      "statistically significantly": 113189,
      "improvements 29": 52798,
      "long scientific": 65087,
      "twolevel hierarchical": 126149,
      "suggest patterns": 115325,
      "modeling story": 72551,
      "highly prone": 50337,
      "unfaithful input": 127066,
      "conducted large": 20932,
      "annotators substantial": 6500,
      "content model": 21908,
      "generating faithful": 47222,
      "humans furthermore": 51080,
      "metrics potentially": 69993,
      "metrics training": 70016,
      "generation studies": 47642,
      "stateoftheart encoderdecoder": 112651,
      "behaviors model": 12676,
      "entailment relation": 37670,
      "article headline": 9134,
      "clear difference": 17722,
      "improvements automatic": 52806,
      "summarization studied": 115570,
      "acquire semantic": 3036,
      "characteristics entities": 16523,
      "produce significantly": 91933,
      "input new": 55375,
      "systems finetuned": 116908,
      "judges rate": 57436,
      "optimization unsupervised": 82217,
      "extraction automatic": 42267,
      "produces shorter": 92007,
      "sentence preserving": 106000,
      "language fluency": 59051,
      "model aspects": 70691,
      "consisting language": 21453,
      "optimization proposed": 82208,
      "art unsupervised": 9099,
      "future evaluation": 46274,
      "summarization requires": 115561,
      "methods experiments": 69489,
      "task shown": 118711,
      "metrics crowdsourced": 69946,
      "annotation methodology": 6337,
      "hierarchical annotation": 49926,
      "use document": 128011,
      "management information": 66904,
      "deal documents": 27856,
      "methods relative": 69719,
      "dont need": 33947,
      "need look": 76831,
      "sentences providing": 106463,
      "process compare": 91438,
      "similar evaluation": 109090,
      "analysis experimental": 5610,
      "sample set": 103564,
      "metrics capture": 69936,
      "effectively tackle": 34849,
      "problem evaluating": 91037,
      "faithfulness generated": 43008,
      "outputs word": 83189,
      "based metric": 11842,
      "similarity learned": 109249,
      "scores especially": 104170,
      "especially highly": 38460,
      "consumer health": 21720,
      "studied area": 114151,
      "area computer": 8868,
      "needs understand": 76903,
      "information example": 54545,
      "information easily": 54511,
      "goldstandard human": 48454,
      "answers provided": 6741,
      "library medicines": 63962,
      "health question": 49624,
      "evaluate single": 38927,
      "leveraging graph": 63679,
      "relations textual": 98270,
      "process multiple": 91532,
      "relations crucial": 98133,
      "easily combined": 34441,
      "attention better": 9800,
      "encoding used": 36732,
      "generate section": 47006,
      "millions news": 70117,
      "news presents": 79357,
      "leverages stateoftheart": 63661,
      "acl 2020": 2994,
      "present demo": 89437,
      "effective contextual": 34652,
      "speech information": 111697,
      "applying supervised": 7273,
      "networkbased methods": 77486,
      "proposed achieved": 93166,
      "answering language": 6645,
      "view paper": 131707,
      "enhance stateoftheart": 37493,
      "contributions threefold": 22794,
      "help alleviate": 49696,
      "bert extra": 13125,
      "structural linguistic": 113779,
      "frequency idf": 45843,
      "finally validate": 44248,
      "dataset comparison": 26807,
      "methods combination": 69374,
      "manner perform": 66955,
      "step used": 113313,
      "showed using": 108394,
      "task humans": 118260,
      "humans remained": 51110,
      "restricting applicability": 101346,
      "present investigation": 89521,
      "sentences drawn": 106288,
      "document introducing": 33027,
      "containing documents": 21787,
      "sentences dataset": 106270,
      "bridges gap": 14601,
      "gap coreference": 46445,
      "publicly shared": 94334,
      "basis future": 12543,
      "models basis": 72821,
      "postediting machine": 88474,
      "sequence recent": 107041,
      "increasingly longer": 53703,
      "performance special": 86728,
      "architectures limited": 8815,
      "limited available": 64212,
      "multiple instances": 75582,
      "different inputs": 31185,
      "significant overlap": 108830,
      "russian news": 103497,
      "news automatic": 79308,
      "russian text": 103504,
      "architecture order": 8717,
      "incorporate entitylevel": 53468,
      "architecture injecting": 8674,
      "model zeroshot": 72343,
      "task primarily": 118562,
      "pipeline approaches": 87524,
      "propagation work": 92400,
      "objectives experimental": 81133,
      "settings demonstrate": 107795,
      "models bilingual": 72855,
      "models news": 73641,
      "including sentiment": 53369,
      "generation special": 47627,
      "special kind": 111364,
      "kind text": 57684,
      "strong natural": 113690,
      "understanding goes": 126856,
      "ability distinguish": 1511,
      "respectively previous": 101157,
      "best score": 13445,
      "researchers face": 100690,
      "library provides": 63963,
      "easily used": 34467,
      "evaluating summarization": 39104,
      "models metrics": 73566,
      "considering text": 21334,
      "wider variety": 132611,
      "lead improvement": 61857,
      "terms generalization": 120330,
      "generalization robustness": 46795,
      "work consists": 134435,
      "suitable dataset": 115395,
      "using bootstrapping": 129419,
      "automatic models": 10609,
      "size using": 109950,
      "leads large": 61943,
      "largest collection": 61528,
      "range automatic": 95544,
      "diverse terms": 32855,
      "collection human": 18475,
      "annotated expert": 6195,
      "expert judges": 41224,
      "crowdsource workers": 25102,
      "work help": 134550,
      "biomedical evidence": 14184,
      "trial reports": 125764,
      "systematic reviews": 116684,
      "models inform": 73400,
      "generated narrative": 47095,
      "reported findings": 99063,
      "presents participation": 89889,
      "regression layer": 97702,
      "embeddings comparison": 35604,
      "embeddings experiment": 35707,
      "variants using": 130884,
      "bert biobert": 13082,
      "results bert": 101549,
      "lstm layer": 65634,
      "obtain sentence": 81318,
      "did improve": 30915,
      "alignment step": 4990,
      "methods typically": 69818,
      "alignment explicit": 4958,
      "task end": 118145,
      "end created": 36801,
      "datasets enabling": 27446,
      "proper evaluation": 92414,
      "utilizing data": 130567,
      "statistical natural": 113138,
      "used check": 128420,
      "obtained approach": 81344,
      "used supplement": 128797,
      "classification end": 17196,
      "awareness global": 11263,
      "way specifically": 132133,
      "attend source": 9778,
      "developed regulate": 30297,
      "search generate": 104318,
      "property global": 92498,
      "search global": 104319,
      "datasets instead": 27529,
      "regularize training": 97766,
      "training furthermore": 123639,
      "jointly extracting": 57343,
      "representations unified": 99951,
      "unified vector": 127131,
      "sentence weighted": 106127,
      "sum word": 115426,
      "embeddings weights": 36022,
      "reflecting content": 97619,
      "allowing users": 5121,
      "direction improving": 31822,
      "summarization proposed": 115553,
      "proposed previous": 93525,
      "highly divergent": 50316,
      "develop endtoend": 30194,
      "information interactive": 54697,
      "released publicly": 98493,
      "evaluating comparing": 39042,
      "comparing baseline": 19500,
      "baseline implementations": 12232,
      "experimentation analysis": 40741,
      "relations long": 98205,
      "graphs containing": 48943,
      "salient entities": 103549,
      "relations important": 98187,
      "develop dataset": 30188,
      "dataset 200k": 26712,
      "develop strong": 30237,
      "baselines task": 12475,
      "loss recall": 65296,
      "resulting higher": 101446,
      "higher f1": 50180,
      "metric measures": 69887,
      "measures content": 67858,
      "embeddings work": 36039,
      "directly measure": 31888,
      "using benchmark": 129386,
      "analysis component": 5538,
      "learning evaluation": 62556,
      "informativeness based": 55163,
      "types negative": 126326,
      "method general": 68850,
      "leverages knowledge": 63646,
      "text retaining": 121266,
      "maximum mean": 67541,
      "mean discrepancy": 67597,
      "work multidocument": 134640,
      "summarization focused": 115507,
      "coverage diversity": 24501,
      "adapt multiple": 3185,
      "multiple kernel": 75583,
      "kernel learning": 57532,
      "learning make": 62704,
      "features knowledge": 43579,
      "based concepts": 11590,
      "efficient modeling": 35094,
      "making general": 66842,
      "sentence filtering": 105874,
      "holds true": 50597,
      "surpass previously": 116094,
      "metrics content": 69942,
      "selection planning": 104815,
      "transformers test": 124464,
      "dataset crosslingual": 26840,
      "human authors": 50757,
      "create goldstandard": 24616,
      "languages aligning": 60397,
      "requiring translation": 100358,
      "translation inference": 124869,
      "meet requirement": 68252,
      "selection surface": 104834,
      "jointly generate": 57344,
      "systems pose": 117068,
      "generation evaluation": 47392,
      "evaluation remains": 39358,
      "unsolved problem": 127564,
      "allows flexible": 5156,
      "pipeline neural": 87549,
      "light important": 63992,
      "important directions": 52145,
      "novel algorithms": 80459,
      "perform sentence": 86062,
      "knowledge points": 58109,
      "findings highlight": 44300,
      "sentences effective": 106289,
      "recently studied": 96769,
      "produce concise": 91880,
      "summaries specifically": 115461,
      "inference module": 54169,
      "diverse information": 32815,
      "combination pointer": 18578,
      "financial narrative": 44256,
      "networks extract": 77599,
      "sentences report": 106476,
      "used paraphrase": 128679,
      "highest precision": 50231,
      "precision scores": 88805,
      "editbased unsupervised": 34539,
      "summarization unsupervised": 115589,
      "research promising": 100592,
      "key modules": 57587,
      "agent language": 4304,
      "agent predicts": 4310,
      "edit actions": 34530,
      "train agent": 122901,
      "methods truly": 69816,
      "data validation": 26615,
      "techniques reinforcement": 119966,
      "analysis providing": 5788,
      "insights future": 55531,
      "understanding summarization": 126979,
      "systems respect": 117118,
      "finegrained syntactic": 44386,
      "bring specific": 14639,
      "pretrained knowledge": 90044,
      "indomain setting": 53981,
      "understanding generalization": 126850,
      "analysis characteristics": 5526,
      "datasets investigate": 27531,
      "goal text": 48391,
      "information known": 54710,
      "compare techniques": 19302,
      "scoring functions": 104229,
      "functions explicitly": 46073,
      "functions fit": 46075,
      "fit human": 44581,
      "insights human": 55535,
      "corpora greatly": 23491,
      "fact existing": 42820,
      "highlevel discourse": 50244,
      "chunks texts": 16970,
      "expanded form": 40360,
      "model outlines": 71623,
      "different segments": 31405,
      "point generate": 87802,
      "generation comes": 47334,
      "comes following": 18779,
      "following merits": 44973,
      "convergence speed": 22925,
      "data easily": 25874,
      "trained pseudo": 123250,
      "effect machine": 34599,
      "datasets understanding": 27772,
      "interpret model": 56210,
      "bart summarization": 11439,
      "factors like": 42896,
      "syntactic distance": 116397,
      "context particularly": 22217,
      "observed effects": 81221,
      "set syntactic": 107595,
      "rules determine": 103393,
      "determine spans": 30134,
      "sentence spans": 106087,
      "contain important": 21741,
      "strong indomain": 113682,
      "allows generalize": 5157,
      "match exceed": 67357,
      "framework guided": 45557,
      "models flexible": 73257,
      "general extensible": 46657,
      "different varieties": 31551,
      "demonstrate different": 28699,
      "network chinese": 77184,
      "encoder target": 36579,
      "parallel computation": 84591,
      "words adopt": 133811,
      "previous characterbased": 90398,
      "characterbased models": 16494,
      "summaries training": 115465,
      "edge weights": 34523,
      "specifically pretrain": 111576,
      "pretrain hierarchical": 89973,
      "objectives experiments": 81135,
      "sentence positions": 105993,
      "linear combination": 64336,
      "aims producing": 4561,
      "largescale labeled": 61453,
      "pretraining leverages": 90278,
      "translation monolingual": 124970,
      "monolingual tasks": 74623,
      "leverage massive": 63602,
      "massive monolingual": 67333,
      "enhance modeling": 37476,
      "language architecture": 58843,
      "scheme effectively": 103919,
      "scores stateoftheart": 104209,
      "2020 unsupervised": 737,
      "representation content": 99195,
      "differs previous": 31602,
      "method empirically": 68797,
      "scientific document": 104000,
      "science society": 103984,
      "bart model": 11438,
      "signals improve": 108701,
      "idea incorporate": 51327,
      "sift large": 108666,
      "context prevent": 22226,
      "identify optimal": 51529,
      "segments important": 104673,
      "modeling power": 72511,
      "avenue research": 11165,
      "token alignments": 122245,
      "measuring information": 67918,
      "result holds": 101377,
      "reliable automatic": 98610,
      "interpretable method": 56245,
      "method evaluating": 68816,
      "does directly": 33342,
      "provided methods": 93974,
      "document corpora": 32983,
      "propose constrained": 92595,
      "extracted source": 42184,
      "feedback form": 43827,
      "constraint used": 21587,
      "text remarkable": 121243,
      "motivated new": 74855,
      "machine text": 65874,
      "benchmark analysis": 12730,
      "efficient evaluation": 35073,
      "generally improve": 46859,
      "zero fewshot": 135474,
      "pretrained selfsupervised": 90180,
      "niche domains": 79476,
      "dataset length": 27001,
      "additional diverse": 3509,
      "10 100": 119,
      "examples compared": 39815,
      "fewshot transfer": 43919,
      "introduce regularization": 56528,
      "unsupervised finetuning": 127639,
      "data analyze": 25599,
      "monolingual bertbased": 74554,
      "include thorough": 53239,
      "models biased": 72848,
      "relevance input": 98513,
      "input focus": 55339,
      "focus biased": 44739,
      "text modifying": 121130,
      "random restarts": 95506,
      "probabilities assigned": 90840,
      "algorithm leads": 4763,
      "faster lighter": 43182,
      "methods similar": 69756,
      "literature thousands": 64778,
      "overcome problem": 83296,
      "contributions context": 22784,
      "findings shared": 44313,
      "model controlled": 70914,
      "separate generation": 106853,
      "generated outputs": 47104,
      "human study": 50969,
      "helpful traditional": 49802,
      "selecting salient": 104757,
      "addition incorporate": 3439,
      "berts ability": 13285,
      "ability natural": 1547,
      "content relative": 21932,
      "reference coverage": 97510,
      "modelbased evaluation": 72351,
      "scores derived": 104166,
      "surprising findings": 116126,
      "documents analysis": 33179,
      "datasets indicates": 27526,
      "problem summarizing": 91256,
      "ways deal": 132151,
      "organize existing": 82472,
      "categories propose": 15747,
      "additional methods": 3550,
      "general flexible": 46659,
      "flexible way": 44666,
      "paper datasets": 83793,
      "example paper": 39789,
      "goal automatically": 48332,
      "text extracting": 120945,
      "users focus": 129120,
      "extracted question": 42172,
      "review information": 102553,
      "popular transformer": 88140,
      "multiple approaches": 75494,
      "approaches proposing": 8297,
      "documentlevel discourse": 33137,
      "extending recently": 41820,
      "framework lightweight": 45601,
      "lightweight alternative": 64009,
      "selfattention approach": 104873,
      "model tree": 72236,
      "approach reaches": 7836,
      "reaches similar": 95916,
      "level despite": 63434,
      "hyperparameter setting": 51230,
      "looking tools": 65224,
      "unique type": 127196,
      "competitive debaters": 19653,
      "particular target": 85455,
      "text vectorization": 121403,
      "vectorization models": 131404,
      "framework flair": 45546,
      "users preferences": 129153,
      "limitation present": 64164,
      "users control": 129105,
      "textual input": 121715,
      "prompts using": 92329,
      "time requiring": 122090,
      "additional human": 3520,
      "quantitatively demonstrate": 94888,
      "summarization setting": 115565,
      "parsbert pretrained": 84878,
      "critical natural": 24818,
      "work generating": 134544,
      "documents prior": 33272,
      "work document": 134481,
      "short summary": 108243,
      "typically case": 126415,
      "legal document": 63313,
      "document book": 32956,
      "book paper": 14412,
      "exploits hierarchical": 41491,
      "results long": 101893,
      "analysis generated": 5631,
      "multitasking approach": 75901,
      "datasets codes": 27350,
      "criteria summary": 24797,
      "quality measures": 94716,
      "better measure": 13626,
      "cast doubt": 15677,
      "doubt view": 33974,
      "best measure": 13363,
      "compression order": 20317,
      "rate quality": 95801,
      "consider scenarios": 21226,
      "compression ii": 20309,
      "algorithms evaluate": 4847,
      "design statistical": 29567,
      "analysis manual": 5691,
      "conduct survey": 20896,
      "little agreement": 64787,
      "method vary": 69220,
      "annotators annotation": 6488,
      "strong impact": 113680,
      "addition highlight": 3435,
      "reliable annotations": 98609,
      "given fixed": 48035,
      "rewriting text": 102701,
      "lose important": 65238,
      "solution model": 110559,
      "contentbased addressing": 21983,
      "language conduct": 58908,
      "short document": 108208,
      "datasets expensive": 27464,
      "available document": 10985,
      "account data": 2021,
      "modern pretrained": 74411,
      "bart lewis": 11435,
      "lewis et": 63719,
      "gpt2 radford": 48567,
      "2019 language": 701,
      "method beats": 68672,
      "human labeling": 50891,
      "lowresource domain": 65496,
      "extensive labeled": 41941,
      "task diverse": 118108,
      "diverse target": 32852,
      "second phase": 104441,
      "domain pretraining": 33615,
      "taskadaptive pretraining": 118869,
      "experiments effectiveness": 40925,
      "correlated similarity": 24208,
      "similarity pretraining": 109283,
      "results illustrate": 101837,
      "highresource settings": 50432,
      "mtl approach": 75013,
      "method code": 68696,
      "generate consistent": 46917,
      "intermediate states": 56144,
      "summary decoder": 115636,
      "applied stateoftheart": 7120,
      "present datasets": 89431,
      "scientific studies": 104021,
      "studies furthermore": 114232,
      "abstracts using": 1828,
      "selected spans": 104735,
      "relevant diverse": 98541,
      "approach gets": 7589,
      "bertscore questeval": 13290,
      "consistency coherence": 21354,
      "approach extractive": 7565,
      "instead ranking": 55677,
      "problems proposing": 91363,
      "hypothesis tests": 51280,
      "evaluating proposed": 39086,
      "demonstrating high": 28974,
      "trained style": 123295,
      "unit prediction": 127207,
      "strong lexical": 113685,
      "languages informative": 60645,
      "multiple variants": 75743,
      "special emphasis": 111360,
      "evaluated experiments": 38975,
      "computational memory": 20403,
      "novel efficient": 80554,
      "existing efficient": 40121,
      "process times": 91577,
      "scores competitive": 104162,
      "competitive comparisons": 19650,
      "masking technique": 67316,
      "time using": 122135,
      "settings importantly": 107807,
      "20 training": 531,
      "bart finetuned": 11434,
      "seldom considering": 104684,
      "help relevant": 49774,
      "dataset semantic": 27181,
      "semantic scholar": 105266,
      "papers different": 84504,
      "graph extensive": 48808,
      "compared pretrained": 19411,
      "crucial better": 25133,
      "content papers": 21919,
      "creating summaries": 24708,
      "summaries lack": 115445,
      "enable accurate": 36348,
      "sources model": 110909,
      "documents content": 33211,
      "content planning": 21922,
      "planning component": 87638,
      "component uses": 20010,
      "model separately": 71983,
      "implemented tested": 51979,
      "framework leads": 45595,
      "equally fluent": 38229,
      "systems started": 117157,
      "performance major": 86518,
      "number annotated": 80836,
      "models assessing": 72777,
      "current techniques": 25378,
      "sources training": 110925,
      "word dependency": 132986,
      "sets simple": 107711,
      "finegrained annotations": 44336,
      "sentencelevel annotations": 106143,
      "annotations synthetic": 6468,
      "medical summaries": 68226,
      "raw counts": 95852,
      "medical report": 68222,
      "quality accuracy": 94585,
      "measure designed": 67789,
      "errors human": 38375,
      "predicting discourse": 88983,
      "discourse trees": 32096,
      "work indicates": 134568,
      "particular generate": 85417,
      "rststyle discourse": 103339,
      "selfattention matrices": 104882,
      "datasets reveal": 27689,
      "works try": 134986,
      "try investigate": 125897,
      "stacking approach": 112133,
      "work highlight": 134552,
      "methods motivates": 69626,
      "provides unified": 94090,
      "base systems": 11483,
      "different application": 30995,
      "scenarios new": 103858,
      "addresses limitations": 3800,
      "limitations traditional": 64195,
      "code provide": 18138,
      "interface use": 56104,
      "interactive visual": 56035,
      "data evaluation": 25910,
      "largely unknown": 61349,
      "limitation introduce": 64161,
      "opensource tool": 82031,
      "tool visualizing": 122406,
      "summarization lexical": 115526,
      "tools offers": 122463,
      "learned entity": 62216,
      "flexible mechanism": 44660,
      "ordered sequences": 82435,
      "summary conditioned": 115633,
      "conditioned entity": 20799,
      "content plan": 21920,
      "applying natural language": 7260,
      "pays little attention": 85817,
      "texts specific domain": 121619,
      "resolve problem propose": 100803,
      "et al 2007": 38690,
      "automatic document summarization": 10522,
      "appropriate training data": 8443,
      "rich data source": 102739,
      "ability humans perform": 1529,
      "syntactic structure sentence": 116486,
      "discourse structure text": 32092,
      "paper present detailed": 84096,
      "suffer information loss": 115225,
      "approach reports new": 7857,
      "documents performed experiments": 33267,
      "using human evaluators": 129744,
      "based summarization methods": 12086,
      "role labeling semantic": 103193,
      "semantic representation text": 105233,
      "spearman correlation coefficient": 111344,
      "good overall performance": 48489,
      "features evaluation results": 43492,
      "encoderdecoder recurrent neural": 36620,
      "corpora propose novel": 23562,
      "rare unseen training": 95745,
      "unseen training time": 127553,
      "shows proposed models": 108620,
      "features work propose": 43792,
      "develop general framework": 30200,
      "hierarchical document encoder": 49957,
      "containing hundreds thousands": 21793,
      "hundreds thousands documentsummary": 51154,
      "thousands documentsummary pairs": 121921,
      "improving information extraction": 53106,
      "domains training data": 33877,
      "model learns select": 71452,
      "learned large corpus": 62226,
      "provide extensive analysis": 93824,
      "summarization opinion summarization": 115544,
      "consists main components": 21482,
      "evaluation generated summaries": 39220,
      "methods perform better": 69671,
      "strong unsupervised baselines": 113730,
      "opinion summarization framework": 82101,
      "era big data": 38265,
      "data deep learning": 25820,
      "information extraction summarization": 54603,
      "using sentence compression": 130146,
      "achieves statistically significant": 2908,
      "sequence seq2seq model": 107050,
      "model gated recurrent": 71222,
      "texts proposed model": 121586,
      "proposed model compared": 93439,
      "generation model model": 47480,
      "experiments method outperforms": 41004,
      "multidocument abstractive summarization": 75075,
      "approach outperforms baselines": 7766,
      "outperforms baselines stateoftheart": 82850,
      "approach achieves promising": 7314,
      "achieves promising results": 2838,
      "controlling output length": 22858,
      "output length neural": 83090,
      "models shown great": 74028,
      "tasks previous work": 119397,
      "networks recently shown": 77725,
      "recently shown effective": 96765,
      "shown effective modeling": 108464,
      "modeling natural languages": 72492,
      "paper aims enhance": 83727,
      "order generate summaries": 82329,
      "models large datasets": 73461,
      "datasets models achieve": 27578,
      "tasks current approaches": 119024,
      "paper address shortcomings": 83710,
      "furthermore propose simple": 46207,
      "rnn based sequence": 102907,
      "investigate performance different": 56794,
      "content deep learning": 21868,
      "analysis work present": 5941,
      "systems real world": 117097,
      "demonstrate approach able": 28665,
      "abstractive summarization machine": 1811,
      "summarization machine translation": 115528,
      "important research area": 52227,
      "text summarization process": 121349,
      "finally discusses briefly": 44174,
      "evaluation datasets task": 39172,
      "using novel crowdsourcing": 129980,
      "novel crowdsourcing approach": 80523,
      "model large scale": 71415,
      "stateoftheart baseline models": 112592,
      "abstractive summarization abstractive": 1809,
      "summarization abstractive summarization": 115473,
      "phrases work propose": 87470,
      "model learns focus": 71446,
      "aims alleviate problem": 4500,
      "model introduce new": 71369,
      "model clearly outperforms": 70834,
      "trained supervised learning": 123298,
      "datasets model obtains": 27574,
      "dataset improvement previous": 26973,
      "important problem natural": 52212,
      "stateoftheart results compared": 112897,
      "based encoderdecoder framework": 11676,
      "based neural model": 11892,
      "semantic similarity texts": 105293,
      "outperforms baseline systems": 82847,
      "significant improvements state": 108805,
      "improvements state art": 52912,
      "practice evidencebased medicine": 88732,
      "brief survey recent": 14618,
      "survey recent years": 116188,
      "data variety sources": 26617,
      "problem domain adaptation": 91018,
      "domain adaptation neural": 33441,
      "neural summarization model": 78692,
      "style target domain": 114597,
      "information available internet": 54388,
      "use complex network": 127946,
      "differently previous works": 31597,
      "significant improvement results": 108775,
      "text summarization text": 121354,
      "summarization text simplification": 115582,
      "text simplification text": 121306,
      "children nonnative speakers": 16727,
      "text simplification aims": 121299,
      "aims reduce linguistic": 4569,
      "reduce linguistic complexity": 97335,
      "based sequencetosequence model": 12030,
      "success text generation": 115133,
      "systems benchmark corpus": 116756,
      "multidocument summarization using": 75083,
      "main ideas document": 66429,
      "ideas document set": 51349,
      "paper present unsupervised": 84145,
      "bag words model": 11364,
      "improve performance model": 52469,
      "results different datasets": 101726,
      "performance gains compared": 86402,
      "widely used evaluation": 132569,
      "models address issue": 72722,
      "correlate better human": 24192,
      "widely used social": 132586,
      "datasets recent years": 27662,
      "select relevant sentences": 104707,
      "relevant sentences context": 98579,
      "summarization mds task": 115530,
      "model nodes represent": 71589,
      "able improve quality": 1654,
      "information used improve": 55076,
      "sentence paper present": 105974,
      "obtain encouraging results": 81281,
      "simple effective mechanism": 109410,
      "networks paper present": 77690,
      "paper present model": 84112,
      "existing neural network": 40232,
      "based sequencetosequence models": 12031,
      "led stateoftheart results": 63288,
      "document summarization models": 33087,
      "summarization models trained": 115540,
      "trained end end": 123128,
      "training data address": 123423,
      "trained generate text": 123152,
      "systems perform standard": 117060,
      "perform standard datasets": 86076,
      "end propose simple": 36827,
      "sequence transduction model": 107091,
      "summarization text summarization": 115583,
      "opens new possibilities": 82011,
      "propose novel training": 92971,
      "reinforcement learning objective": 97818,
      "new challenges field": 78825,
      "important challenging research": 52115,
      "challenging research topic": 16308,
      "representations accurately capture": 99488,
      "approach applying existing": 7361,
      "widely used dataset": 132567,
      "propose fully datadriven": 92690,
      "train evaluate model": 122934,
      "promising results despite": 92293,
      "simple baseline method": 109374,
      "lead higher quality": 61854,
      "summarization methods automatic": 115534,
      "models automatic text": 72794,
      "discourse structure document": 32091,
      "decoder generate summary": 28015,
      "wide range existing": 132505,
      "abstractive summarization methods": 1813,
      "approach makes use": 7702,
      "highly difficult problem": 50315,
      "problem sequencetosequence model": 91223,
      "evaluation method address": 39267,
      "method address problem": 68620,
      "evaluation method does": 39268,
      "improves semantic consistency": 53043,
      "report series experiments": 99042,
      "models better capture": 72846,
      "models used tfidf": 74292,
      "semantic models used": 105117,
      "content social media": 21950,
      "social media dataset": 110370,
      "stateoftheart performances benchmark": 112860,
      "propose unified model": 93130,
      "effectiveness representation learning": 34942,
      "representation learning neural": 99303,
      "previously proposed models": 90612,
      "structured selfattention mechanism": 114040,
      "classification problem model": 17349,
      "hierarchical attention networks": 49934,
      "explore space possible": 41589,
      "based integer linear": 11776,
      "work explore new": 134513,
      "summarization using semantic": 115591,
      "set amr graphs": 107356,
      "knowledge base based": 57764,
      "model attention mechanism": 70702,
      "significant improvements performance": 108800,
      "thoroughly investigated paper": 121905,
      "representation amr semantic": 99168,
      "amr semantic representation": 5396,
      "semantic representation natural": 105229,
      "surface realization step": 116075,
      "data perform experiments": 26231,
      "produced promising results": 91965,
      "source sentences paper": 110823,
      "approach naturally combines": 7733,
      "incorporating sourceside syntactic": 53558,
      "information proposed approach": 54883,
      "approach compares favorably": 7434,
      "compares favorably stateoftheart": 19497,
      "sequencetosequence based neural": 107157,
      "generation tasks proposed": 47673,
      "different previous methods": 31355,
      "model directly predicts": 71000,
      "dataset proposed framework": 27118,
      "using document context": 129603,
      "networks sequence sequence": 77760,
      "abstractive extractive summarization": 1802,
      "using generative models": 129704,
      "models outperform standard": 73682,
      "evaluate proposed techniques": 38903,
      "integer linear programmingbased": 55738,
      "abstractive summarization paper": 1816,
      "proposed framework effective": 93287,
      "semantic analysis approach": 104977,
      "weighting schemes calculating": 132368,
      "schemes calculating input": 103949,
      "calculating input matrix": 15030,
      "account word order": 2044,
      "arabic english datasets": 8514,
      "datasets used evaluate": 27776,
      "linguistic data consortium": 64459,
      "data consortium ldc": 25784,
      "results datasets effectiveness": 101663,
      "datasets effectiveness proposed": 27442,
      "performs comprehensively better": 86998,
      "comprehensively better compared": 20287,
      "better compared stateoftheart": 13543,
      "stateoftheart methods abstractive": 112703,
      "dataset models achieve": 27037,
      "high semantic relevance": 50139,
      "task neural encoderdecoder": 118452,
      "availability large parallel": 10920,
      "automatically acquired web": 10712,
      "paper present initial": 84108,
      "maximal marginal relevance": 67501,
      "human readable form": 50946,
      "existing approaches propose": 40057,
      "relevant parts source": 98568,
      "language model incorporates": 59334,
      "model achieves results": 70586,
      "achieves results comparable": 2849,
      "topicaware convolutional neural": 122604,
      "propose novel abstractive": 92893,
      "articles topics based": 9223,
      "neural networks demonstrate": 78462,
      "models challenged lack": 72895,
      "paper extend previous": 83938,
      "using gold standard": 129710,
      "obtained offtheshelf parser": 81387,
      "trained large corpora": 123178,
      "sentences remove need": 106475,
      "models learn generate": 73484,
      "based human evaluation": 11760,
      "using variational autoencoder": 130350,
      "different summarization data": 31465,
      "higher rouge scores": 50205,
      "deep learning deep": 28267,
      "particular paper focuses": 85438,
      "deep learning reinforcement": 28308,
      "framework deep learning": 45483,
      "used features derived": 128552,
      "bidirectional beam search": 13902,
      "sequence generative models": 106956,
      "knowledge current models": 57848,
      "end propose endtoend": 36824,
      "tackle aforementioned issues": 117284,
      "encoder backward decoder": 36497,
      "approximate inference algorithm": 8455,
      "model enables model": 71068,
      "policy gradient reinforcement": 87954,
      "gradient reinforcement learning": 48605,
      "summaries experimental results": 115437,
      "paper propose training": 84335,
      "summary input text": 115643,
      "pairs training data": 83652,
      "extractive summarization approaches": 42565,
      "approaches obtained encouraging": 8262,
      "obtained encouraging results": 81366,
      "dataset provide baselines": 27127,
      "baselines future research": 12399,
      "code dataset available": 18090,
      "neural abstractive systems": 77826,
      "summarization model learns": 115536,
      "model learns generate": 71448,
      "generate high quality": 46950,
      "model yields significantly": 72340,
      "yields significantly better": 135432,
      "human evaluation confirms": 50808,
      "machine translation rapidly": 66185,
      "techniques abstractive summarization": 119826,
      "various attention mechanisms": 131046,
      "improving state art": 53167,
      "recent methods use": 96472,
      "baselines dataset available": 12381,
      "available languages english": 11026,
      "experiments deep learning": 40873,
      "models results suggest": 73956,
      "shown able outperform": 108440,
      "second propose novel": 104448,
      "memory store information": 68385,
      "using attentionbased deep": 129356,
      "salient sentences document": 103556,
      "question answering forums": 95048,
      "employs variational inference": 36336,
      "stance detection task": 112170,
      "task demonstrate superiority": 118063,
      "existing approaches automatic": 40050,
      "rich internal structure": 102752,
      "summarization based model": 115483,
      "capture longdistance dependencies": 15336,
      "segmentation sentiment analysis": 104633,
      "sentiment analysis effective": 106587,
      "contextual token representations": 22504,
      "recent neural network": 96483,
      "work present neural": 134682,
      "supervision experimental results": 115886,
      "manual evaluation shows": 66995,
      "evaluation shows models": 39399,
      "multitask learning extractive": 75836,
      "learning extractive summarization": 62577,
      "vast amounts training": 131273,
      "situations training data": 109904,
      "available information paper": 11017,
      "language model representations": 59388,
      "language generation pretrained": 59091,
      "range language understanding": 95573,
      "translation abstractive summarization": 124615,
      "state art text": 112480,
      "bert pretrained transformer": 13207,
      "pretrained transformer model": 90197,
      "existing methods controlling": 40181,
      "research field known": 100511,
      "paper novel framework": 84057,
      "supervised extractive summarization": 115754,
      "answer important questions": 6538,
      "strong summarization baselines": 113718,
      "summarization source code": 115567,
      "short natural language": 108223,
      "language descriptions source": 58947,
      "descriptions source code": 29494,
      "lack suitable datasets": 58753,
      "natural language data": 76278,
      "models successful natural": 74122,
      "unlike previous studies": 127443,
      "previous studies learn": 90497,
      "method generate text": 68858,
      "improve rouge scores": 52533,
      "text work proposes": 121415,
      "cost slightly lower": 24374,
      "neural networks apply": 78439,
      "tasks remains unclear": 119457,
      "remains unclear best": 98825,
      "use pretrained lms": 128207,
      "shortage training data": 108272,
      "encoderdecoder models used": 36606,
      "based surface form": 12096,
      "remain poorly understood": 98773,
      "paper attempts bridge": 83747,
      "attempts bridge gap": 9764,
      "work propose solutions": 134745,
      "data available task": 25684,
      "manual evaluation generated": 66993,
      "received lot attention": 96369,
      "introduce opensource toolkit": 56523,
      "propose endtoend model": 92651,
      "release data code": 98446,
      "generation automatically generating": 47319,
      "news headline generation": 79342,
      "human judgments propose": 50885,
      "propose new automatic": 92839,
      "recent years thanks": 96641,
      "paper develop unsupervised": 83866,
      "largescale highquality training": 61442,
      "model better capture": 70774,
      "document experimental results": 33005,
      "strong baselines wide": 113657,
      "baselines wide margin": 12491,
      "global context document": 48230,
      "level paper aim": 63485,
      "paper aim improve": 83723,
      "compare model existing": 19257,
      "human evaluation demonstrate": 50809,
      "outperform traditional approaches": 82767,
      "traditional approaches rely": 122795,
      "best model produces": 13372,
      "scientific papers using": 104018,
      "compared competitive baselines": 19348,
      "summarization task models": 115576,
      "problem proposed improved": 91186,
      "models good performance": 73308,
      "model based deep": 70727,
      "seek better understand": 104516,
      "extractive summarization systems": 42568,
      "simple method postprocessing": 109468,
      "selects salient sentences": 104865,
      "fully supervised baselines": 45972,
      "novel task automatically": 80744,
      "promising performance tasks": 92287,
      "overview recent advances": 83372,
      "recent advances field": 96402,
      "texts recent years": 121591,
      "incorporate domain knowledge": 53466,
      "pretrained deep language": 90022,
      "used measure similarity": 128627,
      "knowledge capture context": 57822,
      "current stateoftheart unsupervised": 25368,
      "progress benchmark datasets": 92142,
      "remains largely unexplored": 98806,
      "conduct extensive analysis": 20866,
      "extensive analysis biases": 41856,
      "bias news articles": 13829,
      "language existing methods": 59025,
      "problem error propagation": 91034,
      "multitask learning lack": 75845,
      "make dataset code": 66649,
      "approaches reinforcement learning": 8316,
      "yield stateoftheart performance": 135357,
      "stateoftheart performance terms": 112850,
      "correlation human ratings": 24234,
      "human ratings previous": 50942,
      "approaches human evaluation": 8181,
      "human evaluation experiments": 50814,
      "function source code": 46044,
      "times faster training": 122177,
      "model code publicly": 70838,
      "information bottleneck principle": 54402,
      "novel approach unsupervised": 80492,
      "approach unsupervised sentence": 7994,
      "direct supervision approach": 31805,
      "outperforms unsupervised models": 83042,
      "model outperforms unsupervised": 71681,
      "including human evaluation": 53300,
      "combining global local": 18728,
      "global local context": 48250,
      "local context paper": 64909,
      "incorporating global context": 53541,
      "task model needs": 118396,
      "generator extensive experiments": 47772,
      "semantic accuracy generated": 104972,
      "produces high quality": 91996,
      "particular propose novel": 85444,
      "statistically significant difference": 113182,
      "tend achieve best": 120140,
      "achieve best performance": 2420,
      "yield improved results": 135342,
      "multiple pretraining tasks": 75646,
      "end end architecture": 36807,
      "models paper stock": 73709,
      "training methods finally": 123708,
      "automatic summarization methods": 10680,
      "variety domains including": 130972,
      "contextual features finally": 22466,
      "study provides new": 114495,
      "sequencetosequence model paper": 107173,
      "paper presents concept": 84161,
      "proposed approach provides": 93208,
      "provides statistically significant": 94081,
      "statistically significant improvements": 113187,
      "improvements stateoftheart models": 52917,
      "datasets human evaluation": 27506,
      "best performing techniques": 13407,
      "linguistically informed features": 64589,
      "attention plays key": 9971,
      "verify effectiveness methods": 131582,
      "generated summary training": 47132,
      "training data generated": 123470,
      "model trained jointly": 72195,
      "using standard datasets": 130211,
      "language inference fact": 59172,
      "convolutional networks experiments": 23170,
      "stateoftheart methods significant": 112713,
      "radiology reports neural": 95452,
      "realworld applications work": 96152,
      "information extraction module": 54580,
      "reward reinforcement learning": 102675,
      "evaluation proposed approach": 39339,
      "automatically collect large": 10737,
      "small fraction model": 110152,
      "fraction model parameters": 45366,
      "based model learns": 11849,
      "datasets demonstrate competitive": 27397,
      "competitive results strong": 19685,
      "results strong baselines": 102217,
      "remains poorly understood": 98819,
      "framework jointly model": 45592,
      "attention mechanism designed": 9895,
      "paper provide detailed": 84376,
      "model gives best": 71252,
      "best results training": 13443,
      "finetuned downstream nlp": 44424,
      "including text summarization": 53396,
      "diverse domains work": 32812,
      "new selfsupervised objective": 79121,
      "using human evaluation": 129743,
      "achieve human performance": 2486,
      "paper propose aggregation": 84216,
      "propose aggregation mechanism": 92537,
      "text representation model": 121247,
      "achieve higher quality": 2481,
      "compared strong baseline": 19468,
      "results compared baseline": 101598,
      "present detailed error": 89447,
      "unsupervised abstractive summarization": 127596,
      "leverage recurrent neural": 63620,
      "recently proposed transformer": 96742,
      "enhance quality generated": 37486,
      "stateoftheart abstractive summarization": 112569,
      "based encoderdecoder models": 11677,
      "encoderdecoder models generate": 36605,
      "model outperformed previous": 71632,
      "outperformed previous models": 82787,
      "attention network sentence": 9947,
      "sequencetosequence seq2seq based": 107203,
      "seq2seq based models": 106889,
      "words unseen words": 134287,
      "generate informative summaries": 46958,
      "furthermore human evaluation": 46180,
      "automatic summarization systems": 10681,
      "reliance handcrafted features": 98630,
      "annotated data lack": 6159,
      "lack large corpora": 58725,
      "corpora news domain": 23542,
      "domain paper propose": 33608,
      "commonly used datasets": 18968,
      "human evaluation metric": 50821,
      "models ability generalize": 72639,
      "attention mechanism computes": 9890,
      "mechanism computes attention": 67959,
      "datasets english language": 27451,
      "score human evaluation": 104083,
      "tasks work investigate": 119610,
      "models compare performance": 72935,
      "new approach automatic": 78785,
      "approach automatic estimation": 7373,
      "hidden states fed": 49913,
      "decoder generate output": 28014,
      "focus capturing local": 44743,
      "sentence level information": 105930,
      "leaving room improvement": 63269,
      "quality paper propose": 94745,
      "specifically proposed model": 111589,
      "wmt machine translation": 132796,
      "language model models": 59350,
      "challenging task generating": 16319,
      "text classification performance": 120754,
      "text experimental results": 120934,
      "model pretrained largescale": 71778,
      "sequencetosequence seq2seq learning": 107204,
      "seq2seq learning problem": 106896,
      "challenging paper presents": 16289,
      "improve performance baselines": 52459,
      "pretrained largescale data": 90117,
      "achieves comparable results": 2764,
      "summarization future research": 115511,
      "pointer generator network": 87822,
      "previous work shows": 90551,
      "extractive summarization baselines": 42566,
      "long document summary": 65071,
      "performance best models": 86181,
      "models achieve results": 72670,
      "achieve results par": 2527,
      "task semantic text": 118676,
      "choices model making": 16916,
      "outperformed stateoftheart models": 82792,
      "heterogeneous graph neural": 49857,
      "code released github": 18145,
      "model paper experiment": 71691,
      "largescale training corpora": 61514,
      "enable new research": 36361,
      "facilitate study task": 42794,
      "collected using novel": 18439,
      "using novel annotation": 129979,
      "effective learning strategy": 34697,
      "strong baselines automated": 113642,
      "publicly available clinical": 94294,
      "datasets approach outperforms": 27316,
      "outperforms strong unsupervised": 83031,
      "comparable stateoftheart supervised": 19182,
      "stateoftheart supervised approaches": 112977,
      "supervised approaches trained": 115725,
      "tasks language modeling": 119240,
      "conducted large scale": 20933,
      "human annotators substantial": 50751,
      "text paper explores": 121164,
      "stateoftheart encoderdecoder model": 112652,
      "encoderdecoder model model": 36601,
      "data used training": 26603,
      "abstractive summarization studied": 1818,
      "results models produce": 101959,
      "better comparable performance": 13540,
      "comparable performance compared": 19156,
      "human judges rate": 50879,
      "sentence model aspects": 105950,
      "language modeling semantic": 59465,
      "semantic similarity metrics": 105286,
      "experiments demonstrate significant": 40894,
      "novel annotation methodology": 80464,
      "analysis experimental results": 5611,
      "existing automatic metrics": 40074,
      "effectively tackle problem": 34850,
      "leverages recent advances": 63658,
      "highly abstractive summaries": 50287,
      "studied area computer": 114152,
      "information easily accessible": 54512,
      "national library medicines": 76214,
      "consumer health question": 21721,
      "health question answering": 49625,
      "medical question answering": 68219,
      "summarizing long documents": 115629,
      "improve summarization performance": 52555,
      "millions news articles": 70118,
      "neural networkbased methods": 78425,
      "bert model proposed": 13172,
      "question answering language": 95059,
      "document frequency idf": 33012,
      "proposed method benchmark": 93349,
      "pretrained transformerbased language": 90202,
      "experiments showed using": 41136,
      "translation automatic postediting": 124658,
      "automatic postediting machine": 10625,
      "postediting machine translation": 88475,
      "recent work introduced": 96573,
      "summarization aims generating": 115476,
      "data existing methods": 25915,
      "pipeline approaches suffer": 87525,
      "approaches suffer error": 8361,
      "error propagation work": 38321,
      "propagation work propose": 92401,
      "reinforcement learning directly": 97801,
      "learning directly optimize": 62505,
      "objectives experimental results": 81134,
      "settings demonstrate effectiveness": 107796,
      "reinforcement learning models": 97817,
      "including sentiment analysis": 53370,
      "strong natural language": 113691,
      "language understanding goes": 60226,
      "meaning individual words": 67642,
      "individual words sentences": 53942,
      "pretrained transformerbased models": 90205,
      "summarization evaluation metrics": 115502,
      "using bootstrapping approach": 129420,
      "models broad range": 72863,
      "range automatic metrics": 95545,
      "hope work help": 50647,
      "finally propose new": 44223,
      "paper presents participation": 84197,
      "obtain word embeddings": 81337,
      "generating training data": 47281,
      "statistical natural language": 113139,
      "beam search generate": 12605,
      "significantly improves stateoftheart": 108956,
      "stateoftheart results unsupervised": 112927,
      "unified vector space": 127132,
      "sum word embeddings": 115427,
      "word embeddings weights": 133244,
      "experiment results multiple": 40502,
      "results multiple benchmark": 101968,
      "allowing users interact": 5122,
      "proposed previous work": 93526,
      "extensive experimentation analysis": 41891,
      "knowledge graphs containing": 57982,
      "human annotations develop": 50747,
      "develop strong baselines": 30238,
      "automatic metric evaluating": 10599,
      "using benchmark datasets": 129387,
      "maximum mean discrepancy": 67542,
      "work multidocument summarization": 134641,
      "multiple kernel learning": 75584,
      "learning make use": 62705,
      "information sources text": 54997,
      "content selection planning": 21940,
      "benchmark dataset crosslingual": 12754,
      "largescale multilingual dataset": 61475,
      "outperforms baseline approaches": 82841,
      "architecture neural text": 8711,
      "sufficient training data": 115290,
      "content selection surface": 21941,
      "selection surface realization": 104835,
      "problem paper present": 91156,
      "paper present empirical": 84098,
      "shed light important": 108153,
      "important directions future": 52146,
      "future research learning": 46304,
      "effects different design": 34982,
      "findings highlight importance": 44301,
      "proposed method consistently": 93358,
      "paper propose methodology": 84263,
      "techniques reinforcement learning": 119967,
      "syntactic semantic level": 116469,
      "giving best results": 48215,
      "perform indepth analysis": 86011,
      "model generalization ability": 71227,
      "significantly better baselines": 108884,
      "comes following merits": 18780,
      "faster convergence speed": 43169,
      "learning framework crosslingual": 62596,
      "data recent studies": 26325,
      "training data easily": 123455,
      "better performance model": 13660,
      "effect machine translation": 34600,
      "models generate text": 73289,
      "set syntactic rules": 107596,
      "experiments demonstrate model": 40887,
      "select important sentences": 104701,
      "method empirically verify": 68798,
      "shared task model": 108065,
      "task model achieves": 118395,
      "model joint learning": 71379,
      "learning framework design": 62598,
      "provide additional contextual": 93753,
      "human evaluation code": 50804,
      "evaluation code available": 39148,
      "paper aim generate": 83722,
      "analysis provides evidence": 5785,
      "evaluation metrics measure": 39290,
      "lack training evaluation": 58764,
      "training evaluation datasets": 123618,
      "paper propose constrained": 84225,
      "constrained decoding technique": 21567,
      "generation automatic human": 47316,
      "generate text remarkable": 47033,
      "remains open research": 98816,
      "open research question": 81930,
      "text paper introduce": 121165,
      "performance highly dependent": 86431,
      "abstractive summarization performance": 1817,
      "models robust noisy": 73967,
      "achieve better comparable": 2428,
      "100 training examples": 170,
      "performance using automatic": 86830,
      "performance different datasets": 86298,
      "model separate generation": 71982,
      "model using additional": 72273,
      "using additional supervision": 129323,
      "sentencelevel extractive summarization": 106159,
      "textual information model": 121713,
      "ability natural language": 1548,
      "modelbased evaluation methods": 72352,
      "experiments proposed methods": 41078,
      "extracted question answering": 42173,
      "reaches similar performance": 95917,
      "sentence level despite": 105927,
      "model sentence level": 71974,
      "text vectorization models": 121404,
      "using single unified": 130182,
      "single unified model": 109809,
      "additional human annotations": 3521,
      "critical natural language": 24819,
      "serve baseline future": 107289,
      "baseline future work": 12229,
      "documents prior work": 33273,
      "focused generating short": 44859,
      "new method generating": 79012,
      "approach present results": 7810,
      "insights future research": 55532,
      "correlation human scores": 24236,
      "manual linguistic quality": 67010,
      "background knowledge paper": 11308,
      "achieving strong improvements": 2989,
      "systems generally rely": 116918,
      "lowresource languages like": 65517,
      "overcome problem propose": 83297,
      "large training datasets": 61302,
      "bart lewis et": 11436,
      "lewis et al": 63720,
      "gpt2 radford et": 48568,
      "lowresource domain adaptation": 65497,
      "domains data available": 33756,
      "data available paper": 25678,
      "available paper present": 11069,
      "similarity pretraining data": 109284,
      "pretraining data target": 90246,
      "learning mtl approach": 62806,
      "auxiliary training objectives": 10898,
      "neural models code": 78169,
      "improve performance approach": 52457,
      "specific language model": 111457,
      "evaluation remains open": 39359,
      "work proposed evaluation": 134751,
      "proposed evaluation metrics": 93274,
      "contains relevant information": 21835,
      "relevant information source": 98557,
      "extensive experiments report": 41930,
      "paper propose graphbased": 84249,
      "datasets results demonstrate": 27686,
      "effectiveness model compared": 34914,
      "evaluating proposed methods": 39087,
      "evaluation human judges": 39234,
      "trained single reference": 123280,
      "existing models use": 40211,
      "indomain crossdomain settings": 53959,
      "outperform prior stateoftheart": 82733,
      "previous work text": 90559,
      "research community paper": 100447,
      "graph extensive experiments": 48809,
      "generate highquality summaries": 46953,
      "lack large parallel": 58727,
      "large parallel training": 61203,
      "different sources model": 31442,
      "compared conventional methods": 19351,
      "method evaluating quality": 68817,
      "embeddings propose new": 35887,
      "previous work indicates": 90532,
      "documentlevel discourse trees": 33138,
      "different application scenarios": 30996,
      "remain largely unknown": 98771,
      "address limitation introduce": 3713,
      "text summarization lexical": 121347,
      "paper present detailed analysis": 84097,
      "semantic role labeling semantic": 105256,
      "containing hundreds thousands documentsummary": 21794,
      "hundreds thousands documentsummary pairs": 51155,
      "human evaluation generated summaries": 50817,
      "automatic evaluation human evaluation": 10529,
      "sequence sequence seq2seq model": 107065,
      "approach achieves promising results": 7315,
      "controlling output length neural": 22859,
      "models shown great success": 74029,
      "neural networks recently shown": 78550,
      "abstractive summarization machine translation": 1812,
      "work propose novel architecture": 134735,
      "using novel crowdsourcing approach": 129981,
      "outperforms stateoftheart baseline models": 83000,
      "abstractive summarization abstractive summarization": 1810,
      "introduce neural network model": 56470,
      "important problem natural language": 52213,
      "method achieves stateoftheart results": 68610,
      "chinese social media text": 16823,
      "significant improvements state art": 108806,
      "aims reduce linguistic complexity": 4570,
      "success text generation tasks": 115134,
      "proposed model outperforms stateoftheart": 93469,
      "model outperforms stateoftheart systems": 71677,
      "stateoftheart systems benchmark corpus": 112987,
      "main ideas document set": 66430,
      "experimental results different datasets": 40606,
      "significant performance gains compared": 108839,
      "correlate better human judgments": 24193,
      "widely used social media": 132587,
      "results approach outperforms baselines": 101516,
      "using neural networks paper": 129959,
      "neural networks paper present": 78531,
      "existing neural network models": 40233,
      "systems perform standard datasets": 117061,
      "important challenging research topic": 52116,
      "approach outperforms existing models": 7772,
      "dataset experimental results model": 26913,
      "model achieves stateoftheart performances": 70603,
      "achieves stateoftheart performances benchmark": 2896,
      "based integer linear programming": 11777,
      "meaning representation amr semantic": 67672,
      "representation amr semantic representation": 99169,
      "amr semantic representation natural": 5397,
      "semantic representation natural language": 105230,
      "endtoend neural network framework": 36946,
      "conduct extensive experiments datasets": 20873,
      "latent semantic analysis approach": 61603,
      "weighting schemes calculating input": 132369,
      "schemes calculating input matrix": 103950,
      "linguistic data consortium ldc": 64460,
      "experimental results datasets effectiveness": 40581,
      "datasets effectiveness proposed model": 27443,
      "performs comprehensively better compared": 86999,
      "comprehensively better compared stateoftheart": 20288,
      "better compared stateoftheart methods": 13544,
      "model achieves results comparable": 70587,
      "paper extend previous work": 83939,
      "deep learning reinforcement learning": 28309,
      "policy gradient reinforcement learning": 87955,
      "using generative adversarial networks": 129703,
      "approaches obtained encouraging results": 8263,
      "yields significantly better performance": 135433,
      "experiments deep learning models": 40874,
      "community question answering forums": 19093,
      "performance comparable stateoftheart systems": 86226,
      "multitask learning extractive summarization": 75837,
      "vast amounts training data": 131274,
      "pretrained language model representations": 90057,
      "range language understanding tasks": 95574,
      "machine translation abstractive summarization": 65882,
      "natural language descriptions source": 76282,
      "language descriptions source code": 58948,
      "models successful natural language": 74123,
      "proposed method improves performance": 93374,
      "variety language understanding tasks": 130986,
      "paper attempts bridge gap": 83748,
      "variety natural language understanding": 131004,
      "new state art performance": 79151,
      "training data available task": 123433,
      "issue propose novel approach": 57026,
      "model achieves competitive results": 70569,
      "novel deep learning method": 80538,
      "automatic human evaluation demonstrate": 10552,
      "novel task automatically generating": 80745,
      "conduct extensive analysis biases": 20867,
      "using deep learning architectures": 129543,
      "correlation human ratings previous": 24235,
      "model code publicly available": 70839,
      "context paper propose novel": 22212,
      "tend achieve best performance": 120141,
      "sequencetosequence model paper presents": 107174,
      "attention plays key role": 9972,
      "natural language inference fact": 76329,
      "graph convolutional networks experiments": 48787,
      "stateoftheart methods significant margin": 112714,
      "small fraction model parameters": 110153,
      "model gives best results": 71253,
      "finetuned downstream nlp tasks": 44425,
      "nlp tasks including text": 79727,
      "tasks including text summarization": 119193,
      "paper propose aggregation mechanism": 84217,
      "present detailed error analysis": 89448,
      "order address issues propose": 82273,
      "model outperformed previous models": 71633,
      "limited training data available": 64293,
      "attention mechanism computes attention": 9891,
      "wmt machine translation tasks": 132797,
      "task demonstrate effectiveness proposed": 118059,
      "text experimental results showed": 120935,
      "sequencetosequence seq2seq learning problem": 107205,
      "enable new research directions": 36362,
      "outperforms strong unsupervised baselines": 83032,
      "performance comparable stateoftheart supervised": 86225,
      "stateoftheart encoderdecoder model model": 112653,
      "proposed method achieves new": 93343,
      "consumer health question answering": 21722,
      "tasks question answering language": 119428,
      "inverse document frequency idf": 56699,
      "effectiveness proposed method benchmark": 34932,
      "pretrained transformerbased language models": 90203,
      "machine translation automatic postediting": 65906,
      "automatic postediting machine translation": 10626,
      "pipeline approaches suffer error": 87526,
      "approaches suffer error propagation": 8362,
      "suffer error propagation work": 115220,
      "error propagation work propose": 38322,
      "reinforcement learning directly optimize": 97802,
      "language models based transformer": 59512,
      "including sentiment analysis question": 53371,
      "strong natural language understanding": 113692,
      "results multiple benchmark datasets": 101969,
      "new benchmark dataset crosslingual": 78813,
      "method significantly outperforms baseline": 69144,
      "content selection surface realization": 21942,
      "effects different design choices": 34983,
      "results proposed method consistently": 102070,
      "proposed method consistently outperforms": 93359,
      "training data recent studies": 123523,
      "datasets model achieves stateoftheart": 27572,
      "shared task model achieves": 108066,
      "human evaluation code available": 50805,
      "lack training evaluation datasets": 58765,
      "stateoftheart abstractive summarization methods": 112570,
      "generation automatic human evaluations": 47317,
      "neural language models generate": 77943,
      "remains open research question": 98817,
      "models robust noisy data": 73968,
      "ability natural language understanding": 1549,
      "using single unified model": 130183,
      "serve baseline future work": 107290,
      "paper present new method": 84122,
      "require large training datasets": 100168,
      "bart lewis et al": 11437,
      "gpt2 radford et al": 48569,
      "similarity pretraining data target": 109285,
      "multitask learning mtl approach": 75853,
      "evaluation remains open research": 39360,
      "results demonstrate effectiveness model": 101676,
      "parallel training data required": 84696,
      "fundamental problem natural language processing": 46108,
      "containing hundreds thousands documentsummary pairs": 21795,
      "model outperforms stateoftheart baseline models": 71671,
      "proposed method achieves stateoftheart results": 93346,
      "experiments proposed model outperforms stateoftheart": 41082,
      "meaning representation amr semantic representation": 67673,
      "representation amr semantic representation natural": 99170,
      "amr semantic representation natural language": 5398,
      "weighting schemes calculating input matrix": 132370,
      "performs comprehensively better compared stateoftheart": 87000,
      "comprehensively better compared stateoftheart methods": 20289,
      "natural language descriptions source code": 76283,
      "variety natural language understanding tasks": 131005,
      "task demonstrate effectiveness proposed approach": 118060,
      "pipeline approaches suffer error propagation": 87527,
      "suffer error propagation work propose": 115221,
      "language models based transformer architecture": 59513,
      "terms automatic metrics human evaluation": 120276,
      "datasets model achieves stateoftheart performance": 27573,
      "gpt2 radford et al 2019": 48570,
      "medstract": 68249,
      "thesauruses": 121844,
      "nile": 79485,
      "unlocking": 127461,
      "650": 1180,
      "visits": 131782,
      "meddra": 68076,
      "pharmacovigilance": 87211,
      "diagnoses": 30501,
      "dive": 32781,
      "curse": 25427,
      "medication": 68232,
      "grey": 49083,
      "banned": 11423,
      "advertised": 4193,
      "semeval2013": 105523,
      "supplies": 115950,
      "iaa": 51297,
      "abounds": 1710,
      "rocauc": 103143,
      "prevented": 90368,
      "deidentified": 28587,
      "4500": 1036,
      "insurance": 55731,
      "therapeutic": 121830,
      "delicate": 28608,
      "0885": 94,
      "dr": 34080,
      "confidencebased": 20998,
      "similaritybased": 109345,
      "retrofitting": 102475,
      "entitypair": 38143,
      "wise": 132738,
      "513": 1094,
      "inpatient": 55289,
      "25k": 834,
      "accumulated": 2052,
      "beginners": 12633,
      "bachelor": 11283,
      "characteraware": 16483,
      "israel": 56986,
      "ade": 3821,
      "vi": 131636,
      "coders": 18196,
      "metamap": 68560,
      "clef": 17743,
      "mantra": 66970,
      "composable": 20056,
      "chemicaldisease": 16709,
      "cdr": 15876,
      "customizable": 25457,
      "misspelling": 70343,
      "maybe": 67558,
      "enrolled": 37572,
      "doubleannotated": 33965,
      "screening": 104246,
      "cohorts": 18337,
      "cad": 15012,
      "shanghai": 107934,
      "knowledgeguided": 58278,
      "inputting": 55500,
      "aiassisted": 4429,
      "novice": 80793,
      "heightened": 49678,
      "highimpact": 50241,
      "longitudinally": 65174,
      "timestamps": 122201,
      "nonsystematic": 80300,
      "062": 45,
      "f1measures": 42703,
      "cohort": 18336,
      "depended": 29052,
      "debugging": 27905,
      "abnormalities": 1707,
      "8701": 1338,
      "disseminated": 32392,
      "10m": 210,
      "bigrucrf": 14012,
      "nonexact": 80199,
      "911": 1384,
      "microaverage": 70039,
      "0815": 79,
      "radiological": 95448,
      "va": 130694,
      "084": 86,
      "treegru": 125673,
      "pr": 88695,
      "curves": 25431,
      "reflex": 97626,
      "nonrelation": 80287,
      "approval": 8449,
      "089": 96,
      "xray": 135218,
      "hedge": 49675,
      "8851": 1349,
      "122": 267,
      "uncertaintybased": 126629,
      "082": 80,
      "060": 43,
      "045": 28,
      "018": 8,
      "035": 19,
      "ser": 107263,
      "nondevelopers": 80190,
      "holistically": 50603,
      "2609": 845,
      "topicattention": 122601,
      "324": 936,
      "sequencing": 107213,
      "triage": 125760,
      "handselected": 49426,
      "0916": 105,
      "highlystructured": 50355,
      "computerbased": 20507,
      "november": 80792,
      "characterizations": 16546,
      "headache": 49581,
      "reframe": 97643,
      "0826": 81,
      "administrative": 3885,
      "top10": 122483,
      "8582": 1327,
      "9224": 1397,
      "scribe": 104250,
      "lognormal": 65048,
      "747": 1244,
      "calm": 15076,
      "rnng": 102950,
      "874": 1341,
      "887": 1351,
      "computeraided": 20502,
      "hassle": 49527,
      "robotics": 103011,
      "nonmedical": 80239,
      "partnership": 85520,
      "stanzas": 112388,
      "variably": 130861,
      "ada": 3164,
      "729": 1230,
      "ccks": 15861,
      "046": 29,
      "upsampled": 127826,
      "063": 47,
      "066": 50,
      "subdomain": 114665,
      "lite": 64726,
      "invested": 56709,
      "depression": 29320,
      "infused": 55179,
      "bookcorpus": 14418,
      "337": 950,
      "221": 778,
      "steadily": 113222,
      "lighten": 64001,
      "imagetext": 51817,
      "monologues": 74637,
      "overconfident": 83311,
      "languageunderstanding": 60997,
      "mixup": 70429,
      "uncertainties": 126612,
      "t1": 117246,
      "clock": 17806,
      "827": 1307,
      "linguisticbased": 64602,
      "runner": 103466,
      "iot": 56924,
      "twosentence": 126156,
      "overestimated": 83313,
      "mutation": 75967,
      "textmining": 121443,
      "consolidating": 21505,
      "drugbank": 34199,
      "kbbert": 57505,
      "plant": 87648,
      "actors": 3138,
      "agriculture": 4415,
      "bulletins": 14962,
      "overlooking": 83347,
      "institutional": 55700,
      "distress": 32606,
      "expeditious": 40407,
      "registered": 97684,
      "synthesise": 116601,
      "streamlining": 113558,
      "intersections": 56305,
      "stochasticity": 113362,
      "resources number": 101023,
      "number language": 80904,
      "type coverage": 126192,
      "aimed developing": 4482,
      "developing tools": 30364,
      "need robust": 76856,
      "robust reliable": 103065,
      "performance information": 86464,
      "data overall": 26202,
      "work include": 134565,
      "extraction proposed": 42453,
      "systems unlike": 117216,
      "proposed semisupervised": 93546,
      "gained increasing": 46362,
      "support applications": 115956,
      "events using": 39610,
      "design performance": 29559,
      "nile natural": 79486,
      "share medical": 107955,
      "community methods": 19082,
      "algorithm named": 4773,
      "prefix suffix": 89254,
      "state machines": 112506,
      "faster existing": 43173,
      "nlp challenge": 79583,
      "data conclusion": 25769,
      "addition nlp": 3454,
      "unified architecture": 127104,
      "symptom severity": 116332,
      "chapter provides": 16412,
      "resources framework": 100979,
      "systems differ": 116851,
      "efforts biomedical": 35189,
      "large publicly": 61226,
      "collaborative work": 18363,
      "synthetic feature": 116628,
      "engineering natural": 37051,
      "sources study": 110921,
      "variables tested": 130859,
      "based length": 11800,
      "text mathematical": 121109,
      "predicting risk": 89007,
      "scalable experiments": 103690,
      "pairs proposed": 83613,
      "field science": 43978,
      "collection analysis": 18457,
      "essential support": 38566,
      "support work": 116020,
      "report analysis": 98975,
      "work manually": 134629,
      "experts manual": 41242,
      "terms time": 120394,
      "processing algorithm": 91611,
      "algorithm able": 4686,
      "analysis practical": 5759,
      "practical viewpoint": 88725,
      "impact quality": 51886,
      "tend occur": 120156,
      "narrative materials": 76174,
      "method injects": 68910,
      "estimate conditional": 38628,
      "output existing": 83073,
      "12 relative": 256,
      "improvement fscore": 52712,
      "prediction target": 89132,
      "cost annotation": 24345,
      "annotation textual": 6388,
      "domains end": 33772,
      "strategies address": 113448,
      "successfully exploit": 115185,
      "15 fscore": 346,
      "newswire domain": 79421,
      "data hand": 25994,
      "demonstrate match": 28778,
      "domain learning": 33569,
      "existing reviews": 40272,
      "extraction deep": 42298,
      "years extracting": 135265,
      "subject research": 114683,
      "challenges relation": 16202,
      "process detecting": 91458,
      "classifying semantic": 17676,
      "use manually": 128137,
      "fed classifier": 43799,
      "methods highly": 69534,
      "dependent quality": 29264,
      "quality user": 94822,
      "cnn learn": 18007,
      "experts knowledge": 41241,
      "features handcrafted": 43533,
      "scope features": 104038,
      "word dependencies": 132985,
      "dependencies sentence": 29111,
      "sentence work": 106132,
      "embedded features": 35354,
      "task indicate": 118290,
      "indicate stateoftheart": 53851,
      "obtained relying": 81402,
      "extraction study": 42496,
      "embeddings sequence": 35933,
      "representation active": 99161,
      "clinical concepts": 17761,
      "reduce manual": 97336,
      "baseline features": 12224,
      "approach comparative": 7430,
      "investigated using": 56845,
      "clinical datasets": 17764,
      "extraction unstructured": 42525,
      "involves extraction": 56892,
      "presents set": 89897,
      "set challenges": 107390,
      "objectives work": 81142,
      "networks extend": 77598,
      "previously studied": 90623,
      "lstmcrf models": 65693,
      "modeling pairwise": 72500,
      "prediction order": 89096,
      "entities distant": 37768,
      "supervision relation": 115911,
      "especially cases": 38437,
      "extraction underexplored": 42524,
      "sentence relation": 106023,
      "extraction core": 42291,
      "core approach": 23309,
      "representation incorporate": 99258,
      "features multiple": 43615,
      "error experiments": 38306,
      "experiments important": 40957,
      "small existing": 110150,
      "existing distant": 40115,
      "supervision paradigm": 115905,
      "relations similar": 98257,
      "possibly leading": 88448,
      "set known": 107471,
      "effects corpora": 34978,
      "initial seeds": 55223,
      "results feature": 101792,
      "texts classify": 121473,
      "classify predefined": 17653,
      "resources difficult": 100966,
      "difficult collect": 31613,
      "lstm crf": 65612,
      "crf decoding": 24747,
      "decoding performing": 28121,
      "semeval2013 task": 105524,
      "workshop natural": 135007,
      "processing challenges": 91635,
      "resources hard": 100985,
      "hard collect": 49454,
      "ranked systems": 95651,
      "systems building": 116763,
      "comprehensive syntactic": 20280,
      "annotations chinese": 6414,
      "texts corresponding": 121487,
      "baselines research": 12455,
      "proposed train": 93573,
      "annotators develop": 6490,
      "using annotation": 129337,
      "annotations partofspeech": 6449,
      "tags entities": 117473,
      "evaluate annotation": 38786,
      "results syntactic": 102247,
      "parsing trees": 85270,
      "corpus number": 23906,
      "introduced improve": 56578,
      "improve corpus": 52360,
      "conclusions study": 20731,
      "method chinese": 68692,
      "corpus nlp": 23903,
      "problem low": 91111,
      "based support": 12092,
      "terms medical": 120346,
      "knowledgebased features": 58251,
      "corpora evaluated": 23473,
      "medical terms": 68227,
      "surpassed stateoftheart": 116101,
      "respectively model": 101150,
      "leading cause": 61887,
      "cause death": 15820,
      "death china": 27885,
      "annotations based": 6411,
      "light annotation": 63987,
      "task included": 118283,
      "annotations applied": 6409,
      "based unstructured": 12147,
      "investigation using": 56866,
      "compared previously": 19422,
      "especially pronounced": 38493,
      "work related": 134767,
      "problems consider": 91308,
      "like benefit": 64019,
      "instead exploit": 55661,
      "exploit raw": 41437,
      "wikipedia knowledge": 132661,
      "learned representation": 62249,
      "predict probable": 88918,
      "classes introduce": 17062,
      "features memory": 43606,
      "memory experiments": 68314,
      "narrative descriptions": 76169,
      "texts according": 121449,
      "narrative documents": 76170,
      "expensive task": 40433,
      "efficient terms": 35118,
      "terms computational": 120296,
      "linear size": 64372,
      "terms proposed": 120365,
      "proposal based": 92519,
      "effort identify": 35175,
      "data items": 26054,
      "techniques represent": 119970,
      "variant word2vec": 130871,
      "feed word": 43820,
      "lstm architecture": 65592,
      "obtained promising": 81397,
      "effects drug": 34984,
      "important concern": 52130,
      "features shown": 43727,
      "utilize word": 130537,
      "embedding latent": 35423,
      "features rely": 43697,
      "rely explicit": 98692,
      "use bidirectional": 127924,
      "bilstm layer": 14096,
      "dataset joint": 26990,
      "relying handcrafted": 98760,
      "proposed lstm": 93326,
      "language umls": 60199,
      "sharing neural": 108136,
      "consistently yields": 21444,
      "strategies exploit": 113462,
      "medical condition": 68198,
      "potentially harmful": 88612,
      "update evaluating": 127791,
      "published literature": 94359,
      "ensemble ranking": 37619,
      "finding important": 44275,
      "walk algorithm": 131996,
      "single views": 109813,
      "use medical": 128139,
      "compared benchmark": 19336,
      "benchmark unsupervised": 12870,
      "identification performance": 51420,
      "metrics performance": 69991,
      "help develop": 49710,
      "readily applied": 95965,
      "applied domains": 7066,
      "biomedical named": 14195,
      "handlabeled data": 49368,
      "like lexicons": 64056,
      "supervision use": 115927,
      "ner taggers": 77086,
      "biomedical ner": 14200,
      "supervised benchmarks": 115733,
      "benchmarks using": 12938,
      "20 teams": 530,
      "networks represent": 77733,
      "network dataset": 77216,
      "extraction recurrent": 42464,
      "network multiple": 77362,
      "sentence comprehension": 105796,
      "model classifies": 70830,
      "methods representation": 69727,
      "identification novel": 51412,
      "aspects associated": 9372,
      "structured databases": 113997,
      "abundant information": 1834,
      "free texts": 45769,
      "methods shown": 69746,
      "unstructured texts": 127587,
      "matrix completion": 67473,
      "approach feature": 7571,
      "vectors learn": 131449,
      "learn projection": 62119,
      "spaces proposed": 111102,
      "exciting field": 39952,
      "techniques ner": 119939,
      "ner relation": 77071,
      "used solving": 128770,
      "means studying": 67769,
      "require preprocessing": 100187,
      "use automated": 127904,
      "systems common": 116792,
      "demonstrate machine": 28775,
      "modeling topic": 72569,
      "used bagofwords": 128405,
      "different measures": 31250,
      "provides efficient": 94022,
      "lstm bidirectional": 65604,
      "results bidirectional": 101554,
      "proposed systems": 93562,
      "embeddings helped": 35737,
      "avoid costly": 11232,
      "need retrained": 76854,
      "short survey": 108245,
      "contradictions disagreements": 22669,
      "distinction contradictions": 32552,
      "motivated need": 74854,
      "transferable neural": 124222,
      "domain lack": 33565,
      "sufficient labeled": 115282,
      "algorithms real": 4877,
      "useful domains": 128879,
      "task source": 118726,
      "source task": 110849,
      "task target": 118775,
      "quite common": 95398,
      "present unified": 89751,
      "aspects related": 9405,
      "related source": 97897,
      "tasks similarity": 119501,
      "choice particular": 16895,
      "framework unified": 45727,
      "unified neural": 127123,
      "specific particular": 111473,
      "task depends": 118067,
      "architectures shown": 8847,
      "contextual morphological": 22482,
      "linear conditional": 64339,
      "label tag": 58417,
      "tag sequence": 117340,
      "dictionary words": 30901,
      "models standard": 74085,
      "crosssentence nary": 25075,
      "relations single": 98258,
      "sentences recent": 106467,
      "general setting": 46714,
      "various intrasentential": 131106,
      "intrasentential intersentential": 56345,
      "intersentential dependencies": 56313,
      "sequential syntactic": 107255,
      "learning distant": 62508,
      "analysis extraction": 5618,
      "deidentification medical": 28581,
      "focuses deidentification": 44894,
      "based conditional": 11591,
      "crfs long": 24767,
      "manually extracted": 67062,
      "decoding layer": 28100,
      "representations medical": 99752,
      "relatedness estimation": 97920,
      "fall categories": 43026,
      "structure fail": 113864,
      "distributional context": 32696,
      "concepts using": 20650,
      "similarity vector": 109329,
      "representations leads": 99722,
      "reported date": 99059,
      "date demonstrate": 27823,
      "demonstrate clear": 28687,
      "clear improvement": 17723,
      "representations compared": 99552,
      "associations using": 9636,
      "mining methods": 70240,
      "diseases genes": 32340,
      "knowledge structured": 58193,
      "rely simple": 98739,
      "tends yield": 120175,
      "spurious associations": 112061,
      "manually curated": 67054,
      "information access": 54352,
      "challenging rewarding": 16309,
      "user intentions": 129000,
      "introduce graphbased": 56432,
      "based multitask": 11872,
      "introduced extract": 56577,
      "extract structured": 42125,
      "structured semantic": 114042,
      "task realworld": 118605,
      "mention pairs": 68424,
      "biological relation": 14161,
      "sentence containing": 105806,
      "single entity": 109733,
      "entity pair": 37979,
      "mention relation": 68428,
      "expressed sentences": 41720,
      "sentences require": 106478,
      "context disambiguate": 22063,
      "selfattention encoder": 104877,
      "mention pair": 68423,
      "training noise": 123752,
      "performance jointly": 86475,
      "predict named": 88904,
      "additional corpus": 3504,
      "weakly labeled": 132191,
      "models kb": 73436,
      "ensembles models": 37626,
      "features additional": 43357,
      "reports written": 99092,
      "written spanish": 135148,
      "annotated entities": 6190,
      "terms relations": 120373,
      "relations corpus": 98131,
      "extraction algorithms": 42261,
      "input use": 55464,
      "make work": 66750,
      "investigate leverage": 56777,
      "learning technologies": 63100,
      "score best": 104055,
      "adapting language": 3310,
      "modeling style": 72556,
      "based overall": 11920,
      "representations written": 100000,
      "design attention": 29517,
      "descriptions corresponding": 29477,
      "090 f1": 101,
      "respectively result": 101160,
      "encoding method": 36711,
      "way provide": 132126,
      "explore usage": 41607,
      "stacked denoising": 112124,
      "representations directly": 99598,
      "input effective": 55327,
      "task gaining": 118230,
      "endtoend need": 36937,
      "rely highquality": 98708,
      "transfer leads": 124072,
      "fewer training": 43892,
      "score identifying": 104084,
      "health conditions": 49617,
      "developing methods": 30347,
      "associations cooccurring": 9631,
      "report preliminary": 99024,
      "results datadriven": 101655,
      "datadriven study": 26679,
      "topics topic": 122659,
      "obtained topics": 81423,
      "augmenting word": 10349,
      "extremely large": 42600,
      "datasets tend": 27749,
      "magnitude smaller": 66388,
      "smaller dataset": 110231,
      "encoding domain": 36689,
      "predict primary": 88916,
      "results imply": 101839,
      "parameters traditional": 84789,
      "meaningful features": 67718,
      "easily adopted": 34435,
      "setting provide": 107775,
      "diagnosis prediction": 30505,
      "medical findings": 68207,
      "discriminating positive": 32199,
      "detect negative": 29813,
      "including public": 53354,
      "corpus general": 23815,
      "datasets demonstrates": 27415,
      "detecting negative": 29843,
      "improvement precision": 52743,
      "extraction significant": 42488,
      "significant problem": 108848,
      "model ade": 70637,
      "inspiration machine": 55557,
      "designing model": 29642,
      "exploit local": 41428,
      "entities extract": 37783,
      "adverse reactions": 4191,
      "interaction text": 55972,
      "visualize understand": 131836,
      "network makes": 77313,
      "endtoend solution": 36964,
      "using seq2seq": 130154,
      "experiments directly": 40911,
      "directly applying": 31864,
      "novel decoder": 80532,
      "mechanism novel": 68021,
      "promote development": 92316,
      "running text": 103470,
      "submission ensemble": 114729,
      "based majority": 11825,
      "task 2017": 117820,
      "methods recurrent": 69710,
      "outperformed conventional": 82780,
      "specific purpose": 111482,
      "purpose tool": 94442,
      "resolves ambiguities": 100813,
      "level annotation": 63420,
      "models external": 73218,
      "provides foundation": 94033,
      "foundation downstream": 45343,
      "advances demonstrated": 3985,
      "demonstrated efficacy": 28915,
      "extracting concepts": 42205,
      "tool includes": 122391,
      "networks chinese": 77533,
      "body parts": 14395,
      "research recent": 100607,
      "knowledge provide": 58128,
      "handling rare": 49420,
      "network different": 77226,
      "representation schemes": 99401,
      "schemes proposed": 103952,
      "proposed handle": 93306,
      "patterns improve": 85750,
      "improve information": 52394,
      "demonstrate features": 28744,
      "models linear": 73510,
      "relevant categories": 98529,
      "mentions texts": 68469,
      "normalization using": 80353,
      "trained multilingual": 123207,
      "trained spanish": 123290,
      "french dutch": 45806,
      "carried french": 15540,
      "showing outperforms": 108420,
      "clef ehealth": 17744,
      "2015 2016": 582,
      "applications medical": 6966,
      "privacy restrictions": 90782,
      "set predefined": 107535,
      "distinguish entities": 32570,
      "model detect": 70979,
      "consists important": 21479,
      "problem english": 91029,
      "words chinese": 133857,
      "characters compound": 16603,
      "units attention": 127232,
      "relationship exists": 98295,
      "molecular structure": 74535,
      "molecular structures": 74536,
      "representation discriminative": 99213,
      "training input": 123663,
      "input dataset": 55316,
      "realistic text": 96107,
      "unconditional conditional": 126640,
      "contain syntactic": 21764,
      "lstm generator": 65625,
      "classification achieved": 17101,
      "domain study": 33662,
      "matrix experiments": 67474,
      "depend external": 29047,
      "common neural": 18899,
      "architectures cnn": 8790,
      "lstm learn": 65636,
      "representations improve": 99683,
      "obtaining stateoftheart": 81450,
      "mining noisy": 70248,
      "sources paper": 110915,
      "given threshold": 48159,
      "dataset prepared": 27100,
      "posts generated": 88517,
      "terms ii": 120337,
      "base fully": 11459,
      "performance significant": 86711,
      "relative simplicity": 98375,
      "simplicity proposed": 109572,
      "literature present": 64765,
      "annotations include": 6439,
      "annotations diverse": 6426,
      "outline set": 82633,
      "heavily manual": 49653,
      "study aim": 114310,
      "train support": 123033,
      "obtained high": 81376,
      "auc score": 10218,
      "categorization novel": 15766,
      "algorithm development": 4719,
      "construction computational": 21685,
      "integrating heterogeneous": 55794,
      "important shown": 52252,
      "achieved combining": 2604,
      "advances challenges": 3977,
      "definitions context": 28534,
      "equivalent entities": 38249,
      "ontologies propose": 81832,
      "definition context": 28519,
      "capable encoding": 15198,
      "addition external": 3429,
      "results f1score": 101791,
      "care research": 15505,
      "patient records": 85703,
      "records standard": 97138,
      "clinical practice": 17784,
      "curated dataset": 25244,
      "ones used": 81708,
      "training enables": 123605,
      "currently lack": 25407,
      "results participation": 102030,
      "participation task": 85367,
      "challenge addressed": 15991,
      "codes french": 18204,
      "dictionarybased approach": 30903,
      "provided task": 93985,
      "achieved fscore": 2627,
      "average score": 11213,
      "leading causes": 61888,
      "eventually lead": 39619,
      "standard technique": 112318,
      "classification rulebased": 17385,
      "rules knowledge": 103412,
      "critical steps": 24834,
      "trigger phrases": 125787,
      "challenge results": 16096,
      "methods relation": 69716,
      "architecture exploits": 8664,
      "advantages cnn": 4071,
      "simultaneously identify": 109679,
      "features cnn": 43402,
      "feature dependencies": 43261,
      "methods datasets": 69419,
      "datasets learning": 27546,
      "spend significant": 111922,
      "significant time": 108872,
      "discuss models": 32269,
      "element extraction": 35268,
      "annotation improve": 6324,
      "expert novice": 41228,
      "elements related": 35281,
      "annotation time": 6389,
      "difference associated": 30939,
      "deal complex": 27855,
      "classification classify": 17145,
      "classify sentences": 17660,
      "problem sequential": 91224,
      "classification structured": 17421,
      "help classify": 49704,
      "domain state": 33657,
      "good learning": 48485,
      "outputs lack": 83172,
      "limited address": 64201,
      "address gap": 3681,
      "task nli": 118459,
      "strategies leverage": 113473,
      "leverage transfer": 63627,
      "data lexical": 26087,
      "using strategies": 130227,
      "characterlevel embeddings": 16564,
      "bilstmcrf models": 14109,
      "leads comparable": 61926,
      "tasks methods": 119296,
      "network conditional": 77205,
      "characters dictionary": 16604,
      "network capture": 77182,
      "neighboring tags": 76995,
      "understanding mental": 126888,
      "similarity entailment": 109227,
      "annotate large": 6129,
      "statistical dialogue": 113093,
      "prone errors": 92332,
      "propose automate": 92565,
      "task learns": 118349,
      "existing nonneural": 40242,
      "nonneural neural": 80255,
      "baselines rouge": 12457,
      "work represents": 134781,
      "entity boundaries": 37909,
      "combination convolutional": 18555,
      "biomedical datasets": 14172,
      "datasets labeled": 27535,
      "costly time": 24390,
      "processing tool": 91841,
      "reduction using": 97465,
      "27 30": 849,
      "labeling results": 58533,
      "viable means": 131643,
      "implementation evaluation": 51943,
      "order text": 82419,
      "text considered": 120813,
      "allowed use": 5106,
      "use dictionaries": 127998,
      "dictionaries knowledge": 30860,
      "actionable information": 3084,
      "relevant wikipedia": 98601,
      "reported baseline": 99058,
      "yields superior": 135447,
      "datasets diverse": 27435,
      "dataset identifying": 26969,
      "generative modeling": 47740,
      "modeling improve": 72444,
      "improve automated": 52337,
      "data field": 25946,
      "problem amplified": 90931,
      "proposed trained": 93574,
      "modeling algorithm": 72372,
      "algorithm automated": 4698,
      "learn discriminative": 62024,
      "additionally trained": 3634,
      "supervised finetuning": 115756,
      "performance compare": 86227,
      "information critical": 54461,
      "history challenging": 50547,
      "exactly matched": 39725,
      "performance labeled": 86478,
      "classification tc": 17463,
      "research used": 100660,
      "containing sentences": 21806,
      "sentences shorter": 106493,
      "embedding deep": 35393,
      "vectors semantic": 131474,
      "cluster center": 17925,
      "vector represents": 131365,
      "words cluster": 133862,
      "finally expand": 44181,
      "respectively evaluate": 101135,
      "domain joint": 33553,
      "detection clinical": 29897,
      "multitask problem": 75888,
      "encoder followed": 36527,
      "followed separate": 44952,
      "performs considerably": 87001,
      "learningbased systems": 63187,
      "dataset exploring": 26918,
      "semisupervised variational": 105633,
      "literature provides": 64768,
      "primarily based": 90638,
      "creating large": 24699,
      "data prohibitively": 26284,
      "methods capable": 69359,
      "reduce burden": 97315,
      "corpora experimental": 23481,
      "dependence labeled": 29054,
      "exploiting unlabeled": 41485,
      "using spelling": 130207,
      "spelling error": 111903,
      "mining increasingly": 70235,
      "biomedical documents": 14173,
      "researchers deep": 100682,
      "effective biomedical": 34637,
      "distribution shift": 32679,
      "corpora article": 23413,
      "bert adapted": 13059,
      "mining domainspecific": 70231,
      "largely outperforms": 61344,
      "bert previous": 13209,
      "comparable previous": 19167,
      "062 f1": 46,
      "corpora helps": 23492,
      "texts make": 121546,
      "advanced natural": 3950,
      "extraction recently": 42463,
      "recently advanced": 96656,
      "pushed stateoftheart": 94463,
      "integrate representations": 55766,
      "extraction including": 42356,
      "semeval 2014": 105496,
      "like elmo": 64036,
      "understand semantic": 126773,
      "encoded contextual": 36472,
      "clinical corpus": 17762,
      "extraction additionally": 42254,
      "encode valuable": 36464,
      "representations relation": 99849,
      "explicit context": 41308,
      "entities raw": 37855,
      "text current": 120847,
      "contextaware representations": 22347,
      "target entities": 117614,
      "complex long": 19830,
      "times target": 122189,
      "help automate": 49697,
      "subtasks corresponding": 114966,
      "subtasks using": 114988,
      "rules supervised": 103446,
      "lexicons developed": 63924,
      "approach allowed": 7346,
      "improved overall": 52623,
      "simple machine": 109462,
      "accuracy automated": 2103,
      "development potential": 30410,
      "length text": 63378,
      "capture prior": 15362,
      "transferred downstream": 124228,
      "tasks embedding": 119085,
      "domains survey": 33865,
      "presents detailed": 89839,
      "detailed review": 29785,
      "processing survey": 91806,
      "comparison popular": 19562,
      "popular embeddings": 88092,
      "methods followed": 69517,
      "solutions various": 110584,
      "way obtain": 132111,
      "approach useful": 7999,
      "taking step": 117553,
      "direction present": 31828,
      "samples positive": 103589,
      "standard used": 112334,
      "constraints results": 21611,
      "data 13": 25549,
      "obtained overall": 81389,
      "participant teams": 85309,
      "requires explicit": 100269,
      "concepts carry": 20606,
      "relations human": 98183,
      "relations fundamental": 98177,
      "extraction tools": 42518,
      "automatically recognize": 10821,
      "tools require": 122470,
      "obtain promising": 81309,
      "learning tools": 63109,
      "relations linguistic": 98204,
      "complexity makes": 19923,
      "texts analyze": 121454,
      "heuristic methods": 49874,
      "logic models": 64991,
      "little manual": 64821,
      "work reports": 134779,
      "particular medical": 85430,
      "sense paper": 105693,
      "fulltext articles": 45937,
      "using suite": 130237,
      "task believe": 117926,
      "texts facilitate": 121519,
      "problem make": 91120,
      "evaluation available": 39129,
      "biomedical embeddings": 14178,
      "models contextualized": 72980,
      "domainspecific corpora": 33893,
      "determine additional": 30119,
      "compare bert": 19228,
      "bert elmo": 13108,
      "elmo trained": 35325,
      "ner nli": 77060,
      "elmo peters": 35320,
      "2018 bert": 656,
      "models minimally": 73568,
      "publiclyavailable pretrained": 94339,
      "address need": 3719,
      "specifically demonstrate": 111537,
      "embeddings domainspecific": 35669,
      "major factors": 66559,
      "factors training": 42902,
      "work recently": 134762,
      "concept frequency": 20572,
      "high stability": 50140,
      "finding propose": 44282,
      "showing linear": 108412,
      "linear correlation": 64344,
      "documents french": 33232,
      "respectively hybrid": 101144,
      "variety document": 130970,
      "gain especially": 46340,
      "near state": 76713,
      "various attributes": 131047,
      "paper outlines": 84062,
      "mechanism handle": 67991,
      "corpus demonstrates": 23749,
      "nlp approaches": 79576,
      "approaches multiple": 8241,
      "remains major": 98808,
      "challenge study": 16106,
      "nlp developed": 79604,
      "092 fscore": 107,
      "growing concern": 49169,
      "concern modern": 20684,
      "introduces large": 56616,
      "sets task": 107716,
      "sentencelevel relation": 106168,
      "respectively approach": 101124,
      "subtasks including": 114969,
      "second teams": 104460,
      "extraction linking": 42375,
      "detailed domain": 29767,
      "demonstrated potential": 28925,
      "notes propose": 80392,
      "propose bridge": 92577,
      "extracted concepts": 42147,
      "text unfortunately": 121384,
      "possible explanations": 88402,
      "domains improve": 33791,
      "task transfer": 118804,
      "datasets inspired": 27528,
      "consists tasks": 21499,
      "evaluate baselines": 38800,
      "ignore information": 51668,
      "contained sentence": 21776,
      "structures target": 114108,
      "joint embedding": 57270,
      "incorporate entity": 53467,
      "extraction sentencelevel": 42481,
      "construct larger": 21635,
      "dataset distant": 26875,
      "f1 measures": 42661,
      "pr curves": 88696,
      "adopting pretrained": 3904,
      "use domainspecific": 128017,
      "progress nlp": 92169,
      "models adopt": 72725,
      "methods pretrained": 69681,
      "domain empirical": 33516,
      "domain natural": 33590,
      "approaches quantitative": 8302,
      "analysis help": 5637,
      "domains systematic": 33866,
      "extraction difficult": 42305,
      "domains ability": 33722,
      "preprocessing training": 89339,
      "area exploiting": 8871,
      "imbalanced data": 51828,
      "data practical": 26254,
      "scenario relation": 103830,
      "pairs relation": 83626,
      "pairs context": 83499,
      "context negative": 22195,
      "negative instances": 76931,
      "positive instances": 88324,
      "negatively affects": 76964,
      "affects models": 4244,
      "architecture jointly": 8680,
      "trains model": 123964,
      "relation identification": 98040,
      "entities relation": 37859,
      "patterns entities": 85740,
      "embeddings enrich": 35694,
      "effective used": 34772,
      "healthrelated information": 49636,
      "recent automatic": 96432,
      "privacy security": 90783,
      "largescale raw": 61503,
      "new setting": 79133,
      "terms extracted": 120321,
      "extracted clinical": 42144,
      "important types": 52285,
      "data surface": 26523,
      "form information": 45095,
      "enables detect": 36380,
      "complementary role": 19752,
      "allow deal": 5077,
      "various settings": 131194,
      "embedding paper": 35467,
      "evidence based": 39636,
      "build training": 14813,
      "purpose build": 94423,
      "dataset investigating": 26988,
      "purpose leverage": 94431,
      "embedding bidirectional": 35383,
      "optimizes performance": 82241,
      "notes using": 80393,
      "text creating": 120842,
      "rulebased heuristics": 103362,
      "trained support": 123301,
      "results 13": 101484,
      "best measures": 13364,
      "respectively svm": 101171,
      "study showed": 114516,
      "interpretable patterns": 56249,
      "training relation": 123809,
      "new facts": 78915,
      "candidate pairs": 15115,
      "relationship expressed": 98296,
      "complete absence": 19761,
      "data experts": 25927,
      "pairs exist": 83537,
      "pairs fully": 83551,
      "available improving": 11013,
      "complexity paper": 19932,
      "performance bilstmcrf": 86184,
      "bilstmcrf model": 14108,
      "domainspecific resources": 33913,
      "presents submissions": 89906,
      "submissions team": 114755,
      "demonstrate incorporating": 28764,
      "incorporating domain": 53533,
      "powerful strategy": 88677,
      "domains medicine": 33816,
      "understanding spatial": 126968,
      "spatial language": 111267,
      "representation framework": 99244,
      "framework annotation": 45424,
      "spatial information": 111266,
      "terms propose": 120364,
      "learningbased natural": 63183,
      "nlp method": 79635,
      "involving word": 56922,
      "word characterlevel": 132946,
      "model extracting": 71161,
      "step analyzing": 113247,
      "different supervised": 31466,
      "approaches annotated": 8061,
      "parameters machine": 84767,
      "sets used": 107723,
      "instance recurrent": 55607,
      "outputs different": 83164,
      "embeddings biomedical": 35580,
      "networks investigate": 77630,
      "small additional": 110128,
      "performs onpar": 87012,
      "art benchmark": 9036,
      "clear benefit": 17719,
      "typically does": 126422,
      "introduce contextual": 56395,
      "context model": 22179,
      "like pretrained": 64081,
      "leveraging semantic": 63702,
      "showed model": 108381,
      "major boost": 66539,
      "bert joint": 13147,
      "extraction chinese": 42279,
      "text entity": 120917,
      "does achieve": 33324,
      "focused attention": 44847,
      "range attention": 95543,
      "layer experimental": 61715,
      "systems contain": 116812,
      "cws fundamental": 25476,
      "fundamental important": 46098,
      "methods greatly": 69526,
      "rely largescale": 98715,
      "efforts devoted": 35196,
      "paper follow": 83957,
      "model including": 71327,
      "interactions adjacent": 55979,
      "adjacent characters": 3849,
      "bigram features": 14006,
      "learning cws": 62481,
      "learning infer": 62652,
      "properties relations": 92476,
      "properties duration": 92447,
      "tackles challenge": 117319,
      "identifies span": 51472,
      "relations relation": 98247,
      "sentences mainly": 106390,
      "contrast proposed": 22710,
      "infer relations": 54102,
      "10 minutes": 142,
      "identifying relations": 51617,
      "jointly inferring": 57346,
      "respectively significantly": 101165,
      "contributions different": 22786,
      "components model": 20033,
      "fed pretrained": 43807,
      "obtain semantic": 81317,
      "embedded graph": 35355,
      "embedding semantic": 35487,
      "structure embeddings": 113849,
      "combined measure": 18658,
      "based order": 11916,
      "entities kb": 37801,
      "finding evidence": 44272,
      "incomplete address": 53420,
      "representations roberta": 99866,
      "application users": 6892,
      "users search": 129167,
      "search evidence": 104314,
      "information reason": 54897,
      "task developed": 118083,
      "subtask submitted": 114956,
      "event related": 39534,
      "entities extraction": 37785,
      "microaveraged f1score": 70043,
      "taskspecific architecture": 119618,
      "solution task": 110565,
      "mentions particular": 68464,
      "predefined relationship": 88834,
      "parameters finetuned": 84749,
      "solve specific": 110619,
      "encode pair": 36442,
      "context consecutive": 22033,
      "special symbol": 111369,
      "linear layer": 64349,
      "layer classify": 61706,
      "considerable class": 21244,
      "dataset examine": 26900,
      "achieve improved": 2490,
      "approach future": 7583,
      "nested named": 77106,
      "nested entities": 77103,
      "objective tokenlevel": 81123,
      "strategy improve": 113521,
      "perform entity": 85988,
      "features ner": 43623,
      "showed competitive": 108371,
      "algorithms developed": 4842,
      "challenging challenges": 16231,
      "challenges include": 16166,
      "dealing noisy": 27870,
      "sentences objective": 106414,
      "objective research": 81113,
      "interpret models": 56211,
      "demonstrate attentionbased": 28675,
      "medical named": 68213,
      "existing open": 40248,
      "source tools": 110857,
      "scalable does": 103688,
      "test procedure": 120480,
      "contained text": 21777,
      "text furthermore": 120970,
      "step effective": 113261,
      "information study": 55011,
      "representations including": 99687,
      "sequence graph": 106957,
      "novel attentionbased": 80502,
      "ways handle": 132159,
      "model conjunction": 70884,
      "learning performs": 62884,
      "recognition er": 96868,
      "second official": 104435,
      "results absolute": 101488,
      "performance improving": 86455,
      "comparable best": 19128,
      "contextualized neural": 22561,
      "patients health": 85707,
      "rules recent": 103433,
      "abbreviation disambiguation": 1465,
      "analysis clinical": 5530,
      "documents main": 33250,
      "sets address": 107646,
      "applied learn": 7087,
      "improved contextualized": 52600,
      "contextualized sentence": 22567,
      "issue existing": 56995,
      "rare senses": 95743,
      "selected public": 104728,
      "disambiguation dataset": 31961,
      "dataset test": 27235,
      "class term": 17051,
      "unbalanced training": 126607,
      "datasets common": 27356,
      "models involves": 73430,
      "mechanisms sequence": 68069,
      "generate single": 47017,
      "models literature": 73513,
      "determine best": 30121,
      "pooling mechanism": 88042,
      "order answer": 82279,
      "mechanisms deep": 68059,
      "semantic latent": 105094,
      "descriptions lack": 29484,
      "unsupervised deep": 127620,
      "shown proposed": 108517,
      "annotation performance": 6349,
      "relations entity": 98149,
      "dependency syntax": 29232,
      "higher recall": 50201,
      "automatically distinguishing": 10763,
      "benchmarks method": 12919,
      "giving stateoftheart": 48218,
      "results literature": 101892,
      "characteristic language": 16516,
      "language major": 59267,
      "component information": 19984,
      "approaches explored": 8159,
      "problem rulebased": 91210,
      "applying transfer": 7277,
      "review previous": 102562,
      "literature addressing": 64743,
      "corpus explore": 23794,
      "bert popular": 13203,
      "datasets trained": 27764,
      "manual verification": 67020,
      "furthermore developed": 46162,
      "information identification": 54660,
      "identification module": 51400,
      "robustness language": 103104,
      "step developing": 113259,
      "triage patients": 125761,
      "dataset contrast": 26830,
      "random test": 95513,
      "sets best": 107652,
      "examining model": 39769,
      "understanding robustness": 126947,
      "contains wealth": 21843,
      "contain valuable": 21768,
      "information effective": 54513,
      "networkbased approach": 77481,
      "extraction specifically": 42491,
      "employs bert": 36329,
      "structure instance": 113884,
      "knowledge respectively": 58155,
      "performance sentences": 86700,
      "overlapping relations": 83338,
      "supplementary information": 115941,
      "phrase extraction": 87348,
      "incorporating local": 53547,
      "context topic": 22300,
      "topic inference": 122530,
      "detected latent": 29825,
      "available results": 11097,
      "terms topic": 120395,
      "extraction published": 42455,
      "published scientific": 94371,
      "additional support": 3577,
      "kbs contain": 57511,
      "relation embeddings": 97973,
      "extraction prior": 42447,
      "extraction firstly": 42340,
      "mutual attention": 75971,
      "embedding knowledge": 35419,
      "extraction experiments": 42335,
      "adding knowledge": 3385,
      "knowledge named": 58072,
      "obtained similar": 81411,
      "dictionarybased systems": 30907,
      "english past": 37236,
      "knowledgebased methods": 58253,
      "text essential": 120920,
      "strong support": 113722,
      "extraction make": 42384,
      "knowledge representations": 58146,
      "including entity": 53292,
      "used control": 128451,
      "embeddings employed": 35685,
      "features shared": 43726,
      "containing additional": 21782,
      "contexts performance": 22417,
      "extraction automatically": 42268,
      "extraction previous": 42446,
      "proposes neural": 93605,
      "networkbased attention": 77483,
      "pair entities": 83429,
      "mechanism employed": 67972,
      "dataset combining": 26800,
      "performance achieve": 86113,
      "results conclusions": 101618,
      "query formulation": 94963,
      "criteria usually": 24798,
      "relations complex": 98125,
      "base training": 11486,
      "extraction dependency": 42300,
      "information prior": 54868,
      "pairs relations": 83627,
      "important syntactic": 52261,
      "improve relation": 52518,
      "convolutional attention": 23150,
      "firstly extract": 44564,
      "extract shortest": 42119,
      "shortest dependency": 108300,
      "path sdp": 85678,
      "sentence includes": 105903,
      "finally order": 44210,
      "order combine": 82296,
      "knowledge play": 58101,
      "important roles": 52248,
      "task knowledgeaware": 118329,
      "methods need": 69634,
      "encoded context": 36471,
      "context experiment": 22090,
      "lead new": 61863,
      "performance leveraging": 86496,
      "efforts paper": 35204,
      "learned knowledge": 62223,
      "embeddings prior": 35882,
      "multiple computational": 75521,
      "computational layers": 20386,
      "medical conditions": 68199,
      "datasets tools": 27762,
      "using official": 129986,
      "performance set": 86707,
      "set random": 107553,
      "entities tend": 37878,
      "tend capture": 120143,
      "example input": 39785,
      "design twostage": 29569,
      "entities like": 37813,
      "task multilabel": 118414,
      "synthesis process": 116600,
      "process essential": 91478,
      "essential achieving": 38539,
      "concept identification": 20576,
      "helpful improve": 49796,
      "information informative": 54686,
      "using pure": 130069,
      "approaches difficult": 8128,
      "ontology semantic": 81849,
      "identify correct": 51488,
      "detection concept": 29905,
      "200 billion": 533,
      "errors natural": 38387,
      "aid human": 4433,
      "human coders": 50775,
      "bert finetune": 13128,
      "80 data": 1288,
      "help human": 49731,
      "additional costs": 3506,
      "motivated goal": 74851,
      "used annotation": 128390,
      "corpus 6k": 23635,
      "train stateoftheart": 123027,
      "results entities": 101760,
      "entities related": 37858,
      "taken models": 117511,
      "applications named": 6969,
      "automatic information": 10567,
      "extraction case": 42274,
      "like named": 64069,
      "relevance detection": 98506,
      "detection additionally": 29870,
      "entities available": 37748,
      "analysis rely": 5804,
      "analyses text": 5471,
      "statistical property": 113155,
      "property language": 92499,
      "dataset methods": 27024,
      "law distributions": 61683,
      "data testing": 26555,
      "reports best": 99080,
      "text benefit": 120684,
      "spanish clinical": 111150,
      "spanish corpus": 111151,
      "research currently": 100455,
      "exhaustive comparison": 39991,
      "presents main": 89865,
      "present large": 89530,
      "merges information": 68480,
      "unique entries": 127185,
      "dictionary entry": 30879,
      "results manual": 101906,
      "achieved unsupervised": 2718,
      "quickly adapt": 95389,
      "drug reaction": 34197,
      "detection user": 30093,
      "consumer reviews": 21723,
      "labels sentences": 58640,
      "include healthrelated": 53226,
      "corpus macro": 23873,
      "task achieved": 117835,
      "achieves macro": 2815,
      "trained russian": 123266,
      "russian data": 103488,
      "corpus pretrained": 23938,
      "domainspecific bert": 33890,
      "number specific": 80970,
      "high capacity": 50039,
      "robust supervised": 103076,
      "demonstrate practice": 28823,
      "gap supervised": 46482,
      "introduced task": 56598,
      "currently major": 25409,
      "major limitation": 66571,
      "forms different": 45241,
      "crossentropy objective": 24915,
      "second dataset": 104399,
      "scores correlation": 104164,
      "outperforms skipgram": 82984,
      "network grammars": 77270,
      "texts evaluated": 121509,
      "models relations": 73909,
      "relations extraction": 98173,
      "bilstm detecting": 14092,
      "learn richer": 62134,
      "specific grammatical": 111447,
      "grammatical characteristics": 48690,
      "designed nlp": 29612,
      "finally derive": 44165,
      "methods support": 69787,
      "build classifiers": 14752,
      "study focusing": 114387,
      "easily implemented": 34456,
      "defines identity": 28509,
      "studies attempted": 114184,
      "using unstructured": 130343,
      "highly significant": 50345,
      "range classification": 95548,
      "models logistic": 73524,
      "lstms transformers": 65718,
      "interpretable framework": 56240,
      "statistics global": 113200,
      "predict relations": 88927,
      "prediction conduct": 89043,
      "dataset framework": 26941,
      "mining multitask": 70241,
      "pairwise mtl": 83677,
      "helpful context": 49794,
      "suitable model": 115404,
      "multichannel cnn": 75041,
      "curated experts": 25246,
      "publications methods": 94282,
      "studies utilizing": 114298,
      "rich textual": 102796,
      "bridge semantic": 14599,
      "provide extra": 93825,
      "knowledge textual": 58207,
      "datasets help": 27501,
      "help knowledge": 49742,
      "knowledge embedding": 57887,
      "multiple channels": 75509,
      "accuracy biomedical": 2110,
      "embeddings enhance": 35692,
      "effectively treat": 34857,
      "disease condition": 32334,
      "database evidence": 26641,
      "prohibitively timeconsuming": 92207,
      "depend manually": 29050,
      "manually compiled": 67045,
      "research end": 100487,
      "instance does": 55599,
      "paper collect": 83762,
      "stronger baseline": 113733,
      "version task": 131619,
      "code new": 18123,
      "improving transfer": 53177,
      "bert performance": 13200,
      "domain medical": 33580,
      "specific bert": 111411,
      "domains publicly": 33842,
      "use terms": 128324,
      "texts information": 121529,
      "texts used": 121643,
      "work combine": 134414,
      "corpus 10k": 23629,
      "filtered using": 44079,
      "technological developments": 120020,
      "knowledge improving": 58009,
      "correlations arguments": 24251,
      "approach addressing": 7336,
      "addressing limitations": 3812,
      "result ranking": 101399,
      "performing soft": 86971,
      "approach challenging": 7415,
      "automatic labeling": 10572,
      "labels appear": 58577,
      "target concept": 117590,
      "normalization helps": 80335,
      "string matching": 113598,
      "research approach": 100409,
      "normalization text": 80350,
      "main drawback": 66417,
      "information learning": 54725,
      "input concept": 55308,
      "mention representation": 68429,
      "approach need": 7734,
      "overcomes drawbacks": 83303,
      "using roberta": 130123,
      "initialized updated": 55243,
      "updated training": 127803,
      "methods standard": 69769,
      "semantically complex": 105365,
      "complex unstructured": 19891,
      "population task": 88169,
      "solution combining": 110550,
      "word2vec embedding": 133670,
      "release resources": 98471,
      "view current": 131704,
      "correct spelling": 24121,
      "text assess": 120663,
      "predicted correctly": 88958,
      "corpus german": 23818,
      "corpora major": 23523,
      "communities english": 19058,
      "data protection": 26298,
      "evaluating existing": 39053,
      "fast adaptation": 43129,
      "efficient tool": 35122,
      "temporal expression": 120103,
      "text extracted": 120944,
      "covid19 patients": 24572,
      "covid19 named": 24568,
      "entities identified": 37792,
      "utilized downstream": 130539,
      "effectiveness contextualized": 34878,
      "relaxed f1score": 98425,
      "accurate syntactic": 2366,
      "largescale unsupervised": 61518,
      "signal detection": 108684,
      "identify potential": 51536,
      "motivating development": 74874,
      "convenient way": 22872,
      "entities capture": 37753,
      "embeddings knowledge": 35770,
      "improvements larger": 52867,
      "reference sets": 97534,
      "corpora additionally": 23404,
      "additionally quantitative": 3626,
      "metric exists": 69881,
      "better test": 13742,
      "test sentence": 120490,
      "intended purpose": 55869,
      "stateoftheart biomedical": 112605,
      "types highly": 126294,
      "accurate robust": 2361,
      "robust variations": 103082,
      "text genre": 121025,
      "nlp framework": 79618,
      "code availability": 18073,
      "flair framework": 44634,
      "event extractors": 39520,
      "covariate shift": 24477,
      "data aligning": 25589,
      "adaptation ada": 3206,
      "model likelihood": 71461,
      "deeper analysis": 28442,
      "semantic shift": 105273,
      "reach f1": 95889,
      "models tried": 74249,
      "traditional sequence": 122866,
      "ensembling models": 37631,
      "diversity final": 32874,
      "ccks 2019": 15862,
      "task solving": 118725,
      "pipeline propose": 87552,
      "propose step": 93084,
      "overlapping triples": 83339,
      "faced challenge": 42748,
      "binary trees": 14139,
      "joint relation": 57315,
      "triples efficiently": 125812,
      "25 f1": 815,
      "protocols using": 93678,
      "data imbalance": 26017,
      "minority classes": 70274,
      "performance described": 86287,
      "described approach": 29375,
      "gradient boosting": 48598,
      "forest rf": 45052,
      "achieved macroaveraged": 2652,
      "labeling framework": 58500,
      "framework implement": 45563,
      "method advantage": 68622,
      "utilizing welldesigned": 130584,
      "decoding processes": 28128,
      "achieving f1scores": 2946,
      "used extraction": 128546,
      "entities common": 37756,
      "study date": 114345,
      "manually construct": 67046,
      "annotating entities": 6267,
      "mentions propose": 68465,
      "predict normalized": 88908,
      "concept results": 20587,
      "hybrid method": 51182,
      "recognition unstructured": 97036,
      "works address": 134918,
      "proposed identify": 93309,
      "recognize unseen": 97058,
      "training difficult": 123587,
      "cost human": 24358,
      "recognize human": 97052,
      "use concepts": 127954,
      "construct dictionary": 21628,
      "dictionary used": 30897,
      "trained classify": 123089,
      "label finally": 58397,
      "finally dictionary": 44168,
      "favorably previous": 43217,
      "demonstrate generalizability": 28746,
      "generalizability method": 46750,
      "increased exponentially": 53628,
      "research challenging": 100442,
      "types concepts": 126262,
      "general corpora": 46647,
      "corpora designed": 23452,
      "loss used": 65304,
      "techniques lower": 119924,
      "text unified": 121385,
      "numerous approaches": 81014,
      "models necessitates": 73618,
      "relations example": 98160,
      "concept relation": 20584,
      "mtl framework": 75016,
      "selfattentive network": 104911,
      "approach prediction": 7807,
      "multiple problems": 75647,
      "shared representation": 108005,
      "path embedding": 85673,
      "mtl models": 75018,
      "baselines deep": 12382,
      "compromising performance": 20340,
      "attention framework": 9842,
      "needed understand": 76880,
      "provide deeper": 93799,
      "identify relationships": 51543,
      "use depression": 127996,
      "model domainspecific": 71027,
      "declarative knowledge": 27972,
      "drug abuse": 34195,
      "sharing mechanism": 108135,
      "model tailored": 72135,
      "ontology used": 81852,
      "used locate": 128616,
      "locate target": 64955,
      "bert extract": 13126,
      "improving low": 53111,
      "encoding target": 36730,
      "providing useful": 94144,
      "output embeddings": 83071,
      "embeddings does": 35667,
      "does improve": 33358,
      "multidomain language": 75093,
      "pretrained generic": 90038,
      "generic data": 47795,
      "additional domain": 3510,
      "generic benchmark": 47792,
      "benchmark glue": 12829,
      "remaining competitive": 98780,
      "furthermore explore": 46174,
      "latent clustering": 61577,
      "using elastic": 129615,
      "language growing": 59115,
      "need applications": 76782,
      "domain address": 33463,
      "20k english": 751,
      "quality best": 94606,
      "met existing": 68529,
      "resources field": 100977,
      "experiments 20": 40753,
      "light ability": 63986,
      "different views": 31559,
      "results typically": 102291,
      "typically presented": 126449,
      "baselines finally": 12396,
      "seeking identify": 104521,
      "success capturing": 115063,
      "capturing syntactic": 15496,
      "semantic world": 105357,
      "new disease": 78885,
      "demonstrating viability": 28984,
      "results observed": 102002,
      "code freely": 18104,
      "led impressive": 63276,
      "architecture incorporating": 8672,
      "embedding prediction": 35468,
      "layers experiments": 61777,
      "tasks comparing": 118995,
      "approaches applying": 8064,
      "context determine": 22054,
      "finegrained spatial": 44385,
      "simple domain": 109399,
      "rules extracted": 103401,
      "planning based": 87637,
      "deidentification task": 28586,
      "random word": 95521,
      "test setting": 120536,
      "domain benchmarks": 33472,
      "works study": 134981,
      "additionally study": 3633,
      "size pretraining": 109940,
      "improvements benchmarks": 52814,
      "larger domain": 61363,
      "understanding domain": 126833,
      "important tool": 52281,
      "learning components": 62448,
      "employed novel": 36306,
      "improving factual": 53097,
      "offer potential": 81570,
      "process report": 91560,
      "errors existing": 38367,
      "performances natural": 86885,
      "complete consistent": 19762,
      "uses natural": 129252,
      "combine novel": 18635,
      "novel use": 80768,
      "semantic equivalence": 105044,
      "datasets substantially": 27736,
      "evaluation qualitative": 39346,
      "knowledge unit": 58221,
      "uncertainty knowledge": 126620,
      "education research": 34569,
      "biomedical fields": 14186,
      "triples extracted": 125813,
      "knowledge general": 57947,
      "knowledge largely": 58038,
      "largely overlooked": 61345,
      "evolves time": 39701,
      "approaches detect": 8122,
      "research wide": 100663,
      "data insufficiency": 26046,
      "ample evidence": 5370,
      "data quite": 26314,
      "problem quantitatively": 91191,
      "classification trained": 17471,
      "stateoftheart rulebased": 112940,
      "used clinical": 128425,
      "relational memory": 98095,
      "memory decoder": 68308,
      "decoder transformer": 28058,
      "work reporting": 134778,
      "knowledge analyses": 57749,
      "tools extract": 122448,
      "transformerbased bert": 124403,
      "classification utilize": 17498,
      "layer finetuning": 61720,
      "finetuning process": 44507,
      "process bert": 91431,
      "improves bert": 52951,
      "addition analysis": 3398,
      "particularly relevant": 85494,
      "incomplete paper": 53423,
      "techniques support": 119992,
      "dataset german": 26956,
      "realworld utility": 96190,
      "demonstrate promising": 28828,
      "attribute extraction": 10180,
      "prediction drug": 89053,
      "drug effects": 34196,
      "applied address": 7043,
      "problem methods": 91127,
      "leading degradation": 61890,
      "reliability estimate": 98606,
      "applied deep": 7057,
      "uncertainty information": 126619,
      "training addition": 123349,
      "confidence penalty": 20993,
      "model retrained": 71926,
      "performance regard": 86665,
      "approaches furthermore": 8168,
      "normalization paper": 80341,
      "close vector": 17838,
      "learning medical": 62716,
      "similarities calculated": 109176,
      "terms relation": 120372,
      "relation triplets": 98077,
      "learning features": 62579,
      "biomedical word": 14215,
      "codes models": 18206,
      "selection suitable": 104833,
      "establish guidelines": 38582,
      "result high": 101376,
      "kappa score": 57493,
      "agreement given": 4393,
      "transformerbased classification": 124406,
      "decoding results": 28130,
      "metrics major": 69983,
      "question ability": 95013,
      "achieving absolute": 2925,
      "sota language": 110689,
      "benefit evaluation": 12966,
      "complete picture": 19766,
      "research explored": 100500,
      "explored classification": 41620,
      "dataset capturing": 26777,
      "contextual aspects": 22444,
      "task 2021": 117825,
      "recently methods": 96708,
      "developed extracting": 30276,
      "degree confidence": 28568,
      "polarity sentences": 87926,
      "modality based": 70485,
      "estimate confidence": 38629,
      "linking aims": 64652,
      "specific challenge": 111415,
      "context biomedical": 22021,
      "morphological variations": 74756,
      "computing resources": 20527,
      "limits applications": 64315,
      "resourcelimited scenarios": 100905,
      "fraction parameters": 45367,
      "parameters bert": 84738,
      "uses simple": 129278,
      "mechanisms capture": 68058,
      "capture variations": 15418,
      "predict properties": 88919,
      "properties representations": 92478,
      "tasks instance": 119213,
      "bert family": 13127,
      "ner tagging": 77087,
      "range linguistic": 95577,
      "traditional bert": 122801,
      "paper going": 83970,
      "runner terms": 103467,
      "embeddings transformer": 35990,
      "methods capture": 69360,
      "global distributional": 48235,
      "architectures designed": 8799,
      "designed optimize": 29614,
      "dynamic embeddings": 34306,
      "added benefit": 3356,
      "paper jointly": 84031,
      "classification objective": 17311,
      "conduct evaluations": 20841,
      "datasets word": 27800,
      "believe offer": 12701,
      "generalize novel": 46816,
      "identifying new": 51606,
      "unseen entity": 127524,
      "effectively handle": 34816,
      "systematically analyze": 116691,
      "generalization current": 46773,
      "cases models": 15657,
      "identify difficulties": 51493,
      "unseen mentions": 127533,
      "debiasing method": 27900,
      "expressed multiple": 41717,
      "concepts including": 20622,
      "literature existing": 64755,
      "detects relations": 30103,
      "lstm attention": 65594,
      "recently transformer": 96772,
      "outperform lstm": 82715,
      "lstm natural": 65647,
      "architecture utilizes": 8771,
      "current token": 25379,
      "mechanism method": 68011,
      "neighbor tokens": 76988,
      "tasks benchmarking": 118960,
      "relation datasets": 97964,
      "entity sentence": 38102,
      "information rapid": 54894,
      "knowledge hidden": 58001,
      "corpus particular": 23926,
      "propose multihead": 92794,
      "utilizes pretrained": 130556,
      "generation sentences": 47612,
      "rules evaluated": 103397,
      "evaluated classification": 38961,
      "applied classification": 7050,
      "highquality generated": 50384,
      "approach measure": 7706,
      "objective work": 81128,
      "work aimed": 134365,
      "aimed demonstrate": 4481,
      "model retrofitting": 71931,
      "algorithm compute": 4707,
      "trained bert": 123079,
      "datasets allow": 27305,
      "allow models": 5085,
      "better finally": 13589,
      "datasets includes": 27516,
      "effective compared": 34647,
      "scores cases": 104160,
      "recognition linking": 96903,
      "tools explore": 122447,
      "users ability": 129088,
      "developed facilitate": 30277,
      "semantic indexing": 105071,
      "does reach": 33382,
      "recognition problems": 96971,
      "problems studied": 91377,
      "low resources": 65397,
      "challenges generalizing": 16164,
      "umls ontology": 126566,
      "f1 pts": 42672,
      "knowledge enables": 57891,
      "aspects improve": 9388,
      "particular french": 85416,
      "information development": 54480,
      "production written": 92063,
      "machines human": 66334,
      "nlp technologies": 79780,
      "bertbased approach": 13267,
      "using selfsupervised": 130138,
      "text prevent": 121200,
      "support especially": 115968,
      "task common": 117975,
      "outcomes based": 82619,
      "integrate knowledge": 55755,
      "public sources": 94274,
      "sources present": 110917,
      "outcome tasks": 82617,
      "reveals strengths": 102522,
      "including current": 53281,
      "capturing dependencies": 15464,
      "makes accurate": 66753,
      "accurate sentence": 2363,
      "limitations new": 64182,
      "ngrams using": 79469,
      "mechanism capture": 67956,
      "datasets comprise": 27368,
      "contain different": 21736,
      "review natural": 102557,
      "study systematically": 114530,
      "nlp applied": 79574,
      "search yields": 104364,
      "review analysis": 102533,
      "studies reporting": 114272,
      "085 f1": 88,
      "important domain": 52151,
      "clinical use": 17796,
      "transformerbased generation": 124411,
      "image model": 51788,
      "sequencetosequence paradigm": 107193,
      "step improve": 113273,
      "stateoftheart benchmark": 112598,
      "sentence objectives": 105960,
      "annotations improve": 6438,
      "relation annotations": 97939,
      "annotations annotated": 6406,
      "087 f1": 92,
      "researchers contribute": 100681,
      "quantitative study": 94885,
      "code prediction": 18131,
      "architecture engineering": 8656,
      "paper conducts": 83781,
      "conducts comprehensive": 20971,
      "distant words": 32482,
      "attention exploit": 9835,
      "information contrary": 54450,
      "demonstrate carefully": 28686,
      "carefully trained": 15529,
      "subset frequent": 114830,
      "directions improving": 31845,
      "rarely discussed": 95757,
      "studies finally": 114228,
      "modern named": 74396,
      "ner methods": 77052,
      "methods determine": 69441,
      "aggregation network": 4365,
      "networks solve": 77777,
      "different medical": 31253,
      "shared modules": 107991,
      "study evaluating": 114371,
      "postediting automatic": 88472,
      "circumvent issue": 16985,
      "study time": 114534,
      "time saving": 122094,
      "present insights": 89519,
      "lessons learnt": 63400,
      "networks surprisingly": 77789,
      "sensitive changes": 105734,
      "groups clinical": 49152,
      "evidence tradeoff": 39675,
      "tradeoff overall": 122778,
      "minority groups": 70275,
      "yield statistically": 135359,
      "analysis medical": 5692,
      "2018 address": 652,
      "utilizes pretraining": 130557,
      "collected variety": 18440,
      "medical sentiment": 68224,
      "models previously": 73796,
      "aimed developing tools": 4483,
      "order improve overall": 82341,
      "information extraction approach": 54566,
      "gained increasing attention": 46363,
      "nile natural language": 79487,
      "finite state machines": 44539,
      "times faster existing": 122173,
      "large publicly available": 61227,
      "engineering natural language": 37052,
      "pairs proposed approach": 83614,
      "task detecting classifying": 118077,
      "apply model output": 7189,
      "model output existing": 71683,
      "results tested method": 102267,
      "suggest proposed method": 115331,
      "apply natural language": 7193,
      "active learning strategies": 3116,
      "extraction deep learning": 42299,
      "neural network recent": 78379,
      "recent years extracting": 96615,
      "semantic relation entities": 105205,
      "existing models task": 40210,
      "biomedical domain use": 14176,
      "highly dependent quality": 50312,
      "handcrafted features based": 49339,
      "crucial role determining": 25162,
      "stateoftheart performance obtained": 112836,
      "benefits word embeddings": 13028,
      "word embeddings features": 133115,
      "manual annotation effort": 66976,
      "using wide range": 130373,
      "results demonstrate significant": 101706,
      "based sequence labeling": 12024,
      "widely used method": 132574,
      "method named entity": 68979,
      "entity recognition information": 38018,
      "recognition information extraction": 96895,
      "neural networks extend": 78482,
      "entities distant supervision": 37769,
      "distant supervision relation": 32476,
      "supervision relation extraction": 115912,
      "relation extraction underexplored": 98032,
      "sentence relation extraction": 106024,
      "core approach graph": 23310,
      "existing distant supervision": 40116,
      "distant supervision paradigm": 32474,
      "propose semisupervised approach": 93033,
      "based small set": 12053,
      "recurrent neural architectures": 97188,
      "classify predefined categories": 17654,
      "networks bidirectional lstm": 77524,
      "bidirectional lstm crf": 13954,
      "workshop natural language": 135008,
      "language processing challenges": 59850,
      "method proposed train": 69080,
      "annotations partofspeech pos": 6450,
      "evaluate annotation quality": 38787,
      "processing information extraction": 91681,
      "annotated corpus results": 6152,
      "model based support": 70746,
      "based support vector": 12093,
      "representations learned unlabeled": 99732,
      "enriching lexical resources": 37567,
      "memory neural networks": 68363,
      "using recent advances": 130085,
      "techniques represent words": 119971,
      "neural network long": 78321,
      "memory lstm architecture": 68323,
      "fed neural network": 43806,
      "obtained promising results": 81398,
      "existing methods model": 40190,
      "handcrafted features features": 49342,
      "similar better performance": 109075,
      "performance existing models": 86352,
      "features paper present": 43645,
      "models perform competitively": 73722,
      "perform competitively stateoftheart": 85969,
      "competitively stateoftheart methods": 19704,
      "medical language umls": 68212,
      "prior knowledge neural": 90709,
      "knowledge neural models": 58085,
      "yields improved performance": 135416,
      "random walk algorithm": 95518,
      "resources like lexicons": 100999,
      "weak supervision use": 132186,
      "labeled datasets training": 58454,
      "achieves competitive scores": 2776,
      "stateoftheart supervised benchmarks": 112978,
      "classification using convolutional": 17486,
      "present approach automatically": 89372,
      "approach automatically classify": 7375,
      "sentence level using": 105934,
      "level using deep": 63519,
      "using deep convolutional": 129539,
      "neural networks represent": 78555,
      "extraction recurrent neural": 42465,
      "language processing propose": 59936,
      "words large corpora": 134028,
      "processing tasks work": 91832,
      "tasks work propose": 119612,
      "propose novel way": 92977,
      "spaces proposed method": 111103,
      "ner relation extraction": 77072,
      "relation extraction paper": 98007,
      "paper provides overview": 84387,
      "demonstrate machine learning": 28776,
      "topic modeling topic": 122547,
      "modeling topic modeling": 72570,
      "commonly used bagofwords": 18967,
      "methods deep learning": 69426,
      "bidirectional lstm bidirectional": 13952,
      "model set baseline": 71992,
      "focus different aspects": 44753,
      "distinction contradictions disagreements": 32553,
      "analysis information retrieval": 5657,
      "network model relation": 77337,
      "model relation classification": 71887,
      "biomedical domain lack": 14175,
      "sufficient labeled data": 115283,
      "lack sufficient training": 58750,
      "transfer learning frameworks": 124095,
      "unified neural architecture": 127124,
      "features specific particular": 43739,
      "work propose unified": 134747,
      "framework using bidirectional": 45734,
      "using bidirectional long": 129398,
      "memory network blstm": 68355,
      "linear conditional random": 64340,
      "use domain specific": 128016,
      "models standard benchmark": 74086,
      "framework outperforms existing": 45640,
      "crosssentence nary relation": 25076,
      "shortterm memory networks": 108344,
      "various intrasentential intersentential": 131107,
      "intrasentential intersentential dependencies": 56346,
      "learning distant supervision": 62509,
      "deidentification medical records": 28582,
      "participating systems team": 85357,
      "based conditional random": 11592,
      "fields crfs long": 43995,
      "crfs long shortterm": 24768,
      "memory networks lstms": 68357,
      "bidirectional lstm network": 13956,
      "similarity relatedness estimation": 109295,
      "based distributional statistics": 11659,
      "text corpora methods": 120829,
      "best results reported": 13442,
      "text mining methods": 121120,
      "based multitask learning": 11873,
      "biological relation extraction": 14162,
      "learning improve models": 62640,
      "improve models performance": 52422,
      "predict named entities": 88905,
      "weakly labeled data": 132192,
      "annotated corpus spanish": 6153,
      "use supervised methods": 128307,
      "work investigate leverage": 134591,
      "deep learning technologies": 28322,
      "f1 score best": 42677,
      "score best baseline": 104056,
      "ones paper propose": 81701,
      "attention mechanism automatically": 9885,
      "stacked denoising autoencoder": 112125,
      "ner fundamental task": 77047,
      "task text mining": 118789,
      "task gaining increasing": 118231,
      "increasing attention recent": 53659,
      "parameters learned endtoend": 84765,
      "approaches rely highquality": 8319,
      "labeled data expensive": 58432,
      "text data improve": 120852,
      "evaluate approach benchmark": 38789,
      "approach benchmark datasets": 7389,
      "leads substantial improvement": 61972,
      "training examples achieve": 123621,
      "f1 score identifying": 42685,
      "augmenting word embeddings": 10350,
      "embeddings surprisingly effective": 35967,
      "surprisingly effective capturing": 116136,
      "training data sizes": 123539,
      "knowledge word embeddings": 58234,
      "embeddings word vectors": 36036,
      "extraction propose new": 42451,
      "propose new algorithm": 92832,
      "model makes use": 71497,
      "network makes use": 77314,
      "makes use local": 66814,
      "task aims automatically": 117865,
      "using seq2seq models": 130155,
      "increasingly important task": 53702,
      "aims promote development": 4563,
      "achieved highest performance": 2641,
      "learning methods recurrent": 62743,
      "deep learningbased methods": 28327,
      "basic language processing": 12527,
      "document sentence level": 33075,
      "entity pair representations": 37980,
      "dataset order magnitude": 27073,
      "magnitude larger existing": 66385,
      "ner sequence tagging": 77077,
      "recent advances demonstrated": 96401,
      "neural networks chinese": 78452,
      "research recent years": 100608,
      "highly competitive performance": 50300,
      "improve information extraction": 52395,
      "translation model trained": 124943,
      "spanish french dutch": 111158,
      "develop neural network": 30217,
      "recurrent units attention": 97271,
      "information propose novel": 54881,
      "features training input": 43766,
      "performance general domain": 86411,
      "word embeddings investigate": 133140,
      "characterbased word representations": 16505,
      "architectures cnn lstm": 8791,
      "models use information": 74276,
      "model learned large": 71436,
      "collection process corpus": 18486,
      "relies heavily manual": 98652,
      "train support vector": 123034,
      "using training dataset": 130305,
      "learning unsupervised learning": 63144,
      "model performance better": 71724,
      "better performance achieved": 13651,
      "use additional information": 127882,
      "provided task organizers": 93986,
      "extract semantic relations": 42114,
      "text classification rulebased": 120762,
      "novel approach combines": 80473,
      "network word embeddings": 77477,
      "method outperforms state": 69027,
      "relation classification convolutional": 97944,
      "methods relation classification": 69717,
      "propose unified architecture": 93128,
      "datasets experiments demonstrate": 27475,
      "spend significant time": 111923,
      "generative models using": 47746,
      "hierarchical neural networks": 49989,
      "overall classification performance": 83220,
      "performance work present": 86864,
      "outperforms stateoftheart results": 83014,
      "models using deep": 74300,
      "data limited address": 26091,
      "leverage transfer learning": 63628,
      "domain knowledge external": 33559,
      "performance gains using": 86404,
      "tasks empirical results": 119090,
      "leads comparable stateoftheart": 61927,
      "comparable stateoftheart performance": 19181,
      "methods achieved significant": 69288,
      "dilated convolutional neural": 31738,
      "neural network conditional": 78290,
      "network conditional random": 77206,
      "dense vector representations": 29020,
      "neural network capture": 78273,
      "sentiment analysis introduce": 106600,
      "analysis introduce new": 5662,
      "embeddings sentence embeddings": 35929,
      "prone errors propose": 92333,
      "radiology reports collected": 95451,
      "existing nonneural neural": 40243,
      "approach named entity": 7724,
      "words introduce novel": 134014,
      "paper present deep": 84092,
      "incorporating deep contextualized": 53526,
      "concept extraction model": 20570,
      "embedding model trained": 35451,
      "using contextual word": 129484,
      "model yields superior": 72341,
      "yields superior performance": 135448,
      "limited quantity quality": 64267,
      "quality labeled data": 94696,
      "learn discriminative features": 62025,
      "competitive baselines large": 19641,
      "word vectors semantic": 133646,
      "conduct experiments public": 20854,
      "general domain joint": 46653,
      "performs considerably better": 87002,
      "considerably better previous": 21267,
      "especially lowresource settings": 38475,
      "provides rich source": 94070,
      "relation extraction aims": 97980,
      "extraction present novel": 42442,
      "extraction model consists": 42392,
      "using multilayer convolutional": 129912,
      "unlabeled data evaluate": 127383,
      "corpora experimental results": 23482,
      "improve performance reduce": 52480,
      "dependence labeled data": 29055,
      "relation extraction results": 98025,
      "exploiting unlabeled data": 41486,
      "improved performance various": 52628,
      "spelling error correction": 111904,
      "improvement compared previous": 52695,
      "mining increasingly important": 70236,
      "investigate recently introduced": 56808,
      "stateoftheart models variety": 112754,
      "advanced natural language": 3951,
      "clinical nlp tasks": 17780,
      "representations elmo bert": 99617,
      "large language model": 61127,
      "language model like": 59343,
      "contextual embeddings pretrained": 22461,
      "representations relation extraction": 99850,
      "entities raw text": 37856,
      "target entities used": 117617,
      "proposed method leads": 93378,
      "rules supervised machine": 103447,
      "results achieved overall": 101494,
      "simple machine learning": 109463,
      "transferred downstream tasks": 124229,
      "representations deep learning": 99586,
      "paper presents detailed": 84164,
      "language processing survey": 59947,
      "present brief overview": 89399,
      "weakly supervised method": 132197,
      "samples positive negative": 103590,
      "relations paper presents": 98233,
      "obtain promising results": 81310,
      "promising results stateoftheart": 92301,
      "present new task": 89593,
      "neural architectures demonstrate": 77839,
      "language models contextualized": 59531,
      "derived pretrained language": 29358,
      "pretraining domainspecific corpora": 90255,
      "performance paper conduct": 86595,
      "downstream task models": 34030,
      "models elmo peters": 73127,
      "elmo peters et": 35321,
      "al 2018 bert": 4629,
      "2018 bert devlin": 657,
      "yields performance improvements": 135427,
      "word embeddings work": 133256,
      "stability word embeddings": 112108,
      "performance supervised model": 86776,
      "ner systems using": 77084,
      "near state art": 76714,
      "method constructed corpus": 68736,
      "build automatic systems": 14745,
      "remains major challenge": 98809,
      "high precision recall": 50106,
      "information extraction models": 54579,
      "test sets task": 120532,
      "sentencelevel relation extraction": 106169,
      "test sets respectively": 120530,
      "clinical notes propose": 17782,
      "propose bridge gap": 92578,
      "domains improve performance": 33792,
      "performance proposed methods": 86633,
      "task transfer learning": 118805,
      "understanding evaluation benchmark": 126838,
      "benchmark facilitate research": 12824,
      "evaluate baselines based": 38801,
      "bert model pretrained": 13171,
      "entity relation extraction": 38092,
      "models usually ignore": 74308,
      "information contained sentence": 54444,
      "target entities paper": 117615,
      "relation extraction sentencelevel": 98027,
      "dataset distant supervision": 26876,
      "stateoftheart baselines terms": 112596,
      "improving performance natural": 53137,
      "domain empirical results": 33517,
      "results demonstrate superiority": 101707,
      "superiority proposed methods": 115709,
      "domain natural language": 33591,
      "methods relation extraction": 69718,
      "relation extraction difficult": 97988,
      "future research area": 46293,
      "embeddings multitask learning": 35817,
      "entity pairs relation": 37982,
      "affects models performance": 4245,
      "mitigate problem propose": 70376,
      "sentence multiple entities": 105955,
      "positive negative instances": 88327,
      "improve performance baseline": 52458,
      "performance baseline model": 86167,
      "outperform stateoftheart models": 82760,
      "used improve models": 128582,
      "texts paper study": 121573,
      "information global context": 54642,
      "experiments case studies": 40818,
      "paper investigate new": 84019,
      "clinical notes using": 17783,
      "trained support vector": 123302,
      "network cnn models": 77196,
      "training relation extraction": 123810,
      "model paper present": 71693,
      "order learn better": 82353,
      "base completion task": 11449,
      "word representations contextualized": 133452,
      "elmo word representations": 35327,
      "paper presents submissions": 84207,
      "uses multitask objective": 129251,
      "work explore different": 134510,
      "explore different strategies": 41540,
      "incorporating domain knowledge": 53534,
      "domain knowledge data": 33558,
      "learningbased natural language": 63184,
      "processing nlp method": 91743,
      "biomedical named entities": 14196,
      "learning algorithms applied": 62324,
      "model improve performance": 71309,
      "instance recurrent neural": 55608,
      "models combined using": 72925,
      "state art benchmark": 112429,
      "provide clear benefit": 93774,
      "typically does provide": 126423,
      "representation model called": 99330,
      "knowledge graph context": 57961,
      "nlp tasks finetuning": 79722,
      "joint entity relation": 57273,
      "relation extraction chinese": 97981,
      "layer experimental results": 61716,
      "recognition relation classification": 96985,
      "learning chinese word": 62431,
      "segmentation cws fundamental": 104582,
      "active learning method": 3114,
      "propose joint model": 92732,
      "embedded graph convolutional": 35356,
      "address challenge present": 3655,
      "model using labeled": 72282,
      "shared task use": 108103,
      "task use data": 118825,
      "networks pretrained language": 77704,
      "language models explore": 59552,
      "word representations semantic": 133471,
      "subtask submitted systems": 114957,
      "embeddings crf layer": 35625,
      "semantic features extracted": 105052,
      "features extracted using": 43510,
      "large unannotated corpus": 61310,
      "word embeddings semantic": 133204,
      "relation extraction pretrained": 98014,
      "pretrained language representations": 90111,
      "shared tasks provide": 108112,
      "task aims extract": 117866,
      "language model learns": 59341,
      "contextual language representations": 22480,
      "large unlabelled corpus": 61317,
      "considerable class imbalance": 21245,
      "relation extraction neural": 98006,
      "nested named entity": 77107,
      "recognition ner relation": 96937,
      "relation extraction proposed": 98020,
      "showed competitive performance": 108372,
      "challenging challenges include": 16232,
      "learning models used": 62796,
      "interpretable text classification": 56254,
      "achieve high classification": 2474,
      "medical named entity": 68214,
      "open source tools": 81938,
      "scalable does require": 103689,
      "information contained text": 54445,
      "data additionally propose": 25579,
      "entity recognition er": 38014,
      "prior best results": 90695,
      "contextualized neural language": 22562,
      "pretrained contextualized language": 90012,
      "contextualized sentence representations": 22569,
      "information improves performance": 54674,
      "relation extraction deep": 97986,
      "achieved stateoftheart performances": 2706,
      "features perform prediction": 43653,
      "employ different strategies": 36260,
      "conduct comprehensive study": 20832,
      "yielding stateoftheart performance": 135386,
      "challenging work propose": 16347,
      "unsupervised deep learning": 127621,
      "relations entity mentions": 98150,
      "entity mentions text": 37974,
      "address problem rulebased": 3750,
      "applying transfer learning": 7278,
      "outperforming previous stateoftheart": 82821,
      "text critical step": 120844,
      "bidirectional transformer bert": 13978,
      "contain valuable information": 21769,
      "use information improve": 128095,
      "results paper propose": 102025,
      "neural networkbased approach": 78421,
      "fed softmax function": 43809,
      "corpus proposed model": 23945,
      "stateoftheart models experimental": 112732,
      "local global context": 64922,
      "results outperforms stateoftheart": 102021,
      "published scientific literature": 94372,
      "entity relation embeddings": 38091,
      "capture important context": 15315,
      "knowledge named entity": 58073,
      "human annotated datasets": 50742,
      "results paper proposes": 102026,
      "proposed model learns": 93453,
      "paper proposes neural": 84359,
      "neural networkbased attention": 78423,
      "use context information": 127959,
      "attention mechanism employed": 9897,
      "comparable results stateoftheart": 19172,
      "knowledge base training": 57792,
      "results method achieve": 101916,
      "effectiveness efficiency proposed": 34887,
      "efficiency proposed method": 35042,
      "relation extraction dependency": 97987,
      "entity pairs relations": 37983,
      "information helps improve": 54655,
      "improve relation extraction": 52519,
      "convolutional attention network": 23151,
      "extraction firstly extract": 42341,
      "extract shortest dependency": 42120,
      "shortest dependency path": 108301,
      "dependency path sdp": 29214,
      "features attention mechanism": 43371,
      "knowledge representations learned": 58147,
      "prior knowledge play": 90710,
      "knowledge play important": 58102,
      "play important roles": 87699,
      "attention mechanism encode": 9898,
      "context experiment results": 22091,
      "dataset proposed approach": 27117,
      "leading new stateoftheart": 61899,
      "multiple computational layers": 75522,
      "multilabel classification problem": 75151,
      "present novel corpus": 89601,
      "errors natural language": 38388,
      "transformers bert finetune": 124445,
      "finetune language model": 44407,
      "automatic information extraction": 10568,
      "information extraction case": 54568,
      "tasks like named": 119258,
      "like named entity": 64070,
      "available annotated dataset": 10941,
      "power law distributions": 88635,
      "perform preliminary experiments": 86049,
      "preliminary experiments using": 89271,
      "results manual evaluation": 101907,
      "adaptation pretrained language": 3251,
      "corpus neural models": 23899,
      "present baseline model": 89392,
      "macro f1 score": 66359,
      "classification task model": 17434,
      "model trained russian": 72206,
      "domainspecific bert models": 33891,
      "models freely available": 73263,
      "large language representation": 61129,
      "methods able reduce": 69274,
      "task specific models": 118729,
      "outperforms skipgram model": 82985,
      "neural network grammars": 78304,
      "previous studies attempted": 90489,
      "models logistic regression": 73525,
      "predict relations entities": 88928,
      "baseline models present": 12264,
      "mining multitask learning": 70242,
      "learning model multiple": 62764,
      "stateoftheart transformer models": 113013,
      "models bert variants": 72840,
      "rich textual information": 102797,
      "demonstrate model outperforms": 28793,
      "stateoftheart models datasets": 112730,
      "model improves accuracy": 71316,
      "provide stronger baseline": 93932,
      "largescale neural language": 61485,
      "specific bert models": 111412,
      "domains publicly available": 33843,
      "scientific texts information": 104026,
      "extraction new task": 42412,
      "scientific information extraction": 104008,
      "approach used enhance": 7997,
      "approach challenging tasks": 7416,
      "requires deep semantic": 100260,
      "deep semantic understanding": 28423,
      "text classification approaches": 120711,
      "jointly learning representations": 57356,
      "randomly initialized updated": 95530,
      "model surpasses existing": 72124,
      "surpasses existing methods": 116105,
      "base population task": 11468,
      "model used predict": 72257,
      "language models applied": 59503,
      "models terms accuracy": 74176,
      "information used data": 55075,
      "corpus german language": 23819,
      "covid19 named entity": 24569,
      "study effectiveness contextualized": 114364,
      "model best model": 70770,
      "language models provide": 59641,
      "syntactic analysis named": 116369,
      "analysis named entity": 5712,
      "provide convenient way": 93794,
      "previously proposed method": 90610,
      "existing models perform": 40207,
      "domain adaptation ada": 33429,
      "performance different types": 86302,
      "using labeled data": 129779,
      "data target domains": 26541,
      "sequence tagging machine": 107078,
      "tagging machine reading": 117403,
      "achieved best result": 2599,
      "relation extraction joint": 97995,
      "task existing works": 118168,
      "25 f1 score": 816,
      "random forest rf": 95495,
      "recognition using bert": 97039,
      "using sequence labeling": 130157,
      "sequence labeling framework": 106970,
      "task sequence labeling": 118697,
      "method method achieves": 68961,
      "entity mentions propose": 37973,
      "methods based pretrained": 69336,
      "learningbased methods proposed": 63181,
      "corpora manually annotated": 23525,
      "human annotation paper": 50745,
      "annotation paper propose": 6346,
      "hybrid method combines": 51183,
      "dictionary used automatically": 30898,
      "distantly supervised training": 32494,
      "learning model trained": 62767,
      "model trained classify": 72183,
      "compares favorably previous": 19496,
      "investigate effect training": 56746,
      "manually annotated training": 67036,
      "compared stateoftheart supervised": 19462,
      "training data ii": 123475,
      "domainspecific language model": 33907,
      "ner benchmark datasets": 77025,
      "benchmark datasets different": 12779,
      "semantic relations entities": 105209,
      "adverse drug reaction": 4189,
      "model relation extraction": 71888,
      "relation extraction problem": 98017,
      "learning mtl framework": 62808,
      "adversarial learning approach": 4131,
      "dependency path embedding": 29212,
      "baselines deep learning": 12383,
      "study demonstrate value": 114347,
      "knowledge learning process": 58051,
      "leverages pretrained bert": 63652,
      "pretrained bert language": 89989,
      "locate target entities": 64956,
      "train language model": 122943,
      "language modeling research": 59463,
      "data improve language": 26022,
      "stateoftheart performance domain": 112820,
      "using elastic weight": 129616,
      "knowledge graph corpus": 57962,
      "shed light ability": 108150,
      "models recently achieved": 73882,
      "recently achieved stateoftheart": 96654,
      "results similar tasks": 102189,
      "like bert shown": 64025,
      "semantic world knowledge": 105358,
      "tasks specifically propose": 119513,
      "specifically propose new": 111583,
      "models including bert": 73378,
      "new sota results": 79141,
      "data code freely": 25734,
      "code freely available": 18105,
      "nlp tasks work": 79778,
      "present novel extension": 89609,
      "new dataset propose": 78869,
      "domain language model": 33567,
      "trained general domain": 123150,
      "improvements previous stateoftheart": 52900,
      "train machine learning": 122952,
      "machine learning components": 65790,
      "novel data sets": 80528,
      "data sets used": 26437,
      "memory lstm based": 68326,
      "performances natural language": 86886,
      "uses natural language": 129253,
      "combine novel use": 18636,
      "information extraction performance": 54591,
      "human evaluation qualitative": 50827,
      "evaluation qualitative analysis": 39347,
      "address problem data": 3732,
      "problem data insufficiency": 90992,
      "design simple effective": 29565,
      "small labeled data": 110167,
      "effect training data": 34616,
      "test set composed": 120499,
      "used clinical practice": 128426,
      "paper propose generate": 84248,
      "transformerbased bert models": 124405,
      "bert models relation": 13181,
      "models relation classification": 73906,
      "finetuning process bert": 44508,
      "bert model best": 13168,
      "model best knowledge": 70769,
      "propose bertbased models": 92575,
      "demonstrate promising performance": 28829,
      "learning knowledge graphs": 62665,
      "presents comprehensive study": 89834,
      "transformerbased classification models": 124407,
      "context sentence paper": 22261,
      "sota language models": 110690,
      "model simpler models": 72046,
      "shared task 2021": 108020,
      "entity linking aims": 37954,
      "entities given knowledge": 37789,
      "attention mechanisms capture": 9923,
      "domain specific bert": 33644,
      "trained predict properties": 123242,
      "range linguistic tasks": 95579,
      "named entity tagging": 76128,
      "gives substantial improvements": 48211,
      "methods capture local": 69361,
      "capture local global": 15333,
      "global distributional properties": 48236,
      "downstream tasks using": 34059,
      "embeddings using skipgram": 36012,
      "current best models": 25266,
      "performance benchmarks based": 86174,
      "detects relations entities": 30104,
      "relations entities multiple": 98147,
      "use graph neural": 128076,
      "memory lstm attention": 68324,
      "lstm attention mechanism": 65595,
      "lstm natural language": 65648,
      "dataset demonstrate potential": 26858,
      "hybrid approach based": 51173,
      "approach based sentence": 7384,
      "using elmo bert": 129620,
      "learn semantic information": 62138,
      "processing nlp technologies": 91761,
      "goal work propose": 48402,
      "finetune pretrained bert": 44412,
      "simple method incorporate": 109467,
      "entity recognition systems": 38076,
      "dependencies sentence boundaries": 29112,
      "attention mechanism capture": 9889,
      "review natural language": 102558,
      "deep learning use": 28324,
      "stateoftheart benchmark datasets": 112599,
      "using hierarchical attention": 129738,
      "modern named entity": 74397,
      "neural networks solve": 78582,
      "notoriously difficult evaluate": 80418,
      "circumvent issue propose": 16986,
      "high overall performance": 50094,
      "yield statistically significant": 135360,
      "model pretrained bert": 71773,
      "pretrained bert devlin": 89984,
      "al 2018 address": 4626,
      "2018 address lack": 653,
      "address lack highquality": 3711,
      "online resources order": 81797,
      "sentiment analysis medical": 106608,
      "data sets training": 26436,
      "order improve overall performance": 82342,
      "engineering natural language processing": 37053,
      "proposed approach outperforms baseline": 93204,
      "apply model output existing": 7190,
      "apply natural language processing": 7194,
      "method named entity recognition": 68980,
      "named entity recognition information": 76069,
      "entity recognition information extraction": 38019,
      "unstructured natural language data": 127577,
      "distant supervision relation extraction": 32477,
      "workshop natural language processing": 135009,
      "natural language processing challenges": 76406,
      "annotations partofspeech pos tags": 6451,
      "model based support vector": 70747,
      "distributed word representations learned": 32638,
      "shortterm memory lstm architecture": 108325,
      "models perform competitively stateoftheart": 73723,
      "unified medical language umls": 127117,
      "text classification using convolutional": 120784,
      "classification using convolutional neural": 17487,
      "using deep convolutional neural": 129540,
      "natural language processing propose": 76477,
      "language processing tasks work": 59973,
      "processing tasks work propose": 91833,
      "tasks work propose novel": 119613,
      "topic modeling topic modeling": 122548,
      "methods deep learning methods": 69427,
      "neural network model relation": 78338,
      "lack sufficient training data": 58751,
      "task empirical results proposed": 118137,
      "proposed framework outperforms existing": 93292,
      "framework outperforms existing models": 45641,
      "crosssentence nary relation extraction": 25077,
      "long shortterm memory networks": 65125,
      "various intrasentential intersentential dependencies": 131108,
      "based conditional random fields": 11593,
      "random fields crfs long": 95487,
      "fields crfs long shortterm": 43996,
      "crfs long shortterm memory": 24769,
      "shortterm memory networks lstms": 108345,
      "learning improve models performance": 62641,
      "f1 score best baseline": 42678,
      "recognition ner fundamental task": 96930,
      "deep learning based approaches": 28255,
      "increasing attention recent years": 53660,
      "evaluate approach benchmark datasets": 38790,
      "embeddings surprisingly effective capturing": 35968,
      "word embeddings word vectors": 133253,
      "convolutional neural network text": 23196,
      "network makes use local": 77315,
      "generation task aims automatically": 47658,
      "traditional machine learning algorithms": 122829,
      "deep learning methods recurrent": 28282,
      "dataset order magnitude larger": 27074,
      "order magnitude larger existing": 82362,
      "recognition natural language processing": 96920,
      "neural translation model trained": 78722,
      "develop neural network model": 30218,
      "gated recurrent units attention": 46520,
      "train support vector machine": 123035,
      "applying natural language processing": 7261,
      "propose novel approach combines": 92898,
      "neural network word embeddings": 78417,
      "method outperforms state art": 69028,
      "outperforms state art methods": 82993,
      "relation classification convolutional neural": 97945,
      "classification convolutional neural network": 17161,
      "bidirectional gated recurrent unit": 13923,
      "leads comparable stateoftheart performance": 61928,
      "neural network conditional random": 78291,
      "convolutional neural network capture": 23180,
      "number natural language processing": 80920,
      "performance deep learning models": 86278,
      "word embeddings sentence embeddings": 133206,
      "approach named entity recognition": 7725,
      "use deep learning techniques": 127985,
      "incorporating deep contextualized word": 53527,
      "word embedding model trained": 133023,
      "model yields superior performance": 72342,
      "competitive baselines large margin": 19642,
      "conduct experiments public data": 20855,
      "task information extraction existing": 118298,
      "performs considerably better previous": 87003,
      "relation extraction present novel": 98013,
      "relation extraction model consists": 98000,
      "using bidirectional long shortterm": 129399,
      "unlabeled data improve performance": 127386,
      "advanced natural language processing": 3952,
      "processing nlp tasks including": 91757,
      "results proposed method leads": 102072,
      "rules supervised machine learning": 103448,
      "natural language processing survey": 76486,
      "derived pretrained language models": 29359,
      "models elmo peters et": 73128,
      "elmo peters et al": 35322,
      "et al 2018 bert": 38719,
      "al 2018 bert devlin": 4630,
      "2018 bert devlin et": 658,
      "near state art results": 76715,
      "study natural language processing": 114452,
      "general language understanding evaluation": 46670,
      "language understanding evaluation benchmark": 60221,
      "target entities paper propose": 117616,
      "improving performance natural language": 53138,
      "pretrained language models transfer": 90104,
      "domain empirical results demonstrate": 33518,
      "empirical results demonstrate superiority": 36184,
      "demonstrate superiority proposed methods": 28884,
      "domain natural language inference": 33592,
      "trained support vector machine": 123303,
      "neural network cnn models": 78286,
      "model paper present simple": 71694,
      "paper present simple effective": 84135,
      "present simple effective method": 89703,
      "knowledge base completion task": 57766,
      "learningbased natural language processing": 63185,
      "language processing nlp method": 59907,
      "instance recurrent neural networks": 55609,
      "joint entity relation extraction": 57274,
      "entity relation extraction task": 38093,
      "entity recognition relation classification": 38066,
      "learning chinese word segmentation": 62432,
      "word segmentation cws fundamental": 133493,
      "embedded graph convolutional network": 35357,
      "improve performance word embeddings": 52494,
      "relation extraction pretrained language": 98015,
      "extraction pretrained language representations": 42445,
      "stateoftheart language model learns": 112684,
      "nested named entity recognition": 77108,
      "entity recognition ner relation": 38045,
      "recognition ner relation extraction": 96938,
      "deep learning models used": 28295,
      "achieve high classification accuracy": 2475,
      "medical named entity recognition": 68215,
      "contextualized neural language models": 22563,
      "pretrained contextualized language models": 90013,
      "models achieved stateoftheart performances": 72695,
      "work propose novel unsupervised": 134740,
      "proposed framework achieves stateoftheart": 93284,
      "stateoftheart models experimental results": 112733,
      "knowledge named entity recognition": 58074,
      "results paper proposes novel": 102027,
      "largescale knowledge bases kbs": 61450,
      "effectiveness efficiency proposed method": 34888,
      "extract shortest dependency path": 42121,
      "shortest dependency path sdp": 108302,
      "dataset method achieves comparable": 27020,
      "method achieves comparable performance": 68596,
      "comparable performance stateoftheart systems": 19162,
      "prior knowledge play important": 90711,
      "leading new stateoftheart performance": 61900,
      "results demonstrate method achieves": 101688,
      "tasks like named entity": 119259,
      "like named entity recognition": 64071,
      "domain adaptation pretrained language": 33446,
      "adaptation pretrained language models": 3252,
      "pretrained language models case": 90069,
      "large language representation models": 61130,
      "recurrent neural network grammars": 97205,
      "relation extraction named entity": 98004,
      "code models publicly available": 18122,
      "results demonstrate model outperforms": 101692,
      "demonstrate model outperforms stateoftheart": 28794,
      "largescale neural language models": 61486,
      "requires deep semantic understanding": 100261,
      "model surpasses existing methods": 72125,
      "knowledge base population task": 57778,
      "covid19 named entity recognition": 24570,
      "syntactic analysis named entity": 116370,
      "analysis named entity recognition": 5713,
      "adversarial domain adaptation ada": 4113,
      "tagging machine reading comprehension": 117404,
      "task sequence labeling problem": 118698,
      "results demonstrate effectiveness method": 101675,
      "methods based pretrained language": 69337,
      "deep learning model trained": 28284,
      "manually annotated training data": 67037,
      "multitask learning mtl framework": 75855,
      "leverages pretrained bert language": 63653,
      "using elastic weight consolidation": 129617,
      "models recently achieved stateoftheart": 73883,
      "recently achieved stateoftheart results": 96655,
      "models like bert shown": 73505,
      "make data code freely": 66643,
      "data code freely available": 25735,
      "shortterm memory lstm based": 108328,
      "performances natural language generation": 86887,
      "human evaluation qualitative analysis": 50828,
      "address problem data insufficiency": 3733,
      "paper presents comprehensive study": 84159,
      "entities given knowledge base": 37790,
      "supervised models trained predict": 115810,
      "models trained predict properties": 74223,
      "best models achieve stateoftheart": 13380,
      "achieve stateoftheart performance benchmarks": 2560,
      "detects relations entities multiple": 30105,
      "relations entities multiple sentences": 98148,
      "shortterm memory lstm attention": 108326,
      "memory lstm attention mechanism": 68325,
      "language processing nlp technologies": 59925,
      "present simple method incorporate": 89707,
      "review natural language processing": 102559,
      "modern named entity recognition": 74398,
      "machine learning deep neural": 65794,
      "yield statistically significant improvements": 135361,
      "pretrained bert devlin et": 89985,
      "et al 2018 address": 38717,
      "al 2018 address lack": 4627,
      "2018 address lack highquality": 654,
      "named entity recognition information extraction": 76070,
      "long shortterm memory lstm architecture": 65112,
      "based long shortterm memory lstm": 11815,
      "text classification using convolutional neural": 120785,
      "classification using convolutional neural networks": 17489,
      "using deep convolutional neural networks": 129541,
      "natural language processing tasks work": 76504,
      "language processing tasks work propose": 59974,
      "proposed framework outperforms existing models": 93293,
      "conditional random fields crfs long": 20786,
      "random fields crfs long shortterm": 95488,
      "fields crfs long shortterm memory": 43997,
      "crfs long shortterm memory networks": 24770,
      "long shortterm memory networks lstms": 65126,
      "entity recognition ner fundamental task": 38039,
      "bidirectional gated recurrent units attention": 13925,
      "applying natural language processing nlp": 7262,
      "convolutional neural network word embeddings": 23198,
      "method outperforms state art methods": 69029,
      "classification convolutional neural network cnn": 17162,
      "approach named entity recognition ner": 7726,
      "incorporating deep contextualized word embeddings": 53528,
      "conduct experiments public data sets": 20856,
      "demonstrate proposed method outperforms stateoftheart": 28842,
      "using bidirectional long shortterm memory": 129400,
      "bidirectional long shortterm memory networks": 13944,
      "language processing nlp tasks including": 59921,
      "models elmo peters et al": 73129,
      "elmo peters et al 2018": 35323,
      "et al 2018 bert devlin": 38720,
      "al 2018 bert devlin et": 4631,
      "2018 bert devlin et al": 659,
      "general language understanding evaluation benchmark": 46671,
      "pretrained language models transfer learning": 90105,
      "convolutional neural network cnn models": 23187,
      "paper present simple effective method": 84136,
      "natural language processing nlp method": 76453,
      "joint entity relation extraction task": 57275,
      "named entity recognition relation classification": 76109,
      "chinese word segmentation cws fundamental": 16837,
      "relation extraction pretrained language representations": 98016,
      "named entity recognition ner relation": 76091,
      "entity recognition ner relation extraction": 38046,
      "extract shortest dependency path sdp": 42122,
      "dataset method achieves comparable performance": 27021,
      "method achieves comparable performance stateoftheart": 68597,
      "experimental results demonstrate method achieves": 40593,
      "tasks like named entity recognition": 119260,
      "model named entity recognition ner": 71556,
      "relation extraction named entity recognition": 98005,
      "demonstrate model outperforms stateoftheart models": 28795,
      "experimental results demonstrate effectiveness method": 40587,
      "achieves competitive performance compared stateoftheart": 2772,
      "models recently achieved stateoftheart results": 73884,
      "pretrained language models like bert": 90084,
      "language models like bert shown": 59588,
      "make data code freely available": 66644,
      "long shortterm memory lstm based": 65114,
      "supervised models trained predict properties": 115811,
      "detects relations entities multiple sentences": 30106,
      "long shortterm memory lstm attention": 65113,
      "shortterm memory lstm attention mechanism": 108327,
      "approach achieves new stateoftheart results": 7312,
      "natural language processing nlp technologies": 76466,
      "pretrained bert devlin et al": 89986,
      "devlin et al 2018 address": 30471,
      "et al 2018 address lack": 38718,
      "al 2018 address lack highquality": 4628,
      "1991": 462,
      "prealigned": 88768,
      "190": 439,
      "triangulation": 125767,
      "globalization": 48282,
      "196": 447,
      "csen": 25187,
      "047": 30,
      "singlepair": 109836,
      "duallearning": 34248,
      "persianitalian": 87090,
      "msnmt": 74949,
      "overtaken": 83359,
      "levantine": 63410,
      "disproportionately": 32382,
      "arabics": 8573,
      "multiagent": 75031,
      "passively": 85629,
      "multi30k": 75029,
      "279": 862,
      "macao": 65737,
      "assert": 9463,
      "inconvenient": 53447,
      "wakesleep": 131993,
      "englishtamil": 37433,
      "wixarika": 132756,
      "yorem": 135450,
      "nokki": 80137,
      "conversions": 23091,
      "estonianenglish": 38678,
      "preorder": 89292,
      "multiparallel": 75478,
      "208": 749,
      "nonzeroshot": 80319,
      "regularizations": 97762,
      "overreliance": 83352,
      "dominantly": 33935,
      "gt": 49206,
      "trilingual": 125802,
      "russianenglish": 103511,
      "tcs": 119676,
      "chinesejapanese": 16866,
      "excerpt": 39934,
      "reassess": 96331,
      "422": 1017,
      "natives": 76238,
      "merit": 68483,
      "223": 779,
      "noresource": 80323,
      "trainingtime": 123960,
      "devote": 30473,
      "323": 935,
      "515": 1095,
      "regularizers": 97770,
      "debatable": 27886,
      "tatar": 119649,
      "glossed": 48295,
      "fourfold": 45354,
      "odin": 81532,
      "arapaho": 8576,
      "2594": 833,
      "temporary": 120136,
      "endow": 36860,
      "commonality": 18953,
      "nigerian": 79482,
      "inequalities": 54078,
      "attenuated": 10131,
      "diacritical": 30496,
      "appreciable": 7282,
      "higherresource": 50220,
      "tofrom": 122243,
      "useless": 128964,
      "exacerbating": 39712,
      "bmass": 14387,
      "bunsetsus": 14967,
      "gebru": 46572,
      "datasheet": 27806,
      "encodersdecoders": 36665,
      "falling": 43032,
      "inspected": 55552,
      "broaden": 14696,
      "imaginative": 51820,
      "sanskrithindi": 103635,
      "anchored": 6086,
      "4m": 1060,
      "czechenglish": 25500,
      "englishbengali": 37356,
      "gaming": 46427,
      "englishmacedonian": 37419,
      "pretrainfinetune": 90229,
      "kazakh": 57499,
      "sinhala": 109864,
      "romanization": 103255,
      "kurmanji": 58366,
      "byncsa": 14989,
      "forwardtranslation": 45334,
      "stabilizes": 112109,
      "sorbian": 110681,
      "237": 797,
      "574": 1128,
      "germanupper": 47945,
      "bpedropout": 14545,
      "volctrans": 131935,
      "sociocultural": 110474,
      "185": 426,
      "defend": 28471,
      "dae": 25502,
      "346": 957,
      "paving": 85806,
      "crossdialect": 24878,
      "293": 871,
      "374": 975,
      "sentences bilingual": 106230,
      "paper algorithm": 83729,
      "algorithm picks": 4785,
      "information certain": 54411,
      "algorithm language": 4761,
      "addition deletion": 3419,
      "algorithm gives": 4748,
      "gives comparable": 48190,
      "does better": 33332,
      "statistical algorithms": 113075,
      "results bilingual": 101556,
      "texts manually": 121547,
      "manually classified": 67042,
      "classified according": 17508,
      "domains collection": 33743,
      "software corpus": 110516,
      "according text": 2006,
      "languages instance": 60646,
      "corpora presented": 23559,
      "presented algorithm": 89775,
      "translation key": 124882,
      "means using": 67771,
      "russian ukrainian": 103508,
      "morphological dictionaries": 74696,
      "dictionaries dictionaries": 30858,
      "terms documents": 120313,
      "used algorithm": 128384,
      "algorithm consideration": 4708,
      "30 thousand": 908,
      "open architecture": 81881,
      "texts multilingual": 121555,
      "texts linguistic": 121543,
      "processing parallel": 91768,
      "project build": 92212,
      "resources required": 101036,
      "specific languages": 111458,
      "extraction parallel": 42425,
      "novel languageindependent": 80614,
      "languages exploiting": 60573,
      "vocabulary overlap": 131890,
      "shows absolute": 108544,
      "improvement best": 52686,
      "overall method": 83239,
      "factor 25": 42853,
      "approaches machine": 8223,
      "translation purpose": 125161,
      "largest corpora": 61529,
      "use english": 128028,
      "french results": 45823,
      "smt english": 110281,
      "language chinese": 58875,
      "spanish propose": 111166,
      "combination using": 18592,
      "pair languages": 83437,
      "sentencealigned corpus": 106136,
      "domain terminology": 33671,
      "engine using": 37037,
      "large target": 61288,
      "translation high": 124848,
      "translation new": 125016,
      "research effective": 100481,
      "work endtoend": 134490,
      "availability high": 10907,
      "abundant monolingual": 1835,
      "compared phrasebased": 19409,
      "baseline obtain": 12278,
      "chat messages": 16659,
      "tasks parallel": 119366,
      "dialect arabic": 30517,
      "different dialects": 31093,
      "dialects arabic": 30535,
      "selection demonstrate": 104780,
      "small highly": 110155,
      "targeted data": 117764,
      "quality critical": 94624,
      "systems nlp": 117024,
      "approach developed": 7489,
      "comparable corpus": 19140,
      "implements various": 51986,
      "sentence comparison": 105790,
      "synonyms semantic": 116352,
      "text processed": 121203,
      "processed using": 91602,
      "matrix language": 67476,
      "model estimation": 71102,
      "practical situations": 88714,
      "previous training": 90512,
      "data allows": 25594,
      "experiments billion": 40808,
      "slightly improve": 110023,
      "results use": 102297,
      "set surprisingly": 107594,
      "scenario training": 103834,
      "data mix": 26133,
      "source given": 110763,
      "sources based": 110879,
      "based relevant": 11991,
      "role boosting": 103160,
      "boosting fluency": 14443,
      "work combines": 134416,
      "models note": 73649,
      "capacity learn": 15231,
      "learn information": 62069,
      "strategies train": 113491,
      "data obtain": 26186,
      "obtain substantial": 81332,
      "wmt 15": 132767,
      "finetuning indomain": 44465,
      "indomain monolingual": 53971,
      "improvements iwslt": 52862,
      "scores range": 104197,
      "idea train": 51338,
      "word replacement": 133434,
      "syntax based": 116531,
      "exceeding performance": 39906,
      "machine translate": 65876,
      "used novel": 128667,
      "coverage especially": 24504,
      "exploit monolingual": 41430,
      "corpora improve": 23499,
      "corpora data": 23449,
      "serve encoder": 107294,
      "morphological constraints": 74694,
      "common solution": 18928,
      "corpora source": 23591,
      "technique introduces": 119795,
      "combination baseline": 18551,
      "corpus fixed": 23806,
      "fixed network": 44601,
      "sentence finetune": 105875,
      "highly similar": 50346,
      "idea design": 51322,
      "latent bilingual": 61575,
      "remarkably improve": 98852,
      "sentences obtain": 106416,
      "learn unlabeled": 62178,
      "data duallearning": 25872,
      "mechanism inspired": 67999,
      "dual task": 34241,
      "frenchtoenglish translation": 45832,
      "primal dual": 90636,
      "task agent": 117857,
      "likelihood output": 64123,
      "iteratively update": 57148,
      "update models": 127796,
      "learning monolingual": 62801,
      "data 10": 25548,
      "decoder paper": 28041,
      "employ attentionbased": 36251,
      "special treatment": 111373,
      "minimal number": 70170,
      "points addition": 87847,
      "results applied": 101511,
      "translation pivot": 125122,
      "approaches delivered": 8119,
      "performance resourcerich": 86673,
      "boosting neural": 14445,
      "networks need": 77673,
      "high computation": 50043,
      "propose alleviate": 92541,
      "network imitates": 77277,
      "learning difficult": 62503,
      "difficult concepts": 31615,
      "translation simplified": 125260,
      "introduce idea": 56435,
      "ability deep": 1504,
      "using simplified": 130179,
      "performance boosted": 86189,
      "reference automatic": 97507,
      "translations used": 125504,
      "persianitalian parallel": 87091,
      "effectiveness statistical": 34953,
      "build accurate": 14740,
      "bilingual persianitalian": 14051,
      "nonparallel comparable": 80263,
      "scores source": 104206,
      "sentences candidate": 106233,
      "candidate selection": 15120,
      "google distance": 48520,
      "extensions baseline": 41846,
      "corpus significantly": 23998,
      "translation msnmt": 124976,
      "modifying neural": 74462,
      "sentence train": 106110,
      "solution gives": 110555,
      "gives competitive": 48191,
      "nmt leverages": 79885,
      "leverages multilingual": 63648,
      "visualizing attention": 131841,
      "effective various": 34775,
      "issues study": 57080,
      "enhanced results": 37517,
      "improvement neural": 52731,
      "languages additional": 60391,
      "serve new": 107297,
      "sentences unlike": 106533,
      "view properties": 131709,
      "contribution present": 22775,
      "pairs second": 83631,
      "sentences comparable": 106246,
      "given parallel": 48076,
      "corpus written": 24070,
      "spoken dialect": 111973,
      "generate data": 46921,
      "translation generated": 124837,
      "data proven": 26299,
      "nmt introduce": 79880,
      "selection nmt": 104809,
      "applying technique": 7274,
      "gradual finetuning": 48615,
      "approaches improved": 8190,
      "improved state": 52645,
      "data produce": 26282,
      "reordering translation": 98902,
      "addition novel": 3455,
      "evaluation lowresource": 39257,
      "nmt investigate": 79881,
      "produce acceptable": 91871,
      "pair method": 83442,
      "transfer parameters": 124173,
      "including source": 53377,
      "training second": 123832,
      "experiments transfer": 41184,
      "bpe baseline": 14538,
      "decoding phase": 28122,
      "witnessed rapid": 132753,
      "state space": 112515,
      "training address": 123351,
      "certain limitations": 15946,
      "sampling propose": 103605,
      "quality standard": 94797,
      "addressing data": 3809,
      "engineering using": 37057,
      "networks results": 77740,
      "extracted parallel": 42170,
      "systems believe": 116754,
      "investigate deep": 56739,
      "solutions problem": 110581,
      "simple lexical": 109454,
      "module jointly": 74496,
      "sizes ranging": 109961,
      "public domain": 94257,
      "domain new": 33597,
      "new parallel": 79070,
      "corpus preprocessed": 23933,
      "asian language": 9288,
      "2017 corpus": 631,
      "noncommercial research": 80170,
      "zeroshot neural": 135517,
      "strategies applied": 113450,
      "tackle zeroshot": 117315,
      "modeling case": 72386,
      "resources small": 101044,
      "modeling morphology": 72480,
      "differences dialects": 30953,
      "bleu model": 14295,
      "corpus major": 23874,
      "baseline statistical": 12315,
      "propose interactive": 92727,
      "interactive multimodal": 56023,
      "multimodal framework": 75429,
      "image description": 51780,
      "captioning neural": 15264,
      "multi30k datasets": 75030,
      "methods universal": 69824,
      "new universal": 79233,
      "approach focusing": 7578,
      "transferlearning approach": 124226,
      "encoders languages": 36654,
      "higher resource": 50202,
      "using tiny": 130293,
      "bleu strong": 14338,
      "20 bleu": 493,
      "data monolingual": 26142,
      "quality statistical": 94800,
      "learning sourcetotarget": 63041,
      "targettosource nmt": 117805,
      "starts initial": 112417,
      "models expected": 73194,
      "systems enhanced": 116870,
      "training including": 123658,
      "comparable texts": 19187,
      "12 billion": 247,
      "good example": 48475,
      "cases data": 15639,
      "researchers shown": 100709,
      "create synthetic": 24644,
      "train highquality": 122940,
      "translation capabilities": 124684,
      "capabilities nmt": 15174,
      "using backtranslated": 129379,
      "amounts backtranslated": 5330,
      "analyse resulting": 5435,
      "useful neural": 128909,
      "embeddings proven": 35893,
      "nmt extensively": 79866,
      "nmt tasks": 79986,
      "setting phrasebased": 107770,
      "achieve near": 2499,
      "simpler having": 109557,
      "having fewer": 49559,
      "outperforming state": 82829,
      "art 11": 9031,
      "semisupervised supervised": 105626,
      "parallel crosslingual": 84629,
      "work translation": 134861,
      "identify lack": 51514,
      "add source": 3353,
      "transfer achieve": 124033,
      "baseline training": 12326,
      "language secondly": 60071,
      "accuracy qualitative": 2251,
      "incorporate explicit": 53469,
      "embeddings create": 35622,
      "classify french": 17650,
      "produces comparable": 91989,
      "comparable bleu": 19134,
      "nmt use": 80003,
      "independent encoders": 53767,
      "sharing partial": 108139,
      "responsible extracting": 101310,
      "highlevel representations": 50248,
      "networks gans": 77607,
      "translation requires": 125216,
      "text learn": 121083,
      "resources source": 101045,
      "address challenging": 3665,
      "including semantic": 53364,
      "knowledge translation": 58215,
      "effectiveness multitask": 34918,
      "english present": 37248,
      "articles collected": 9177,
      "turkish news": 125963,
      "corpus modeling": 23885,
      "problem classification": 90960,
      "classify data": 17648,
      "original texts": 82555,
      "translation synthetic": 125306,
      "despite impressive": 29695,
      "lowresource outofdomain": 65549,
      "outofdomain scenarios": 82659,
      "difficult cases": 31611,
      "directions language": 31846,
      "data requiring": 26350,
      "augmented parallel": 10332,
      "incorporate source": 53487,
      "models reduce": 73896,
      "effectively reduces": 34846,
      "embeddings impact": 35745,
      "effect using": 34618,
      "embeddings nmt": 35834,
      "strategies results": 113486,
      "suggest pretrained": 115327,
      "nmt especially": 79861,
      "exploit multiple": 41431,
      "multiple inputs": 75580,
      "corpora complete": 23437,
      "explicitly handle": 41371,
      "situations study": 109902,
      "multiencoder nmt": 75102,
      "examines simple": 39767,
      "missing source": 70333,
      "higher translation": 50212,
      "commonly employed": 18959,
      "work offer": 134659,
      "inference generative": 54147,
      "wakesleep algorithm": 131994,
      "work effectively": 134484,
      "socalled zeroshot": 110321,
      "zeroshot inference": 135506,
      "pairs seen": 83632,
      "provides quantitative": 94062,
      "analysis translations": 5909,
      "systems ii": 116944,
      "dominant neural": 33933,
      "closeness languages": 17870,
      "leverages multiple": 63649,
      "focuses automatic": 44891,
      "bleu ter": 14340,
      "errors neural": 38389,
      "problem context": 90979,
      "pairs multilingual": 83587,
      "paper used": 84486,
      "wikipedia subsequently": 132678,
      "improved bleu": 52596,
      "nmt phrasebased": 79946,
      "pairs englishhindi": 83532,
      "compared training": 19483,
      "task gets": 118243,
      "gets complicated": 47953,
      "nahuatl wixarika": 76002,
      "wixarika yorem": 132757,
      "yorem nokki": 135451,
      "projected shared": 92227,
      "shared layer": 107989,
      "layer model": 61729,
      "close possible": 17828,
      "published studies": 94375,
      "studies require": 114273,
      "testing method": 120595,
      "transfer introduce": 124063,
      "task zeroshot": 118861,
      "allows perform": 5177,
      "provides reasonable": 94065,
      "17 bleu": 400,
      "sentences randomly": 106464,
      "high prediction": 50107,
      "prediction loss": 89073,
      "wmt news": 132804,
      "augment parallel": 10263,
      "synthetic source": 116644,
      "stronger training": 113736,
      "effects finally": 34986,
      "monolingual sentences": 74614,
      "bleu wmt14": 14344,
      "nmt typically": 80002,
      "fixed language": 44596,
      "model tm": 72173,
      "data added": 25572,
      "capacity modeling": 15233,
      "sentence rely": 106025,
      "architecture simpler": 8749,
      "alternative ways": 5251,
      "ways utilize": 132176,
      "data backtranslation": 25687,
      "learning proven": 62926,
      "technique neural": 119802,
      "training lowresource": 123693,
      "translation rare": 125195,
      "present generic": 89507,
      "using phrasebased": 130014,
      "demonstrate benefit": 28677,
      "outofdomain translation": 82668,
      "resources improving": 100989,
      "directions english": 31837,
      "translate multiple": 124530,
      "single target": 109800,
      "sources systems": 110922,
      "systems necessary": 117016,
      "corpora rarely": 23567,
      "language sequence": 60082,
      "capturing lexical": 15474,
      "mixture component": 70417,
      "selects training": 104867,
      "baseline standard": 12313,
      "decoding unsupervised": 28145,
      "new dedicated": 78875,
      "architectures rely": 8844,
      "synthetic bilingual": 116612,
      "stateoftheart architectures": 112588,
      "generated unsupervised": 47146,
      "incrementally improved": 53746,
      "germanenglish news": 47938,
      "encoder represent": 36558,
      "represent source": 99143,
      "decoder using": 28060,
      "reduce large": 97333,
      "lefttoright righttoleft": 63310,
      "models resourcerich": 73944,
      "divergence multilingual": 32784,
      "languages transfer": 60928,
      "nmt train": 79990,
      "divergent word": 32789,
      "learning little": 62695,
      "match word": 67371,
      "capable obtaining": 15208,
      "representations means": 99750,
      "performing translation": 86978,
      "languageindependent meaning": 60364,
      "representation efficiently": 99220,
      "scheduled training": 103887,
      "translation demonstrates": 124743,
      "importance properly": 52073,
      "popular online": 88116,
      "generation inconsistent": 47438,
      "english specific": 37290,
      "texts lowresource": 121544,
      "learned multilingual": 62232,
      "representations prior": 99824,
      "posterior regularization": 88488,
      "sources remains": 110919,
      "provide general": 93836,
      "sources features": 110891,
      "aim better": 4448,
      "exploit limited": 41426,
      "loss compares": 65258,
      "reconstructed inputs": 97112,
      "bleu improvements": 14287,
      "pairs translation": 83653,
      "outperforms alternative": 82836,
      "translation dynamic": 124772,
      "initial model": 55220,
      "vocabulary items": 131878,
      "transfer mechanism": 124149,
      "evaluated scenarios": 39004,
      "add new": 3351,
      "spanning languages": 111188,
      "sentences significant": 106494,
      "reach higher": 95893,
      "training steps": 123886,
      "work multilingual": 134642,
      "performance zeroshot": 86867,
      "directions observed": 31850,
      "setting propose": 107773,
      "iterative training": 57137,
      "contain mixed": 21747,
      "mixed language": 70398,
      "language shared": 60086,
      "directions multilingual": 31848,
      "offtheshelf neural": 81625,
      "apply models": 7191,
      "vocabulary translation": 131914,
      "using sparse": 130202,
      "simple thresholding": 109532,
      "methods produced": 69689,
      "unsupervised translation": 127744,
      "pseudo data": 94178,
      "data inevitably": 26037,
      "models iterative": 73431,
      "backtranslation process": 11340,
      "models wordlevel": 74356,
      "creation hindienglish": 24717,
      "learning applications": 62335,
      "hindienglish parallel": 50491,
      "task verify": 118839,
      "face paucity": 42732,
      "word spelling": 133584,
      "encoding semantic": 36723,
      "nmt baselines": 79840,
      "languages achieving": 60385,
      "make substantial": 66733,
      "backtranslation data": 11336,
      "pose difficult": 88251,
      "graphs enhance": 48948,
      "quality investigate": 94693,
      "graph augmented": 48762,
      "augmented neural": 10329,
      "translation multiple": 124992,
      "languages multiple": 60747,
      "models translating": 74246,
      "corpus massively": 23879,
      "settings outperforming": 107827,
      "fails fully": 42982,
      "findings confirm": 44289,
      "effective new": 34724,
      "usually low": 130436,
      "translation loss": 124914,
      "loss evaluate": 65263,
      "embeddings regularization": 35904,
      "value joint": 130780,
      "learn distributional": 62030,
      "unseen sentences": 127541,
      "portuguese english": 88197,
      "demonstrate capability": 28684,
      "source scientific": 110808,
      "automatically aligned": 10715,
      "lot traction": 65330,
      "promising interesting": 92277,
      "representations open": 99790,
      "proposed order": 93515,
      "quality lack": 94698,
      "exploration paper": 41503,
      "paper serve": 84421,
      "incremental training": 53740,
      "improvements 28": 52797,
      "results larger": 101884,
      "languages previously": 60813,
      "translations new": 125477,
      "data auxiliary": 25665,
      "formulation propose": 45295,
      "samples target": 103592,
      "augmentation sentence": 10300,
      "segmentation lowresource": 104598,
      "nmt proven": 79951,
      "results depend": 101713,
      "size quality": 109942,
      "japanesechinese scientific": 57206,
      "paper excerpt": 83905,
      "excerpt corpus": 39935,
      "performance revisiting": 86685,
      "settings discuss": 107798,
      "especially helpful": 38457,
      "helpful lowresource": 49801,
      "settings resulting": 107835,
      "resulting set": 101472,
      "different amounts": 30989,
      "far data": 43075,
      "data previously": 26275,
      "previously claimed": 90592,
      "translating language": 124590,
      "property training": 92505,
      "approach translates": 7983,
      "language ids": 59152,
      "source decoded": 110742,
      "422 bleu": 1018,
      "approach translating": 7984,
      "given rough": 48108,
      "uncover latent": 126648,
      "demand using": 28632,
      "unsupervised techniques": 127735,
      "languages obtaining": 60769,
      "creates new": 24691,
      "set difficult": 107418,
      "selection synthetic": 104836,
      "work carried": 134407,
      "experiments shed": 41132,
      "regularization methods": 97752,
      "robust zeroshot": 103085,
      "zeroshot conditions": 135492,
      "points 12": 87842,
      "12 language": 253,
      "paper submission": 84447,
      "representations uses": 99964,
      "architecture trained": 8761,
      "obtain multilingual": 81300,
      "promising methods": 92281,
      "additional gains": 3519,
      "14 bleu": 310,
      "bleu respectively": 14319,
      "low noresource": 65376,
      "study methods": 114438,
      "parallel datasets": 84662,
      "total size": 122700,
      "models taskspecific": 74167,
      "parameters fully": 84751,
      "tasks bilingual": 118967,
      "allows scale": 5185,
      "modules allowing": 74519,
      "shows close": 108560,
      "multistage finetuning": 75799,
      "nmt taking": 79984,
      "benchmarking solutions": 12881,
      "scenarios multilingual": 103857,
      "exploit outofdomain": 41433,
      "pseudoparallel data": 94192,
      "37 bleu": 971,
      "simple automatic": 109368,
      "developing machine": 30345,
      "certain domains": 15940,
      "techniques successful": 119990,
      "pair work": 83468,
      "corpus arabic": 23667,
      "arabic news": 8537,
      "suggest approach": 115302,
      "handling syntactic": 49421,
      "empirical successes": 36209,
      "divergent languages": 32788,
      "source used": 110861,
      "experiments simulated": 41144,
      "multiagent learning": 75032,
      "translation conventional": 124712,
      "training additional": 123350,
      "introducing diverse": 56632,
      "tasks combining": 118985,
      "synthetic sentences": 116643,
      "mt approaches": 74956,
      "particular investigate": 85424,
      "performances training": 86901,
      "update existing": 127792,
      "utility neural": 130488,
      "extra time": 42064,
      "pair mean": 83440,
      "point training": 87816,
      "new lowresource": 78999,
      "language fixed": 59050,
      "approach subword": 7943,
      "generalizes languages": 46839,
      "languages variety": 60962,
      "families translation": 43054,
      "competing methods": 19618,
      "required learn": 100218,
      "words shared": 134211,
      "transformer selfattentive": 124376,
      "networks focus": 77603,
      "learn translation": 62172,
      "incrementally improves": 53747,
      "relying monolingual": 98763,
      "nmt outperforms": 79936,
      "recurrent nmt": 97257,
      "quality indomain": 94689,
      "face challenges": 42730,
      "words dealing": 133898,
      "neural conditional": 77868,
      "information order": 54815,
      "produce higher": 91900,
      "work relatively": 134770,
      "reduce parameter": 97349,
      "sacrificing performance": 103527,
      "hardware resources": 49500,
      "architecture approach": 8617,
      "typical nmt": 126403,
      "different researchers": 31387,
      "use seed": 128258,
      "times training": 122190,
      "previous nmt": 90439,
      "work including": 134566,
      "demonstrate largescale": 28768,
      "pretraining significantly": 90324,
      "performance able": 86107,
      "performance larger": 86491,
      "larger sets": 61382,
      "set accordingly": 107344,
      "demonstrate advantages": 28658,
      "translation domains": 124770,
      "strong stateoftheart": 113715,
      "improve clustering": 52351,
      "experiment unsupervised": 40516,
      "russian translation": 103506,
      "respectively survey": 101170,
      "methods leverage": 69591,
      "semisupervised unsupervised": 105631,
      "adapting multilingual": 3315,
      "translation unseen": 125406,
      "languages lrl": 60710,
      "languages hrl": 60623,
      "linguistic assumptions": 64429,
      "distance extensively": 32418,
      "settings zeroshot": 107847,
      "vocabulary results": 131898,
      "adaptation experiments": 3223,
      "related data": 97851,
      "grade levels": 48586,
      "train multitask": 122973,
      "grade level": 48585,
      "secondly use": 104476,
      "present metrics": 89556,
      "using accuracy": 129315,
      "accuracy matching": 2206,
      "noise errors": 80049,
      "allows unsupervised": 5194,
      "corpus recent": 23961,
      "recent machine": 96470,
      "corpora availability": 23417,
      "achieved surpassed": 2712,
      "surpassed performance": 116100,
      "training initial": 123662,
      "widely regarded": 132551,
      "nontrivial challenge": 80312,
      "algorithm decoding": 4714,
      "decoding making": 28104,
      "approaches tackling": 8368,
      "tackling challenge": 117322,
      "add information": 3350,
      "extra embeddings": 42059,
      "tokens tokens": 122343,
      "explored crosslingual": 41621,
      "methods involving": 69565,
      "language far": 59043,
      "idea make": 51331,
      "built encoder": 14919,
      "applied zeroshot": 7148,
      "task dialect": 118086,
      "diversity source": 32895,
      "reveal methods": 102499,
      "simplified traditional": 109607,
      "chinese conversion": 16749,
      "translation finetuning": 124825,
      "nmt conducts": 79847,
      "limited model": 64253,
      "decoder capture": 27999,
      "languagespecific information": 60993,
      "translation better": 124674,
      "gains wide": 46414,
      "variety machine": 130991,
      "complete model": 19765,
      "demonstrate adding": 28656,
      "types transfer": 126379,
      "train final": 122937,
      "addition studying": 3477,
      "novel setup": 80727,
      "target pair": 117687,
      "data exist": 25913,
      "data naturally": 26157,
      "framework novel": 45631,
      "loss term": 65300,
      "wmt16 englishgerman": 132829,
      "directions particular": 31851,
      "direction neural": 31823,
      "nigerian pidgin": 79483,
      "experiments building": 40811,
      "script languages": 104257,
      "report improvement": 99005,
      "translation africa": 124631,
      "communication importance": 19035,
      "accurate machine": 2341,
      "create stateoftheart": 24642,
      "issues facing": 57049,
      "african nlp": 4276,
      "step creating": 113257,
      "robust translation": 103080,
      "french research": 45822,
      "creation large": 24718,
      "model publicly": 71829,
      "available low": 11035,
      "shown improvements": 108486,
      "collected available": 18403,
      "evaluation comparing": 39151,
      "baseline single": 12311,
      "multilingual modeling": 75283,
      "points translation": 87880,
      "experimental data": 40529,
      "future works": 46327,
      "works addressing": 134919,
      "challenges nmt": 16185,
      "languages incorporating": 60636,
      "leveraging target": 63707,
      "size available": 109912,
      "sentences low": 106386,
      "technique address": 119764,
      "automatically expands": 10766,
      "model maintaining": 71489,
      "reference language": 97524,
      "translation long": 124913,
      "long tradition": 65142,
      "supervised learningbased": 115785,
      "clues available": 17917,
      "paradigm propose": 84547,
      "auxiliary language": 10875,
      "usefulness proposed": 128963,
      "supervision improves": 115890,
      "results big": 101555,
      "nmt parallel": 79941,
      "leverage weakly": 63637,
      "information parallel": 54828,
      "nmt empirical": 79856,
      "quality bleu": 94607,
      "score benchmark": 104054,
      "translation usage": 125411,
      "generation constraint": 47342,
      "zeroshot multilingual": 135516,
      "translation able": 124613,
      "order fulfill": 82325,
      "learning subword": 63068,
      "review approaches": 102534,
      "pair target": 83457,
      "english danish": 37103,
      "danish swedish": 25536,
      "corpora achieved": 23399,
      "unmt systems": 127473,
      "adequate training": 3835,
      "unmt based": 127468,
      "performance case": 86195,
      "conventional unmt": 22912,
      "succeed fail": 115057,
      "target corpora": 117594,
      "translation relies": 125213,
      "328 bleu": 939,
      "closes gap": 17880,
      "lifelong learning": 63982,
      "effect overall": 34605,
      "learns weight": 63244,
      "weight training": 132329,
      "conditions unsupervised": 20821,
      "experiments supervised": 41160,
      "unsupervised results": 127707,
      "ways improve": 132160,
      "pairs varying": 83664,
      "typological characteristics": 126482,
      "overcome bottleneck": 83274,
      "architectures identify": 8808,
      "settings improves": 107808,
      "comparable data": 19143,
      "comparable training": 19188,
      "involves using": 56908,
      "set images": 107461,
      "pivot source": 87584,
      "captions images": 15268,
      "languages independently": 60640,
      "translation dictionary": 124751,
      "valid translations": 130712,
      "mt training": 75002,
      "reference sentence": 97532,
      "distribution possible": 32674,
      "english improvements": 37168,
      "improvements 12": 52787,
      "mt task": 74998,
      "new ones": 79062,
      "focus finetuning": 44765,
      "25 languages": 817,
      "parameters important": 84753,
      "outofdomain training": 82666,
      "scarcity large": 103810,
      "corpora important": 23498,
      "output distributions": 83068,
      "lm objective": 64867,
      "does compromise": 33335,
      "work requires": 134783,
      "clear improvements": 17724,
      "translating monolingual": 124592,
      "analyse impact": 5433,
      "time maintaining": 122048,
      "highquality mt": 50396,
      "method taking": 69177,
      "typically addressed": 126409,
      "addressed transfer": 3789,
      "account linguistic": 2031,
      "syntactic analyzers": 116371,
      "focuses japanese": 44905,
      "units called": 127235,
      "competitive better": 19648,
      "pretraining gives": 90262,
      "individual methods": 53917,
      "hierarchical multilingual": 49979,
      "learning empirical": 62540,
      "experimentation dataset": 40742,
      "gebru et": 46573,
      "languagespecific encodersdecoders": 60991,
      "modular architecture": 74465,
      "retraining existing": 102372,
      "decoder modules": 28033,
      "representation task": 99433,
      "training beneficial": 123373,
      "hindi translation": 50477,
      "shot translation": 108357,
      "available spanish": 11108,
      "architecture tested": 8759,
      "model englishhindi": 71078,
      "pairs order": 83595,
      "pipelines nmt": 87567,
      "reduce memory": 97337,
      "translation cycles": 124731,
      "sanskrithindi parallel": 103636,
      "corpus 300": 23633,
      "construction parallel": 21694,
      "information state": 55004,
      "mt based": 74959,
      "sentences refer": 106469,
      "motivated ability": 74842,
      "uses bilingual": 129200,
      "pairs approaches": 83490,
      "embedding transformation": 35520,
      "achieving performances": 2969,
      "new release": 79103,
      "methodology gathering": 69261,
      "noisy sentence": 80117,
      "outperform google": 82710,
      "corpora collection": 23434,
      "online sources": 81807,
      "corpus augmented": 23670,
      "corpus preparation": 23932,
      "corpus sentiment": 23989,
      "training character": 123380,
      "finetuning recent": 44509,
      "creating model": 24702,
      "pretraining demonstrated": 90251,
      "multilingual finetuning": 75256,
      "finetuning instead": 44467,
      "instead finetuning": 55663,
      "starting pretrained": 112413,
      "available demonstrate": 10980,
      "reproducible research": 100089,
      "finetuning improves": 44463,
      "average bleu": 11183,
      "orthographic information": 82595,
      "explored different": 41622,
      "researchers started": 100710,
      "research regarding": 100610,
      "utilised improve": 130470,
      "underlying assumptions": 126675,
      "demonstrate recent": 28857,
      "current efforts": 25279,
      "widespread adoption": 132614,
      "expensive develop": 40419,
      "using lowcost": 129839,
      "translate api": 124523,
      "paracrawl dataset": 84530,
      "winning submission": 132723,
      "task submission": 118749,
      "gpu days": 48577,
      "corpora unsupervised": 23609,
      "lm finetuned": 64861,
      "finetuned languages": 44431,
      "reuse pretrained": 102488,
      "extension method": 41836,
      "called codeswitching": 15052,
      "replaces words": 98946,
      "perform lexicon": 86026,
      "induction unsupervised": 54038,
      "adopts encoderdecoder": 3910,
      "codemixed sentence": 18179,
      "decoder predicts": 28045,
      "able pretrain": 1672,
      "pretrainfinetune discrepancy": 90230,
      "unsupervised parallel": 127688,
      "sentencelevel translation": 106180,
      "translation costly": 124716,
      "created humans": 24671,
      "training source": 123870,
      "achieves close": 2755,
      "results wmt16": 102340,
      "methods combined": 69377,
      "score previous": 104108,
      "set 1000": 107340,
      "pairs extensive": 83543,
      "way future": 132086,
      "reached impressive": 95908,
      "pairs englishfrench": 83529,
      "settings involving": 107814,
      "robustness model": 103110,
      "analyze factors": 5974,
      "scripts neural": 104270,
      "translation optimal": 125073,
      "optimal transfer": 82173,
      "vocabulary case": 131867,
      "transferring languages": 124240,
      "improve transfer": 52565,
      "neighbor classifier": 76983,
      "search approach": 104300,
      "requires additional": 100243,
      "examples test": 39888,
      "highly expressive": 50321,
      "search improves": 104321,
      "stateoftheart germanenglish": 112669,
      "results average": 101535,
      "pairs improvements": 83561,
      "relevant examples": 98547,
      "attempt create": 9737,
      "websites manually": 132304,
      "pairs major": 83581,
      "kurdish sorani": 58364,
      "sorani kurmanji": 110680,
      "cc byncsa": 15853,
      "byncsa 40": 14990,
      "40 license": 1007,
      "method soft": 69156,
      "decoder experiments": 28009,
      "gains 18": 46380,
      "datasets exploiting": 27476,
      "consists phases": 21488,
      "strong nmt": 113695,
      "analyses reveal": 5463,
      "reveal approach": 102495,
      "models resulting": 73952,
      "final models": 44112,
      "approach pretrain": 7813,
      "pretrain universal": 89979,
      "phrases similar": 87459,
      "languages closer": 60447,
      "pairs jointly": 83566,
      "models carry": 72887,
      "diverse settings": 32848,
      "medium rich": 68246,
      "compared directly": 19358,
      "multiple lowresource": 75609,
      "utilized improve": 130544,
      "corpus code": 23693,
      "decoder achieves": 27992,
      "similar previous": 109123,
      "model converges": 70916,
      "original unlabeled": 82560,
      "german upper": 47924,
      "upper sorbian": 127823,
      "training synthetic": 123897,
      "data achieved": 25564,
      "improvement 10": 52660,
      "germanupper sorbian": 47946,
      "trained online": 123225,
      "additionally experiment": 3606,
      "wmt20 shared": 132869,
      "module based": 74479,
      "model mining": 71516,
      "module adopts": 74477,
      "scores followed": 104174,
      "decoders current": 28064,
      "rely limited": 98717,
      "method entirely": 68808,
      "compared bilingual": 19340,
      "bilingual baseline": 14021,
      "using adapter": 129319,
      "architecture bleu": 8630,
      "explore transferability": 41601,
      "transfer grounded": 124057,
      "aspects multilingual": 9396,
      "papers reported": 84513,
      "hard predict": 49477,
      "limits performance": 64319,
      "data zeroshot": 26635,
      "segmentation recent": 104627,
      "train parallel": 122994,
      "model ignores": 71297,
      "language tag": 60141,
      "bias english": 13802,
      "effectively reduced": 34845,
      "challenges automated": 16136,
      "present parallel": 89639,
      "cope data": 23272,
      "just pretraining": 57468,
      "online neural": 81784,
      "modeling pretraining": 72513,
      "vast quantities": 131279,
      "tasks hand": 119154,
      "magnitude slower": 66387,
      "make inference": 66686,
      "channel approach": 16405,
      "approach fast": 7569,
      "strong pretraining": 113704,
      "pretraining results": 90315,
      "results achieving": 101497,
      "language classifier": 58877,
      "available end": 10990,
      "employing parallel": 36318,
      "operations used": 82075,
      "configuration experiments": 21004,
      "improve mt": 52424,
      "available set": 11102,
      "years challenge": 135244,
      "models smallscale": 74062,
      "corpora case": 23431,
      "appears important": 6812,
      "use static": 128295,
      "order training": 82421,
      "learn test": 62165,
      "different sized": 31431,
      "translation disentangling": 124760,
      "conceptually attractive": 20680,
      "generalizing new": 46844,
      "demonstrate main": 28777,
      "main factor": 66418,
      "points zeroshot": 87887,
      "integration new": 55813,
      "layer outputs": 61739,
      "outputs approach": 83158,
      "language resulting": 60062,
      "set originally": 107519,
      "source monolingual": 110793,
      "originally target": 82570,
      "token vocabulary": 122271,
      "understanding role": 126948,
      "transport ot": 125538,
      "ot problem": 82607,
      "diverse scenarios": 32839,
      "bleu gain": 14283,
      "gpu hours": 48579,
      "increasingly powerful": 53709,
      "developing parallel": 30357,
      "bilingual translation": 14066,
      "corpora composed": 23438,
      "aligned bilingual": 4909,
      "language usually": 60315,
      "data shown": 26446,
      "leveraging small": 63705,
      "pretrained gpt2": 90042,
      "data preliminary": 26258,
      "extreme adaptation": 42591,
      "adaptation personalized": 3249,
      "humans machine": 51089,
      "integrate new": 55764,
      "knowledge exploitation": 57918,
      "model morphology": 71532,
      "morphology source": 74804,
      "framework assess": 45429,
      "important address": 52097,
      "translate new": 124533,
      "shared encoderdecoder": 107979,
      "translation unlike": 125403,
      "nmt jointly": 79882,
      "architecture advantage": 8613,
      "internal language": 56167,
      "restricted size": 101343,
      "size performance": 109939,
      "architecture universal": 8766,
      "public multilingual": 94265,
      "results universal": 102293,
      "little effort": 64805,
      "newly added": 79263,
      "languages achieves": 60384,
      "mitigate negative": 70374,
      "existing strategies": 40298,
      "usage training": 127871,
      "data clean": 25730,
      "intuition good": 56664,
      "wmt language": 132794,
      "approaches employing": 8141,
      "using realworld": 130083,
      "novel zeroshot": 80782,
      "pretrained nmt": 90174,
      "supervised bilingual": 115734,
      "work look": 134620,
      "seek exploit": 104518,
      "easily generalize": 34453,
      "word lookup": 133354,
      "unseen source": 127546,
      "representation new": 99356,
      "artificial noise": 9264,
      "exploiting models": 41473,
      "translation vocabulary": 125424,
      "vocabulary adaptation": 131866,
      "language capacity": 58865,
      "paving way": 85807,
      "unseen scripts": 127540,
      "minor degradation": 70271,
      "degradation translation": 28554,
      "excellent resource": 39919,
      "extremely noisy": 42606,
      "single features": 109735,
      "presents alternative": 89816,
      "learns weights": 63245,
      "filter sentences": 44075,
      "corpora effectively": 23465,
      "effectively provide": 34840,
      "beats strong": 12623,
      "effect domain": 34594,
      "evaluating mt": 39075,
      "corpus special": 24009,
      "benchmarks compare": 12890,
      "analyze effect": 5967,
      "affects final": 4243,
      "successful methods": 115161,
      "employ crosslingual": 36254,
      "modeling sequence": 72544,
      "enhance bilingual": 37464,
      "embeddings empirical": 35681,
      "45 bleu": 1034,
      "paper efforts": 83888,
      "efforts make": 35200,
      "corpus combined": 23697,
      "improvements 24": 52795,
      "evaluations assess": 39453,
      "average quality": 11210,
      "quality ranking": 94768,
      "potential usefulness": 88594,
      "languages conduct": 60462,
      "data compute": 25764,
      "languages researchers": 60846,
      "using modest": 129894,
      "compute resources": 20469,
      "improve supervised": 52557,
      "techniques improving": 119908,
      "translating lowresource": 124591,
      "small quantities": 110190,
      "publiclyavailable parallel": 94337,
      "web resulting": 132250,
      "sentences scanned": 106483,
      "sentences 11": 106190,
      "research nmt": 100572,
      "sparse transformer": 111239,
      "models negative": 73621,
      "language interference": 59221,
      "enhance translation": 37499,
      "based insights": 11774,
      "selects different": 104861,
      "benchmarks largescale": 12912,
      "indispensable training": 53893,
      "solve issue": 110598,
      "translation apply": 124643,
      "language multiple": 59698,
      "improvement 58": 52677,
      "translation target language": 125327,
      "statistical analysis using": 113079,
      "comparable results existing": 19171,
      "generation machine translation": 47466,
      "according text encoding": 2007,
      "different languages instance": 31203,
      "multilingual parallel texts": 75326,
      "parallel texts multilingual": 84691,
      "extraction parallel sentences": 42426,
      "parallel sentences using": 84685,
      "used domain adaptation": 128500,
      "adaptation machine translation": 3230,
      "propose novel languageindependent": 92933,
      "novel languageindependent approach": 80615,
      "spelling word order": 111918,
      "using additional data": 129322,
      "translation language pair": 124890,
      "approaches machine translation": 8224,
      "pair languages given": 83438,
      "linguistic data given": 64461,
      "large target vocabulary": 61289,
      "results compared existing": 101602,
      "recent success neural": 96537,
      "propose method allows": 92762,
      "use large target": 128116,
      "performance measured bleu": 86526,
      "machine translation recent": 66187,
      "translation recent work": 125201,
      "translation shown promising": 125253,
      "availability high quality": 10908,
      "high quality parallel": 50122,
      "abundant monolingual corpora": 1836,
      "machine translation statistical": 66240,
      "extracting parallel data": 42227,
      "report preliminary experiments": 99025,
      "data additional training": 25576,
      "training data usually": 123565,
      "training data allows": 123426,
      "experiments billion words": 40809,
      "scenario training data": 103835,
      "heldout test data": 49690,
      "data taking account": 26537,
      "models monolingual data": 73582,
      "stateofthe art performance": 112562,
      "language pairs using": 59785,
      "important role boosting": 52238,
      "role boosting fluency": 103161,
      "information language model": 54712,
      "obtain substantial improvements": 81333,
      "framework neural machine": 45625,
      "significantly improves bleu": 108945,
      "learning method improve": 62721,
      "semisupervised learning neural": 105609,
      "machine translation endtoend": 65970,
      "parallel corpora usually": 84611,
      "monolingual corpora improve": 74558,
      "training nmt models": 123751,
      "encoder decoder respectively": 36513,
      "improvements stateoftheart smt": 52918,
      "stateoftheart smt nmt": 112957,
      "smt nmt systems": 110293,
      "corpora source target": 23592,
      "paper examine use": 83903,
      "parallel data source": 84653,
      "data source target": 26474,
      "encoderdecoder neural network": 36613,
      "neural network neural": 78364,
      "bilingual training data": 14065,
      "training data obtained": 123504,
      "learn unlabeled data": 62179,
      "unlabeled data duallearning": 127381,
      "approach neural machine": 7737,
      "learning monolingual data": 62802,
      "bleu points addition": 14304,
      "large data training": 61071,
      "propose alleviate problem": 92542,
      "neural network imitates": 78306,
      "englishtofrench englishtogerman translation": 37442,
      "englishtogerman translation tasks": 37450,
      "translation tasks using": 125368,
      "effectiveness statistical machine": 34954,
      "corpus used training": 24058,
      "corpus study english": 24023,
      "sentence similarity metric": 106074,
      "improve quality extracted": 52507,
      "source sentences multiple": 110822,
      "machine translation msnmt": 66062,
      "translation nmt architecture": 125030,
      "gives competitive results": 48192,
      "synthetic parallel corpus": 116637,
      "pseudo parallel corpus": 94183,
      "bidirectional translation tasks": 13983,
      "representation words sentences": 99470,
      "parallel sentences comparable": 84681,
      "sentences comparable corpora": 106247,
      "shared task using": 108104,
      "suffers data scarcity": 115251,
      "language pairs domains": 59746,
      "parallel corpora available": 84595,
      "significantly improves baseline": 108943,
      "various language pairs": 131115,
      "novel approach generate": 80478,
      "generate synthetic data": 47026,
      "given parallel corpus": 48077,
      "target language parallel": 117647,
      "baseline neural machine": 12271,
      "improved state art": 52646,
      "model present empirical": 71767,
      "related languages neural": 97875,
      "method improve neural": 68885,
      "language pair method": 59730,
      "machine translation decoding": 65945,
      "past years witnessed": 85658,
      "years witnessed rapid": 135313,
      "witnessed rapid progress": 132754,
      "translation quality standard": 125186,
      "addressing data sparsity": 3810,
      "sentences different languages": 106278,
      "domain specific feature": 33647,
      "specific feature engineering": 111442,
      "neural networks results": 78559,
      "performance statistical machine": 86751,
      "learning parallel sentence": 62876,
      "available noncommercial research": 11058,
      "knowledge largest publicly": 58041,
      "translation paper proposed": 125094,
      "problem low resourced": 91112,
      "small parallel corpus": 110182,
      "languages results suggest": 60852,
      "parallel corpus major": 84620,
      "essential tasks natural": 38569,
      "processing machine translation": 91707,
      "multilingual parallel corpora": 75324,
      "corpora paper introduce": 23553,
      "baseline statistical machine": 12316,
      "image captioning neural": 51771,
      "languages limited parallel": 60699,
      "data proposed approach": 26295,
      "bleu strong baseline": 14339,
      "furthermore proposed approach": 46209,
      "quality statistical machine": 94801,
      "leveraging monolingual data": 63689,
      "learning sourcetotarget targettosource": 63042,
      "sourcetotarget targettosource nmt": 110954,
      "targettosource nmt models": 117806,
      "nmt models language": 79916,
      "results chineseenglish englishgerman": 101572,
      "recently researchers shown": 96755,
      "data used create": 26601,
      "translation capabilities nmt": 124685,
      "using backtranslated data": 129380,
      "word embeddings useful": 133235,
      "useful neural machine": 128910,
      "20 bleu points": 494,
      "language pairs work": 59787,
      "simpler having fewer": 109558,
      "outperforming state art": 82830,
      "state art 11": 112424,
      "11 bleu points": 216,
      "language data effective": 58933,
      "named entity translation": 76129,
      "demonstrate model learns": 28792,
      "models approach produces": 72763,
      "comparable bleu scores": 19135,
      "machine translation aims": 65891,
      "adversarial networks gans": 4153,
      "translation new approach": 125017,
      "approach achieve significant": 7296,
      "achieve significant improvements": 2537,
      "machine translation requires": 66200,
      "parallel training text": 84697,
      "address challenging problem": 3666,
      "including semantic parsing": 53365,
      "semantic parsing syntactic": 105168,
      "parsing syntactic parsing": 85253,
      "knowledge translation model": 58216,
      "effectiveness multitask learning": 34919,
      "news articles collected": 79291,
      "problem classification problem": 90961,
      "feature extraction methods": 43279,
      "texts different languages": 121501,
      "annotated corpus publicly": 6150,
      "machine translation synthetic": 66246,
      "nmt improve performance": 79875,
      "augmented parallel data": 10333,
      "incorporate source target": 53488,
      "word embeddings impact": 133127,
      "results suggest pretrained": 102234,
      "indomain monolingual data": 53972,
      "transformer recurrent neural": 124374,
      "translation recently neural": 125204,
      "recently neural machine": 96716,
      "single multilingual nmt": 109770,
      "shared representation space": 108006,
      "pairs seen training": 83633,
      "nmt model capable": 79897,
      "morphology word order": 74811,
      "errors neural machine": 38390,
      "problem context machine": 90980,
      "improved bleu scores": 52597,
      "phrasebased smt systems": 87390,
      "language pairs englishhindi": 59753,
      "languages challenging task": 60437,
      "nahuatl wixarika yorem": 76003,
      "wixarika yorem nokki": 132758,
      "crosslingual document embedding": 24943,
      "method improves current": 68889,
      "translation nmt framework": 125033,
      "documents different languages": 33219,
      "training objective function": 123757,
      "new method called": 79009,
      "different languages training": 31214,
      "training data includes": 123478,
      "machine translation new": 66085,
      "new languages paper": 78983,
      "training data related": 123524,
      "achieving bleu scores": 2937,
      "17 bleu points": 401,
      "data synthetic data": 26532,
      "sentences randomly sampled": 106465,
      "wmt news translation": 132805,
      "translation task method": 125335,
      "12 bleu points": 249,
      "effective method improve": 34706,
      "machine translation monolingual": 66060,
      "translation monolingual data": 124971,
      "parallel training corpus": 84693,
      "stronger training signal": 113737,
      "translation nmt typically": 125062,
      "fixed language model": 44597,
      "translation model tm": 124942,
      "training data added": 123421,
      "source sentence rely": 110814,
      "monolingual data backtranslation": 74566,
      "machine translation transfer": 66276,
      "translation transfer learning": 125392,
      "technique neural machine": 119803,
      "translation rare words": 125196,
      "using phrasebased models": 130015,
      "lexical resources improving": 63802,
      "information multiple sources": 54780,
      "multiple sources systems": 75700,
      "translation sequence sequence": 125242,
      "models capturing lexical": 72884,
      "capturing lexical syntactic": 15475,
      "paper develop novel": 83865,
      "novel sequence sequence": 80723,
      "model improves translation": 71320,
      "decoding unsupervised neural": 28146,
      "synthetic bilingual data": 116613,
      "data training propose": 26575,
      "improved using backtranslation": 52654,
      "preliminary experiments approach": 89269,
      "wmt16 germanenglish news": 132833,
      "germanenglish news translation": 47939,
      "translation task translation": 125344,
      "task translation directions": 118812,
      "machine translation current": 65935,
      "encoder decoder using": 36515,
      "improvements conventional nmt": 52836,
      "translation nmt train": 125061,
      "nmt train nmt": 79994,
      "train nmt model": 122991,
      "language different word": 58964,
      "different word order": 31571,
      "target language available": 117632,
      "match word order": 67372,
      "used transfer learning": 128828,
      "set model achieves": 107497,
      "report experiments english": 98999,
      "significant bleu score": 108729,
      "bleu score improvements": 14327,
      "significant progress recently": 108852,
      "remains challenge work": 98788,
      "challenge work propose": 16119,
      "model experiments chineseenglish": 71132,
      "better exploit limited": 13583,
      "language pairs translation": 59784,
      "machine translation dynamic": 65962,
      "significant performance gain": 108837,
      "nmt model scratch": 79899,
      "competitive performance respect": 19666,
      "surprisingly good performance": 116138,
      "observed training time": 81238,
      "models analysis shows": 72750,
      "translation smt models": 125271,
      "built pretrained language": 14933,
      "tasks method outperforms": 119295,
      "method outperforms strong": 69034,
      "baseline achieves new": 12181,
      "parallel corpus generation": 84618,
      "hindienglish parallel corpus": 50492,
      "improvements lowresource languages": 52870,
      "face paucity data": 42733,
      "consistent improvements strong": 21389,
      "knowledge graphs enhance": 57983,
      "semantic feature extraction": 105050,
      "translation quality investigate": 125175,
      "investigate different strategies": 56744,
      "different strategies incorporating": 31453,
      "knowledge graph augmented": 57955,
      "graph augmented neural": 48763,
      "consistent improvements bleu": 21384,
      "translation multiple source": 124993,
      "multilingual nmt models": 75317,
      "experiments largescale dataset": 40983,
      "issue propose alternative": 57023,
      "proposed approach consistently": 93188,
      "languages article present": 60412,
      "corpus training statistical": 24046,
      "aligned sentences parallel": 4920,
      "gained lot traction": 46367,
      "distributed representations open": 32620,
      "survey existing literature": 116173,
      "various approaches based": 131040,
      "machine translation common": 65925,
      "language representation neural": 60046,
      "improve lowresource neural": 52408,
      "translation nmt proven": 125051,
      "achieve impressive results": 2489,
      "japanesechinese scientific paper": 57207,
      "scientific paper excerpt": 104015,
      "paper excerpt corpus": 83906,
      "machine translation case": 65917,
      "requiring large amounts": 100345,
      "nmt systems recent": 79979,
      "especially helpful lowresource": 38458,
      "best practices lowresource": 13412,
      "propose use simple": 93145,
      "422 bleu points": 1019,
      "machine translation given": 65993,
      "data selection methods": 26387,
      "synthetic data nmt": 116623,
      "selection synthetic data": 104837,
      "experiments shed light": 41133,
      "standard transformer model": 112329,
      "transformer model architecture": 124340,
      "bleu points 12": 14303,
      "12 language pairs": 254,
      "submission achieved best": 114726,
      "performance multilingual translation": 86550,
      "shows close results": 108561,
      "multilingual transfer learning": 75391,
      "lowresource scenarios multilingual": 65559,
      "exploit outofdomain data": 41434,
      "translation nmt standard": 125056,
      "models language pairs": 73455,
      "syntactically divergent languages": 116511,
      "machine translation conventional": 65931,
      "results nist chineseenglish": 101991,
      "absolute improvements strong": 1741,
      "models achieve best": 72658,
      "parallel data used": 84660,
      "augmenting training set": 10348,
      "synthetic data using": 116624,
      "nmt statistical machine": 79966,
      "results reveal models": 102139,
      "starting point training": 112411,
      "years neural machine": 135278,
      "experiments languages english": 40976,
      "languages english italian": 60541,
      "monolingual data results": 74580,
      "high quality indomain": 50119,
      "indomain parallel corpora": 53977,
      "model learns combine": 71441,
      "sources information order": 110899,
      "produce higher quality": 91901,
      "models work relatively": 74362,
      "reduce parameter size": 97351,
      "recent progress neural": 96501,
      "progress neural machine": 92167,
      "model paper propose": 71697,
      "modifications model architecture": 74443,
      "model architecture approach": 70683,
      "approach does need": 7505,
      "achieve better translation": 2434,
      "paper investigate problem": 84020,
      "largest dataset date": 61532,
      "training time propose": 123926,
      "pretraining significantly improves": 90325,
      "significantly improves nmt": 108949,
      "improves nmt performance": 53001,
      "significant performance boost": 108833,
      "test set accordingly": 120493,
      "translation recent years": 125202,
      "available supervised data": 11119,
      "training models trained": 123716,
      "improve quality machine": 52510,
      "translation current stateoftheart": 124728,
      "current stateoftheart machine": 25355,
      "paper explore effect": 83917,
      "methods leverage monolingual": 69592,
      "language pairs large": 59764,
      "learning techniques paper": 63096,
      "unseen languages multilingual": 127531,
      "languages multilingual neural": 60744,
      "lowresource languages lrl": 65518,
      "highresource languages hrl": 50426,
      "work explore problem": 134515,
      "selection method outperforms": 104801,
      "method outperforms current": 69014,
      "outperforms current approaches": 82874,
      "high quality dataset": 50114,
      "demonstrate new approach": 28809,
      "new approach neural": 78795,
      "present metrics evaluating": 89557,
      "parallel corpora limited": 84604,
      "performance comparable model": 86223,
      "reduced training time": 97376,
      "challenge paper propose": 16079,
      "paper propose explore": 84237,
      "beam search algorithm": 12599,
      "search algorithm decoding": 104295,
      "lack parallel training": 58734,
      "syntactic semantic levels": 116470,
      "experimental results reveal": 40690,
      "results reveal methods": 102138,
      "simplified traditional chinese": 109608,
      "richresource language pairs": 102818,
      "limited model capacity": 64254,
      "translation source language": 125278,
      "translation paper demonstrates": 125086,
      "performance gains wide": 86406,
      "effective method generate": 34704,
      "generate large number": 46963,
      "parallel sentences training": 84684,
      "improved neural machine": 52620,
      "shown outperform standard": 108501,
      "data used train": 26602,
      "available propose novel": 11083,
      "machine translation africa": 65889,
      "accurate machine translation": 2342,
      "model publicly available": 71830,
      "available low resource": 11036,
      "model semisupervised learning": 71971,
      "model report results": 71904,
      "reference language based": 97525,
      "leverage weakly supervised": 63638,
      "make use information": 66744,
      "translation quality bleu": 125164,
      "significant improvements machine": 108795,
      "improvements machine translation": 52873,
      "machine translation usage": 66289,
      "machine translation able": 65880,
      "languages monolingual data": 60735,
      "translation tasks english": 125353,
      "scenarios unsupervised neural": 103871,
      "language pairs proposed": 59777,
      "pairs proposed methods": 83616,
      "conventional unmt systems": 22913,
      "extensive empirical evaluation": 41870,
      "source target corpora": 110831,
      "languages use different": 60945,
      "use different scripts": 128002,
      "machine translation relies": 66198,
      "328 bleu points": 940,
      "training sets languages": 123856,
      "overall performance paper": 83247,
      "consistently outperforms heuristic": 21433,
      "terms average performance": 120279,
      "stateoftheart unsupervised methods": 113020,
      "methods neural machine": 69637,
      "language pairs varying": 59786,
      "single reference translation": 109786,
      "task paper investigates": 118505,
      "focus finetuning model": 44766,
      "parallel corpora important": 84601,
      "knowledge language models": 58032,
      "previous work requires": 90544,
      "mt systems used": 74997,
      "addressed transfer learning": 3790,
      "release code pretrained": 98440,
      "lowresource languages transfer": 65528,
      "transfer learning empirical": 124091,
      "lowresource languages use": 65529,
      "gebru et al": 46574,
      "popular machine translation": 88098,
      "encoder decoder modules": 36509,
      "resource language pair": 100855,
      "zero shot translation": 135482,
      "language pairs order": 59770,
      "dimensionality reduction word": 31761,
      "reduce memory usage": 97338,
      "sanskrithindi parallel corpus": 103637,
      "using parallel sentences": 130004,
      "translation mt based": 124979,
      "uses bilingual dictionary": 129201,
      "target language experiments": 117638,
      "language pairs approaches": 59739,
      "crosslingual word embedding": 25044,
      "work available public": 134394,
      "indian languages present": 53825,
      "translation current work": 124729,
      "corpus sentiment analysis": 23990,
      "sentiment analysis sentences": 106652,
      "character based neural": 16418,
      "finetuning recent work": 44510,
      "work multilingual pretraining": 134643,
      "multilingual pretraining demonstrated": 75342,
      "compared multilingual models": 19396,
      "models trained scratch": 74227,
      "data particularly important": 26224,
      "training evaluation data": 123617,
      "machine translation applications": 65894,
      "different languages recently": 31212,
      "recently researchers started": 96756,
      "machine translation article": 65899,
      "previous work area": 90519,
      "knowledge improves performance": 58008,
      "considerable attention given": 21243,
      "information different levels": 54487,
      "languages given particular": 60611,
      "despite widespread adoption": 29743,
      "google translate api": 48525,
      "shared task results": 108090,
      "data models code": 26138,
      "language model languages": 59339,
      "data available language": 25674,
      "new language propose": 78977,
      "bleu points translation": 14314,
      "new pretraining method": 79081,
      "pretraining method called": 90284,
      "source sentence translation": 110817,
      "target language specifically": 117654,
      "perform lexicon induction": 86027,
      "lexicon induction unsupervised": 63888,
      "unsupervised word embedding": 127753,
      "mapping source target": 67148,
      "adopts encoderdecoder framework": 3911,
      "close performance supervised": 17827,
      "machine translation evaluate": 65974,
      "pave way future": 85805,
      "languages data code": 60482,
      "machine translation rare": 66186,
      "language pairs englishfrench": 59750,
      "pairs englishfrench englishgerman": 83530,
      "stateoftheart supervised model": 112981,
      "superior performance proposed": 115690,
      "languages different scripts": 60505,
      "nearest neighbor classifier": 76720,
      "approach requires additional": 7860,
      "stateoftheart germanenglish translation": 112670,
      "language pairs improvements": 59761,
      "pairs improvements bleu": 83562,
      "english german chinese": 37148,
      "efficient machine translation": 35088,
      "kurdish sorani kurmanji": 58365,
      "cc byncsa 40": 15854,
      "byncsa 40 license": 14991,
      "nmt models translate": 79923,
      "wmt14 englishgerman englishfrench": 132814,
      "consistently significantly improves": 21441,
      "performance strong nmt": 86760,
      "models extensive analyses": 73214,
      "better generalization capability": 13595,
      "machine translation leveraging": 66029,
      "language pairs obtain": 59769,
      "multiple lowresource language": 75610,
      "able improve translation": 1655,
      "code data pretrained": 18083,
      "data pretrained models": 26271,
      "tasks closely related": 118980,
      "german upper sorbian": 47925,
      "parallel data achieved": 84632,
      "improvement 10 bleu": 52661,
      "10 bleu points": 125,
      "machine translation shared": 66213,
      "shared task language": 108059,
      "current endtoend approaches": 25281,
      "experiments different languages": 40908,
      "translation knowledge transfer": 124885,
      "small parallel data": 110183,
      "challenges neural machine": 16182,
      "present parallel data": 89640,
      "parallel data set": 84652,
      "cope data scarcity": 23273,
      "nlp tasks hand": 79724,
      "recently shown achieve": 96764,
      "models order magnitude": 73670,
      "results achieving new": 101498,
      "data improve mt": 26025,
      "success recent years": 115126,
      "order training samples": 82422,
      "machine translation source": 66233,
      "test set test": 120523,
      "test set originally": 120510,
      "originally target language": 82571,
      "optimal transport ot": 82175,
      "transport ot problem": 125539,
      "monolingual data target": 74581,
      "especially lowresource language": 38471,
      "indomain parallel data": 53978,
      "data preliminary experiments": 26259,
      "effectiveness method generate": 34906,
      "extreme adaptation personalized": 42592,
      "translation using bilingual": 125417,
      "using bilingual dictionaries": 129410,
      "led significant improvements": 63283,
      "morphology source target": 74805,
      "learning methods neural": 62740,
      "word representations important": 133456,
      "achieves competitive better": 2769,
      "competitive better results": 19649,
      "training process using": 123795,
      "alignment training data": 4996,
      "training data clean": 123443,
      "brings significant improvement": 14651,
      "test data different": 120435,
      "data different domains": 25842,
      "pretrained nmt model": 90175,
      "translation crosslingual word": 124724,
      "crosslingual embeddings word": 24949,
      "encoder sentence representation": 36574,
      "target language use": 117659,
      "machine translation vocabulary": 66298,
      "languages unseen scripts": 60942,
      "evaluating mt models": 39076,
      "benchmarks compare performance": 12891,
      "benchmark future research": 12828,
      "language models unsupervised": 59682,
      "methods unsupervised neural": 69826,
      "embeddings empirical results": 35682,
      "demonstrate improved performance": 28761,
      "empirical study neural": 36203,
      "new stateoftheart result": 79166,
      "data paper address": 26207,
      "results machine learning": 101898,
      "sentence pairs english": 105969,
      "publiclyavailable parallel corpora": 94338,
      "train model learn": 122958,
      "strong baselines multiple": 113651,
      "translation parallel corpora": 125098,
      "language pairs exist": 59755,
      "domain adaptation machine translation": 33438,
      "propose novel languageindependent approach": 92934,
      "recent success neural machine": 96538,
      "use large target vocabulary": 128117,
      "neural machine translation recent": 78086,
      "machine translation recent work": 66190,
      "machine translation shown promising": 66216,
      "translation shown promising results": 125254,
      "statistical machine translation statistical": 113118,
      "machine translation statistical machine": 66241,
      "data additional training data": 25577,
      "plays important role boosting": 87732,
      "important role boosting fluency": 52239,
      "framework neural machine translation": 45626,
      "transfer learning method improve": 124101,
      "translation paper propose novel": 125093,
      "neural machine translation endtoend": 77998,
      "improvements stateoftheart smt nmt": 52919,
      "experiments demonstrate method significantly": 40886,
      "approach neural machine translation": 7738,
      "suffer data scarcity problem": 115214,
      "englishtofrench englishtogerman translation tasks": 37443,
      "effectiveness statistical machine translation": 34955,
      "neural machine translation msnmt": 78034,
      "machine translation nmt architecture": 66094,
      "parallel sentences comparable corpora": 84682,
      "target language parallel corpus": 117648,
      "baseline neural machine translation": 12272,
      "require large amounts training": 100159,
      "related languages neural machine": 97876,
      "translation present simple method": 125136,
      "neural machine translation decoding": 77986,
      "past years witnessed rapid": 85659,
      "domain specific feature engineering": 33648,
      "performance statistical machine translation": 86752,
      "freely available noncommercial research": 45784,
      "knowledge largest publicly available": 58042,
      "machine translation paper proposed": 66146,
      "essential tasks natural language": 38570,
      "language processing machine translation": 59884,
      "baseline statistical machine translation": 12317,
      "quality statistical machine translation": 94802,
      "sourcetotarget targettosource nmt models": 110955,
      "results chineseenglish englishgerman translation": 101573,
      "useful neural machine translation": 128911,
      "translation machine translation systems": 124923,
      "outperforming state art 11": 82831,
      "generative adversarial networks gans": 47714,
      "neural machine translation requires": 78095,
      "including semantic parsing syntactic": 53366,
      "semantic parsing syntactic parsing": 105169,
      "annotated corpus publicly available": 6151,
      "neural machine translation synthetic": 78109,
      "machine translation recently neural": 66193,
      "translation recently neural machine": 125205,
      "recently neural machine translation": 96717,
      "errors neural machine translation": 38391,
      "problem context machine translation": 90981,
      "nahuatl wixarika yorem nokki": 76004,
      "machine translation nmt framework": 66097,
      "crosslingual document classification task": 24942,
      "different languages training data": 31215,
      "machine translation monolingual data": 66061,
      "machine translation nmt typically": 66125,
      "machine translation transfer learning": 66277,
      "technique neural machine translation": 119804,
      "machine translation sequence sequence": 66210,
      "translation sequence sequence seq2seq": 125243,
      "models capturing lexical syntactic": 72885,
      "wmt16 germanenglish news translation": 132834,
      "germanenglish news translation task": 47940,
      "news translation task translation": 79390,
      "translation task translation directions": 125345,
      "neural machine translation current": 77982,
      "machine translation nmt train": 66124,
      "nmt train nmt model": 79995,
      "different word order source": 31572,
      "significant improvement translation quality": 108780,
      "significant bleu score improvements": 108730,
      "remains challenge work propose": 98789,
      "neural machine translation dynamic": 77995,
      "machine translation smt models": 66228,
      "built pretrained language models": 14934,
      "baseline achieves new stateoftheart": 12182,
      "knowledge graph augmented neural": 57956,
      "translation multiple source languages": 124994,
      "aligned sentences parallel corpus": 4921,
      "training paper propose new": 123772,
      "improve lowresource neural machine": 52409,
      "machine translation nmt proven": 66114,
      "japanesechinese scientific paper excerpt": 57208,
      "scientific paper excerpt corpus": 104016,
      "neural machine translation case": 77979,
      "machine translation case study": 65918,
      "machine translation nmt standard": 66119,
      "nmt models language pairs": 79917,
      "propose simple effective solution": 93058,
      "neural machine translation conventional": 77981,
      "experimental results nist chineseenglish": 40662,
      "nmt neural machine translation": 79929,
      "translation nmt models achieve": 125040,
      "nmt statistical machine translation": 79967,
      "used starting point training": 128779,
      "recent years neural machine": 96624,
      "years neural machine translation": 135279,
      "progress neural machine translation": 92168,
      "paper propose simple method": 84325,
      "achieve better translation quality": 2435,
      "significantly improves nmt performance": 108950,
      "machine translation recent years": 66191,
      "improve quality machine translation": 52511,
      "machine translation current stateoftheart": 65936,
      "current stateoftheart machine translation": 25356,
      "methods leverage monolingual data": 69593,
      "neural machine translation stateoftheart": 78107,
      "languages multilingual neural machine": 60745,
      "data selection method outperforms": 26386,
      "method outperforms current approaches": 69015,
      "multilingual models data augmentation": 75287,
      "new approach neural machine": 78796,
      "beam search algorithm decoding": 12600,
      "datasets approach significantly outperforms": 27318,
      "approach significantly outperforms strong": 7917,
      "machine translation paper demonstrates": 66140,
      "significant performance gains wide": 108840,
      "improved neural machine translation": 52621,
      "data available propose novel": 25680,
      "shows significant performance improvements": 108630,
      "data achieve promising results": 25562,
      "method significantly improves translation": 69139,
      "significantly improves translation quality": 108960,
      "improves translation quality bleu": 53061,
      "significant improvements machine translation": 108796,
      "scenarios unsupervised neural machine": 103872,
      "results language pairs proposed": 101876,
      "language pairs proposed methods": 59778,
      "pairs proposed methods substantially": 83617,
      "languages use different scripts": 60946,
      "overall performance paper propose": 83248,
      "performance paper propose method": 86598,
      "methods neural machine translation": 69638,
      "training data lowresource languages": 123488,
      "data work propose novel": 26634,
      "quality mt systems used": 94725,
      "release code pretrained models": 98441,
      "low resource language pair": 65388,
      "sentences paper propose new": 106429,
      "machine translation mt based": 66065,
      "various language pairs approaches": 131116,
      "work available public use": 134395,
      "machine translation current work": 65937,
      "character based neural machine": 16419,
      "machine translation model using": 66049,
      "using automated metrics bleu": 129365,
      "work multilingual pretraining demonstrated": 134644,
      "multilingual machine translation models": 75276,
      "competitive performance stateoftheart models": 19668,
      "new language propose novel": 78978,
      "mapping source target languages": 67149,
      "achieves significant improvements baselines": 2858,
      "language pairs englishfrench englishgerman": 59751,
      "language pairs improvements bleu": 59762,
      "cc byncsa 40 license": 15855,
      "improve performance neural machine": 52475,
      "consistently significantly improves performance": 21442,
      "neural machine translation leveraging": 78018,
      "multiple lowresource language pairs": 75611,
      "code data pretrained models": 18084,
      "data pretrained models available": 26272,
      "language models trained monolingual": 59672,
      "improvement 10 bleu points": 52662,
      "machine translation shared task": 66214,
      "task paper describes submission": 118503,
      "challenges neural machine translation": 16183,
      "neural machine translation shown": 78099,
      "neural machine translation source": 78105,
      "test set test set": 120524,
      "optimal transport ot problem": 82176,
      "monolingual data target language": 74582,
      "especially lowresource language pairs": 38472,
      "led significant improvements machine": 63284,
      "neural machine translation able": 77966,
      "datasets proposed model achieves": 27642,
      "proposed model achieves competitive": 93427,
      "model achieves competitive better": 70566,
      "methods unsupervised neural machine": 69827,
      "embeddings empirical results demonstrate": 35683,
      "achieves new stateoftheart result": 2824,
      "recent success neural machine translation": 96539,
      "neural machine translation recent work": 78089,
      "machine translation shown promising results": 66217,
      "statistical machine translation statistical machine": 113119,
      "machine translation statistical machine translation": 66242,
      "plays important role boosting fluency": 87733,
      "framework neural machine translation nmt": 45627,
      "machine translation paper propose novel": 66145,
      "problems neural machine translation nmt": 91351,
      "effectiveness statistical machine translation smt": 34956,
      "neural machine translation nmt architecture": 78045,
      "results proposed method significantly improves": 102075,
      "languages neural machine translation nmt": 60765,
      "require large amounts training data": 100160,
      "related languages neural machine translation": 97877,
      "machine translation present simple method": 66167,
      "natural language processing machine translation": 76434,
      "results chineseenglish englishgerman translation tasks": 101574,
      "stateoftheart neural machine translation nmt": 112781,
      "performance neural machine translation nmt": 86567,
      "machine translation machine translation systems": 66039,
      "including semantic parsing syntactic parsing": 53367,
      "neural machine translation recently neural": 78092,
      "machine translation recently neural machine": 66194,
      "translation recently neural machine translation": 125206,
      "recently neural machine translation nmt": 96718,
      "neural machine translation nmt framework": 78048,
      "neural machine translation nmt typically": 78066,
      "machine translation sequence sequence seq2seq": 66211,
      "translation sequence sequence seq2seq models": 125244,
      "wmt16 germanenglish news translation task": 132835,
      "news translation task translation directions": 79391,
      "approaches neural machine translation nmt": 8256,
      "neural machine translation nmt train": 78065,
      "knowledge neural machine translation nmt": 58084,
      "improve lowresource neural machine translation": 52410,
      "lowresource neural machine translation nmt": 65546,
      "neural machine translation nmt proven": 78057,
      "japanesechinese scientific paper excerpt corpus": 57209,
      "improving neural machine translation nmt": 53130,
      "neural machine translation nmt standard": 78061,
      "nmt neural machine translation nmt": 79930,
      "machine translation nmt models achieve": 66104,
      "nmt statistical machine translation smt": 79968,
      "statistical machine translation smt models": 113115,
      "recent years neural machine translation": 96625,
      "years neural machine translation nmt": 135280,
      "neural machine translation recent years": 78090,
      "neural machine translation current stateoftheart": 77983,
      "languages multilingual neural machine translation": 60746,
      "new approach neural machine translation": 78797,
      "approach neural machine translation nmt": 7739,
      "neural machine translation paper demonstrates": 78071,
      "scenarios unsupervised neural machine translation": 103873,
      "experimental results language pairs proposed": 40629,
      "results language pairs proposed methods": 101877,
      "language pairs proposed methods substantially": 59779,
      "models neural machine translation nmt": 73629,
      "character based neural machine translation": 16420,
      "based neural machine translation model": 11890,
      "neural machine translation model using": 78027,
      "code data pretrained models available": 18085,
      "language models trained monolingual data": 59673,
      "led significant improvements machine translation": 63285,
      "methods unsupervised neural machine translation": 69828,
      "discourage": 32035,
      "evokes": 39689,
      "happily": 49441,
      "suicide": 115384,
      "emission": 36076,
      "deduce": 28184,
      "modernday": 74428,
      "lives": 64853,
      "tension": 120185,
      "bws": 14988,
      "commerce": 18833,
      "acted": 3073,
      "improvised": 53185,
      "sad": 103531,
      "intramodality": 56339,
      "intermodality": 56156,
      "volatile": 131933,
      "gestures": 47949,
      "stl": 113352,
      "301": 914,
      "regressors": 97720,
      "fulfillment": 45927,
      "instruments": 55717,
      "blacklivesmatter": 14260,
      "threelayer": 121938,
      "604": 1155,
      "messenger": 68526,
      "ekmans": 35222,
      "lexicondriven": 63920,
      "051": 35,
      "svr": 116250,
      "741": 1238,
      "metaphorical": 68564,
      "intricacies": 56347,
      "bimodal": 14114,
      "zadeh": 135467,
      "intermodal": 56155,
      "596": 1136,
      "unweighted": 127776,
      "supposedly": 116045,
      "reusability": 102484,
      "printed": 90687,
      "worsened": 135069,
      "accountable": 2046,
      "tells": 120071,
      "iiidyt": 51723,
      "underestimates": 126657,
      "adheres": 3843,
      "bearing": 12615,
      "sentimentspecific": 106837,
      "deepmoji": 28466,
      "stark": 112392,
      "proceeded": 91410,
      "sentiwordnet": 106840,
      "featurelevel": 43344,
      "014": 6,
      "surprised": 116124,
      "2step": 894,
      "feels": 43865,
      "affection": 4233,
      "confronts": 21047,
      "boil": 14401,
      "postures": 88532,
      "humanizing": 51045,
      "testsets": 120616,
      "mystery": 75999,
      "witnessing": 132755,
      "evoke": 39687,
      "depicted": 29297,
      "enjoyment": 37545,
      "foremost": 45046,
      "uitvsmec": 126546,
      "mediating": 68191,
      "ecb": 34499,
      "reserve": 100731,
      "scrutiny": 104274,
      "president": 89944,
      "moral": 74660,
      "democracy": 28644,
      "extractthenclassify": 42583,
      "aesthetic": 4207,
      "anxiety": 6762,
      "childhood": 16721,
      "highconfidence": 50151,
      "bruteforce": 14730,
      "extrapolating": 42587,
      "unattainable": 126592,
      "forethought": 45055,
      "regularizes": 97771,
      "cautionary": 15844,
      "carriers": 15547,
      "morality": 74661,
      "divisive": 32922,
      "altering": 5214,
      "instill": 55696,
      "positivity": 88356,
      "bangladesh": 11417,
      "transgender": 124478,
      "medial": 68187,
      "cohens": 18288,
      "0876": 93,
      "experiencer": 40445,
      "liking": 64150,
      "insufficiently": 55729,
      "pulling": 94384,
      "emotionaware": 36125,
      "spanprediction": 111194,
      "coexist": 18251,
      "1985": 457,
      "multidimension": 75066,
      "secured": 104496,
      "alters": 5256,
      "2stage": 892,
      "emotions text": 36131,
      "higher dimensional": 50178,
      "represent richer": 99134,
      "richer set": 102808,
      "different notions": 31306,
      "sentiment different": 106732,
      "association lexicon": 9621,
      "polarity words": 87933,
      "words positive": 134127,
      "annotation crowdsourcing": 6295,
      "solutions address": 110572,
      "help obtain": 49760,
      "level word": 63521,
      "higher interannotator": 50183,
      "agreement obtained": 4397,
      "way propose": 132123,
      "unprecedented amounts": 127490,
      "paper sentiment": 84418,
      "collections text": 18504,
      "search using": 104361,
      "using google": 129711,
      "cooccurring words": 23248,
      "lexicon crowdsourcing": 63878,
      "use compare": 127944,
      "suicide notes": 115385,
      "genders use": 46627,
      "visualizations help": 131834,
      "personality detection": 87129,
      "person pronouns": 87105,
      "verbs sentiment": 131562,
      "using coarse": 129447,
      "features svm": 43750,
      "features leads": 43588,
      "leads statistically": 61966,
      "analysis typical": 5916,
      "audio textual": 10240,
      "approach perform": 7792,
      "videos based": 131664,
      "visual clues": 131786,
      "stateoftheart computational": 112615,
      "facial expressions": 42758,
      "speech fundamental": 111688,
      "accuracy 84": 2075,
      "task demonstrating": 118065,
      "relies shared": 98659,
      "success measured": 115097,
      "automatic understanding": 10706,
      "support groups": 115971,
      "need incorporate": 76822,
      "fashion train": 43122,
      "feature combinations": 43257,
      "facebook pages": 42740,
      "speech considered": 111661,
      "trained sequence": 123273,
      "function allows": 45997,
      "allows consider": 5135,
      "comparison recent": 19568,
      "quality proposed": 94765,
      "human verbal": 50987,
      "problem integrating": 91088,
      "affective information": 4237,
      "lstm long": 65638,
      "memory language": 68319,
      "categories proposed": 15748,
      "existing manually": 40170,
      "automatic lexicons": 10579,
      "scaling bws": 103772,
      "reliable finegrained": 98614,
      "building natural": 14868,
      "beings recognize": 12685,
      "analysis design": 5570,
      "protocol automatic": 93674,
      "context behavioral": 22020,
      "analysis designed": 5571,
      "language levels": 59249,
      "fusion automatic": 46232,
      "automatic selection": 10648,
      "selection relevant": 104822,
      "features high": 43539,
      "space automatic": 110973,
      "speech emotion": 111679,
      "recognition study": 97015,
      "performance strongly": 86762,
      "depends type": 29295,
      "features furthermore": 43521,
      "detection textual": 30082,
      "states generated": 113045,
      "communicate using": 19020,
      "messaging applications": 68524,
      "based embeddings": 11671,
      "work evaluated": 134496,
      "setup relevant": 107856,
      "paper pose": 84072,
      "pose problem": 88253,
      "analysis modeling": 5701,
      "create datasets": 24610,
      "create benchmark": 24602,
      "manifest language": 66919,
      "present shared": 89695,
      "learning setups": 63020,
      "dataset shared": 27194,
      "improve understanding": 52579,
      "recognition wild": 97042,
      "using aggregated": 129327,
      "improve generalisation": 52388,
      "use gender": 128063,
      "networks method": 77657,
      "singletask learning": 109845,
      "learning stl": 63058,
      "using gender": 129696,
      "space obtained": 111034,
      "speech image": 111692,
      "introduces corpus": 56609,
      "dialogue deep": 30659,
      "suggest types": 115336,
      "leverage sequence": 63623,
      "given difficulty": 48020,
      "answering approach": 6603,
      "emotion expressed": 36103,
      "text difficult": 120875,
      "advances using": 4025,
      "new mechanism": 79004,
      "store relevant": 113382,
      "extract word": 42132,
      "outperforming number": 82812,
      "number competitive": 80855,
      "successfully identify": 115189,
      "22 shared": 775,
      "finally combines": 44153,
      "unique resource": 127193,
      "events peoples": 39592,
      "sentiment dictionaries": 106730,
      "make similar": 66726,
      "learn improve": 62067,
      "linguistic reflexes": 64540,
      "application called": 6836,
      "linguistic descriptions": 64465,
      "current lexical": 25293,
      "resources annotation": 100944,
      "annotation detection": 6298,
      "textbased dialogue": 121424,
      "order maintain": 82364,
      "features order": 43640,
      "increased depth": 53627,
      "new pipeline": 79077,
      "results shows": 102178,
      "capacity learning": 15232,
      "result state": 101407,
      "features natural": 43619,
      "representational space": 99481,
      "different academic": 30984,
      "academic fields": 1857,
      "research showing": 100623,
      "features predict": 43661,
      "predict finegrained": 88887,
      "emotions capture": 36128,
      "capture similarity": 15397,
      "nlp demonstrated": 79601,
      "scarcity annotated": 103802,
      "mutual benefit": 75972,
      "second using": 104465,
      "features auxiliary": 43374,
      "simultaneously predict": 109690,
      "emojis using": 36091,
      "direction future": 31819,
      "complex social": 19878,
      "given growing": 48036,
      "growing assortment": 49160,
      "assortment sentiment": 9641,
      "sentiment measuring": 106758,
      "measuring instruments": 67919,
      "contribute classification": 22748,
      "ability provide": 1564,
      "provide richer": 93912,
      "dictionarybased methods": 30905,
      "briefly examine": 14621,
      "examine 20": 39733,
      "20 methods": 512,
      "generally robust": 46875,
      "complex narratives": 19839,
      "study limits": 114428,
      "social data": 110334,
      "data tell": 26549,
      "mental illness": 68411,
      "conclude review": 20717,
      "separately analyze": 106869,
      "languages 100": 60380,
      "100 years": 172,
      "technologies help": 120024,
      "help incorporate": 49740,
      "daily lives": 25510,
      "behaviors increase": 12675,
      "goal building": 48337,
      "demonstrate need": 28806,
      "multimodal learning": 75438,
      "based sentencelevel": 12018,
      "text temporal": 121364,
      "allowing optimal": 5118,
      "weighted accuracy": 132335,
      "corpus multiparty": 23890,
      "multimodal resources": 75454,
      "textual datasets": 121687,
      "emotion labels": 36107,
      "labels individual": 58611,
      "flow emotions": 44681,
      "tv scripts": 126005,
      "facebook messenger": 42739,
      "affect tweets": 4227,
      "testing sets": 120603,
      "sets tasks": 107717,
      "approaches arabic": 8066,
      "arabic version": 8567,
      "regression task": 97716,
      "similar characteristics": 109077,
      "languages gain": 60604,
      "mechanism inside": 67998,
      "affective computing": 4236,
      "affect human": 4217,
      "support emotional": 115967,
      "complexity ambiguity": 19900,
      "learning demonstrated": 62492,
      "demonstrated paper": 28923,
      "paper specific": 84435,
      "networks regard": 77731,
      "analysis output": 5734,
      "findings considerable": 44290,
      "problem speech": 91246,
      "potential benefit": 88538,
      "benefit combining": 12962,
      "recognition acoustic": 96802,
      "affective state": 4240,
      "state speaker": 112517,
      "resulting low": 101452,
      "features relevant": 43696,
      "architectures utilize": 8858,
      "inferred training": 54263,
      "accuracy f1score": 2162,
      "character wordlevel": 16482,
      "wordlevel recurrent": 133750,
      "intensity detection": 55884,
      "tweet messages": 126014,
      "text lack": 121076,
      "ensemble neural": 37614,
      "processing input": 91684,
      "input character": 55304,
      "ranks place": 95704,
      "tweets english": 126030,
      "classified different": 17509,
      "performs domain": 87007,
      "achieved 1st": 2590,
      "pushing stateoftheart": 94467,
      "deeplearning models": 28459,
      "tweets propose": 126050,
      "bilstm architecture": 14091,
      "mechanism attention": 67949,
      "performance allows": 86135,
      "salient words": 103557,
      "gain insight": 46344,
      "features limited": 43595,
      "2nd subtask": 891,
      "regression achieved": 97690,
      "results subtasks": 102226,
      "task affect": 117855,
      "distantly labeled": 32487,
      "knowledge exploiting": 57920,
      "vector regression": 131339,
      "regression svr": 97715,
      "tasks placed": 119381,
      "features represent": 43700,
      "heterogeneous inputs": 49863,
      "architecture attention": 8619,
      "datasets demonstrated": 27414,
      "demonstrated model": 28922,
      "modeling flow": 72433,
      "power make": 88636,
      "readers makes": 95961,
      "model flow": 71196,
      "book using": 14417,
      "quantify usefulness": 94857,
      "best weighted": 13470,
      "weighted f1score": 132347,
      "success multitask": 115100,
      "data predicting": 26257,
      "data translating": 26582,
      "ii applying": 51696,
      "models subtasks": 74117,
      "simply averaging": 109622,
      "did result": 30922,
      "attentionbased lstm": 10077,
      "propose long": 92754,
      "labeling tweets": 58554,
      "initialize model": 55241,
      "text review": 121268,
      "science psychology": 103981,
      "interaction artificial": 55939,
      "work identifying": 134559,
      "created detect": 24663,
      "detect emotion": 29802,
      "methods insufficient": 69556,
      "essential need": 38562,
      "design architecture": 29516,
      "lead think": 61876,
      "attention linguistic": 9871,
      "pooling techniques": 88049,
      "network sentiment": 77426,
      "classification understanding": 17476,
      "video segments": 131660,
      "multimodal research": 75453,
      "deals various": 27879,
      "2017 present": 636,
      "sequence segments": 107046,
      "generate rich": 47003,
      "leveraging richer": 63701,
      "polarity scores": 87924,
      "scores text": 104215,
      "baselines state": 12469,
      "predict multiple": 88903,
      "multiple emotions": 75548,
      "modeling word": 72587,
      "respectively modeling": 101151,
      "seed word": 104508,
      "historical texts": 50537,
      "language including": 59160,
      "sets seed": 107708,
      "emotional meaning": 36121,
      "lexicon construction": 63877,
      "representation format": 99243,
      "problems come": 91307,
      "results reliable": 102124,
      "reliable human": 98615,
      "results generate": 101809,
      "13 typologically": 290,
      "quality representation": 94772,
      "lexicons past": 63935,
      "semantic polarity": 105180,
      "polarity positive": 87921,
      "developed various": 30319,
      "largescale investigation": 61447,
      "languages evidence": 60561,
      "modalities combined": 70475,
      "multimodal models": 75443,
      "particular build": 85385,
      "learning individual": 62651,
      "diachronic changes": 30487,
      "changes word": 16397,
      "large diachronic": 61078,
      "combines stateoftheart": 18699,
      "sentiment recognition": 106787,
      "partly solve": 85517,
      "annotating new": 6270,
      "dialogues social": 30843,
      "corpus deep": 23746,
      "opensource license": 82021,
      "sentiment intensity": 106748,
      "intensity prediction": 55885,
      "paper multitask": 84050,
      "coarsegrained finegrained": 18056,
      "posts news": 88519,
      "aims leverage": 4547,
      "obtain performance": 81308,
      "improvement 23": 52673,
      "literature mainly": 64759,
      "turn given": 125972,
      "literature research": 64769,
      "reflected language": 97616,
      "bias certain": 13796,
      "improving existing": 53096,
      "sentence levels": 105935,
      "improved representations": 52636,
      "help build": 49700,
      "classification like": 17253,
      "analysis abusive": 5482,
      "corpus texts": 24040,
      "pseudo task": 94185,
      "trained millions": 123199,
      "millions tweets": 70125,
      "supervised labels": 115761,
      "models conducting": 72958,
      "measure reduce": 67818,
      "composed single": 20072,
      "words bidirectional": 133838,
      "network bilstm": 77172,
      "representations context": 99564,
      "vectors dense": 131425,
      "representations label": 99711,
      "official submission": 81604,
      "code replicating": 18146,
      "detection understanding": 30092,
      "psychological complexity": 94211,
      "computational work": 20444,
      "multiple forms": 75568,
      "studied general": 114163,
      "context sentiment": 22263,
      "methods discrete": 69451,
      "discrete classes": 32163,
      "words incorporate": 133995,
      "classification addition": 17104,
      "term weighting": 120247,
      "emotions expressed": 36130,
      "neural transfer": 78716,
      "given tweet": 48166,
      "work experiment": 134503,
      "experiment neural": 40478,
      "embeddings set": 35934,
      "models team": 74168,
      "ranked 3rd": 95637,
      "participants achieving": 85313,
      "task past": 118525,
      "propose shared": 93048,
      "systems infer": 116954,
      "word masked": 133356,
      "tweets collected": 126023,
      "30 teams": 907,
      "submitted results": 114766,
      "results range": 102109,
      "words bigrams": 133839,
      "obtains f1": 81464,
      "subtle textual": 115001,
      "textual clues": 121677,
      "sarcasm classification": 103639,
      "personality recognition": 87130,
      "sentimentspecific word": 106838,
      "performances stateoftheart": 86898,
      "recognition conversations": 96846,
      "database containing": 26640,
      "sentiment labels": 106753,
      "modalities propose": 70481,
      "propose strong": 93087,
      "information emotion": 54521,
      "multimodal speech": 75461,
      "features building": 43386,
      "classifiers paper": 17622,
      "sources predict": 110916,
      "features extensive": 43502,
      "strong data": 113665,
      "data techniques": 26548,
      "annotation formats": 6313,
      "stark contrast": 112393,
      "survey results": 116191,
      "polish portuguese": 87982,
      "100 data": 160,
      "embeddings main": 35798,
      "achieving results": 2973,
      "accordingly key": 2014,
      "humans recognize": 51109,
      "datasets requiring": 27676,
      "including opinion": 53344,
      "debates argumentation": 27894,
      "argumentation mining": 8966,
      "mining understanding": 70266,
      "track individual": 122730,
      "linguistic study": 64560,
      "substantial research": 114876,
      "categories linguistic": 15741,
      "lesser extent": 63397,
      "far investigated": 43086,
      "extensive interdisciplinary": 41939,
      "interdisciplinary literature": 56060,
      "based pointwise": 11932,
      "variants word": 130885,
      "ngram corpus": 79430,
      "conjunction datasets": 21068,
      "amazon reviews": 5271,
      "including strong": 53383,
      "tweets topics": 126061,
      "outperforms general": 82907,
      "dataset significant": 27202,
      "margin perform": 67199,
      "investigating properties": 56856,
      "algorithms achieving": 4825,
      "results realworld": 102115,
      "classification multimodal": 17292,
      "paper predict": 84074,
      "workshop held": 135006,
      "held conjunction": 49684,
      "present multimodal": 89562,
      "features bestperforming": 43380,
      "features achieved": 43351,
      "concordance correlation": 20735,
      "respectively achieved": 101118,
      "processing works": 91860,
      "event context": 39500,
      "reason lack": 96198,
      "releasing dataset": 98500,
      "tasks baseline": 118954,
      "performance demonstrates": 86284,
      "noncontextualized word": 80184,
      "input based": 55301,
      "capability handling": 15182,
      "classification finegrained": 17211,
      "enriches word": 37564,
      "extensions based": 41845,
      "crowdsourced corpus": 25106,
      "annotations tweets": 6475,
      "containing focus": 21790,
      "deliver different": 28618,
      "different emotions": 31125,
      "emotions different": 36129,
      "hard effectively": 49457,
      "captured propose": 15435,
      "wordlevel inputs": 133737,
      "individual features": 53909,
      "models attain": 72780,
      "respectively particularly": 101155,
      "stateoftheart conversational": 112619,
      "video audio": 131654,
      "learning multimodal": 62815,
      "predicting sentiment": 89011,
      "offers improvement": 81587,
      "multigenre corpus": 75110,
      "information trained": 55051,
      "utilize different": 130502,
      "represent features": 99109,
      "global label": 48245,
      "identification aims": 51360,
      "aims identifying": 4539,
      "proposed address": 93169,
      "based manually": 11827,
      "task attempt": 117903,
      "automatically capture": 10730,
      "capture causal": 15271,
      "content text": 21960,
      "information relative": 54915,
      "global labels": 48246,
      "elements text": 35285,
      "unified endtoend": 127106,
      "information incorporated": 54679,
      "aims discovering": 4521,
      "proposed solve": 93554,
      "based rnns": 11999,
      "rnns encode": 102968,
      "ways encode": 132154,
      "encode relative": 36448,
      "information transformer": 55056,
      "transformer capture": 124315,
      "efficient finally": 35075,
      "finally achieve": 44142,
      "pair extraction": 83432,
      "extracting potential": 42228,
      "certain emotions": 15941,
      "years wide": 135311,
      "applications realworld": 7001,
      "propose 2step": 92524,
      "2step approach": 895,
      "extraction multitask": 42398,
      "learning conduct": 62454,
      "approach analysis": 7352,
      "communication channels": 19028,
      "numerous studies": 81029,
      "studies literature": 114249,
      "automatic storytelling": 10675,
      "complex highly": 19816,
      "short stories": 108240,
      "body postures": 14396,
      "learning understanding": 63131,
      "embedding neural": 35462,
      "translation cove": 124721,
      "classification detecting": 17180,
      "hierarchical approach": 49927,
      "emotional state": 36123,
      "featurebased classifiers": 43334,
      "gaussian processes": 46553,
      "experiments hierarchical": 40955,
      "hierarchical architectures": 49929,
      "architectures consistently": 8795,
      "classification textual": 17468,
      "learning classical": 62434,
      "academia industries": 1851,
      "biggest challenge": 14003,
      "question regarding": 95209,
      "history evolution": 50553,
      "resources building": 100955,
      "simple rulebased": 109509,
      "approach notice": 7743,
      "architecture utilize": 8770,
      "utilize available": 130495,
      "consider contextual": 21193,
      "imbalance problem": 51826,
      "problem alleviate": 90926,
      "loss furthermore": 65276,
      "domain adaptability": 33427,
      "adaptability model": 3202,
      "softmax classifier": 110506,
      "ranks 3rd": 95702,
      "wolf et": 132887,
      "approach finetunes": 7574,
      "response language": 101220,
      "prediction dialogue": 89050,
      "detection evaluate": 29943,
      "framework bert": 45440,
      "adapt bert": 3172,
      "utterance present": 130611,
      "distribution based": 32641,
      "dataset confirm": 26816,
      "outperforms multitask": 82925,
      "shows high": 108584,
      "interpretability model": 56225,
      "recent line": 96468,
      "media existing": 68110,
      "researches explore": 100721,
      "authors similar": 10386,
      "gender location": 46604,
      "easy obtain": 34482,
      "connect similar": 21121,
      "results usefulness": 102299,
      "relations method": 98208,
      "using japanese": 129770,
      "japanese data": 57185,
      "improves supervised": 53052,
      "dialogues using": 30846,
      "teams registered": 119731,
      "registered participate": 97685,
      "microf1 score": 70055,
      "analysis hindi": 5641,
      "hindi text": 50475,
      "characters story": 16622,
      "sentence annotated": 105771,
      "native hindi": 76222,
      "challenges annotation": 16126,
      "baseline classifiers": 12202,
      "task presence": 118547,
      "focused modeling": 44866,
      "purely supervised": 94419,
      "corpora task": 23598,
      "investigate generative": 56760,
      "model multiturn": 71549,
      "source transfer": 110858,
      "classifier target": 17585,
      "encoders approach": 36632,
      "intersentential context": 56312,
      "idea perform": 51334,
      "datasets improvement": 27513,
      "better validation": 13766,
      "performances significantly": 86896,
      "help recognize": 49770,
      "topic understanding": 122582,
      "intelligence area": 55826,
      "area unlike": 8886,
      "training larger": 123679,
      "encoder code": 36502,
      "utterance encoder": 130596,
      "masked utterance": 67308,
      "task carried": 117954,
      "boosts model": 14449,
      "users expressed": 129118,
      "majority existing": 66600,
      "used resources": 128731,
      "coverage issues": 24511,
      "did contribute": 30913,
      "resulted better": 101426,
      "recognition attention": 96818,
      "critical build": 24804,
      "build memory": 14785,
      "historical context": 50516,
      "unidirectional gated": 127091,
      "units grus": 127244,
      "ordering information": 82437,
      "particularly propose": 85493,
      "historical utterances": 50540,
      "analysis demonstrating": 5566,
      "recognition vietnamese": 97041,
      "vietnamese social": 131692,
      "task result": 118654,
      "terms polarity": 120360,
      "negative form": 76928,
      "vietnamese lowresource": 131682,
      "performance weighted": 86850,
      "emotional language": 36119,
      "different type": 31522,
      "process address": 91417,
      "develop language": 30208,
      "content dialogue": 21871,
      "models detailed": 73055,
      "metric score": 69899,
      "series analysis": 107269,
      "using emotion": 129623,
      "time span": 122106,
      "present textbased": 89740,
      "biases learned": 13875,
      "parameterfree model": 84731,
      "historical shifts": 50534,
      "sentiment concepts": 106725,
      "apply methodology": 7184,
      "change scale": 16369,
      "detection short": 30052,
      "sentence express": 105865,
      "polarity sentiment": 87927,
      "leverage relatedness": 63621,
      "outperform individual": 82713,
      "extraction learning": 42373,
      "underlying causes": 126676,
      "twostage method": 126161,
      "critical issues": 24815,
      "hinder effectiveness": 50449,
      "effectiveness high": 34895,
      "limit practical": 64157,
      "application method": 6863,
      "learns link": 63218,
      "cause analysis": 15819,
      "research cases": 100439,
      "emotional information": 36118,
      "possible causes": 88388,
      "causes paper": 15838,
      "regarded special": 97654,
      "spanbased extractthenclassify": 111144,
      "target span": 117711,
      "span boundaries": 111118,
      "boundaries corresponding": 14515,
      "classified using": 17514,
      "localized context": 64948,
      "task respectively": 118653,
      "aesthetic emotions": 4208,
      "literature news": 64762,
      "domains focus": 33782,
      "responses consider": 101260,
      "text intended": 121060,
      "evaluate novel": 38878,
      "scale analysis": 103704,
      "emotion categorization": 36097,
      "formulated text": 45288,
      "psychological theories": 94215,
      "approaches need": 8252,
      "need learn": 76830,
      "learn properties": 62120,
      "properties events": 92452,
      "propose make": 92756,
      "classification encoded": 17195,
      "extending previous": 41819,
      "detection previous": 30025,
      "work topic": 134854,
      "concentrated english": 20554,
      "explore crosslingual": 41532,
      "transfer approaches": 124037,
      "data millions": 26128,
      "language studies": 60129,
      "crucial improving": 25144,
      "improving understanding": 53181,
      "factors impact": 42890,
      "thousand poems": 121915,
      "language comes": 58890,
      "information age": 54369,
      "gender differences": 46595,
      "extend current": 41782,
      "encode sentence": 36452,
      "uniform label": 127139,
      "distributions computed": 32755,
      "training introduced": 123667,
      "regularization unsupervised": 97761,
      "augmented sentences": 10337,
      "multimodal language": 75437,
      "analysis human": 5643,
      "modalities including": 70479,
      "facial gestures": 42759,
      "recent multimodal": 96474,
      "adjusts weights": 3882,
      "representations differently": 99597,
      "importance individual": 52060,
      "features weight": 43783,
      "single input": 109742,
      "expressed language": 41716,
      "harmful online": 49505,
      "finegrained typology": 44392,
      "comments labeled": 18816,
      "conduct transfer": 20905,
      "dataset generalizes": 26952,
      "select words": 104719,
      "semantics select": 105459,
      "consists components": 21473,
      "components content": 20021,
      "emotion based": 36096,
      "texts changes": 121471,
      "style interact": 114581,
      "particularly case": 85473,
      "based cognitive": 11581,
      "psycholinguistic norms": 94205,
      "using numeric": 129985,
      "numeric scores": 81005,
      "processing commonly": 91638,
      "used sentence": 128747,
      "model annotations": 70666,
      "captures relationships": 15452,
      "approach predicting": 7806,
      "scores various": 104218,
      "led computer": 63274,
      "overview sentiment": 83373,
      "tasks challenges": 118973,
      "label semantics": 58410,
      "emotional reactions": 36122,
      "typically seen": 126459,
      "attention representing": 9992,
      "representing input": 100051,
      "mechanisms track": 68072,
      "low rank": 65382,
      "multimodal sequences": 75460,
      "individually work": 53945,
      "express emotional": 41699,
      "lowrank factorization": 65485,
      "present transformerbased": 89748,
      "architecture overparameterization": 8720,
      "represent latent": 99116,
      "methods multimodal": 69629,
      "recognition results": 96990,
      "comparably larger": 19196,
      "analysis showed": 5846,
      "examine bias": 39737,
      "rate current": 95790,
      "location information": 64964,
      "personal narratives": 87122,
      "used everyday": 128528,
      "narrative understanding": 76183,
      "provide finegrained": 93828,
      "state explore": 112493,
      "systems increasing": 116952,
      "lives important": 64854,
      "field study": 43985,
      "novel datadriven": 80529,
      "significant room": 108865,
      "likelihood functions": 64120,
      "approach predicts": 7808,
      "intuitive judgment": 56669,
      "methodology builds": 69251,
      "work contextualized": 134437,
      "independent datasets": 53763,
      "perceive process": 85910,
      "core aspect": 23312,
      "labels furthermore": 58606,
      "furthermore examine": 46169,
      "examine role": 39755,
      "predicts speakers": 89223,
      "extracted textual": 42188,
      "task multimodal": 118422,
      "modalities contributes": 70477,
      "better joint": 13613,
      "representation speakers": 99417,
      "recognition despite": 96857,
      "task problems": 118566,
      "issues use": 57081,
      "mapping functions": 67131,
      "makes predictions": 66800,
      "directly adapt": 31858,
      "knowledge including": 58012,
      "express emotions": 41700,
      "knowledge construct": 57839,
      "conducted benchmark": 20912,
      "improved integrating": 52610,
      "integrating pretrained": 55799,
      "vietnamese textual": 131699,
      "preprocessing affects": 89322,
      "method textual": 69189,
      "different preprocessing": 31346,
      "preprocessing techniques": 89337,
      "flat structure": 44638,
      "responses mimic": 101282,
      "varying degree": 131256,
      "positivity negativity": 88357,
      "relevance response": 98516,
      "importance factors": 52058,
      "task unlike": 118822,
      "sentencelevel text": 106178,
      "accessible specifically": 1931,
      "task finally": 118212,
      "finally finetune": 44188,
      "datasets particularly": 27615,
      "build learn": 14781,
      "conversation current": 22944,
      "methods encounter": 69473,
      "shift detection": 108175,
      "related emotion": 97857,
      "learning distinct": 62510,
      "recognition different": 96862,
      "dataset emotional": 26884,
      "stories articles": 113397,
      "national language": 76212,
      "methods manual": 69614,
      "linguistic expertise": 64474,
      "classification standard": 17419,
      "relevant feature": 98549,
      "target aspect": 117584,
      "aspect detection": 9326,
      "analysis previous": 5767,
      "english accomplish": 37063,
      "classification performs": 17333,
      "set high": 107458,
      "study manually": 114434,
      "media comments": 68090,
      "comments posted": 18820,
      "cohens kappa": 18289,
      "accuracy 085": 2062,
      "dataset speech": 27213,
      "speech visual": 111849,
      "applying similar": 7268,
      "networks emotion": 77579,
      "learning active": 62307,
      "dominant method": 33931,
      "implement neural": 51936,
      "architectures address": 8779,
      "vector fed": 131307,
      "fed fully": 43803,
      "arabic speech": 8556,
      "22 absolute": 771,
      "recently semantic": 96759,
      "category prediction": 15790,
      "unclear semantic": 126637,
      "multiple corpora": 75528,
      "classification decision": 17173,
      "literature corpora": 64748,
      "dataset sentiment": 27187,
      "detection introduce": 29975,
      "languages providing": 60826,
      "annotate dataset": 6127,
      "multilabel multiclass": 75155,
      "multiclass dataset": 75055,
      "carefully evaluated": 15524,
      "ethical considerations": 38747,
      "research created": 100454,
      "considerations involved": 21284,
      "resources use": 101061,
      "recognition pretraining": 96969,
      "increasingly studied": 53713,
      "represent data": 99103,
      "response intents": 101219,
      "human social": 50964,
      "generate dialogues": 46927,
      "approach produce": 7820,
      "socially unacceptable": 110463,
      "controllable interpretable": 22827,
      "key element": 57565,
      "dialog intent": 30566,
      "modelling development": 72602,
      "goal produce": 48382,
      "information visualization": 55090,
      "opendomain conversations": 81963,
      "conversation generation": 22951,
      "multisource information": 75784,
      "emotion flow": 36104,
      "predict suitable": 88943,
      "relevant conversation": 98534,
      "current speaker": 25338,
      "generate satisfactory": 47005,
      "linguistic inquiry": 64493,
      "responses inference": 101276,
      "subjective evaluation": 114691,
      "subjective objective": 114698,
      "objective demonstrate": 81068,
      "conversations address": 23046,
      "research nlp": 100570,
      "advances area": 3974,
      "improve interpretability": 52398,
      "method introduce": 68921,
      "types based": 126255,
      "baselines address": 12352,
      "different subtasks": 31462,
      "dataset causal": 26780,
      "contextual pretrained": 22486,
      "used preprocess": 128690,
      "accuracy 65": 2070,
      "analyze quality": 5996,
      "resultant dataset": 101423,
      "task conventional": 118021,
      "current method": 25297,
      "method models": 68970,
      "consider local": 21214,
      "modeling binary": 72384,
      "excellent ability": 39915,
      "transformer blocks": 124314,
      "training analysis": 123357,
      "emotion personality": 36108,
      "personality trait": 87132,
      "especially massive": 38477,
      "massive digital": 67329,
      "researchers demonstrated": 100683,
      "demonstrated strong": 28932,
      "evaluate discuss": 38824,
      "quality learning": 94705,
      "author profiling": 10367,
      "learn associations": 61991,
      "sentence furthermore": 105882,
      "learning meaningful": 62713,
      "unified multimodal": 127121,
      "design label": 29544,
      "multimodal natural": 75446,
      "data reading": 26320,
      "researchers understand": 100714,
      "processing signals": 91794,
      "beneficial machine": 12949,
      "signal beneficial": 108683,
      "eeg features": 34581,
      "embedding types": 35521,
      "outperforms multiple": 82923,
      "detection research": 30035,
      "needed finally": 76874,
      "text annotate": 120644,
      "setting enables": 107744,
      "enables annotators": 36375,
      "measuring interannotator": 67920,
      "variables results": 130858,
      "results knowledge": 101871,
      "leads competitive": 61929,
      "corpora domain": 23464,
      "tool understanding": 122403,
      "introduce number": 56521,
      "movie characters": 74893,
      "understanding stories": 126976,
      "characters use": 16627,
      "negative words": 76960,
      "intensity annotation": 55883,
      "humans judge": 51085,
      "content texts": 21961,
      "texts implicitly": 121526,
      "contemporary american": 21845,
      "neutral sentences": 78757,
      "computational understanding": 20443,
      "macrof1 score": 66365,
      "respectively shared": 101164,
      "task leaderboard": 118344,
      "4th rank": 1066,
      "2nd rank": 890,
      "single modality": 109757,
      "features fixed": 43519,
      "crossmodal attention": 25061,
      "extraction experimental": 42333,
      "results fully": 101802,
      "transformer efficient": 124321,
      "past utterances": 85652,
      "addressing task": 3817,
      "certain time": 15959,
      "transformerbased network": 124429,
      "network task": 77451,
      "improved performances": 52629,
      "quantitative error": 94866,
      "analyses support": 5470,
      "aims change": 4510,
      "2stage training": 893,
      "data note": 26182,
      "perform style": 86080,
      "style linguistic": 114583,
      "learn disentangle": 62026,
      "new interactive": 78960,
      "transformers language": 124451,
      "extraction context": 42289,
      "emerging task": 36074,
      "pipeline identifies": 87541,
      "information limits": 54731,
      "alleviate limitations": 5039,
      "networks contextual": 77548,
      "words positive negative": 134128,
      "level word level": 63522,
      "higher interannotator agreement": 50184,
      "interannotator agreement obtained": 56045,
      "analysis emotion detection": 5593,
      "paper sentiment analysis": 84419,
      "association lexicon crowdsourcing": 9622,
      "leads statistically significant": 61967,
      "datasets compare performance": 27361,
      "current stateoftheart techniques": 25367,
      "test models existing": 120476,
      "quality proposed method": 94766,
      "shortterm memory language": 108322,
      "memory language model": 68320,
      "categories proposed model": 15749,
      "bestworst scaling bws": 13492,
      "building natural language": 14869,
      "human beings recognize": 50768,
      "speech emotion recognition": 111680,
      "prior work proposed": 90762,
      "models feature sets": 73243,
      "work conduct extensive": 134429,
      "learning objective function": 62854,
      "machine learning baselines": 65784,
      "popular research area": 88124,
      "language paper pose": 59791,
      "introduce novel model": 56516,
      "features useful detecting": 43776,
      "present shared task": 89696,
      "particularly useful task": 85503,
      "dataset shared task": 27195,
      "mismatch training test": 70306,
      "tasks deep neural": 119031,
      "based state art": 12069,
      "models using gender": 74303,
      "outperform existing approaches": 82707,
      "document classification present": 32966,
      "classification present new": 17338,
      "network models attention": 77344,
      "given difficulty task": 48021,
      "question answering approach": 95023,
      "extraction aims identify": 42259,
      "using deep memory": 129545,
      "networks propose new": 77711,
      "propose new mechanism": 92861,
      "proposed approach extract": 93196,
      "outperforming number competitive": 82814,
      "number competitive baselines": 80856,
      "representation learning approach": 99282,
      "able successfully identify": 1693,
      "22 shared task": 776,
      "combines lexical syntactic": 18690,
      "recent work models": 96576,
      "present method learn": 89550,
      "achieves better result": 2750,
      "result state art": 101408,
      "features natural language": 43620,
      "natural language representations": 76524,
      "scarcity annotated data": 103803,
      "humanlevel performance using": 51052,
      "direction future research": 31820,
      "using sentiment analysis": 130152,
      "complex social systems": 19879,
      "given growing assortment": 48037,
      "growing assortment sentiment": 49161,
      "assortment sentiment measuring": 9642,
      "sentiment measuring instruments": 106759,
      "dictionarybased methods applied": 30906,
      "briefly examine 20": 14622,
      "examine 20 methods": 39734,
      "publicly available paper": 94315,
      "results demonstrate need": 101696,
      "recognition spoken language": 97012,
      "present novel deep": 89604,
      "dataset result shows": 27165,
      "shows promising performance": 108612,
      "labels individual words": 58612,
      "words sentences documents": 134198,
      "based textual content": 12118,
      "tweets paper describes": 126045,
      "results crosslingual multilingual": 101650,
      "neural network perform": 78371,
      "attention mechanism inside": 9901,
      "neural networks transfer": 78593,
      "annotated data limited": 6163,
      "sentiment analysis result": 106645,
      "learning methods propose": 62741,
      "wordlevel recurrent neural": 133751,
      "performs domain adaptation": 87008,
      "final prediction proposed": 44119,
      "outperform baseline model": 82691,
      "semeval2018 task predicting": 105541,
      "transfer learning paper": 124119,
      "english tweets propose": 37323,
      "self attention mechanism": 104870,
      "attention mechanism attention": 9884,
      "embeddings trained large": 35985,
      "trained large collection": 123177,
      "dataset semeval 2017": 27183,
      "ranked 1st subtask": 95635,
      "semeval2018 task exploiting": 105537,
      "semeval2018 task affect": 105535,
      "task affect tweets": 117856,
      "network models feature": 77346,
      "support vector regression": 116016,
      "vector regression svr": 131340,
      "multiple data sources": 75533,
      "paper model flow": 84047,
      "best weighted f1score": 13471,
      "training data translating": 123557,
      "detection spoken language": 30061,
      "distant supervision automatically": 32463,
      "expanding vocabulary size": 40367,
      "bidirectional lstm model": 13955,
      "lstm model attention": 65640,
      "select informative words": 104703,
      "pay attention linguistic": 85811,
      "including information retrieval": 53303,
      "datasets demonstrates proposed": 27416,
      "al 2017 present": 4621,
      "text present results": 121195,
      "present results model": 89673,
      "overcome limitation present": 83290,
      "small sets seed": 110202,
      "sets seed words": 107709,
      "polarity positive negative": 87922,
      "present largescale investigation": 89533,
      "sentiment score prediction": 106792,
      "multitask learning models": 75851,
      "experiments sentiment analysis": 41126,
      "diachronic text corpora": 30491,
      "research social media": 100627,
      "hierarchical recurrent network": 49994,
      "dialogues social media": 30844,
      "emotion sentiment intensity": 36114,
      "sentiment intensity prediction": 106749,
      "reflected language use": 97617,
      "representations like word2vec": 99740,
      "representations word sentence": 99985,
      "word sentence levels": 133540,
      "text classification like": 120734,
      "learning sentencelevel representations": 63005,
      "memory network bilstm": 68354,
      "finally present experimental": 44217,
      "context sentiment analysis": 22264,
      "models team ranked": 74169,
      "participants achieving f1": 85314,
      "propose shared task": 93049,
      "task systems predict": 118772,
      "teams submitted results": 119735,
      "obtains f1 score": 81465,
      "multitask learning different": 75833,
      "sentimentspecific word embedding": 106839,
      "results tasks using": 102258,
      "recently gained popularity": 96691,
      "emotion recognition using": 36112,
      "recognition challenging task": 96829,
      "proposed model proposed": 93477,
      "learning training data": 63116,
      "opendomain conversation models": 81959,
      "paper new method": 84053,
      "classification model outperforms": 17273,
      "extensive interdisciplinary literature": 41940,
      "based pointwise mutual": 11933,
      "used conjunction datasets": 128448,
      "real world text": 96094,
      "features bestperforming model": 43381,
      "bestperforming model used": 13476,
      "concordance correlation coefficient": 20736,
      "language processing works": 59988,
      "main reason lack": 66456,
      "noncontextualized word embeddings": 80185,
      "word2vec fasttext embeddings": 133673,
      "recognition dialogue systems": 96861,
      "corpus paper present": 23922,
      "features train model": 43763,
      "rule based methods": 103350,
      "proposed address problem": 93170,
      "propose model based": 92785,
      "model based neural": 70737,
      "embedding learning algorithm": 35429,
      "including rulebased methods": 53363,
      "task previous work": 118559,
      "work propose joint": 134724,
      "improve f1 score": 52385,
      "recent years wide": 96642,
      "extraction multitask learning": 42399,
      "multitask learning conduct": 75829,
      "studies mainly use": 114252,
      "paper explore various": 83927,
      "embedding neural machine": 35463,
      "stateoftheart endtoend neural": 112655,
      "approach does rely": 7506,
      "does rely handcrafted": 33385,
      "consider contextual information": 21194,
      "class imbalance problem": 17040,
      "stateoftheart model shows": 112721,
      "classification paper describes": 17321,
      "model known bert": 71398,
      "large corpus model": 61064,
      "indomain data performance": 53964,
      "wolf et al": 132888,
      "learning approach finetunes": 62347,
      "al 2018 dataset": 4632,
      "social media existing": 110377,
      "previous researches explore": 90454,
      "experimental results usefulness": 40716,
      "minimally supervised learning": 70188,
      "using discourse relations": 129583,
      "experiments using japanese": 41194,
      "labeled data improves": 58433,
      "teams registered participate": 119732,
      "hindi text corpus": 50476,
      "provide detailed analysis": 93802,
      "challenging task presence": 16330,
      "available corpora task": 10966,
      "model multiturn conversations": 71550,
      "perform experiments multiple": 85997,
      "artificial intelligence area": 9250,
      "pretraining task named": 90334,
      "boosts model performance": 14450,
      "finetuning pretrained bert": 44493,
      "response problem propose": 101229,
      "adversarial learning framework": 4134,
      "prior work commonly": 90743,
      "unidirectional gated recurrent": 127092,
      "recurrent units grus": 97274,
      "vietnamese social media": 131693,
      "level analysis results": 63419,
      "vietnamese lowresource language": 131683,
      "language generation process": 59093,
      "process address problem": 91418,
      "detailed experiments approach": 29775,
      "time series analysis": 122101,
      "diachronic word embeddings": 30493,
      "paper propose leverage": 84259,
      "related tasks multitask": 97903,
      "processing task aims": 91810,
      "learning model extract": 62760,
      "link prediction task": 64638,
      "attracted attention researchers": 10145,
      "propose spanbased extractthenclassify": 93079,
      "supervision target span": 115922,
      "target span boundaries": 117712,
      "span boundaries corresponding": 111119,
      "large scale analysis": 61242,
      "experiments based bert": 40790,
      "results high quality": 101822,
      "previous work topic": 90560,
      "concentrated english texts": 20555,
      "explore crosslingual transfer": 41533,
      "use parallel data": 128192,
      "uniform label distributions": 127140,
      "analysis human language": 5644,
      "multiple sources information": 75698,
      "multiple downstream tasks": 75547,
      "conduct transfer learning": 20906,
      "transfer learning experiments": 124092,
      "content style interact": 21955,
      "numeric scores used": 81006,
      "language processing commonly": 59852,
      "commonly used sentence": 18972,
      "overview sentiment analysis": 83374,
      "sentiment analysis research": 106644,
      "sentiment analysis potential": 106629,
      "performance compared baselines": 86230,
      "accuracy rate current": 2256,
      "recent work contextualized": 96568,
      "translation propose new": 125151,
      "makes predictions based": 66801,
      "experiments conducted benchmark": 40838,
      "different preprocessing techniques": 31347,
      "accessible specifically propose": 1932,
      "results demonstrate pretraining": 101697,
      "conversations paper address": 23064,
      "short stories articles": 108241,
      "data collection methods": 25743,
      "manual annotation process": 66978,
      "sentiment analysis previous": 106630,
      "analysis previous work": 5768,
      "evaluate different approaches": 38820,
      "english data sets": 37105,
      "sets different domains": 107663,
      "different domains results": 31115,
      "results sequence labeling": 102160,
      "media data natural": 68098,
      "mental health conditions": 68410,
      "social media comments": 110359,
      "shown significant progress": 108524,
      "representation proposed method": 99387,
      "proposed method achieved": 93338,
      "neural networks emotion": 78472,
      "convolutional layers extract": 23160,
      "fed fully connected": 43804,
      "recently semantic role": 96760,
      "structures text answer": 114111,
      "dataset sentiment analysis": 27188,
      "languagespecific bert models": 60990,
      "use lexical resources": 128124,
      "lexical resources use": 63807,
      "train neural models": 122984,
      "hybrid method improves": 51184,
      "learning methods automatically": 62739,
      "experiments automatic human": 40786,
      "achieve consistent improvement": 2452,
      "generation paper introduce": 47528,
      "dialogues propose novel": 30841,
      "multimodal natural language": 75447,
      "language processing signals": 59945,
      "outperforms multiple baselines": 82924,
      "lot attention natural": 65317,
      "leads competitive performance": 61930,
      "paper introduce framework": 83992,
      "corpus contemporary american": 23725,
      "contemporary american english": 21846,
      "feature extraction algorithms": 43276,
      "crossmodal attention mechanism": 25062,
      "extraction experimental results": 42334,
      "introduce novel task": 56519,
      "proposed framework perform": 93297,
      "improvement stateoftheart baselines": 52766,
      "objective subjective evaluation": 81118,
      "sentiment analysis aims": 106565,
      "results method performs": 101924,
      "performs better baselines": 86985,
      "sentiment analysis emotion detection": 106591,
      "deep recurrent neural network": 28405,
      "long shortterm memory language": 65109,
      "shortterm memory language model": 108323,
      "traditional machine learning baselines": 122831,
      "model outperforms stateoftheart approaches": 71669,
      "training development test sets": 123584,
      "mismatch training test data": 70307,
      "convolutional neural networks propose": 23204,
      "neural networks propose new": 78543,
      "outperforming number competitive baselines": 82815,
      "given growing assortment sentiment": 48038,
      "growing assortment sentiment measuring": 49162,
      "assortment sentiment measuring instruments": 9643,
      "briefly examine 20 methods": 14623,
      "neural networks transfer learning": 78594,
      "neural network trained endtoend": 78409,
      "wordlevel recurrent neural network": 133752,
      "ntuaslp semeval2018 task predicting": 80821,
      "word embeddings trained large": 133228,
      "dataset semeval 2017 task": 27184,
      "semeval2018 task affect tweets": 105536,
      "support vector regression svr": 116017,
      "datasets demonstrates proposed model": 27417,
      "et al 2017 present": 38712,
      "small sets seed words": 110203,
      "emotion sentiment intensity prediction": 36115,
      "contextualized word representations paper": 22583,
      "finally present experimental results": 44218,
      "participants achieving f1 score": 85315,
      "speech emotion recognition using": 111681,
      "based pointwise mutual information": 11934,
      "features bestperforming model used": 43382,
      "natural language processing works": 76515,
      "gated recurrent neural network": 46514,
      "model based neural network": 70738,
      "embedding neural machine translation": 35464,
      "does rely handcrafted features": 33386,
      "rely handcrafted features lexicons": 98706,
      "outperforms previous stateoftheart model": 82949,
      "previous stateoftheart model shows": 90477,
      "classification paper describes approach": 17322,
      "wolf et al 2019": 132889,
      "largescale pretrained language model": 61495,
      "et al 2018 dataset": 38721,
      "new pretraining task named": 79083,
      "analysis introduce new dataset": 5663,
      "pretrained bert model achieves": 89992,
      "model achieves best results": 70557,
      "dialogues experimental results proposed": 30833,
      "gated recurrent units grus": 46523,
      "vietnamese social media text": 131694,
      "tasks multitask learning framework": 119319,
      "language processing task aims": 59951,
      "supervision target span boundaries": 115923,
      "target span boundaries corresponding": 117713,
      "tasks sentiment analysis emotion": 119491,
      "conduct transfer learning experiments": 20907,
      "natural language processing commonly": 76408,
      "work propose novel task": 134739,
      "builds recent work contextualized": 14910,
      "machine translation propose new": 66176,
      "extensive experiments conducted benchmark": 41901,
      "dataset verify effectiveness proposed": 27273,
      "experimental results demonstrate pretraining": 40598,
      "sentiment analysis previous work": 106631,
      "data sets different domains": 26423,
      "social media data natural": 110367,
      "media data natural language": 68099,
      "recently semantic role labeling": 96761,
      "popular natural language processing": 88108,
      "machine learning methods automatically": 65800,
      "experiments automatic human evaluation": 40787,
      "received lot attention natural": 96370,
      "lot attention natural language": 65318,
      "corpus contemporary american english": 23726,
      "lead suboptimal performance paper": 61872,
      "long shortterm memory language model": 65110,
      "experiments model outperforms stateoftheart approaches": 41017,
      "convolutional neural networks propose new": 23205,
      "given growing assortment sentiment measuring": 48039,
      "growing assortment sentiment measuring instruments": 49163,
      "classification natural language processing nlp": 17303,
      "does rely handcrafted features lexicons": 33387,
      "model outperforms previous stateoftheart model": 71659,
      "propose new pretraining task named": 92881,
      "natural language processing task aims": 76490,
      "supervision target span boundaries corresponding": 115924,
      "social media data natural language": 110368,
      "media data natural language processing": 68100,
      "received lot attention natural language": 96371,
      "lot attention natural language processing": 65319,
      "headless": 49591,
      "resourcelight": 100902,
      "mimics": 70141,
      "epistemic": 38215,
      "panel": 83691,
      "dagbased": 25504,
      "reviving": 102658,
      "overcomplete": 83309,
      "generalises": 46744,
      "permit": 87047,
      "nonzero": 80318,
      "documentwise": 33321,
      "umbc": 126563,
      "positiondependent": 88306,
      "900": 1371,
      "ganitkevitch": 46432,
      "wordsimilarity": 134336,
      "informationwe": 55126,
      "ppmi": 88693,
      "relearning": 98427,
      "conflates": 21030,
      "commodity": 18852,
      "desiderata": 29510,
      "336": 949,
      "injury": 55271,
      "gp": 48548,
      "humandesigned": 51018,
      "googlenews": 48528,
      "intruder": 56661,
      "redefining": 97306,
      "rightcontext": 102841,
      "sketching": 109968,
      "discriminatory": 32228,
      "spherical": 111928,
      "corpuswide": 24088,
      "clac": 17006,
      "sys": 116656,
      "2way": 896,
      "shannons": 107937,
      "entropies": 38159,
      "heterogeneously": 49868,
      "hubness": 50712,
      "covariance": 24475,
      "pu": 94230,
      "penguin": 85854,
      "microscopic": 70057,
      "mechanics": 67938,
      "distributionbased": 32752,
      "compensated": 19602,
      "crossvalidated": 25086,
      "kitchen": 57706,
      "conjectured": 21064,
      "poincar": 87792,
      "arithmetics": 9011,
      "explorer": 41641,
      "jaccard": 57173,
      "sif": 108664,
      "slovenian": 110099,
      "decays": 27915,
      "preferentially": 89247,
      "scalefree": 103766,
      "intergroup": 56114,
      "earth": 34402,
      "revisits": 102656,
      "maltese": 66888,
      "verbtonoun": 131566,
      "textttword2vec": 121671,
      "representability": 99156,
      "obscurity": 81161,
      "portraying": 88191,
      "measures purpose": 67891,
      "classification similarity": 17406,
      "introduction novel": 56656,
      "presenting literature": 89807,
      "association norms": 9625,
      "implementation model": 51947,
      "approaches measuring": 8232,
      "similarity short": 109312,
      "short context": 108204,
      "context problems": 22230,
      "methods fact": 69505,
      "contexts share": 22427,
      "demonstrated possible": 28924,
      "extract meaning": 42098,
      "correlate word": 24204,
      "word words": 133661,
      "present specific": 89713,
      "studies performed": 114260,
      "traditionally performed": 122890,
      "performed poorly": 86929,
      "work hybrid": 134555,
      "relatedness automatic": 97916,
      "measures use": 67900,
      "merits limitations": 68485,
      "overcome drawbacks": 83284,
      "paper concludes": 83773,
      "comparison distributional": 19538,
      "estimation word": 38671,
      "computing continuous": 20517,
      "representations measured": 99751,
      "measured word": 67842,
      "lower computational": 65425,
      "similarity knowledge": 109247,
      "investigate source": 56814,
      "word counts": 132979,
      "adaptation semantic": 3262,
      "elements various": 35287,
      "elements according": 35277,
      "designing intelligent": 29640,
      "ability compare": 1497,
      "abstract concrete": 1762,
      "proposes comprehensive": 93596,
      "measure comparison": 67785,
      "cognitive sciences": 18280,
      "sciences linguistics": 103988,
      "enable semantic": 36365,
      "discusses existing": 32310,
      "based structure": 12079,
      "similarity approaches": 109198,
      "measures help": 67867,
      "practitioners select": 88746,
      "terms provide": 120366,
      "composition different": 20091,
      "decision semantic": 27943,
      "better average": 13515,
      "tends outperform": 120174,
      "cognitively plausible": 18287,
      "plausible approach": 87680,
      "methods information": 69555,
      "ontology terms": 81850,
      "values model": 130797,
      "model functional": 71210,
      "motivates work": 74872,
      "used alternative": 128385,
      "semantic sequence": 105272,
      "sequence similarity": 107069,
      "attracted great": 10146,
      "preventing researchers": 90370,
      "update equations": 127790,
      "semantic lexicons": 105101,
      "paraphrase database": 84808,
      "method refining": 69098,
      "using relational": 130104,
      "similar vector": 109167,
      "representations makes": 99747,
      "vector training": 131397,
      "relational semantics": 98098,
      "hypothesis word": 51282,
      "improvements cases": 52822,
      "representations aka": 99497,
      "syntactic regularities": 116454,
      "based stochastic": 12075,
      "encoding function": 36695,
      "unsupervised pos": 127694,
      "role unsupervised": 103222,
      "value problem": 130786,
      "multinomial distributions": 75470,
      "gaussian distributions": 46543,
      "distributions word": 32776,
      "downstream pos": 34021,
      "domain known": 33562,
      "representation sequence": 99407,
      "evaluation popular": 39325,
      "methods context": 69402,
      "updating word": 127809,
      "lead improvements": 61858,
      "words domain": 133926,
      "surprisingly results": 116145,
      "difference different": 30942,
      "tasks analogy": 118927,
      "representations sensitive": 99875,
      "sensitive semantic": 105745,
      "exist words": 40032,
      "extract lexical": 42097,
      "patterns cooccurrence": 85735,
      "pairs use": 83658,
      "train data": 122917,
      "sampled corpus": 103571,
      "corpus requiring": 23971,
      "manual intervention": 67005,
      "theories lexical": 121812,
      "synonymy hypernymy": 116355,
      "methods transform": 69813,
      "scale natural": 103740,
      "systems commonly": 116793,
      "ignore morphological": 51669,
      "exceeding best": 39905,
      "error margin": 38307,
      "margin furthermore": 67194,
      "parameters trained": 84790,
      "sum parts": 115425,
      "phrases different": 87417,
      "propose average": 92570,
      "representations classical": 99544,
      "representations distributed": 99602,
      "vectors relatively": 131469,
      "lowdimensional space": 65414,
      "shown encode": 108468,
      "evidence using": 39676,
      "words mapped": 134047,
      "automatically ranked": 10820,
      "cca method": 15857,
      "dimension data": 31744,
      "derive word": 29341,
      "context way": 22323,
      "evaluating utility": 39109,
      "differences lexical": 30963,
      "relation learning": 98049,
      "capturing different": 15465,
      "relations despite": 98137,
      "supervision prior": 115907,
      "range lexical": 95576,
      "paper carry": 83757,
      "evaluation learning": 39254,
      "range relations": 95597,
      "experimental approach": 40525,
      "direct relations": 31799,
      "vary word": 131251,
      "artificial word": 9269,
      "corpora learn": 23517,
      "learn similar": 62149,
      "present similar": 89698,
      "occurrence statistics": 81512,
      "paper incorporate": 83981,
      "shows superiority": 108638,
      "embeddings reflect": 35903,
      "developed consisting": 30261,
      "wordsentence pairs": 134332,
      "faster alternative": 43161,
      "available semantic": 11100,
      "good capturing": 48468,
      "regularities language": 97738,
      "language allows": 58825,
      "natural ask": 76240,
      "question design": 95146,
      "reveal neural": 102501,
      "provide preliminary": 93897,
      "proposes model": 93604,
      "contexts based": 22380,
      "based partofspeech": 11924,
      "element natural": 35269,
      "utilize pos": 130522,
      "paper paper": 84065,
      "matrices experiments": 67469,
      "identifying word": 51634,
      "contexts target": 22430,
      "words weighting": 134313,
      "according rank": 1996,
      "contexts dependency": 22386,
      "occur similar": 81501,
      "contexts compared": 22382,
      "best setting": 13451,
      "mutually dependent": 75989,
      "depending number": 29273,
      "approach computational": 7442,
      "present ensemble": 89472,
      "ppdb ganitkevitch": 88685,
      "ganitkevitch et": 46433,
      "information common": 54424,
      "embeddings produces": 35885,
      "problems evaluation": 91318,
      "tasks lacking": 119238,
      "words nlp": 134080,
      "relied heavily": 98639,
      "heavily word": 49663,
      "problems associated": 91302,
      "summarize existing": 115604,
      "embeddings hierarchical": 35738,
      "lack structured": 58747,
      "informationwe propose": 55127,
      "relatedness entities": 97919,
      "framework handle": 45558,
      "quantify similarity": 94856,
      "random initializations": 95500,
      "intrinsic property": 56368,
      "tasks believe": 118957,
      "robustness important": 103101,
      "consider developing": 21197,
      "certain topic": 15960,
      "various categories": 131055,
      "work discuss": 134480,
      "provide list": 93868,
      "commonalities differences": 18952,
      "believe proposed": 12702,
      "respect human": 101081,
      "sampling negative": 103604,
      "syntactic performance": 116448,
      "need store": 76859,
      "scalable word": 103699,
      "embeddings mapping": 35804,
      "words natural": 134073,
      "based weighted": 12159,
      "scale learning": 103735,
      "embeddings incrementally": 35759,
      "strategy greatly": 113520,
      "greatly reduces": 49063,
      "vocabulary learn": 131881,
      "vocabulary word": 131919,
      "produces embeddings": 91993,
      "embedding tools": 35518,
      "vectors features": 131437,
      "correlations downstream": 24254,
      "method word": 69227,
      "remarkably effective": 98850,
      "dimensions word": 31775,
      "structures natural": 114089,
      "introduce information": 56437,
      "information entropy": 54540,
      "solving word": 110654,
      "space represent": 111049,
      "utility method": 130483,
      "identify fundamental": 51504,
      "difficulties nlp": 31686,
      "resulting vectors": 101478,
      "operations vectors": 82076,
      "simultaneously trained": 109697,
      "require vectors": 100207,
      "stored memory": 113390,
      "transfer paper": 124171,
      "training vectors": 123947,
      "solutions evaluate": 110575,
      "proposed benchmark": 93236,
      "vectorspace representations": 131500,
      "fixedlength vectors": 44618,
      "vectors containing": 131419,
      "word result": 133484,
      "mining techniques": 70264,
      "words automatic": 133830,
      "selection context": 104775,
      "concerned identifying": 20687,
      "word class": 132950,
      "selected context": 104723,
      "paper intend": 83986,
      "context relevance": 22246,
      "popular test": 88136,
      "uncertainty word": 126627,
      "spaces paper": 111100,
      "techniques successfully": 119991,
      "introduce generic": 56431,
      "work information": 134569,
      "capture meaningful": 15344,
      "concepts evaluate": 20616,
      "embeddings competitive": 35605,
      "news vectors": 79395,
      "vectors cases": 131412,
      "building work": 14902,
      "performance measure": 86524,
      "reliability evaluation": 98607,
      "account learning": 2029,
      "performance applications": 86142,
      "applications learning": 6955,
      "far aware": 43071,
      "paragraph embedding": 84564,
      "considering words": 21335,
      "motivated observations": 74858,
      "major contributions": 66551,
      "information produce": 54874,
      "lowdimensional vector": 65416,
      "increasing importance": 53673,
      "importance spoken": 52080,
      "model inherits": 71344,
      "simple supervised": 109525,
      "single important": 109741,
      "important goal": 52168,
      "faster surprisingly": 43191,
      "way embeddings": 132072,
      "addition recent": 3470,
      "importance learning": 52067,
      "argue focus": 8921,
      "representation evaluation": 99229,
      "evaluation selected": 39381,
      "selected word": 104740,
      "brings new": 14649,
      "new insight": 78957,
      "performance characteristics": 86201,
      "analysis scripts": 5832,
      "using genetic": 129706,
      "genetic programming": 47821,
      "corpus recently": 23962,
      "embedding able": 35369,
      "operations defined": 82069,
      "vectors general": 131438,
      "composition semantic": 20103,
      "spaces instead": 111096,
      "able reproduce": 1684,
      "reproduce behavior": 100080,
      "robustness approach": 103090,
      "embedding technique": 35512,
      "explainable word": 41269,
      "link structure": 64639,
      "structure test": 113970,
      "items evaluate": 57109,
      "relevant query": 98572,
      "query order": 94973,
      "report improvements": 99006,
      "empirical theoretical": 36212,
      "learn vector": 62185,
      "choice context": 16887,
      "context takes": 22290,
      "model component": 70869,
      "fully understood": 45976,
      "set distinct": 107419,
      "hashing propose": 49523,
      "using hash": 129727,
      "pairwise distances": 83673,
      "weighting methods": 132364,
      "discriminatory power": 32229,
      "tools reasoning": 122469,
      "set objects": 107514,
      "relationships recent": 98330,
      "spaces shown": 111105,
      "cognitive scientists": 18281,
      "modern word": 74427,
      "providing detailed": 94111,
      "approach captures": 7411,
      "similarity judgments": 109246,
      "recent benchmark": 96433,
      "domains analysis": 33729,
      "identify assumptions": 51478,
      "allowing multiple": 5117,
      "multiple example": 75561,
      "significant semantic": 108866,
      "semantic challenges": 105001,
      "ontology classes": 81837,
      "selecting representative": 104756,
      "converted vector": 23101,
      "proposes methodology": 93603,
      "needs major": 76896,
      "goal automated": 48331,
      "considers problem": 21340,
      "posts web": 88527,
      "discusses approach": 32304,
      "discussion thread": 32326,
      "learning supervised": 63074,
      "composed word": 20074,
      "propose comparison": 92587,
      "posts popular": 88521,
      "model representing": 71909,
      "tasks traditional": 119566,
      "representation existing": 99231,
      "help representation": 49775,
      "texts experimental": 121514,
      "results topical": 102275,
      "deals using": 27878,
      "relations pairs": 98229,
      "incremental updating": 53741,
      "updating models": 127808,
      "training texts": 123920,
      "evaluated task": 39011,
      "important work": 52295,
      "work remains": 134775,
      "applications constructing": 6907,
      "effectively combines": 34801,
      "construct word": 21651,
      "12 standard": 257,
      "achieving similar": 2977,
      "metric natural": 69891,
      "achieved fmeasure": 2626,
      "compositional phrases": 20124,
      "provide meaningful": 93871,
      "common natural": 18896,
      "portuguese corpus": 88196,
      "tagging sentence": 117445,
      "entities important": 37794,
      "relations pair": 98228,
      "compositional approach": 20106,
      "using attributes": 129359,
      "word indirectly": 133319,
      "common relations": 18914,
      "relations hold": 98182,
      "insights semantics": 55545,
      "words consequently": 133875,
      "signal learning": 108687,
      "texts common": 121476,
      "theoretical relationship": 121800,
      "objective motivated": 81096,
      "relationship does": 98291,
      "despite data": 29683,
      "embeddings range": 35896,
      "concept hierarchies": 20574,
      "popular topic": 88139,
      "embedding seen": 35485,
      "hash function": 49521,
      "dimensional representation": 31750,
      "representation token": 99441,
      "shared pool": 108003,
      "easily deal": 34445,
      "tokens using": 122351,
      "dictionary training": 30896,
      "training perform": 123777,
      "embeddings exhibit": 35704,
      "embeddings wide": 36023,
      "parameters needed": 84772,
      "embedding standard": 35507,
      "standard embeddings": 112233,
      "distributional inclusion": 32702,
      "inclusion vector": 53409,
      "detection modeling": 29993,
      "limits coverage": 64316,
      "coverage models": 24517,
      "models addressed": 72723,
      "nonnegative vector": 80251,
      "word contexts": 132969,
      "space experimental": 111001,
      "precision previous": 88796,
      "highest average": 50223,
      "tfidf sentence": 121762,
      "fitted using": 44589,
      "word entropies": 133265,
      "length textual": 63379,
      "textual gender": 121709,
      "hyperparameter selection": 51229,
      "inference unseen": 54245,
      "model reached": 71852,
      "reached stateoftheart": 95910,
      "cases semantic": 15666,
      "embedding dimensions": 35395,
      "makes interpretation": 66789,
      "method quantify": 69087,
      "word intrusion": 133326,
      "relation vectors": 98085,
      "rely cooccurrence": 98681,
      "corpus learn": 23859,
      "modeling relationships": 72530,
      "method directly": 68778,
      "learns relation": 63229,
      "vectors relation": 131468,
      "space novel": 111033,
      "word space": 133579,
      "linguistic items": 64498,
      "model applying": 70673,
      "method implementing": 68882,
      "similarity definition": 109220,
      "feature ranking": 43300,
      "current literature": 25295,
      "task expect": 118169,
      "continuous distributed": 22616,
      "collections wikipedia": 18507,
      "used main": 128621,
      "realvalued word": 96146,
      "representations able": 99483,
      "models proposing": 73829,
      "representations gained": 99665,
      "presents extensive": 89850,
      "field word": 43988,
      "vectors better": 131410,
      "learned text": 62264,
      "text proved": 121221,
      "proved successful": 93715,
      "attempt represent": 9752,
      "structure neighborhood": 113914,
      "adequate context": 3831,
      "surge research": 116086,
      "embedding improving": 35415,
      "representation attempt": 99174,
      "obtained distributional": 81363,
      "representations helps": 99674,
      "handcrafted lexical": 49349,
      "resource used": 100886,
      "issues resource": 57077,
      "devices like": 30457,
      "furthermore training": 46218,
      "space precision": 111044,
      "unsupervised vector": 127746,
      "block language": 14360,
      "classification current": 17168,
      "implicitly assume": 52023,
      "regardless context": 97664,
      "word importance": 133313,
      "distribution context": 32645,
      "accurate unsupervised": 2371,
      "classification requiring": 17379,
      "requiring little": 100347,
      "words outperforms": 134097,
      "unsupervised stateoftheart": 127723,
      "category word": 15796,
      "exhibits significant": 40015,
      "improvements 10": 52786,
      "hyperparameters training": 51237,
      "vector dimensions": 131303,
      "datasets conduct": 27371,
      "empirically measure": 36236,
      "hyperparameters english": 51235,
      "attempt focus": 9743,
      "focus outofvocabulary": 44801,
      "provide solutions": 93923,
      "based related": 11989,
      "words attribute": 133829,
      "attribute prediction": 10182,
      "showing results": 108429,
      "deeper look": 28447,
      "compared basic": 19334,
      "stanford dependency": 112367,
      "work exploring": 134523,
      "frequency words": 45851,
      "meaning nlp": 67653,
      "nlp semantic": 79686,
      "arbitrary distributional": 8579,
      "update vectors": 127799,
      "occurring external": 81516,
      "external lexicons": 42029,
      "words intact": 134011,
      "space specialisation": 111064,
      "knowledge seen": 58163,
      "seen words": 104556,
      "yields considerable": 135407,
      "considerable gains": 21248,
      "intrinsic word": 56371,
      "tracking lexical": 122758,
      "lexical text": 63832,
      "demonstrating importance": 28975,
      "pu learning": 94231,
      "component downstream": 19975,
      "embeddings positive": 35867,
      "results confirming": 101623,
      "confirming importance": 21026,
      "rare outofvocabulary": 95741,
      "meaning polarity": 67659,
      "combination individual": 18563,
      "overall polarity": 83249,
      "combination words": 18598,
      "words depending": 133906,
      "new states": 79182,
      "word complex": 132960,
      "inspired models": 55572,
      "derive meaning": 29339,
      "embeddings information": 35762,
      "key point": 57591,
      "allows words": 5199,
      "concatenating word": 20539,
      "representation quality": 99389,
      "new weighting": 79248,
      "modified versions": 74451,
      "experimentally evaluated": 40738,
      "algorithm encourages": 4725,
      "original semantic": 82541,
      "semantic learning": 105095,
      "align words": 4907,
      "verify proposed": 131591,
      "benchmark tests": 12867,
      "unsupervised relation": 127700,
      "words form": 133963,
      "combination ad": 18546,
      "explicitly encoding": 41366,
      "encoding relational": 36720,
      "aspects word": 9418,
      "finally test": 44238,
      "loss compared": 65257,
      "embedding shows": 35490,
      "richer semantic": 102806,
      "predict set": 88937,
      "modeling tools": 72568,
      "surface semantic": 116078,
      "evaluated results": 39003,
      "psychological process": 94213,
      "unable model": 126570,
      "words semantics": 134191,
      "words correspond": 133891,
      "potentially different": 88608,
      "techniques sense": 119981,
      "popularity research": 88154,
      "topic evaluation": 122514,
      "exist specifically": 40030,
      "dynamic semantics": 34325,
      "models surpassed": 74137,
      "performance ceiling": 86197,
      "stanford contextual": 112360,
      "highlight shortcomings": 50273,
      "context dataset": 22048,
      "generic evaluation": 47798,
      "propose explicit": 92666,
      "allowing direct": 5110,
      "embeddings tell": 35975,
      "judgments experimental": 57445,
      "approach tests": 7970,
      "supervised classifiers": 115742,
      "based vector": 12153,
      "properties relevant": 92477,
      "perceptual information": 85934,
      "properties captured": 92437,
      "embeddings understanding": 35997,
      "arithmetic operators": 9007,
      "provide formal": 93833,
      "word weighting": 133657,
      "information theoretic": 55041,
      "created equal": 24667,
      "latent hierarchical": 61589,
      "structure generation": 113869,
      "unsupervised learned": 127651,
      "embeddings reveal": 35919,
      "propose embed": 92645,
      "hyperbolic spaces": 51209,
      "learn unsupervised": 62180,
      "embeddings type": 35994,
      "experiments prove": 41087,
      "trained largest": 123187,
      "processing related": 91784,
      "performances achieved": 86870,
      "achieved neural": 2656,
      "usually costly": 130414,
      "short paper": 108227,
      "address unsupervised": 3774,
      "including edit": 53289,
      "tfidf based": 121759,
      "word models": 133364,
      "representation independent": 99260,
      "measure changes": 67783,
      "shown different": 108460,
      "performance consistency": 86254,
      "extrinsic intrinsic": 42619,
      "improve simple": 52540,
      "weighted cosine": 132342,
      "use common": 127943,
      "compare embeddings": 19244,
      "domain word": 33696,
      "embeddings studied": 35955,
      "domain usually": 33693,
      "nlp processes": 79672,
      "processes contrast": 91604,
      "end manually": 36816,
      "evaluation various": 39439,
      "neighbors word": 76999,
      "extent word": 41988,
      "models preserve": 73777,
      "reflected distances": 97615,
      "proxy syntactic": 94156,
      "syntactically valid": 116519,
      "contexts investigate": 22405,
      "benchmarks correlate": 12893,
      "inform future": 54334,
      "selection word": 104847,
      "model suggesting": 72111,
      "appropriate selection": 8435,
      "size parameter": 109938,
      "tasks evaluating": 119105,
      "efforts focusing": 35198,
      "problem constructing": 90977,
      "representations discovered": 99601,
      "analysis possible": 5757,
      "successfully predict": 115192,
      "embeddings high": 35739,
      "important machine": 52182,
      "direct way": 31810,
      "correlation principal": 24242,
      "structure representation": 113943,
      "singular value": 109861,
      "value decomposition": 130777,
      "provided additional": 93958,
      "secondorder information": 104480,
      "success variety": 115140,
      "nodes graphs": 80037,
      "large graphs": 61101,
      "graphs explore": 48952,
      "learns dense": 63202,
      "embeddings opposed": 35841,
      "directly computing": 31868,
      "trained indonesian": 123161,
      "indonesian online": 53995,
      "yielding significant": 135384,
      "semantics distributional": 105410,
      "far better": 43073,
      "importantly study": 52301,
      "study role": 114508,
      "tasks probes": 119401,
      "use negative": 128166,
      "corpora existing": 23480,
      "analogy datasets": 5419,
      "predefined relations": 88833,
      "morphological relations": 74726,
      "relations named": 98215,
      "knowledge wordlevel": 58235,
      "analogy dataset": 5418,
      "representations embed": 99618,
      "achieve global": 2462,
      "hypotheses apply": 51249,
      "apply statistical": 7220,
      "methods strong": 69771,
      "performance extrinsic": 86373,
      "pmibased word": 87774,
      "extracted english": 42152,
      "design used": 29572,
      "approximation methods": 8470,
      "semantics developed": 105407,
      "words come": 133866,
      "linguistic probes": 64526,
      "probes propose": 90883,
      "comparative approach": 19204,
      "similar set": 109145,
      "compare quality": 19286,
      "crosslingual analogy": 24926,
      "embeddings ensemble": 35695,
      "producing word": 92023,
      "corpus modern": 23887,
      "tools built": 122443,
      "direct method": 31792,
      "representations currently": 99582,
      "currently exists": 25405,
      "disambiguation account": 31952,
      "english number": 37227,
      "language exhibits": 59022,
      "algorithm paper": 4779,
      "generated systematically": 47134,
      "scale used": 103761,
      "scales different": 103768,
      "significantly depending": 108904,
      "identify subset": 51561,
      "importance developing": 52053,
      "embeddings skipgram": 35942,
      "roles word": 103239,
      "analysis called": 5521,
      "called semantic": 15062,
      "fact word": 42839,
      "semantic groups": 105068,
      "semantic subspaces": 105312,
      "far known": 43089,
      "extent commonly": 41973,
      "relationships sentences": 98332,
      "stateoftheart solution": 112958,
      "processing learn": 91693,
      "exhibit interesting": 40001,
      "interesting semantic": 56092,
      "hypothesis states": 51277,
      "given contexts": 48005,
      "work starting": 134815,
      "semantic overlap": 105131,
      "earth movers": 34403,
      "baselines source": 12466,
      "different algorithms": 30986,
      "training processes": 123796,
      "known relation": 58320,
      "quantify distance": 94846,
      "distance different": 32412,
      "embeddings metric": 35809,
      "usually applied": 130407,
      "applied using": 7138,
      "cases text": 15669,
      "paper revisits": 84415,
      "correct bias": 24094,
      "unlabeled pu": 127404,
      "generalize concept": 46805,
      "key elements": 57566,
      "annotation set": 6374,
      "set context": 107404,
      "popular methods": 88101,
      "technique data": 119780,
      "average multiple": 11202,
      "usage words": 127873,
      "sentence despite": 105820,
      "way work": 132145,
      "introduces explores": 56610,
      "explores universal": 41653,
      "unit uniform": 127218,
      "uniform vector": 127142,
      "space taskindependent": 111068,
      "taskindependent evaluation": 118879,
      "constructing analogy": 21671,
      "datasets terms": 27750,
      "experiment multiple": 40474,
      "multiple representation": 75666,
      "examine geometric": 39746,
      "geometric properties": 47862,
      "models incorporated": 73386,
      "class flexibility": 17037,
      "grammatical categories": 48689,
      "37 languages": 972,
      "semantic variation": 105350,
      "contextualized models": 22558,
      "variability word": 130835,
      "dynamic word": 34334,
      "words function": 133966,
      "involving semantic": 56919,
      "semantic variability": 105349,
      "quantitative analyses": 94861,
      "semantic aspect": 104993,
      "real complex": 96058,
      "work investigating": 134597,
      "shown word": 108539,
      "component vectors": 20013,
      "postprocessing step": 88509,
      "contradicts prior": 22674,
      "embeddings requiring": 35914,
      "requiring access": 100334,
      "nlp understanding": 79789,
      "given rich": 48106,
      "widely various": 132599,
      "volumes text": 131944,
      "representations utilized": 99968,
      "algorithms variety": 4890,
      "nlp related": 79679,
      "based classifiers": 11575,
      "embeddings techniques": 35974,
      "space moving": 111029,
      "technique create": 119777,
      "different methodologies": 31255,
      "nature word": 76673,
      "associations different": 9632,
      "semisupervised clustering": 105596,
      "method cluster": 68695,
      "annotated proper": 6221,
      "bias results": 13842,
      "report propose": 99031,
      "coverage embedding": 24503,
      "algorithms work": 4892,
      "distance word": 32436,
      "precise understanding": 88783,
      "possible combinations": 88390,
      "explicit linguistic": 41330,
      "generate embeddings": 46937,
      "corpus quantify": 23955,
      "similarity measures purpose": 109263,
      "language processing article": 59845,
      "compared previously best": 19423,
      "previously best performing": 90591,
      "different types neural": 31534,
      "types neural networks": 126328,
      "accuracy lower computational": 2202,
      "lower computational cost": 65426,
      "stateoftheart performance test": 112851,
      "syntactic semantic word": 116479,
      "extensive experiments large": 41912,
      "experiments large number": 40978,
      "knowledge based word": 57797,
      "mimic human ability": 70133,
      "paper proposes comprehensive": 84352,
      "knowledge sources order": 58181,
      "paper discusses existing": 83878,
      "computational measures semantic": 20402,
      "geographic information retrieval": 47852,
      "challenge address issue": 15990,
      "attracted great attention": 10147,
      "useful various nlp": 128946,
      "neural networks understanding": 78598,
      "optimization techniques including": 82215,
      "using relational information": 130105,
      "similar vector representations": 109168,
      "semantic evaluation tasks": 105046,
      "word representations trained": 133475,
      "semantic syntactic regularities": 105319,
      "semantic representations word": 105239,
      "features supervised learning": 43748,
      "supervised learning problems": 115775,
      "gaussian distributions word": 46544,
      "embeddings distributed word": 35663,
      "updating word representations": 127810,
      "capture semantic relations": 15381,
      "unsupervised method learning": 127667,
      "represent semantic relations": 99136,
      "pairs proposed method": 83615,
      "pairs use train": 83659,
      "statistically significantly outperforms": 113190,
      "current stateoftheart word": 25369,
      "scale natural language": 103741,
      "tasks propose new": 119409,
      "embedding model produces": 35450,
      "margin furthermore model": 67195,
      "continuous vector spaces": 22645,
      "shown capture semantic": 108449,
      "novel model jointly": 80646,
      "jointly learns word": 57362,
      "words phrases different": 134117,
      "common semantic space": 18920,
      "quality word representations": 94830,
      "representations distributed representations": 99603,
      "encode semantic information": 36451,
      "derive word embeddings": 29342,
      "relation learning recent": 98050,
      "types word embeddings": 126387,
      "word embeddings proposed": 133192,
      "similarity words phrases": 109342,
      "data set evaluating": 26406,
      "representation models results": 99342,
      "embeddings paper proposes": 35853,
      "paper proposes model": 84358,
      "element natural language": 35270,
      "natural language stateoftheart": 76537,
      "demonstrated effectiveness proposed": 28914,
      "similarity vector space": 109330,
      "target words weighting": 117755,
      "test set improvement": 120506,
      "vector space present": 131379,
      "present ensemble method": 89473,
      "knowledge semantic networks": 58168,
      "ppdb ganitkevitch et": 88686,
      "ganitkevitch et al": 46434,
      "using word similarity": 130384,
      "word similarity evaluation": 133561,
      "lack structured knowledge": 58748,
      "embedding methods word": 35444,
      "methods word embeddings": 69852,
      "different random initializations": 31370,
      "matches outperforms stateoftheart": 67385,
      "word embeddings mapping": 133160,
      "words natural language": 134074,
      "embedding learning method": 35430,
      "learning method based": 62719,
      "online learning algorithm": 81780,
      "word embeddings large": 133143,
      "word embedding tools": 133038,
      "semantic syntactic tasks": 105322,
      "processing tasks existing": 91813,
      "tasks existing models": 119115,
      "structures natural language": 114090,
      "quality word embedding": 94829,
      "word representations specifically": 133473,
      "capture semantic relationships": 15383,
      "proposed benchmark dataset": 93237,
      "data mining techniques": 26132,
      "space representations words": 111052,
      "best previously proposed": 13417,
      "test sets results": 120531,
      "work information extraction": 134570,
      "datasets word similarity": 27801,
      "major contributions paper": 66552,
      "paragraph embedding method": 84565,
      "model inherits advantages": 71345,
      "specifically propose evaluation": 111580,
      "selected word embeddings": 104741,
      "based methods proposed": 11840,
      "address different types": 3675,
      "different types language": 31532,
      "word embedding technique": 133036,
      "embedding technique called": 35513,
      "vector representing word": 131364,
      "proposed word embedding": 93591,
      "existing methods neural": 40191,
      "problem present simple": 91168,
      "learn vector representations": 62186,
      "fully understood paper": 45977,
      "embedding learning techniques": 35432,
      "large benchmark dataset": 61040,
      "word embeddings standard": 133211,
      "recent benchmark datasets": 96434,
      "current word embedding": 25390,
      "used train machine": 128820,
      "word representations study": 133474,
      "performance different word": 86303,
      "models train models": 74191,
      "lexical grammatical features": 63768,
      "probabilistic generative model": 90807,
      "help representation learning": 49776,
      "texts experimental results": 121515,
      "relations word embeddings": 98283,
      "paper deals using": 83795,
      "relations pairs words": 98230,
      "proposed approach evaluated": 93194,
      "used downstream applications": 128502,
      "embeddings dimensionality reduction": 35656,
      "applications paper present": 6986,
      "evaluating semantic similarity": 39096,
      "common natural language": 18897,
      "suffers data sparseness": 115252,
      "benchmark datasets word": 12809,
      "different relation representations": 31378,
      "propose learning objective": 92747,
      "learning objective motivated": 62855,
      "objective motivated theoretical": 81097,
      "relationship does hold": 98292,
      "word embeddings finally": 133116,
      "demonstrate ensemble model": 28738,
      "number parameters needed": 80935,
      "embeddings constructed using": 35611,
      "distributional inclusion vector": 32703,
      "supervised learning labeled": 115770,
      "unsupervised sentence representations": 127719,
      "sentence representations word": 106047,
      "representations word information": 99983,
      "advances word embeddings": 4027,
      "representations unlabeled text": 99953,
      "reached stateoftheart performance": 95911,
      "systems semantic structure": 117132,
      "embeddings proposed method": 35890,
      "cooccurrence statistics large": 23234,
      "vector representations word": 131361,
      "pretrained word representations": 90226,
      "available pretrained models": 11079,
      "outperform current state": 82703,
      "word representations able": 133442,
      "representations able capture": 99484,
      "popularity recent years": 88153,
      "paper presents extensive": 84172,
      "word vectors better": 133634,
      "word set words": 133549,
      "dense word vectors": 29023,
      "helps improving performance": 49820,
      "word similarity relatedness": 133565,
      "handcrafted lexical resources": 49350,
      "word vectors using": 133650,
      "train word vectors": 123052,
      "building block language": 14825,
      "classification requiring little": 17380,
      "modeling word embeddings": 72588,
      "classification propose novel": 17356,
      "joint model word": 57293,
      "model word embeddings": 72320,
      "downstream applications using": 33993,
      "words words appear": 134322,
      "high frequency words": 50075,
      "word vector spaces": 133631,
      "vectors words occurring": 131494,
      "words occurring external": 134092,
      "occurring external lexicons": 81517,
      "vector space specialisation": 131384,
      "unseen words propose": 127557,
      "words propose novel": 134146,
      "model yields considerable": 72339,
      "intrinsic word similarity": 56372,
      "tasks downstream tasks": 119077,
      "state tracking lexical": 112529,
      "lexical text simplification": 63833,
      "processing natural languages": 91722,
      "languages paper study": 60786,
      "word embeddings positive": 133184,
      "information word embeddings": 55097,
      "subword information learning": 115021,
      "rare outofvocabulary words": 95742,
      "achieve better performances": 2432,
      "better performances stateoftheart": 13667,
      "approaches based word": 8083,
      "word embeddings preserving": 133186,
      "semantic properties words": 105187,
      "syntactic relations words": 116456,
      "words semantically related": 134189,
      "experimental results manual": 40636,
      "based word vector": 12168,
      "combination ad hoc": 18547,
      "richer semantic information": 102807,
      "model loss function": 71478,
      "methods word similarity": 69853,
      "computational models word": 20414,
      "models using different": 74301,
      "using different combinations": 129569,
      "contextualized embeddings proposed": 22548,
      "stanford contextual word": 112361,
      "paper propose explicit": 84235,
      "semantic information captured": 105073,
      "semantic features word": 105054,
      "features word embedding": 43786,
      "results provide initial": 102088,
      "embedding models skipgram": 35456,
      "latent hierarchical structure": 61590,
      "word embeddings reveal": 133202,
      "propose embed words": 92646,
      "learn unsupervised word": 62181,
      "word embeddings largescale": 133144,
      "language processing related": 59940,
      "trained labeled data": 123167,
      "labeled data supervised": 58445,
      "including edit distance": 53290,
      "correlation analysis study": 24224,
      "set word vectors": 107638,
      "weighted cosine similarity": 132343,
      "domain word embeddings": 33697,
      "word embeddings studied": 133214,
      "trained large text": 123182,
      "nearest neighbors word": 76725,
      "extent word embedding": 41989,
      "use speech pos": 128292,
      "inform future research": 54335,
      "future research applications": 46292,
      "context window size": 22325,
      "english word embeddings": 37343,
      "predict speech pos": 88941,
      "important machine translation": 52183,
      "correlation principal component": 24243,
      "singular value decomposition": 109862,
      "introduce simple efficient": 56539,
      "indonesian online news": 53996,
      "downstream tasks pretrained": 34049,
      "distributional semantics distributional": 32725,
      "relations named entities": 98216,
      "paper present multimodal": 84113,
      "increasingly important role": 53701,
      "benchmark tasks glue": 12865,
      "text processing deep": 121208,
      "different tasks datasets": 31478,
      "sense disambiguation account": 105661,
      "task evaluation metrics": 118159,
      "models applied language": 72758,
      "fact word embeddings": 42840,
      "capture semantic relationship": 15382,
      "representations word vectors": 99986,
      "extent commonly used": 41974,
      "work starting point": 134816,
      "earth movers distance": 34404,
      "baselines source code": 12467,
      "different embedding spaces": 31123,
      "different sets word": 31423,
      "embeddings different algorithms": 35649,
      "word embeddings lowresource": 133155,
      "using large amounts": 129786,
      "positive unlabeled pu": 88347,
      "unlabeled pu learning": 127405,
      "obtains competitive results": 81463,
      "method generate word": 68859,
      "used word embedding": 128851,
      "similarity benchmark task": 109204,
      "dependency parsing experiments": 29180,
      "better word embedding": 13770,
      "semantic change word": 105003,
      "results work present": 102347,
      "present experimental study": 89482,
      "representation individual word": 99262,
      "learning universal representations": 63137,
      "work introduces explores": 134584,
      "introduces explores universal": 56611,
      "explores universal representation": 41654,
      "universal representation learning": 127325,
      "representation learning embeddings": 99285,
      "learning embeddings different": 62536,
      "different levels linguistic": 31232,
      "levels linguistic unit": 63549,
      "linguistic unit uniform": 64573,
      "unit uniform vector": 127219,
      "uniform vector space": 127143,
      "vector space taskindependent": 131385,
      "space taskindependent evaluation": 111069,
      "present approach constructing": 89374,
      "approach constructing analogy": 7455,
      "constructing analogy datasets": 21672,
      "analogy datasets terms": 5420,
      "datasets terms words": 27752,
      "terms words phrases": 120409,
      "phrases sentences experiment": 87455,
      "sentences experiment multiple": 106300,
      "experiment multiple representation": 40475,
      "multiple representation models": 75667,
      "representation models examine": 99338,
      "models examine geometric": 73182,
      "examine geometric properties": 39747,
      "geometric properties learned": 47863,
      "properties learned vector": 92459,
      "achieves highest accuracy": 2803,
      "tasks different language": 119061,
      "word class flexibility": 132951,
      "method builds recent": 68684,
      "qualitative quantitative analyses": 94571,
      "unlike previous works": 127445,
      "shown word embeddings": 108540,
      "language models word": 59689,
      "used widely various": 128849,
      "large volumes text": 61328,
      "capturing semantic information": 15486,
      "applications word embeddings": 7039,
      "models recent word": 73876,
      "word embeddings techniques": 133221,
      "combination word embeddings": 18597,
      "specific task paper": 111497,
      "static word embedding": 113067,
      "word embeddings lexical": 133149,
      "single words word": 109817,
      "used improve quality": 128584,
      "natural language processing article": 76401,
      "different types neural networks": 31535,
      "accuracy lower computational cost": 2203,
      "useful various nlp tasks": 128947,
      "embeddings distributed word representations": 35664,
      "propose unsupervised method learning": 93136,
      "shown capture semantic syntactic": 108450,
      "model jointly learns word": 71385,
      "relation learning recent work": 98051,
      "semantic similarity words phrases": 105298,
      "document representation models results": 33066,
      "stateoftheart word embedding models": 113030,
      "ppdb ganitkevitch et al": 88687,
      "ganitkevitch et al 2013": 46435,
      "word embedding learning method": 133016,
      "language processing tasks existing": 59954,
      "processing tasks existing models": 91814,
      "vector space representations words": 131383,
      "word similarity tasks word": 133572,
      "learn vector representations words": 62187,
      "processing information retrieval tasks": 91683,
      "embeddings used downstream applications": 36005,
      "word embeddings dimensionality reduction": 133084,
      "common natural language processing": 18898,
      "propose learning objective motivated": 92748,
      "learning objective motivated theoretical": 62856,
      "sentence representations word information": 106048,
      "word embeddings proposed method": 133193,
      "outperform current state art": 82704,
      "word representations able capture": 133443,
      "gained popularity recent years": 46371,
      "text classification propose novel": 120758,
      "distributional word vector spaces": 32748,
      "vectors words occurring external": 131495,
      "words occurring external lexicons": 134093,
      "models achieve better performances": 72660,
      "stanford contextual word similarity": 112362,
      "natural language processing related": 76481,
      "propose new method called": 92865,
      "trained large text corpora": 123183,
      "deep neural networks use": 28384,
      "word sense disambiguation account": 133518,
      "different sets word embeddings": 31424,
      "positive unlabeled pu learning": 88348,
      "work introduces explores universal": 134585,
      "introduces explores universal representation": 56612,
      "explores universal representation learning": 41655,
      "universal representation learning embeddings": 127326,
      "representation learning embeddings different": 99286,
      "learning embeddings different levels": 62537,
      "embeddings different levels linguistic": 35651,
      "different levels linguistic unit": 31233,
      "levels linguistic unit uniform": 63550,
      "linguistic unit uniform vector": 64574,
      "unit uniform vector space": 127220,
      "vector space taskindependent evaluation": 131386,
      "present approach constructing analogy": 89375,
      "approach constructing analogy datasets": 7456,
      "constructing analogy datasets terms": 21673,
      "analogy datasets terms words": 5421,
      "datasets terms words phrases": 27753,
      "terms words phrases sentences": 120410,
      "words phrases sentences experiment": 134120,
      "phrases sentences experiment multiple": 87456,
      "sentences experiment multiple representation": 106301,
      "experiment multiple representation models": 40476,
      "multiple representation models examine": 75668,
      "representation models examine geometric": 99339,
      "models examine geometric properties": 73183,
      "examine geometric properties learned": 39748,
      "geometric properties learned vector": 47864,
      "properties learned vector space": 92460,
      "pretrained word embeddings shown": 90223,
      "various machine learning ml": 131127,
      "ppdb ganitkevitch et al 2013": 88688,
      "natural language processing tasks existing": 76493,
      "language processing tasks existing models": 59955,
      "language processing information retrieval tasks": 59874,
      "propose learning objective motivated theoretical": 92749,
      "popular natural language processing nlp": 88109,
      "vectors words occurring external lexicons": 131496,
      "work introduces explores universal representation": 134586,
      "introduces explores universal representation learning": 56613,
      "explores universal representation learning embeddings": 41656,
      "universal representation learning embeddings different": 127327,
      "representation learning embeddings different levels": 99287,
      "learning embeddings different levels linguistic": 62538,
      "embeddings different levels linguistic unit": 35652,
      "different levels linguistic unit uniform": 31234,
      "levels linguistic unit uniform vector": 63551,
      "linguistic unit uniform vector space": 64575,
      "present approach constructing analogy datasets": 89376,
      "approach constructing analogy datasets terms": 7457,
      "constructing analogy datasets terms words": 21674,
      "analogy datasets terms words phrases": 5422,
      "datasets terms words phrases sentences": 27754,
      "terms words phrases sentences experiment": 120411,
      "words phrases sentences experiment multiple": 134121,
      "phrases sentences experiment multiple representation": 87457,
      "sentences experiment multiple representation models": 106302,
      "experiment multiple representation models examine": 40477,
      "multiple representation models examine geometric": 75669,
      "representation models examine geometric properties": 99340,
      "models examine geometric properties learned": 73184,
      "examine geometric properties learned vector": 39749,
      "geometric properties learned vector space": 47865,
      "verbnoun": 131547,
      "appreciated": 7284,
      "v11": 130688,
      "831": 1309,
      "2018s": 691,
      "sesame": 107333,
      "httpsgithubcomnamisanmtdnn": 50703,
      "837": 1314,
      "amazing": 5262,
      "widelystudied": 132601,
      "linearization": 64397,
      "sanh": 103627,
      "structuredependent": 114053,
      "singlelanguage": 109832,
      "syntaxsensitive": 116591,
      "selfsimilarity": 104932,
      "precursor": 88815,
      "hire": 50510,
      "feather": 43245,
      "subjectobject": 114707,
      "662": 1185,
      "strubell": 113752,
      "corrupts": 24332,
      "reside": 100734,
      "undermined": 126711,
      "commits": 18848,
      "protorole": 93681,
      "adapterhubml": 3299,
      "intermediatetask": 56151,
      "humicroedit": 51136,
      "memorizing": 68283,
      "dichotomy": 30850,
      "modulo": 74533,
      "punta": 94405,
      "cana": 15093,
      "sprung": 112057,
      "concentrating": 20557,
      "robertabase": 103003,
      "fineturning": 44527,
      "maskpredict": 67317,
      "increment": 53719,
      "upgrading": 127815,
      "fixedsized": 44624,
      "pairings": 83477,
      "constructional": 21702,
      "farreaching": 43108,
      "robertas": 103007,
      "anisotropy": 6120,
      "drivers": 34158,
      "electric": 35243,
      "syntaxenhanced": 116584,
      "confine": 21010,
      "535": 1103,
      "arm": 9012,
      "recurrency": 97157,
      "datascarce": 26699,
      "onestage": 81718,
      "forgotten": 45065,
      "collaboratively": 18364,
      "obfuscated": 81044,
      "unsettled": 127559,
      "rosita": 103278,
      "985": 1449,
      "pruningbased": 94174,
      "mediators": 68193,
      "specialize": 111382,
      "lead faster": 61852,
      "object study": 81055,
      "study general": 114391,
      "evaluating analyzing": 39035,
      "analyzing performance": 6046,
      "knowledge tasks": 58202,
      "detailed linguistic": 29778,
      "separate model": 106854,
      "developing general": 30341,
      "general robust": 46707,
      "model filters": 71180,
      "order optimize": 82376,
      "model employ": 71061,
      "fscore gain": 45911,
      "model highest": 71285,
      "highest fscore": 50227,
      "improvement linguistic": 52719,
      "using real": 130082,
      "art wide": 9104,
      "architecture lstm": 8688,
      "lstm cnn": 65609,
      "cnn self": 18025,
      "vary network": 131245,
      "network depth": 77221,
      "longer range": 65159,
      "independent architecture": 53761,
      "learned auxiliary": 62202,
      "investigate properties": 56802,
      "type linguistic": 126210,
      "ability induce": 1533,
      "induce syntactic": 54005,
      "syntactic partofspeech": 116446,
      "make fair": 66670,
      "pretraining deep": 90249,
      "bidirectional representations": 13974,
      "text jointly": 121068,
      "point absolute": 87796,
      "accuracy 867": 2076,
      "squad v11": 112080,
      "pretraining sentence": 90320,
      "effective language": 34695,
      "similar improvements": 109098,
      "sesame street": 107334,
      "seen surge": 104543,
      "pretrained variants": 90209,
      "variants language": 130876,
      "modeling especially": 72424,
      "intermediate task": 56146,
      "task negatively": 118449,
      "syntactic abilities": 116360,
      "english syntactic": 37298,
      "phenomena using": 87248,
      "replaced words": 98939,
      "agreement reflexive": 4403,
      "reflexive anaphora": 97629,
      "anaphora phenomena": 6064,
      "phenomena bert": 87220,
      "multiple natural": 75621,
      "regularization effect": 97747,
      "2015 incorporating": 590,
      "snli scitail": 110317,
      "tasks pushing": 119424,
      "substantially fewer": 114889,
      "labels pretrained": 58630,
      "available httpsgithubcomnamisanmtdnn": 11011,
      "set nlp": 107510,
      "tasks suggesting": 119536,
      "encode useful": 36463,
      "light linguistic": 63994,
      "study representations": 114502,
      "produced recent": 91966,
      "addition better": 3404,
      "understand makes": 126758,
      "task yields": 118860,
      "task lead": 118343,
      "appropriate model": 8424,
      "model syntactic": 72128,
      "computational expense": 20378,
      "primary contribution": 90649,
      "baselines provide": 12447,
      "provide foundation": 93834,
      "works leverage": 134958,
      "compression approaches": 20307,
      "leading incomparable": 61896,
      "incomparable results": 53416,
      "generalized knowledge": 46829,
      "conduct exhaustive": 20842,
      "finetuning methods": 44477,
      "methods bert": 69343,
      "steps traditional": 113337,
      "tasks building": 118971,
      "longrange phenomena": 65182,
      "phenomena existing": 87224,
      "translation produce": 125147,
      "produce strong": 91938,
      "small improvements": 110156,
      "improvements semantic": 52907,
      "noncontextual baseline": 80179,
      "shown contextual": 108456,
      "require sensitivity": 100197,
      "using diagnostic": 129563,
      "anaphora does": 6063,
      "networks bert": 77522,
      "great recent": 49019,
      "outputs language": 83173,
      "model surprisal": 72126,
      "works propose": 134969,
      "methods analyzing": 69309,
      "apply bert": 7163,
      "correspond linguistic": 24266,
      "linguistic notions": 64514,
      "notions syntax": 80414,
      "heads attend": 49604,
      "remarkably high": 98851,
      "use demonstrate": 127990,
      "inducing syntactic": 54018,
      "main verb": 66475,
      "sentence ungrammatical": 106116,
      "understanding challenging": 126811,
      "approach multitask": 7721,
      "singletask multitask": 109848,
      "impact final": 51871,
      "replication study": 98968,
      "impact key": 51874,
      "bert significantly": 13230,
      "published best": 94349,
      "recently reported": 96751,
      "code predicting": 18130,
      "prosodic labels": 93645,
      "train number": 122993,
      "models 10": 72633,
      "dataset light": 27002,
      "light results": 63999,
      "presented based": 89779,
      "distinguishing feature": 32589,
      "use grammar": 128073,
      "higher results": 50203,
      "exploration work": 41506,
      "bert new": 13195,
      "leverage language": 63593,
      "levels language": 63545,
      "outperforming published": 82826,
      "techniques pretrained": 119953,
      "general algorithm": 46631,
      "interaction pretraining": 55961,
      "sequentially applied": 107262,
      "applied data": 7054,
      "structure dependence": 113832,
      "relations learning": 98201,
      "accurately model": 2386,
      "structure recent": 113940,
      "extent stateoftheart": 41986,
      "structuredependent phenomena": 114054,
      "number agreement": 80835,
      "verbs evaluate": 131556,
      "26 languages": 841,
      "languages singlelanguage": 60879,
      "capture syntaxsensitive": 15409,
      "representations comparing": 99553,
      "embeddings replacing": 35907,
      "produced models": 91960,
      "finite number": 44536,
      "layers elmo": 61772,
      "representations semanticsaware": 99874,
      "enables series": 36402,
      "series success": 107283,
      "incorporating structured": 53560,
      "pretrained semantic": 90181,
      "labeling introduce": 58505,
      "bert backbone": 13068,
      "facilitate wide": 42799,
      "particular encode": 85409,
      "distributional knowledge": 32707,
      "corpora incorporated": 23503,
      "work complement": 134421,
      "remaining consistent": 98781,
      "lexical simplification": 63820,
      "simplification task": 109593,
      "mean representations": 67609,
      "structure just": 113887,
      "tasks construction": 119008,
      "tasks learned": 119249,
      "control task": 22820,
      "accuracy second": 2269,
      "small bert": 110136,
      "large fraction": 61093,
      "vocabulary embedding": 131873,
      "models reduced": 73897,
      "model smaller": 72062,
      "distilled bert": 32525,
      "understanding benchmarks": 126805,
      "models prevalent": 73792,
      "generalpurpose language": 46884,
      "larger counterparts": 61358,
      "introduce triple": 56561,
      "representations considerably": 99559,
      "simplified version": 109609,
      "experiment datasets": 40458,
      "performance pretraining": 86617,
      "showed models": 108382,
      "struggle understand": 114129,
      "understand rare": 126770,
      "embeddings problem": 35883,
      "separately learning": 106874,
      "bert leads": 13155,
      "performance increases": 86458,
      "approaches advantage": 8054,
      "taking single": 117551,
      "power pretrained": 88649,
      "pairwise word": 83685,
      "embeddingbased approaches": 35534,
      "model pairwise": 71689,
      "pairwise interactions": 83674,
      "interactions word": 56007,
      "models disregard": 73092,
      "instead modeling": 55672,
      "bert tasks": 13239,
      "demonstrate consistent": 28693,
      "improvements quality": 52903,
      "adding explicit": 3375,
      "interaction module": 55955,
      "serve strong": 107302,
      "limits use": 64321,
      "embeddings leveraging": 35787,
      "strong representation": 113706,
      "disambiguating context": 31950,
      "results seven": 102165,
      "large variability": 61321,
      "trained multiple": 123208,
      "times dataset": 122169,
      "dataset evaluated": 26896,
      "generalization natural": 46782,
      "remarkably consistent": 98849,
      "consistent accuracy": 21371,
      "contrast models": 22698,
      "performance example": 86348,
      "final layers": 44106,
      "quality downstream": 94635,
      "tasks naturally": 119334,
      "examine recent": 39754,
      "entailment semantic": 37674,
      "study resulting": 114504,
      "layers does": 61769,
      "stored pretrained": 113392,
      "adequately learn": 3839,
      "learn humanlike": 62066,
      "improving bert": 53073,
      "start new": 112400,
      "new chapter": 78827,
      "practice apply": 88730,
      "tune model": 125929,
      "biased embedding": 13856,
      "process finetuning": 91488,
      "normalization methods": 80338,
      "witnessed development": 132751,
      "sufficient number": 115284,
      "large pretraining": 61223,
      "strubell et": 113753,
      "knowledge syntactic": 58198,
      "supervised selfattention": 115830,
      "coreference information": 23348,
      "performance complex": 86245,
      "coreference auxiliary": 23341,
      "auxiliary supervision": 10884,
      "variants model": 130877,
      "suggesting future": 115346,
      "techniques problems": 119955,
      "bert distributional": 13102,
      "embeddings vector": 36015,
      "produces contextualized": 91991,
      "track syntactic": 122740,
      "individual attention": 53898,
      "implicitly capture": 52025,
      "maximum attention": 67530,
      "extract implicit": 42086,
      "parsed english": 84898,
      "overall dependency": 83225,
      "incorporating bert": 53518,
      "power variety": 88657,
      "effectively apply": 34789,
      "think better": 121865,
      "leverage bert": 63577,
      "direction propose": 31829,
      "extract representations": 42109,
      "representations fused": 99664,
      "decoder nmt": 28037,
      "documentlevel translations": 33171,
      "translation achieve": 124619,
      "seven benchmark": 107867,
      "toolkit multitask": 122412,
      "rapid customization": 95711,
      "supports multitask": 116036,
      "compress deep": 20293,
      "universal feature": 127306,
      "bert natural": 13192,
      "improving deep": 53086,
      "learning broad": 62412,
      "prevent models": 90365,
      "datasets leading": 27544,
      "observe finetuning": 81195,
      "bert specific": 13232,
      "bert works": 13257,
      "pushed state": 94461,
      "understanding success": 126978,
      "studies popular": 114262,
      "source toolkit": 110856,
      "glue superglue": 48317,
      "recently natural": 96711,
      "novel pretrained": 80687,
      "2019 proposed": 706,
      "obtain stateofthe": 81325,
      "performance numerous": 86581,
      "finetuning representations": 44511,
      "set task": 107600,
      "released multilingual": 98487,
      "obtained impressive": 81378,
      "community started": 19101,
      "generate abundant": 46897,
      "number bert": 80849,
      "language tested": 60160,
      "potential mbert": 88573,
      "provide immediate": 93846,
      "website used": 132298,
      "model reconstruct": 71868,
      "tasks generally": 119143,
      "amounts compute": 5331,
      "input approach": 55298,
      "network instead": 77286,
      "models example": 73186,
      "model gpu": 71263,
      "differences accuracy": 30950,
      "reflect differences": 97603,
      "accuracy respect": 2262,
      "length mdl": 63366,
      "effectively transmit": 34856,
      "needed achieve": 76870,
      "achieve quality": 2519,
      "probing model": 90894,
      "results informative": 101864,
      "adaptive inference": 3331,
      "speed inference": 111868,
      "chinese datasets": 16755,
      "datasets able": 27290,
      "distillation bert": 32503,
      "studies showing": 114279,
      "smaller size": 110243,
      "parameters current": 84741,
      "bert distillation": 13101,
      "simple lstm": 109461,
      "representations recently": 99842,
      "introduce learning": 56449,
      "case learning": 15591,
      "problem settings": 91225,
      "settings unsupervised": 107842,
      "pretraining different": 90253,
      "verify generality": 131588,
      "understanding discourse": 126832,
      "maintaining comparable": 66508,
      "models common": 72928,
      "achieves successful": 2917,
      "various supervised": 131210,
      "applying bert": 7233,
      "called transformerbased": 15066,
      "benefits deep": 13006,
      "task times": 118793,
      "interpreted context": 56281,
      "task related": 118621,
      "alternative finetuning": 5230,
      "present efficient": 89461,
      "finetuning smaller": 44513,
      "information necessary": 54791,
      "produce models": 91909,
      "data objective": 26183,
      "mechanism focus": 67987,
      "finetuning bertlarge": 44450,
      "did achieve": 30910,
      "bridge performance": 14597,
      "tasks successfully": 119534,
      "formal text": 45179,
      "bert implemented": 13141,
      "great effectiveness": 49001,
      "step involves": 113275,
      "sequences time": 107142,
      "structure capable": 113814,
      "fixedsize representations": 44622,
      "embeddings finetuning": 35725,
      "little understood": 64833,
      "makes significant": 66807,
      "lead catastrophic": 61846,
      "particular dependency": 85403,
      "processing finally": 91673,
      "outofdomain sentences": 82660,
      "sentences suggesting": 106509,
      "suggesting room": 115352,
      "improvement model": 52725,
      "exploring latent": 41663,
      "encoders elmo": 36646,
      "bert brought": 13084,
      "learn encode": 62041,
      "encode meaningful": 36439,
      "existing classifierbased": 40088,
      "representations interpretable": 99702,
      "form experiments": 45086,
      "new evidence": 78913,
      "encoders including": 36650,
      "earlier methods": 34378,
      "work showed": 134798,
      "method relation": 69099,
      "tree construction": 125585,
      "construction approach": 21684,
      "produces significantly": 92008,
      "bert introducing": 13146,
      "set additional": 107351,
      "tasks taken": 119546,
      "technique analyzing": 119767,
      "task improvement": 118281,
      "linguistic formalism": 64484,
      "contextualized encoders": 22549,
      "role semantics": 103217,
      "important dimension": 52143,
      "studies commonly": 114193,
      "crosslingual experimental": 24956,
      "extraction stage": 42492,
      "learn task": 62162,
      "learned multiple": 62233,
      "knowledge different": 57860,
      "traditional strategies": 122873,
      "strategies finetuning": 113465,
      "code adapters": 18070,
      "adapters available": 3301,
      "available adapterhubml": 10934,
      "intermediatetask transfer": 56153,
      "gains natural": 46398,
      "intermediatetask training": 56152,
      "strongly correlated": 113743,
      "highlighting need": 50280,
      "benchmarks observe": 12928,
      "forgetting knowledge": 45062,
      "benchmarks little": 12916,
      "recent attention": 96431,
      "neural encoders": 77908,
      "extent information": 41977,
      "actually used": 3155,
      "showing models": 108416,
      "embeddings play": 35864,
      "encoding properties": 36717,
      "properties training": 92488,
      "task highlighting": 118258,
      "highlighting importance": 50279,
      "importance careful": 52048,
      "experiments finally": 40946,
      "determining entity": 30147,
      "entity represented": 38098,
      "modeling specifically": 72546,
      "specifically bert": 111530,
      "contextual properties": 22487,
      "present fully": 89497,
      "performance established": 86344,
      "especially applied": 38432,
      "lms bert": 64880,
      "forgetting distributional": 45060,
      "corpus respectively": 23974,
      "code experiments": 18101,
      "structure distillation": 113839,
      "bidirectional encoders": 13920,
      "success downstream": 115078,
      "success remains": 115128,
      "bert fully": 13132,
      "syntactically informative": 116513,
      "set structured": 107592,
      "mixed results": 70401,
      "understanding syntactic": 126980,
      "benchmarks natural": 12923,
      "assessing ability": 9506,
      "models roberta": 73965,
      "roberta distilbert": 102995,
      "distilbert albert": 32497,
      "models humor": 73346,
      "results offer": 102011,
      "offer alternative": 81560,
      "assessed measuring": 9502,
      "measuring influence": 67917,
      "influence causal": 54302,
      "possible partofspeech": 88419,
      "multiple studies": 75709,
      "wordlevel linguistic": 133743,
      "representations extract": 99650,
      "information threat": 55043,
      "considerable efforts": 21247,
      "test words": 120553,
      "seen versus": 104554,
      "versus unseen": 131632,
      "used explicitly": 128537,
      "efficient resourceintensive": 35104,
      "words follow": 133961,
      "semantic signals": 105275,
      "occur time": 81503,
      "employed enhance": 36297,
      "models reaching": 73859,
      "reaching performance": 95922,
      "attached supplementary": 9691,
      "global selfattention": 48267,
      "cost attention": 24346,
      "attention block": 9801,
      "design build": 29524,
      "lower training": 65450,
      "assessing linguistic": 9513,
      "diagnostic classifier": 30508,
      "2019 showed": 711,
      "classification insufficient": 17235,
      "criteria lead": 24794,
      "task boosting": 117944,
      "boosting bert": 14442,
      "bert dependencies": 13094,
      "network incorporate": 77282,
      "tuned language": 125933,
      "classifying sentence": 17677,
      "behaviors pretrained": 12677,
      "set randomly": 107554,
      "predict tokens": 88946,
      "improve pretraining": 52501,
      "model varies": 72297,
      "necessarily provide": 76747,
      "provide model": 93875,
      "punta cana": 94406,
      "contextualised representations": 22533,
      "novel bertbased": 80505,
      "probing paper": 90899,
      "paper draw": 83884,
      "enable intrinsic": 36356,
      "estimate word": 38638,
      "bert shows": 13228,
      "word greater": 133306,
      "word versus": 133655,
      "prime words": 90659,
      "lower probabilities": 65446,
      "studying word": 114562,
      "possible parallels": 88418,
      "phrasal representation": 87326,
      "representation composition": 99193,
      "tasks new": 119342,
      "representation phrases": 99375,
      "sophisticated composition": 110666,
      "phrasal representations": 87327,
      "results control": 101638,
      "phrase representation": 87366,
      "models relies": 73919,
      "types make": 126317,
      "models compressing": 72945,
      "assistants amazon": 9576,
      "alexa google": 4671,
      "learn compositional": 62003,
      "performances semantic": 86895,
      "parsing performances": 85192,
      "recipe training": 96790,
      "training analyze": 123358,
      "empirical assessment": 36154,
      "humans process": 51102,
      "sequence encoded": 106936,
      "partial output": 85296,
      "output provided": 83111,
      "contexts generated": 22396,
      "learn represent": 62125,
      "new englishlanguage": 78900,
      "diagnostic set": 30512,
      "use test": 128325,
      "model prefers": 71765,
      "models scratch": 73980,
      "success large": 115086,
      "sparked probing": 111215,
      "probing representations": 90905,
      "analysis typologically": 5918,
      "effects tasks": 35005,
      "lms display": 64881,
      "bert widely": 13253,
      "certain aspects": 15931,
      "structured manner": 114013,
      "large predefined": 61214,
      "hierarchically organized": 50021,
      "bert layers": 13154,
      "lmtc datasets": 64893,
      "scale pretrained": 103746,
      "remains nontrivial": 98810,
      "taking different": 117547,
      "different bert": 31029,
      "simple lightweight": 109455,
      "taskspecific dataset": 119623,
      "flexible efficient": 44657,
      "easily conduct": 34443,
      "autoregressive baselines": 10861,
      "contrastive pretraining": 22737,
      "discrepancy pretraining": 32157,
      "present contrastive": 89419,
      "learn noise": 62104,
      "focuses particular": 44910,
      "modality comprehensive": 70486,
      "understanding crossmodal": 126822,
      "surprisingly consistent": 116132,
      "tasks application": 118932,
      "application models": 6866,
      "focusing bert": 44922,
      "linear clustering": 64335,
      "scheme works": 103944,
      "explore robustness": 41582,
      "work finds": 134530,
      "distribution statistics": 32680,
      "finding word": 44285,
      "different use": 31543,
      "features bert": 43378,
      "injection method": 55268,
      "predict missing": 88902,
      "sentence knowledge": 105916,
      "knowledge lexical": 58052,
      "embeddings layer": 35777,
      "information beneficial": 54394,
      "model qualitative": 71833,
      "cases involving": 15648,
      "clustering text": 17958,
      "syntactic formalisms": 116408,
      "report competitive": 98982,
      "induction resource": 54032,
      "make right": 66720,
      "prediction taking": 89131,
      "shows methods": 108597,
      "language consists": 58912,
      "form function": 45088,
      "shown bert": 108446,
      "linguistic dimensions": 64467,
      "studies exploring": 114226,
      "results allow": 101503,
      "linguists typically": 64629,
      "observation potentially": 81165,
      "characterlevel contextual": 16558,
      "representation structure": 99421,
      "model encoder": 71070,
      "separate encoder": 106849,
      "adding individual": 3382,
      "noncontextual embeddings": 80180,
      "performance subset": 86768,
      "language modelbased": 59421,
      "similarity vectors": 109331,
      "vectors demonstrate": 131424,
      "distinguish word": 32581,
      "difficult incorporate": 31634,
      "bert blackbox": 13083,
      "models draw": 73111,
      "encoder block": 36501,
      "models internal": 73419,
      "offer insights": 81566,
      "data nlp": 26171,
      "transformer lms": 124336,
      "lms learn": 64885,
      "data adopt": 25583,
      "finetuning nlu": 44486,
      "learning curves": 62480,
      "encode syntactic": 36458,
      "features test": 43756,
      "needed order": 76876,
      "necessary language": 76757,
      "understanding large": 126875,
      "multitask benchmarks": 75813,
      "pretraining transfer": 90345,
      "benchmarks focus": 12904,
      "new multitask": 79042,
      "available build": 10956,
      "using representational": 130109,
      "language typically": 60197,
      "aspects context": 9376,
      "context captured": 22024,
      "studies investigate": 114242,
      "representation encodes": 99223,
      "word determined": 132989,
      "reflect linguistic": 97607,
      "language core": 58920,
      "significant breakthroughs": 108732,
      "tasks attracted": 118945,
      "gains bert": 46385,
      "specifically decompose": 111535,
      "phases phase": 87216,
      "reducing performance": 97432,
      "process stages": 91573,
      "layers gradually": 61781,
      "depth model": 29322,
      "model adding": 70628,
      "stage train": 112154,
      "added parameters": 3366,
      "improved experimental": 52603,
      "tasks considerable": 119004,
      "considerable performance": 21255,
      "gains large": 46397,
      "small student": 110210,
      "models studies": 74109,
      "knowledge domains": 57884,
      "domains achieve": 33724,
      "public multidomain": 94264,
      "recently leveraging": 96705,
      "suitability approach": 115390,
      "approach low": 7694,
      "points work": 87886,
      "learning speed": 63045,
      "gain model": 46346,
      "layers language": 61784,
      "suitable lowresource": 115403,
      "crucial performance": 25158,
      "use genetic": 128070,
      "algorithm ga": 4743,
      "accelerate search": 1871,
      "chosen evaluation": 16940,
      "evaluation obtaining": 39314,
      "strategy proposed": 113534,
      "tokens occur": 122321,
      "problem leveraging": 91105,
      "leveraging syntactic": 63706,
      "text enhance": 120913,
      "syntax text": 116563,
      "stage finetuning": 112147,
      "limits application": 64314,
      "syntaxaware attention": 116574,
      "considers dependency": 21337,
      "tree text": 125617,
      "tokens dependency": 122303,
      "performance public": 86638,
      "datasets major": 27556,
      "major findings": 66560,
      "text improves": 121046,
      "global syntactic": 48277,
      "larger performance": 61376,
      "focuses linguistic": 44907,
      "specific granularity": 111448,
      "specifically extract": 111549,
      "segments based": 104671,
      "incorporate different": 53463,
      "chinese including": 16774,
      "evaluation finally": 39215,
      "unified pretraining": 127126,
      "finegrained coarsegrained": 44342,
      "attention pretrained": 9976,
      "attention local": 9872,
      "restrict attention": 101336,
      "propose syntaxaware": 93102,
      "focus syntactically": 44827,
      "bert benchmark": 13074,
      "forward field": 45326,
      "context single": 22270,
      "embeddings difficult": 35653,
      "exists small": 40350,
      "embeddings successfully": 35963,
      "large new": 61173,
      "derived bert": 29346,
      "semantic differences": 105031,
      "additionally provides": 3625,
      "features represented": 43702,
      "development large": 30396,
      "greatly increased": 49057,
      "network binary": 77173,
      "inherits good": 55199,
      "drop compared": 34167,
      "glue squad": 48315,
      "pressing issue": 89956,
      "datasets sentiment": 27699,
      "analysis compare": 5533,
      "furthermore test": 46216,
      "provides solid": 94077,
      "roberta achieved": 102991,
      "tasks memory": 119292,
      "viable solution": 131645,
      "representation syntax": 99430,
      "understanding key": 126871,
      "better integrate": 13607,
      "gets competitive": 47951,
      "task domainspecific": 118118,
      "outperform generic": 82709,
      "generic pretrained": 47806,
      "extraction requires": 42469,
      "domain bert": 33473,
      "typical method": 126401,
      "scarce training": 103795,
      "automatically augments": 10725,
      "better original": 13639,
      "robust training": 103078,
      "inference obtaining": 54184,
      "tradeoff performances": 122780,
      "problem early": 91020,
      "onestage joint": 81719,
      "datasets performed": 27622,
      "structure let": 113897,
      "novel multihead": 80654,
      "focused exclusively": 44855,
      "exclusively english": 39959,
      "decoding experiments": 28093,
      "baseline accuracy": 12178,
      "furthermore attempt": 46148,
      "address recent": 3760,
      "experiment finetuning": 40467,
      "parameters frozen": 84750,
      "solve supervised": 110622,
      "application pretrained": 6874,
      "usually contain": 130412,
      "modeling various": 72586,
      "parameters online": 84774,
      "scenarios knowledge": 103854,
      "provide best": 93769,
      "structural aspect": 113758,
      "tasks providing": 119422,
      "source files": 110761,
      "variable names": 130845,
      "researchers focus": 100693,
      "opening black": 81993,
      "box models": 14532,
      "employ syntactic": 36287,
      "head results": 49578,
      "grammar types": 48662,
      "similarity sms": 109313,
      "tasks explicit": 119124,
      "explicit consideration": 41307,
      "information proved": 54884,
      "proved crucial": 93710,
      "applications incorporate": 6939,
      "trees effectively": 125688,
      "applicable arbitrary": 6823,
      "various datasets": 131071,
      "effectiveness syntax": 34959,
      "improvement multiple": 52729,
      "hinders application": 50454,
      "faced problem": 42749,
      "work primarily": 134703,
      "critical question": 24825,
      "response question": 101233,
      "methods weight": 69844,
      "findings best": 44288,
      "large sparse": 61277,
      "sparse models": 111229,
      "contain small": 21761,
      "assignment problem": 9562,
      "problem allowing": 90929,
      "equal number": 38225,
      "scheme improves": 103929,
      "training requiring": 123813,
      "exploring role": 41666,
      "results studies": 102218,
      "achieved training": 2715,
      "obtained different": 81361,
      "model encoding": 71073,
      "space bert": 110975,
      "distinct meaningful": 32538,
      "explain reasons": 41258,
      "specific token": 111502,
      "grammatical number": 48714,
      "probing neural": 90896,
      "properties train": 92487,
      "encoded model": 36478,
      "models lower": 73531,
      "complexity finally": 19911,
      "lowerlevel tasks": 65457,
      "captured lower": 15431,
      "discover linguistic": 32117,
      "considers family": 21338,
      "reveal linguistic": 102498,
      "projecting embeddings": 92229,
      "objectives dependency": 81132,
      "reveal possible": 102503,
      "visualization results": 131827,
      "subsequent work": 114820,
      "facts training": 42924,
      "important question": 52221,
      "bound possible": 14512,
      "learn training": 62168,
      "surprisingly training": 116150,
      "conduct set": 20894,
      "predominantly focused": 89230,
      "relations experiment": 98164,
      "languages discourse": 60508,
      "capturing discourse": 15466,
      "layers best": 61764,
      "processing bert": 91632,
      "required perform": 100223,
      "controlling context": 22852,
      "tasks possible": 119386,
      "context lengths": 22166,
      "representations present simple": 99814,
      "tasks limited training": 119266,
      "separate model task": 106855,
      "limited available data": 64213,
      "state art wide": 112485,
      "effective paper present": 34729,
      "present detailed empirical": 89446,
      "cnn self attention": 18026,
      "learn representations vary": 62128,
      "representations vary network": 99973,
      "vary network depth": 131246,
      "word embedding layer": 133014,
      "ability induce syntactic": 1534,
      "make fair comparison": 66671,
      "data results suggest": 26360,
      "transfer learning applications": 124076,
      "pretraining deep bidirectional": 90250,
      "new language representation": 78979,
      "deep bidirectional representations": 28206,
      "stateoftheart models wide": 112755,
      "language modeling especially": 59431,
      "recently introduced bert": 96696,
      "subjectverb agreement reflexive": 114713,
      "agreement reflexive anaphora": 4404,
      "multitask deep neural": 75816,
      "multiple natural language": 75622,
      "bidirectional transformer language": 13979,
      "publicly available httpsgithubcomnamisanmtdnn": 94308,
      "set nlp tasks": 107511,
      "model syntactic structures": 72129,
      "classification model using": 17276,
      "model using bert": 72275,
      "foundation future work": 45345,
      "understanding paper explores": 126913,
      "use knowledge distillation": 128105,
      "ensemble learning improve": 37598,
      "model train single": 72180,
      "results question answering": 102104,
      "challenging practical problem": 16293,
      "leading incomparable results": 61897,
      "models significant speedup": 74037,
      "model bert bidirectional": 70762,
      "results language understanding": 101878,
      "conduct exhaustive experiments": 20843,
      "finally proposed solution": 44226,
      "classical nlp pipeline": 17093,
      "range syntactic semantic": 95607,
      "language modeling translation": 59479,
      "perform tasks require": 86088,
      "tasks require sensitivity": 119462,
      "outputs language model": 83174,
      "language model surprisal": 59404,
      "correspond linguistic notions": 24267,
      "language understanding challenging": 60211,
      "evaluate approach multitask": 38793,
      "glue benchmark method": 48311,
      "published best model": 94350,
      "number different models": 80870,
      "paper describes model": 83832,
      "respectively result new": 101161,
      "different levels language": 31231,
      "future research make": 46305,
      "structure recent work": 113941,
      "representations word different": 99980,
      "word different contexts": 132992,
      "cosine similarity different": 24339,
      "existing language representation": 40154,
      "character word embeddings": 16477,
      "explicit contextual semantics": 41310,
      "range downstream nlp": 95563,
      "downstream nlp applications": 34014,
      "word embeddings particular": 133178,
      "text corpora incorporated": 120828,
      "external lexical knowledge": 42027,
      "language modeling sentence": 59466,
      "sentence prediction objectives": 105997,
      "small bert models": 110137,
      "language understanding benchmarks": 60208,
      "largescale pretrained models": 61497,
      "word embeddings problem": 133189,
      "task downstream tasks": 118121,
      "student model experiment": 114140,
      "language models transformerbased": 59677,
      "transformerbased pretrained language": 124436,
      "output final layer": 83077,
      "pairwise word interaction": 83686,
      "contextualized embeddings bert": 22546,
      "nlp tasks outperforming": 79746,
      "embeddings word representations": 36032,
      "word representations training": 133476,
      "test set performance": 120513,
      "generalization natural language": 46783,
      "finetuning pretrained transformerbased": 44502,
      "textual entailment semantic": 121698,
      "similarity sentiment analysis": 109310,
      "knowledge stored pretrained": 58186,
      "apply pretrained bert": 7204,
      "fine tune model": 44326,
      "pretraining data tasks": 90247,
      "strubell et al": 113754,
      "knowledge syntactic structure": 58199,
      "existing model improve": 40202,
      "model trained scratch": 72207,
      "tiny fraction parameters": 122213,
      "word embeddings vector": 133239,
      "embeddings vector representations": 36016,
      "individual attention heads": 53899,
      "maximum attention weight": 67531,
      "dependency ud trees": 29252,
      "using methods results": 129880,
      "models known learn": 73444,
      "new algorithm named": 78768,
      "use bert extract": 127922,
      "encoder decoder nmt": 36510,
      "model attention mechanisms": 70703,
      "machine translation achieve": 65883,
      "seven benchmark datasets": 107868,
      "multitask learning paradigm": 75860,
      "bert natural language": 13193,
      "effective feature extractors": 34674,
      "bert specific task": 13233,
      "pushed state art": 94462,
      "open source toolkit": 81937,
      "transfer learning multitask": 124111,
      "multitask training experiments": 75896,
      "benchmark tasks demonstrate": 12864,
      "recently natural language": 96712,
      "pretrained contextual representation": 90008,
      "contextual representation models": 22491,
      "proposed model called": 93437,
      "obtain stateofthe art": 81326,
      "art performance numerous": 9080,
      "performance numerous nlp": 86582,
      "universal language model": 127310,
      "language model model": 59349,
      "nlp community started": 79589,
      "model reconstruct original": 71869,
      "produce good results": 91896,
      "pretraining task called": 90333,
      "experiments demonstrate new": 40888,
      "description length mdl": 29465,
      "english chinese datasets": 37085,
      "lstm based model": 65599,
      "experimental results multiple": 40657,
      "results multiple nlp": 101972,
      "tasks existing approaches": 119113,
      "negative log likelihood": 76933,
      "ability learn text": 1537,
      "effectively capture contextual": 34794,
      "contextual semantic information": 22495,
      "maintaining comparable performance": 66509,
      "comparable performance previous": 19157,
      "performance previous models": 86620,
      "semantic similarity task": 105291,
      "task related features": 118622,
      "yields performance comparable": 135426,
      "pretraining tasks downstream": 90336,
      "understanding tasks recent": 126990,
      "recent pretrained models": 96492,
      "pretrained models improved": 90141,
      "bridge performance gap": 14598,
      "pretrained model bert": 90130,
      "representations input sequences": 99697,
      "using large scale": 129790,
      "large scale natural": 61254,
      "solve downstream tasks": 110597,
      "lead catastrophic forgetting": 61847,
      "encoders elmo bert": 36647,
      "previous work showed": 90548,
      "supervised manner using": 115798,
      "approach evaluating language": 7544,
      "tasks introduce additional": 119222,
      "sentiment classification task": 106714,
      "downstream tasks recent": 34052,
      "results suggest linguistic": 102232,
      "knowledge extraction knowledge": 57931,
      "representations learned multiple": 99728,
      "code adapters available": 18071,
      "adapters available adapterhubml": 3302,
      "learning pretrained models": 62907,
      "gains natural language": 46399,
      "intermediate task finetuning": 56147,
      "task finetuning target": 118220,
      "given target task": 48146,
      "forgetting knowledge learned": 45063,
      "results nlp benchmarks": 101994,
      "representations learned neural": 99729,
      "model perform task": 71721,
      "case study natural": 15621,
      "models learn encode": 73482,
      "word embeddings play": 133182,
      "task determining entity": 118081,
      "models lms bert": 73519,
      "tasks recent work": 119440,
      "catastrophic forgetting distributional": 15711,
      "forgetting distributional knowledge": 45061,
      "models substantially outperform": 74115,
      "code experiments open": 18102,
      "experiments open sourced": 41049,
      "answer question introduce": 6561,
      "bert models masked": 13180,
      "data contribute better": 25793,
      "benchmarks natural language": 12924,
      "semeval2020 task assessing": 105552,
      "roberta distilbert albert": 102996,
      "extensive experiments models": 41916,
      "inference based approaches": 54118,
      "growing body work": 49167,
      "propose simple general": 93062,
      "seen versus unseen": 104555,
      "training method used": 123706,
      "rare words usually": 95752,
      "help model understand": 49754,
      "impressive performance various": 52320,
      "large memory footprint": 61149,
      "lower training cost": 65451,
      "produces better results": 91988,
      "semeval 2020 shared": 105509,
      "shared task classifying": 108028,
      "behaviors pretrained language": 12678,
      "does necessarily provide": 33371,
      "use pretrained contextual": 128204,
      "nlp tasks new": 79745,
      "humans paper present": 51097,
      "present systematic analysis": 89733,
      "stateoftheart pretrained transformers": 112872,
      "assistants amazon alexa": 9577,
      "alexa google assistant": 4672,
      "stateoftheart performances semantic": 112864,
      "features use features": 43772,
      "remains unclear extent": 98826,
      "context work present": 22333,
      "work present systematic": 134689,
      "analysis typologically diverse": 5919,
      "diverse languages different": 32821,
      "bert widely used": 13254,
      "used nlp community": 128662,
      "large scale pretrained": 61258,
      "scale pretrained language": 103747,
      "success various natural": 115142,
      "decoding algorithm named": 28072,
      "different prior work": 31358,
      "language models recently": 59650,
      "word embeddings layer": 133145,
      "performance improvements multiple": 86451,
      "model qualitative analysis": 71834,
      "report strong results": 99051,
      "popular language model": 88095,
      "context present novel": 22224,
      "discourse representation structure": 32082,
      "sequencetosequence model encoder": 107169,
      "data sets english": 26424,
      "closely related information": 17861,
      "distinguish word senses": 32582,
      "nlp tasks inspired": 79728,
      "achieve strong performance": 2575,
      "language generation evaluation": 59071,
      "pretraining transfer learning": 90346,
      "new multitask benchmark": 79043,
      "models language generation": 73450,
      "compare model performance": 19258,
      "source code dataset": 110718,
      "embeddings using representational": 36009,
      "using representational similarity": 130110,
      "results demonstrate ability": 101671,
      "generally involves huge": 46864,
      "involves huge number": 56894,
      "greatly reduce training": 49061,
      "model adding new": 70629,
      "significant performance degradation": 108835,
      "models applied various": 72759,
      "considerable performance gains": 21256,
      "help knowledge distillation": 49743,
      "effectiveness superiority proposed": 34958,
      "approach low resource": 7695,
      "layers language model": 61785,
      "lowresource settings improving": 65563,
      "small student model": 110211,
      "genetic algorithm ga": 47819,
      "using algorithm based": 129329,
      "text introduce new": 121062,
      "tokens dependency tree": 122304,
      "stateoftheart performance public": 112840,
      "english chinese including": 37088,
      "paper propose syntaxaware": 84331,
      "bert benchmark datasets": 13075,
      "information words context": 55105,
      "embeddings difficult interpret": 35654,
      "conduct extensive analyses": 20865,
      "datasets sentiment analysis": 27700,
      "tasks work provides": 119614,
      "bert roberta achieved": 13221,
      "based models use": 11854,
      "better sentence representation": 13720,
      "gets competitive performance": 47952,
      "scarce training data": 103796,
      "proposed method generates": 93369,
      "dramatically improved state": 34093,
      "downstream tasks providing": 34051,
      "natural language code": 76274,
      "opening black box": 81994,
      "black box models": 14246,
      "qualitative quantitative analysis": 94572,
      "trees pretrained language": 125705,
      "performances various nlp": 86904,
      "nlp tasks explicit": 79721,
      "address problem proposing": 3747,
      "problem proposing novel": 91188,
      "experiments various datasets": 41199,
      "model previous work": 71786,
      "previous work primarily": 90540,
      "approaches typically use": 8387,
      "neural networks linguistic": 78506,
      "models internal representations": 73420,
      "achieves higher accuracy": 2801,
      "captured lower layers": 15432,
      "methods learn training": 69585,
      "learn training data": 62169,
      "models existing work": 73193,
      "understanding natural language understanding": 126900,
      "tasks limited training data": 119267,
      "learn representations vary network": 62129,
      "representations vary network depth": 99974,
      "new language representation model": 78980,
      "language representation model called": 60041,
      "recently introduced bert model": 96697,
      "subjectverb agreement reflexive anaphora": 114714,
      "multitask deep neural networks": 75818,
      "multitask deep neural network": 75817,
      "multiple natural language understanding": 75623,
      "liu et al 2015": 64842,
      "bidirectional transformer language model": 13980,
      "models publicly available httpsgithubcomnamisanmtdnn": 73844,
      "deep neural networks knowledge": 28373,
      "experiment results method significantly": 40497,
      "model bert bidirectional encoder": 70763,
      "contextualized word representations contextualized": 22582,
      "perform tasks require sensitivity": 86089,
      "natural language understanding challenging": 76554,
      "best model achieves stateoftheart": 13370,
      "new natural language processing": 79047,
      "shared task paper presents": 108080,
      "understanding recently pretrained language": 126939,
      "representations word different contexts": 99981,
      "existing language representation models": 40155,
      "wide range downstream nlp": 132503,
      "student model experiment results": 114141,
      "transformerbased pretrained language models": 124437,
      "finetuning pretrained transformerbased language": 44503,
      "language models bert roberta": 59517,
      "tasks textual entailment semantic": 119561,
      "strubell et al 2018": 113755,
      "word embeddings vector representations": 133240,
      "embeddings vector representations words": 36017,
      "universal dependency ud trees": 127302,
      "propose new algorithm named": 92833,
      "opensource natural language understanding": 82024,
      "recently natural language processing": 96713,
      "pretrained contextual representation models": 90009,
      "art performance numerous nlp": 9081,
      "performance numerous nlp tasks": 86583,
      "minimum description length mdl": 70207,
      "model achieves promising results": 70585,
      "results english chinese datasets": 101755,
      "results multiple nlp tasks": 101973,
      "experiments method achieves stateoftheart": 41001,
      "contextualized language models bert": 22556,
      "large scale natural language": 61255,
      "code adapters available adapterhubml": 18072,
      "transfer learning pretrained models": 124123,
      "intermediate task finetuning target": 56148,
      "case study natural language": 15622,
      "language models lms bert": 59593,
      "lead catastrophic forgetting distributional": 61848,
      "catastrophic forgetting distributional knowledge": 15712,
      "code experiments open sourced": 18103,
      "data contribute better understanding": 25794,
      "benchmarks natural language understanding": 12925,
      "neural machine translation encoder": 77996,
      "code pretrained models released": 18137,
      "semeval 2020 shared task": 105510,
      "natural language understanding reasoning": 76578,
      "taskoriented semantic parsing models": 118903,
      "analysis typologically diverse languages": 5920,
      "typologically diverse languages different": 126499,
      "end propose novel approach": 36826,
      "large scale pretrained language": 61259,
      "scale pretrained language models": 103748,
      "success various natural language": 115143,
      "translation tasks proposed method": 125363,
      "different prior work focuses": 31359,
      "pretrained language models recently": 90100,
      "results wide range nlp": 102328,
      "embeddings using representational similarity": 36010,
      "using representational similarity analysis": 130111,
      "generally involves huge number": 46865,
      "greatly reduce training time": 49062,
      "results proposed method achieve": 102069,
      "results natural language understanding": 101979,
      "achieves stateoftheart performance public": 2891,
      "datasets english chinese including": 27450,
      "translation question answering paper": 125192,
      "task experimental results model": 118175,
      "experiments demonstrate proposed method": 40892,
      "model extensive experiments realworld": 71155,
      "dramatically improved state art": 34094,
      "trees pretrained language models": 125706,
      "performances various nlp tasks": 86905,
      "paper address problem proposing": 83707,
      "address problem proposing novel": 3748,
      "methods learn training data": 69586,
      "language understanding natural language understanding": 60242,
      "learn representations vary network depth": 62130,
      "pretrained models publicly available httpsgithubcomnamisanmtdnn": 90150,
      "multiple natural language understanding tasks": 75624,
      "experiment results method significantly outperform": 40498,
      "model bert bidirectional encoder representations": 70764,
      "best model achieves stateoftheart results": 13371,
      "language understanding recently pretrained language": 60264,
      "finetuning pretrained transformerbased language models": 44504,
      "pretrained language models bert roberta": 90066,
      "word embeddings vector representations words": 133241,
      "recently natural language processing nlp": 96714,
      "art performance numerous nlp tasks": 9082,
      "experiments method achieves stateoftheart performance": 41002,
      "lead catastrophic forgetting distributional knowledge": 61849,
      "large pretrained language models lms": 61220,
      "analysis typologically diverse languages different": 5921,
      "large scale pretrained language models": 61260,
      "results wide range nlp tasks": 102329,
      "embeddings using representational similarity analysis": 36011,
      "experimental results proposed method achieve": 40675,
      "results natural language understanding tasks": 101980,
      "machine translation question answering paper": 66184,
      "demonstrate proposed method significantly outperforms": 28844,
      "paper address problem proposing novel": 83708,
      "kbp": 57506,
      "longshorttermmemory": 65190,
      "virtues": 131754,
      "award": 11253,
      "awards": 11255,
      "entitypairs": 38144,
      "couplings": 24463,
      "ds": 34203,
      "decently": 27920,
      "infoboxes": 54331,
      "9th": 1453,
      "ea": 34366,
      "suffices": 115267,
      "falsepositive": 43047,
      "nre": 80808,
      "formalisation": 45182,
      "dolce": 33421,
      "humanevaluated": 51020,
      "25m": 835,
      "dags": 25505,
      "aliases": 4894,
      "wellvalidated": 132446,
      "easynegative": 34490,
      "docred": 32942,
      "errorcascading": 38346,
      "populating": 88163,
      "dots": 33959,
      "270": 854,
      "interdependency": 56056,
      "defects": 28469,
      "aggregator": 4371,
      "noisylabeled": 80136,
      "bigdata": 13999,
      "multilingualbert": 75406,
      "machineunderstandable": 66352,
      "cornerstones": 23393,
      "morgan": 74663,
      "leak": 61978,
      "unmatched": 127465,
      "resorting": 100825,
      "historybased": 50575,
      "gda": 46570,
      "634": 1171,
      "hands": 49425,
      "scierc": 104031,
      "featured": 43339,
      "setbased": 107642,
      "multiplied": 75764,
      "fn": 44730,
      "suppressing": 116047,
      "capsules": 15250,
      "vlsp": 131856,
      "theyre": 121858,
      "sensor": 105756,
      "gathers": 46530,
      "nto1": 80817,
      "783": 1269,
      "tac kbp": 117277,
      "effective algorithms": 34624,
      "techniques achieved": 119828,
      "improvements mainly": 52874,
      "multilingual open": 75322,
      "identify relation": 51541,
      "sentence relying": 106026,
      "english heavy": 37165,
      "set present": 107540,
      "languages release": 60840,
      "using inference": 129753,
      "contain errors": 21738,
      "data distant": 25856,
      "bootstrapping using": 14464,
      "using lists": 129821,
      "distant labeling": 32450,
      "data way": 26625,
      "reducing noise": 97429,
      "labels example": 58598,
      "constraints small": 21612,
      "networks traditional": 77797,
      "filling address": 44033,
      "sentence parts": 105980,
      "approaches relation": 8317,
      "classification finally": 17210,
      "better individual": 13606,
      "demonstrate effect": 28704,
      "finally combining": 44154,
      "voting scheme": 131955,
      "results semeval": 102146,
      "semeval 2010": 105494,
      "solve related": 110615,
      "structure small": 113958,
      "propagation graph": 92392,
      "large unstructured": 61319,
      "soft constraint": 110490,
      "type second": 126224,
      "argument relation": 8953,
      "classification associated": 17126,
      "applications artificial": 6897,
      "features solve": 43735,
      "parts target": 85591,
      "entities plays": 37838,
      "parsing used": 85276,
      "neural relation": 78647,
      "novel relational": 80706,
      "text predict": 121189,
      "relation pair": 98058,
      "rely direct": 98686,
      "propose pathbased": 92987,
      "sentences inference": 106355,
      "noise mitigation": 80058,
      "apply finegrained": 7175,
      "models comparable": 72932,
      "embeddings entities": 35697,
      "tasks performs": 119380,
      "simplification approach": 109577,
      "open approaches": 81880,
      "easier process": 34422,
      "subsequently applied": 114822,
      "specific challenges": 111416,
      "preliminary evaluation": 89265,
      "embedding relation": 35480,
      "textual relation": 121728,
      "textual relations": 121730,
      "base relations": 11481,
      "learned textual": 62265,
      "best existing": 13330,
      "survey deep": 116167,
      "noise generated": 80051,
      "generated training": 47141,
      "thoroughly evaluate": 121901,
      "approach wide": 8016,
      "stateoftheart various": 113023,
      "approach relation": 7851,
      "extraction fundamental": 42346,
      "overcome drawback": 83282,
      "drawback propose": 34120,
      "given instance": 48047,
      "performance attentive": 86155,
      "extract values": 42130,
      "types attributes": 126254,
      "dnn architecture": 32929,
      "slot type": 110069,
      "absolute fscore": 1728,
      "gain slot": 46352,
      "attention relation": 9991,
      "rnnbased model": 102947,
      "focuses important": 44901,
      "layers attention": 61762,
      "networks joint": 77631,
      "linearchain conditional": 64390,
      "predicting entity": 88984,
      "classify relation": 17656,
      "entities sentence": 37866,
      "regularization propose": 97754,
      "approaches chinese": 8095,
      "relation instances": 98044,
      "relation seeds": 98069,
      "statistics entities": 113199,
      "entities predict": 37839,
      "effective relation": 34736,
      "distributional patternbased": 32713,
      "types methods": 126319,
      "complementary supervision": 19754,
      "build effective": 14762,
      "module training": 74513,
      "informative patterns": 55145,
      "patterns pattern": 85770,
      "module framework": 74490,
      "supervision ds": 115880,
      "term pair": 120237,
      "identify problems": 51538,
      "widely varying": 132600,
      "interannotator disagreement": 56049,
      "model requiring": 71913,
      "chinese literature": 16784,
      "partially lack": 85306,
      "propose tagging": 93106,
      "including heuristic": 53297,
      "extraction distant": 42306,
      "extraction efficient": 42317,
      "corpora widely": 23620,
      "task modeling": 118400,
      "lowdimensional spaces": 65415,
      "intrasentence intersentence": 56341,
      "embedding sentences": 35489,
      "use informative": 128096,
      "setting addition": 107730,
      "extraction develop": 42303,
      "embeddings alleviate": 35551,
      "individual feature": 53908,
      "papers focus": 84505,
      "focus classification": 44745,
      "network entity": 77245,
      "wikipedia infoboxes": 132658,
      "features performs": 43655,
      "reliably detecting": 98626,
      "detecting relevant": 29849,
      "entities unstructured": 37887,
      "resource knowledge": 100851,
      "ensemble convolutional": 37587,
      "detailed explanations": 29776,
      "features analyze": 43361,
      "included input": 53246,
      "benefit word": 12999,
      "models distantly": 73093,
      "extraction bidirectional": 42271,
      "secondly introduce": 104475,
      "supervision dataset": 115878,
      "dataset relation": 27149,
      "multiple realworld": 75654,
      "parsers performance": 85039,
      "crf conditional": 24745,
      "identify multiple": 51523,
      "news biomedical": 79309,
      "paths semantic": 85692,
      "path embeddings": 85674,
      "clean noisy": 17702,
      "respectively furthermore": 101142,
      "language explanations": 59031,
      "explanations training": 41300,
      "training accurate": 123344,
      "language explanation": 59030,
      "parser converts": 84931,
      "providing explanations": 94113,
      "just labels": 57464,
      "specific entity": 111438,
      "bottleneck performance": 14497,
      "generate falsepositive": 46940,
      "reduction strategies": 97460,
      "learn sentencelevel": 62145,
      "rely distant": 98688,
      "labeling problems": 58529,
      "efficient flexible": 35078,
      "flexible applied": 44655,
      "shown experiments": 108470,
      "denoising method": 28999,
      "perform error": 85991,
      "terms texts": 120393,
      "lightweight semantic": 64015,
      "interpretation model": 56267,
      "currently semantic": 25416,
      "relations evaluated": 98154,
      "relations used": 98276,
      "ability express": 1520,
      "alternative semantic": 5243,
      "traditionally focuses": 122889,
      "techniques dealing": 119861,
      "observations text": 81184,
      "patterns experiments": 85743,
      "experiment demonstrate": 40459,
      "potential knowledge": 88569,
      "relations neural": 98220,
      "bases automatically": 12494,
      "resulting unacceptable": 101477,
      "035 039": 20,
      "work adversarial": 134362,
      "extraction adversarial": 42256,
      "training regularization": 123808,
      "particular demonstrate": 85402,
      "allows improving": 5164,
      "sentences typical": 106532,
      "typical methods": 126402,
      "integrating various": 55805,
      "model rich": 71934,
      "uses parallel": 129259,
      "crowdsourcing semantic": 25123,
      "method performing": 69045,
      "text known": 121073,
      "annotations sentences": 6462,
      "expanding number": 40363,
      "insufficient learning": 55724,
      "2d matrix": 881,
      "matrix row": 67481,
      "row vector": 103328,
      "using weak": 130365,
      "introduce weakly": 56567,
      "linguistic clues": 64437,
      "contextual usage": 22511,
      "obtaining high": 81440,
      "resourcescarce scenarios": 101071,
      "trees improves": 125695,
      "dependencybased models": 29256,
      "computationally inefficient": 20456,
      "structures efficiently": 114071,
      "content apply": 21856,
      "novel pruning": 80693,
      "pruning strategy": 94171,
      "input trees": 55461,
      "trees keeping": 125697,
      "path entities": 85675,
      "mechanism attends": 67948,
      "average attention": 11181,
      "implicit relation": 52015,
      "inference approach": 54115,
      "predictions entity": 89165,
      "implicit global": 52010,
      "constraints capture": 21592,
      "capture implicit": 15313,
      "predictions improve": 89176,
      "improve various": 52583,
      "sentence keeps": 105913,
      "current works": 25393,
      "works terms": 134984,
      "assuming sentence": 9659,
      "relations properly": 98240,
      "relations ignored": 98184,
      "difficult identify": 31632,
      "tackle new": 117299,
      "network acts": 77127,
      "sentence better": 105778,
      "improvement relation": 52754,
      "information aliases": 54372,
      "ignore readily": 51670,
      "available extensive": 10996,
      "paper firstly": 83945,
      "propose linear": 92752,
      "importance words": 52093,
      "sentence respect": 106051,
      "entities words": 37898,
      "capture relevance": 15369,
      "relevance sentences": 98518,
      "hidden relations": 49904,
      "latent entity": 61586,
      "ner addition": 77019,
      "mechanisms fully": 68061,
      "features effectively": 43473,
      "generated parameters": 47107,
      "learning field": 62581,
      "perform qualitative": 86051,
      "discover accurate": 32113,
      "considers interactions": 21339,
      "paths entities": 85688,
      "used update": 128835,
      "accurately predicts": 2389,
      "labeling present": 58522,
      "simple bertbased": 109376,
      "labeling recent": 58532,
      "years stateoftheart": 135303,
      "paper extensive": 83942,
      "search applied": 104299,
      "applied obtain": 7105,
      "globally consistent": 48287,
      "seven times": 107874,
      "highly advantageous": 50290,
      "entities stateoftheart": 37874,
      "nlp classification": 79586,
      "general issue": 46665,
      "propose distant": 92627,
      "supervision used": 115928,
      "examples obtained": 39861,
      "examples easynegative": 39827,
      "easynegative examples": 34491,
      "neural layer": 77961,
      "layer different": 61709,
      "rely set": 98738,
      "additional annotated": 3489,
      "annotated language": 6205,
      "introduces additional": 56605,
      "limitations introduce": 64178,
      "instead explicit": 55660,
      "selfattentive transformer": 104914,
      "dependencies entity": 29083,
      "tacred semeval": 117331,
      "documentlevel relation": 33159,
      "intersentence relations": 56310,
      "handled existing": 49407,
      "dataset documentlevel": 26880,
      "methods conduct": 69392,
      "based detailed": 11639,
      "experiments discuss": 40912,
      "discuss multiple": 32270,
      "multiple promising": 75648,
      "classification achieving": 17103,
      "recognizing diverse": 97068,
      "setting finetuning": 107749,
      "types high": 126293,
      "levels improving": 63543,
      "improving crossdomain": 53079,
      "trees shown": 125712,
      "trees potential": 125703,
      "potential problem": 88579,
      "semantics relations": 105456,
      "control information": 22808,
      "input entity": 55332,
      "provides potential": 94060,
      "potential way": 88598,
      "help ir": 49741,
      "conduct inference": 20882,
      "entity tuple": 38114,
      "multiple relation": 75664,
      "relation facts": 98036,
      "types latent": 126309,
      "relation classes": 97942,
      "based loss": 11816,
      "connections relations": 21148,
      "furthermore deal": 46159,
      "labels extensive": 58602,
      "address situation": 3767,
      "situation introduce": 109888,
      "aims facilitate": 4529,
      "facilitate information": 42774,
      "neural tagging": 78697,
      "extraction open": 42418,
      "challenge obtain": 16071,
      "text conventional": 120823,
      "methods heavily": 69528,
      "learning structures": 63065,
      "training exploration": 123631,
      "difficulties build": 31680,
      "way design": 132070,
      "design tagging": 29568,
      "relation tagging": 98072,
      "encode potential": 36444,
      "mechanism including": 67994,
      "multilingual resource": 75353,
      "heavily biased": 49649,
      "leveraging resource": 63700,
      "zeroshot relation": 135520,
      "language intelligence": 59214,
      "corpus entity": 23777,
      "entity label": 37947,
      "achieve f1score": 2460,
      "model ensembling": 71084,
      "extend knowledge": 41785,
      "based technology": 12110,
      "addition uses": 3484,
      "use handlabeled": 128082,
      "extraction wide": 42537,
      "yielding substantial": 135389,
      "existing joint": 40145,
      "entity level": 37950,
      "triple level": 125810,
      "tagging framework": 117392,
      "extraction ee": 42315,
      "ee task": 34578,
      "relational triples": 98100,
      "framework compared": 45458,
      "entities participating": 37835,
      "given relation": 48103,
      "relation list": 98053,
      "overview various": 83382,
      "extraction complex": 42283,
      "complex human": 19817,
      "relationships named": 98324,
      "using multiinstance": 129909,
      "neural rule": 78658,
      "rules rules": 103438,
      "automatically mined": 10802,
      "mechanism prior": 68028,
      "exact matching": 39721,
      "matching module": 67418,
      "supervision addition": 115864,
      "analysis public": 5789,
      "methods user": 69834,
      "efficiency human": 35026,
      "comparable models": 19151,
      "relations test": 98267,
      "community future": 19075,
      "triples sentence": 125819,
      "methods encoder": 69472,
      "module uses": 74516,
      "employing pretrained": 36319,
      "respectively indepth": 101145,
      "strategy joint": 113525,
      "solves task": 110637,
      "process extracting": 91483,
      "strategy model": 113529,
      "stateoftheart public": 112874,
      "labels clean": 58584,
      "cope challenge": 23271,
      "probabilistic modelling": 90822,
      "learning empirically": 62541,
      "labels leveraging": 58617,
      "noisy labeled": 80108,
      "supervision exploits": 115888,
      "unfortunately work": 127080,
      "supervised based": 115730,
      "outperform various": 82771,
      "strong model": 113687,
      "predict entities": 88884,
      "predict specific": 88939,
      "modularity extensibility": 74470,
      "making easy": 66837,
      "knowledgedriven applications": 58266,
      "networks slot": 77773,
      "filling slot": 44053,
      "classification particular": 17328,
      "arguments neural": 8989,
      "including joint": 53306,
      "typeaware neural": 126233,
      "lead best": 61842,
      "pipeline joint": 87542,
      "training performs": 123779,
      "performs comparable": 86990,
      "particular importance": 85421,
      "constraints constraints": 21593,
      "strong assumption": 113620,
      "rich contextual": 102734,
      "using selective": 130136,
      "positive given": 88322,
      "new representations": 79105,
      "combined appropriate": 18648,
      "using fixed": 129683,
      "model alleviates": 70657,
      "triggers experiments": 125793,
      "mtl effective": 75015,
      "investigate issue": 56773,
      "taskspecific layers": 119631,
      "layers separately": 61809,
      "magnitude fewer": 66382,
      "sota architecture": 110688,
      "previous solutions": 90463,
      "network documentlevel": 77230,
      "aggregating multiple": 4355,
      "inference information": 54153,
      "transformation applied": 124261,
      "entitylevel information": 38141,
      "hierarchical aggregation": 49925,
      "representations substantially": 99910,
      "data relations": 26330,
      "component human": 19982,
      "explosion web": 41670,
      "utilize data": 130501,
      "look existing": 65214,
      "extraction despite": 42302,
      "encode source": 36455,
      "classifying relation": 17674,
      "tasks targeting": 119548,
      "40 different": 1005,
      "architecture linguistic": 8686,
      "task bias": 117936,
      "results contrast": 101636,
      "improves considerably": 52962,
      "performance entity": 86342,
      "propagation propose": 92399,
      "agent cooperative": 4299,
      "group evaluates": 49138,
      "used adjust": 128382,
      "adjust training": 3874,
      "datasets verify": 27791,
      "instance proposed": 55606,
      "main motivations": 66444,
      "experiments result": 41118,
      "recognition dependency": 96855,
      "designed annotation": 29579,
      "phrases experimental": 87423,
      "syntax traditional": 116564,
      "traditional endtoend": 122814,
      "algorithm experiment": 4730,
      "2018 competition": 660,
      "won place": 132897,
      "inputs pretrained": 55490,
      "lm focus": 64863,
      "using internal": 129765,
      "internal attention": 56159,
      "features evidence": 43493,
      "representing information": 100050,
      "way semantic": 132129,
      "patterns extract": 85744,
      "relations deep": 98134,
      "learning second": 62987,
      "april 2021": 8477,
      "chapter book": 16410,
      "models consensus": 72961,
      "consensus reached": 21169,
      "entity span": 38103,
      "comprehensive search": 20269,
      "based rc": 11969,
      "baseline bert": 12196,
      "model ablation": 70505,
      "understanding type": 127001,
      "context entity": 22082,
      "heuristics entity": 49882,
      "mentions contribute": 68449,
      "analyses propose": 5461,
      "longtail relations": 65195,
      "learning handle": 62626,
      "model operates": 71611,
      "balance training": 11399,
      "objective predict": 81106,
      "relation sentence": 98070,
      "margin terms": 67202,
      "superior capability": 115679,
      "relations contrast": 98130,
      "representation form": 99242,
      "form table": 45132,
      "encoder designed": 36516,
      "classification study": 17422,
      "supervision sources": 115919,
      "weak labels": 132179,
      "source reliability": 110806,
      "model benchmarks": 70759,
      "sentiment topic": 106808,
      "methods consistently": 69397,
      "extraction language": 42366,
      "completion stateoftheart": 19794,
      "models readily": 73860,
      "readily used": 95969,
      "input corpus": 55314,
      "constrained inference": 21570,
      "model enable": 71066,
      "study novel": 114456,
      "based table": 12107,
      "table representation": 117258,
      "presented using": 89803,
      "complicated handcrafted": 19955,
      "models ace05": 72656,
      "extraction used": 42528,
      "classification mainly": 17261,
      "pretraining setup": 90323,
      "setup significantly": 107857,
      "spanish turkish": 111176,
      "relations wikipedia": 98281,
      "relations fewshot": 98175,
      "mechanisms learn": 68062,
      "word specified": 133583,
      "attention filter": 9838,
      "accuracy furthermore": 2167,
      "study verify": 114548,
      "sentencelevel counterpart": 106151,
      "occurs multiple": 81522,
      "multiple possible": 75642,
      "classification prior": 17340,
      "cdr gda": 15877,
      "extraction manual": 42385,
      "great research": 49020,
      "research interests": 100537,
      "tackle limitation": 117294,
      "labeled ones": 58462,
      "directly mapping": 31887,
      "use kinds": 128103,
      "entity reference": 38088,
      "approach entity": 7534,
      "relations recent": 98245,
      "performing multitask": 86960,
      "pipelined approach": 87562,
      "stateoftheart standard": 112970,
      "previous joint": 90418,
      "context finally": 22106,
      "requires pass": 100308,
      "difficulty identifying": 31702,
      "relations shared": 98255,
      "usually involve": 130429,
      "truth conditions": 125879,
      "extraction token": 42516,
      "extraction achieves": 42250,
      "datasets metalearning": 27563,
      "metalearning neural": 68555,
      "select subset": 104715,
      "training suffer": 123891,
      "guide selection": 49245,
      "propose metalearning": 92760,
      "metalearning based": 68548,
      "demonstrate reference": 28858,
      "comparing existing": 19503,
      "achieved certain": 2602,
      "certain success": 15956,
      "inevitably wrong": 54084,
      "problem build": 90953,
      "annotators label": 6495,
      "ranking lists": 95671,
      "previous seq2seq": 90460,
      "parallel decoding": 84663,
      "autoregressive approaches": 10858,
      "order proposed": 82392,
      "final set": 44126,
      "bipartite matching": 14224,
      "matching loss": 67409,
      "networks accurate": 77500,
      "entities experiments": 37781,
      "training code": 123385,
      "harm performance": 49502,
      "scientific technical": 104022,
      "russian paper": 103499,
      "scientific advances": 103991,
      "efficient processing": 35103,
      "actively used": 3124,
      "corpus scientific": 23980,
      "entities semantic": 37865,
      "produces stateoftheart": 92010,
      "articles build": 9175,
      "benchmarks english": 12901,
      "generated datasets": 47070,
      "appropriate relations": 8428,
      "high f1": 50072,
      "bag sentences": 11359,
      "usually represent": 130450,
      "supervision models": 115900,
      "loss effectively": 65261,
      "prediction relation": 89117,
      "set paper": 107521,
      "generality approach": 46747,
      "complex relation": 19871,
      "identify target": 51566,
      "traditional binary": 122803,
      "including supervised": 53386,
      "tasks variants": 119594,
      "applications practice": 6989,
      "words relation": 134166,
      "prediction recent": 89115,
      "bertbased sequence": 13279,
      "relationships tokens": 98336,
      "possible relationships": 88424,
      "supervision makes": 115896,
      "sparse noisy": 111230,
      "supplement training": 115939,
      "extraction relations": 42467,
      "attention capsule": 9803,
      "explicitly encourage": 41367,
      "encourage diversity": 36752,
      "conducted widely": 20959,
      "extraction empirical": 42320,
      "task vlsp": 118841,
      "vlsp 2020": 131858,
      "bert entity": 13119,
      "improved single": 52643,
      "used encode": 128508,
      "encode entity": 36428,
      "inference improve": 54150,
      "exploit dependency": 41409,
      "problem suffered": 91255,
      "mutual dependency": 75974,
      "category representations": 15791,
      "based design": 11638,
      "mechanism enhance": 67977,
      "bidirectional interaction": 13928,
      "annotations model": 6445,
      "trigger information": 125786,
      "module improves": 74494,
      "far perfect": 43096,
      "paper revisit": 84413,
      "improvements supervised": 52925,
      "sentences syntactically": 106514,
      "obtained distant": 81362,
      "nlg data": 79517,
      "searchbased approach": 104367,
      "extraction documents": 42311,
      "focus local": 44784,
      "gathers relevant": 46531,
      "suggest joint": 115315,
      "approach par": 7788,
      "based dual": 11664,
      "confirmed proposed": 21024,
      "models f1score": 73227,
      "techniques jointly": 119913,
      "performing entity": 86952,
      "approach adopt": 7338,
      "adopt joint": 3893,
      "techniques performance": 119949,
      "dataset survey": 27227,
      "field information": 43956,
      "recently advances": 96657,
      "difficult track": 31667,
      "performance publicly": 86639,
      "attentionbased approach": 10058,
      "sentence usually": 106122,
      "existing methodologies": 40174,
      "pattern extraction": 85714,
      "automatic knowledge": 10570,
      "thesis propose": 121857,
      "frameworks based": 45744,
      "hierarchical entity": 49961,
      "spoken language systems": 111993,
      "based wikipedia links": 12161,
      "dependency parsers present": 29165,
      "method language independent": 68928,
      "annotated test set": 6244,
      "test set present": 120514,
      "distant supervision using": 32481,
      "generate large amounts": 46962,
      "novel method detecting": 80631,
      "performance relation extraction": 86668,
      "distantly supervised data": 32490,
      "neural networks traditional": 78591,
      "slot filling address": 110038,
      "relation classification paper": 97951,
      "relation classification model": 97949,
      "models using sequence": 74306,
      "dependency parsing used": 29208,
      "neural relation extraction": 78648,
      "distantly supervised relation": 32492,
      "entities propose pathbased": 37848,
      "experimental results realworld": 40688,
      "results realworld datasets": 102116,
      "realworld datasets model": 96165,
      "performance stateoftheart open": 86747,
      "syntactically complex sentences": 116507,
      "extraction study problem": 42497,
      "textual relation embedding": 121729,
      "cooccurrence statistics textual": 23235,
      "statistics textual knowledge": 113207,
      "textual knowledge base": 121718,
      "knowledge base relations": 57789,
      "learned textual relation": 62266,
      "survey deep learning": 116168,
      "generated training data": 47142,
      "noise training data": 80071,
      "approach wide range": 8017,
      "approach consistently improves": 7448,
      "approach relation extraction": 7852,
      "extraction fundamental task": 42347,
      "overcome drawback propose": 83283,
      "performance attentive neural": 86156,
      "absolute fscore gain": 1729,
      "gain slot filling": 46353,
      "focuses important parts": 44902,
      "neural networks joint": 78495,
      "relation classification introduce": 97948,
      "linearchain conditional random": 64391,
      "training data unsupervised": 123560,
      "existing stateoftheart techniques": 40295,
      "task field natural": 118209,
      "nlp paper present": 79656,
      "learn relation representations": 62122,
      "model helps improve": 71279,
      "important task text": 52273,
      "models paper study": 73710,
      "data distant supervision": 25857,
      "distant supervision ds": 32468,
      "relation extraction text": 98031,
      "extraction task identify": 42504,
      "preliminary results using": 89279,
      "research dataset available": 100458,
      "shared task semantic": 108093,
      "classification task simple": 17437,
      "neural network entity": 78298,
      "task present novel": 118549,
      "provide new evidence": 93880,
      "reduction training time": 97464,
      "entities unstructured text": 37888,
      "based ensemble convolutional": 11682,
      "task provide detailed": 118584,
      "shared task approach": 108025,
      "models distantly supervised": 73094,
      "distant supervision dataset": 32467,
      "dataset relation extraction": 27150,
      "extensive experiments multiple": 41917,
      "multiple realworld datasets": 75655,
      "effectiveness proposed methods": 34935,
      "paper propose joint": 84255,
      "joint neural model": 57303,
      "manually extracted features": 67063,
      "crf conditional random": 24746,
      "extensive experimental setup": 41888,
      "outperforms previous neural": 82941,
      "dependency path embeddings": 29213,
      "natural language explanations": 76298,
      "classification work propose": 17504,
      "natural language explanation": 76297,
      "semantic parser converts": 105134,
      "recent stateoftheart approaches": 96518,
      "generative adversarial training": 47715,
      "extraction distant supervision": 42307,
      "sentence level paper": 105933,
      "rely distant supervision": 98689,
      "methods proposed method": 69694,
      "improvements compared stateoftheart": 52830,
      "representations neural relation": 99779,
      "investigate use different": 56823,
      "perform error analysis": 85992,
      "gain better understanding": 46337,
      "identification semantic relations": 51436,
      "knowledge bases automatically": 57799,
      "identify important semantic": 51508,
      "improve robustness neural": 52529,
      "extraction using graph": 42533,
      "standard benchmark model": 112206,
      "benchmark model shows": 12842,
      "experiments significant improvement": 41141,
      "propose novel multilevel": 92946,
      "datasets proposed framework": 27637,
      "using weak supervision": 130366,
      "introduce weakly supervised": 56568,
      "absence labeled data": 1718,
      "dependency trees improves": 29245,
      "extraction dependency trees": 42301,
      "capture longrange relations": 15340,
      "different tree structures": 31518,
      "input trees keeping": 55462,
      "resulting model achieves": 101454,
      "stateoftheart performance largescale": 112827,
      "detailed analysis model": 29755,
      "attention mechanism attends": 9883,
      "extraction task identifying": 42505,
      "joint inference framework": 57281,
      "chinese datasets proposed": 16756,
      "proposed framework effectively": 93288,
      "improvement relation extraction": 52755,
      "relevant information aliases": 98555,
      "usually ignore readily": 130424,
      "ignore readily available": 51671,
      "syntactic information text": 116423,
      "available extensive experiments": 10997,
      "words sentence respect": 134195,
      "method semantic relation": 69120,
      "bidirectional lstm networks": 13957,
      "semantic relations entity": 105210,
      "relations entity pairs": 98151,
      "entity pairs sentences": 37984,
      "processing nlp previous": 91749,
      "attention mechanisms fully": 9925,
      "issues propose novel": 57069,
      "classification task demonstrate": 17428,
      "features graph neural": 43531,
      "reasoning machine learning": 96265,
      "machine learning field": 65795,
      "improvements compared baselines": 52828,
      "perform qualitative analysis": 86052,
      "using external tools": 129663,
      "distant supervision approach": 32462,
      "outperforming number baselines": 82813,
      "purely neural model": 94417,
      "role labeling present": 103190,
      "labeling present simple": 58523,
      "role labeling recent": 103192,
      "stateoftheart performance achieved": 112806,
      "using neural models": 129955,
      "syntactic features partofspeech": 116405,
      "using external features": 129658,
      "simple bertbased model": 109377,
      "model achieve stateoftheart": 70541,
      "future research neural": 46306,
      "beam search applied": 12601,
      "language model entity": 59316,
      "tasks relation classification": 119450,
      "paper propose model": 84265,
      "propose model leverages": 92787,
      "distant supervision used": 32480,
      "examples easynegative examples": 39828,
      "methods typically rely": 69819,
      "feature extraction models": 43280,
      "deep language representations": 28243,
      "explicit linguistic features": 41331,
      "model longrange dependencies": 71474,
      "documentlevel relation extraction": 33160,
      "new dataset constructed": 78864,
      "multiple sentences document": 75690,
      "detailed analysis experiments": 29754,
      "experiments discuss multiple": 40913,
      "discuss multiple promising": 32271,
      "multiple promising directions": 75649,
      "recognizing diverse set": 97069,
      "dependency trees shown": 29248,
      "trees shown useful": 125713,
      "useful source information": 128933,
      "current deep learning": 25273,
      "dependency trees potential": 29247,
      "paper introduces novel": 84010,
      "control information flow": 22809,
      "benchmark datasets proposed": 12795,
      "relation extraction knowledge": 97996,
      "retrieval ir systems": 102401,
      "labels extensive experiments": 58603,
      "annotated data training": 6171,
      "data training systems": 26576,
      "address situation introduce": 3768,
      "facilitate information extraction": 42775,
      "extraction knowledge graph": 42365,
      "types relations including": 126349,
      "neural tagging model": 78698,
      "methods heavily depend": 69529,
      "deep learning structures": 28317,
      "build largescale highquality": 14780,
      "tagging model employs": 117408,
      "different topics different": 31503,
      "new multilingual dataset": 79041,
      "information entity recognition": 54539,
      "scale knowledge graphs": 103727,
      "knowledge graphs using": 57993,
      "paper propose fully": 84243,
      "propose fully automated": 92689,
      "relation extraction wide": 98034,
      "extraction ee task": 42316,
      "sequence labeling problems": 106981,
      "proposed framework compared": 93285,
      "framework compared stateoftheart": 45459,
      "graphbased neural models": 48905,
      "using multiinstance learning": 129910,
      "semantically similar sentences": 105378,
      "extensive experiments analysis": 41894,
      "unseen relations test": 127538,
      "relations test time": 98268,
      "extraction task propose": 42506,
      "pretrained bert encoder": 89988,
      "respectively indepth analysis": 101146,
      "code data released": 18088,
      "prior work typically": 90764,
      "text corpora human": 120827,
      "significantly improves current": 108946,
      "distant supervision exploits": 32471,
      "given target entity": 48145,
      "increasing attention recently": 53661,
      "apply pretrained language": 7205,
      "benefit various downstream": 12998,
      "information retrieval question": 54946,
      "neural networks slot": 78580,
      "networks slot filling": 77774,
      "slot filling slot": 110055,
      "filling slot filling": 44054,
      "neural networks large": 78501,
      "years seen rapid": 135298,
      "experiments english chinese": 40931,
      "relative position information": 98364,
      "selfattention mechanism capture": 104884,
      "capture rich contextual": 15371,
      "rich contextual dependencies": 102735,
      "novel approach improve": 80479,
      "approach improve performance": 7621,
      "contrast previous studies": 22704,
      "make use unlabeled": 66746,
      "consistent improvements compared": 21385,
      "neural network incorporate": 78307,
      "effective method learning": 34708,
      "problem jointly learning": 91096,
      "extraction propose novel": 42452,
      "stateoftheart sota results": 112963,
      "sota results tasks": 110697,
      "task competitive results": 117991,
      "order magnitude fewer": 82359,
      "sentence level document": 105928,
      "previous work paper": 90538,
      "work paper propose": 134666,
      "information different granularities": 54485,
      "demonstrate using bert": 28900,
      "using bert representations": 129389,
      "bert representations substantially": 13219,
      "encode source sentence": 36456,
      "makes difficult understand": 66774,
      "distant supervision based": 32464,
      "supervision based methods": 115871,
      "human annotation efforts": 50744,
      "unsupervised relation extraction": 127701,
      "named entities raw": 76034,
      "using named entities": 129932,
      "extraction task extracting": 42503,
      "persian language different": 87080,
      "entity recognition dependency": 38009,
      "phrases experimental results": 87424,
      "algorithm improves accuracy": 4754,
      "extraction challenging task": 42277,
      "help pretrained language": 49766,
      "model using internal": 72281,
      "language model focus": 59326,
      "relations scientific publications": 98251,
      "deep learning large": 28276,
      "relations deep learning": 98135,
      "entity span identification": 38104,
      "method automatically discover": 68654,
      "model ablation study": 70506,
      "improve performance models": 52470,
      "based analyses propose": 11504,
      "robustness neural models": 103115,
      "relation extraction recent": 98023,
      "knowledge work propose": 58240,
      "balance training data": 11400,
      "improves prior stateoftheart": 53032,
      "large margin terms": 61145,
      "superior capability handling": 115680,
      "information learning process": 54726,
      "representation learning process": 99309,
      "learning process experiments": 62916,
      "standard datasets model": 112224,
      "neural text classification": 78706,
      "text classification study": 120770,
      "neural text classifiers": 78707,
      "weak supervision sources": 132185,
      "soft attention mechanism": 110486,
      "language model predictions": 59371,
      "representations entity mentions": 99633,
      "complicated handcrafted features": 19956,
      "confirm proposed method": 21018,
      "training data human": 123474,
      "baseline model based": 12255,
      "multilingual pretraining setup": 75344,
      "relation classification english": 97947,
      "achieves comparable better": 2758,
      "study verify effectiveness": 114549,
      "classification prior work": 17341,
      "extraction manual annotation": 42386,
      "existing work focuses": 40337,
      "representations entities relations": 99631,
      "suffer problem exposure": 115237,
      "extraction achieves stateoftheart": 42251,
      "training data augmented": 123430,
      "achieved certain success": 2603,
      "problem build new": 90954,
      "based models require": 11853,
      "trained models available": 123203,
      "pretraining tasks experimental": 90337,
      "model capture useful": 70809,
      "capture useful information": 15415,
      "information extraction entity": 54569,
      "produces stateoftheart results": 92011,
      "work present methodology": 134681,
      "high quality training": 50124,
      "news articles build": 79290,
      "encoder trained corpus": 36583,
      "generates training data": 47189,
      "final model performance": 44111,
      "performance previous work": 86621,
      "methods usually represent": 69838,
      "prediction relation extraction": 89118,
      "problems closely related": 91306,
      "multiview learning framework": 75935,
      "makes possible automatically": 66799,
      "additional information sources": 3527,
      "sources knowledge bases": 110905,
      "multihead attention capsule": 75123,
      "explicitly encourage diversity": 41368,
      "experiments conducted widely": 40850,
      "conducted widely used": 20960,
      "extraction empirical study": 42321,
      "using pretrained bert": 130043,
      "task vlsp 2020": 118842,
      "ensemble model combines": 37605,
      "improved single models": 52644,
      "perform entity recognition": 85989,
      "propagation problem suffered": 92398,
      "performance makes model": 86521,
      "task release code": 118627,
      "remains key challenge": 98800,
      "annotated data data": 6158,
      "results suggest joint": 102231,
      "parameters training steps": 84793,
      "semantic relationships entities": 105222,
      "sentence previous studies": 106003,
      "previous studies focused": 90495,
      "multiple entities sentence": 75554,
      "mechanism proposed model": 68031,
      "field information extraction": 43957,
      "techniques deep neural": 119864,
      "embeddings deep neural": 35633,
      "future research present": 46308,
      "extraction using deep": 42532,
      "performance publicly available": 86640,
      "extraction documentlevel relation": 42309,
      "tasks including entity": 119177,
      "received attention nlp": 96359,
      "work propose attentionbased": 134714,
      "relation pair entities": 98059,
      "automatic knowledge base": 10571,
      "network model finding": 77328,
      "manually annotated test set": 67035,
      "benchmark datasets different domains": 12780,
      "distantly supervised relation extraction": 32493,
      "neural relation extraction model": 78649,
      "experimental results realworld datasets": 40689,
      "cooccurrence statistics textual knowledge": 23236,
      "statistics textual knowledge base": 113208,
      "textual knowledge base relations": 121719,
      "learned textual relation embedding": 62267,
      "models significantly improve performance": 74042,
      "extensive experimental results demonstrate": 41886,
      "linearchain conditional random field": 64392,
      "outperforms existing stateoftheart techniques": 82902,
      "task field natural language": 118210,
      "processing nlp paper present": 91747,
      "nlp paper present novel": 79657,
      "tasks knowledge base completion": 119236,
      "effectiveness neural network approaches": 34923,
      "multiple realworld datasets demonstrate": 75656,
      "demonstrate effectiveness proposed methods": 28722,
      "outperforms previous neural models": 82942,
      "achieve significant improvements compared": 2538,
      "significant improvements compared stateoftheart": 108789,
      "standard benchmark model shows": 112207,
      "resulting model achieves stateoftheart": 101455,
      "experimental results english chinese": 40615,
      "chinese datasets proposed framework": 16757,
      "datasets proposed framework effectively": 27638,
      "models usually ignore readily": 74309,
      "usually ignore readily available": 130425,
      "method semantic relation classification": 69121,
      "semantic relations entity pairs": 105211,
      "relations entity pairs sentences": 98152,
      "language processing nlp previous": 59913,
      "address issues propose novel": 3708,
      "relation classification task demonstrate": 97958,
      "classification task demonstrate model": 17429,
      "features graph neural networks": 43532,
      "semantic role labeling present": 105253,
      "role labeling present simple": 103191,
      "semantic role labeling recent": 105255,
      "model achieve stateoftheart performance": 70542,
      "pretrained language model entity": 90051,
      "paper propose model leverages": 84266,
      "pretrained bert language model": 89990,
      "experiments discuss multiple promising": 40914,
      "discuss multiple promising directions": 32272,
      "multiple promising directions future": 75650,
      "current deep learning models": 25274,
      "experiments benchmark datasets proposed": 40799,
      "benchmark datasets proposed model": 12797,
      "datasets proposed model outperforms": 27643,
      "information retrieval ir systems": 54936,
      "information extraction knowledge graph": 54576,
      "hybrid neural network model": 51190,
      "paper propose fully automated": 84244,
      "approach knowledge base completion": 7659,
      "benchmark demonstrate effectiveness proposed": 12813,
      "unseen relations test time": 127539,
      "large text corpora human": 61292,
      "approach significantly improves current": 7902,
      "extensive experimental results benchmark": 41885,
      "applications information retrieval question": 6944,
      "information retrieval question answering": 54947,
      "convolutional neural networks slot": 23208,
      "neural networks slot filling": 78581,
      "slot filling slot filling": 110056,
      "recent years seen rapid": 96634,
      "significant consistent improvements compared": 108744,
      "relation extraction propose novel": 98019,
      "propose novel neural architecture": 92952,
      "stateoftheart sota results tasks": 112964,
      "documentlevel relation extraction documentlevel": 33161,
      "sentence level document level": 105929,
      "previous work paper propose": 90539,
      "experimental results method achieves": 40638,
      "results method achieves stateoftheart": 101918,
      "demonstrate using bert representations": 28901,
      "using bert representations substantially": 129390,
      "representations neural relation extraction": 99780,
      "distant supervision based methods": 32465,
      "significantly outperforms stateoftheart entity": 109034,
      "named entities raw text": 76035,
      "essential natural language processing": 38560,
      "extraction challenging task requires": 42278,
      "capture different types information": 15292,
      "pretrained language models perform": 90090,
      "suffer problem exposure bias": 115238,
      "extraction achieves stateoftheart performance": 42252,
      "approach consistently improves performance": 7449,
      "problem build new dataset": 90955,
      "code trained models available": 18160,
      "pretraining tasks experimental results": 90338,
      "experiments method achieves new": 40999,
      "experiments conducted widely used": 40851,
      "dataset proposed model significantly": 27125,
      "error propagation problem suffered": 38320,
      "attention mechanism proposed model": 9911,
      "proposed model outperformed previous": 93457,
      "word embeddings deep neural": 133075,
      "performance publicly available datasets": 86641,
      "relation extraction documentlevel relation": 97990,
      "extraction documentlevel relation extraction": 42310,
      "cooccurrence statistics textual knowledge base": 23237,
      "statistics textual knowledge base relations": 113209,
      "using convolutional neural networks cnns": 129501,
      "task field natural language processing": 118211,
      "language processing nlp paper present": 59911,
      "multiple realworld datasets demonstrate effectiveness": 75657,
      "chinese datasets proposed framework effectively": 16758,
      "models usually ignore readily available": 74310,
      "extensive experiments benchmark datasets demonstrate": 41897,
      "semantic relations entity pairs sentences": 105212,
      "natural language processing nlp previous": 76458,
      "semantic role labeling present simple": 105254,
      "experiments discuss multiple promising directions": 40915,
      "discuss multiple promising directions future": 32273,
      "multiple promising directions future research": 75651,
      "extensive experiments benchmark datasets proposed": 41899,
      "experiments benchmark datasets proposed model": 40800,
      "benchmark datasets proposed model outperforms": 12798,
      "datasets proposed model outperforms existing": 27644,
      "proposed model outperforms existing methods": 93465,
      "applications information retrieval question answering": 6945,
      "convolutional neural networks slot filling": 23209,
      "achieves significant consistent improvements compared": 2855,
      "experimental results method achieves stateoftheart": 40639,
      "results method achieves stateoftheart performance": 101919,
      "demonstrate using bert representations substantially": 28902,
      "model significantly outperforms stateoftheart entity": 72035,
      "experiments method achieves new stateoftheart": 41000,
      "relation extraction documentlevel relation extraction": 97991,
      "aggression": 4373,
      "obscene": 81157,
      "racism": 95429,
      "requesting": 100105,
      "harassing": 49446,
      "alliance": 5064,
      "tech": 119740,
      "bullying": 14963,
      "damage": 25525,
      "unregistered": 127500,
      "positing": 88280,
      "provocative": 94150,
      "responders": 101184,
      "offensiveness": 81558,
      "cyberbullying": 25480,
      "commit": 18845,
      "evade": 38778,
      "researching": 100724,
      "25000": 823,
      "git": 47974,
      "countermeasures": 24433,
      "artists": 9276,
      "censorship": 15887,
      "obfuscating": 81045,
      "judiciary": 57454,
      "hostile": 50663,
      "destination": 29746,
      "dark": 25542,
      "timeseries": 122195,
      "ramifications": 95469,
      "reannotated": 96192,
      "innocuous": 55280,
      "0805": 77,
      "deliberate": 28605,
      "chicago": 16716,
      "colour": 18537,
      "pending": 85853,
      "956": 1425,
      "monitored": 74543,
      "realism": 96099,
      "732": 1232,
      "16k": 392,
      "olid": 81646,
      "tighter": 121968,
      "benevolent": 13029,
      "immigrants": 51849,
      "hateval": 49546,
      "migrants": 70075,
      "combating": 18544,
      "stakes": 112165,
      "incidental": 53215,
      "taylor": 119671,
      "swift": 116266,
      "0730": 63,
      "sixth": 109906,
      "commentsposts": 18832,
      "defending": 28472,
      "investing": 56870,
      "gay": 46559,
      "garbage": 46490,
      "polarization": 87934,
      "operationalization": 82062,
      "aae": 1459,
      "acc": 1866,
      "sns": 110319,
      "tn": 122232,
      "prompted": 92325,
      "stresses": 113584,
      "equalize": 38226,
      "equalizing": 38227,
      "hominem": 50610,
      "vegan": 131503,
      "subareas": 114648,
      "trusted": 125874,
      "pcl": 85828,
      "referral": 97569,
      "earned": 34400,
      "harms": 49510,
      "overestimating": 83315,
      "nearstateoftheart": 76741,
      "dedicate": 28178,
      "prejudice": 89260,
      "dialectic": 30531,
      "disturbing": 32780,
      "selfharm": 104921,
      "flagging": 44630,
      "reputation": 100100,
      "misogyny": 70314,
      "medias": 68189,
      "openais": 81951,
      "allied": 5065,
      "reannotate": 96191,
      "nonoffensive": 80259,
      "euphemisms": 38761,
      "pot": 88533,
      "household": 50687,
      "violates": 131734,
      "dice": 30848,
      "chrome": 16943,
      "entitybased": 38134,
      "results russian": 102142,
      "aggression detection": 4374,
      "study aimed": 114311,
      "studying ways": 114560,
      "messages used": 68520,
      "obtained english": 81367,
      "compare random": 19287,
      "classifier convolutional": 17533,
      "approach discourse": 7497,
      "neutral terms": 78758,
      "data augmenting": 25661,
      "general words": 46731,
      "remove bias": 98873,
      "dictionary created": 30875,
      "dutch text": 34283,
      "machines using": 66345,
      "dictionary obtained": 30885,
      "obtained fscore": 81375,
      "performance statistically": 86753,
      "dataset online": 27069,
      "existing public": 40260,
      "terms size": 120382,
      "dataset increase": 26980,
      "corresponding improvements": 24289,
      "online platforms": 81789,
      "learning analyze": 62331,
      "methodology english": 69255,
      "allows explore": 5154,
      "media conversations": 68095,
      "conversations social": 23068,
      "goal facilitating": 48352,
      "corpus suggest": 24026,
      "rate degree": 95791,
      "align opinion": 4904,
      "language key": 59228,
      "tend low": 120154,
      "classify messages": 17652,
      "particular terms": 85457,
      "used crowdsourced": 128467,
      "tweets categories": 126021,
      "speech offensive": 111717,
      "multiclass classifier": 75054,
      "classifier distinguish": 17540,
      "distinguish different": 32569,
      "close analysis": 17809,
      "tweets likely": 126036,
      "offensive tweets": 81557,
      "detection subtasks": 30069,
      "detection analysis": 29875,
      "actions taken": 3094,
      "detection subtask": 30068,
      "twostep classification": 126170,
      "language classifying": 58878,
      "onestep approach": 81721,
      "sexism racism": 107891,
      "dataset approximately": 26743,
      "news sports": 79370,
      "reliable solutions": 98621,
      "using weakly": 130367,
      "including corpus": 53279,
      "language detecting": 58950,
      "detecting online": 29845,
      "overlooked existing": 83346,
      "baseline f1": 12222,
      "media social": 68165,
      "rely annotated": 98675,
      "keywords classification": 57643,
      "meaning used": 67702,
      "collection data": 18469,
      "community detection": 19071,
      "determine possible": 30129,
      "word used": 133624,
      "detection contextual": 29908,
      "list keywords": 64693,
      "existing classification": 40087,
      "speech social": 111805,
      "detect hate": 29806,
      "lexical baselines": 63732,
      "task applying": 117885,
      "uses character": 129203,
      "ngrams word": 79470,
      "accuracy identifying": 2178,
      "identifying posts": 51610,
      "classes results": 17074,
      "challenge lies": 16049,
      "quality annotated": 94592,
      "corpus essential": 23778,
      "types type": 126380,
      "types using": 126384,
      "years increasing": 135268,
      "drawn significant": 34136,
      "developed automated": 30254,
      "classify textual": 17664,
      "characteristics types": 16542,
      "typical datasets": 126396,
      "dataset difficult": 26874,
      "performing method": 86956,
      "simple dictionarybased": 109398,
      "use examples": 128035,
      "examples study": 39882,
      "approach outline": 7761,
      "employ supervised": 36283,
      "classification set": 17401,
      "representations apply": 99502,
      "apply approaches": 7159,
      "stacked generalization": 112126,
      "task require": 118638,
      "majority cases": 66595,
      "protect individuals": 93658,
      "comments news": 18817,
      "news propose": 79358,
      "adaptation text": 3276,
      "real use": 96083,
      "cases outperforms": 15660,
      "experimental methodology": 40542,
      "web predicting": 132244,
      "sentiment recent": 106786,
      "need better": 76785,
      "usually try": 130463,
      "types events": 126283,
      "behavior analyze": 12646,
      "media different": 68107,
      "describes created": 29393,
      "tweets created": 126027,
      "posts number": 88520,
      "number topics": 80984,
      "build classifier": 14751,
      "reveals strongest": 102523,
      "dataset measuring": 27015,
      "kaggle competition": 57486,
      "recent data": 96442,
      "people willing": 85903,
      "perform annotation": 85945,
      "systems wellknown": 117232,
      "scores best": 104157,
      "spread social": 112051,
      "pressing problem": 89958,
      "models prior": 73800,
      "architecture important": 8668,
      "important type": 52284,
      "methods effective": 69461,
      "training does": 123590,
      "attacks using": 9718,
      "using wordlevel": 130388,
      "features comparative": 43410,
      "data extremely": 25940,
      "extremely difficult": 42597,
      "studied datasets": 114155,
      "conduct comparative": 20827,
      "various learning": 131118,
      "gru networks": 49201,
      "topic clustering": 122503,
      "neural characterbased": 77856,
      "abuse detection": 1841,
      "advent social": 4092,
      "media recent": 68160,
      "abusive content": 1843,
      "words users": 134294,
      "oov vocabulary": 81867,
      "detection datasets": 29919,
      "twitter wikipedia": 126132,
      "media post": 68152,
      "online expression": 81771,
      "content users": 21970,
      "users recent": 129165,
      "used collect": 128428,
      "corpora shows": 23585,
      "share certain": 107947,
      "qualitative study": 94577,
      "commonly defined": 18958,
      "usergenerated web": 129079,
      "steadily increasing": 113223,
      "years online": 135286,
      "sentences manually": 106394,
      "allows annotators": 5126,
      "labelling paper": 58571,
      "method partially": 69040,
      "process consists": 91443,
      "consists steps": 21496,
      "method data": 68756,
      "discussion platform": 32324,
      "field recently": 43976,
      "directions research": 31853,
      "end compare": 36797,
      "comment dataset": 18798,
      "propose ensemble": 92658,
      "enable perform": 36363,
      "dataset labels": 26993,
      "ngram tfidf": 79455,
      "todays world": 122238,
      "exponential increase": 41677,
      "use internet": 128099,
      "classify tweets": 17666,
      "tweets twitter": 126065,
      "tfidf values": 121763,
      "multiple machine": 75612,
      "media platform": 68141,
      "language constructs": 58915,
      "contextual attention": 22445,
      "corpus complete": 23704,
      "large online": 61195,
      "new perspectives": 79075,
      "previously estimated": 90597,
      "designed language": 29605,
      "data chinese": 25728,
      "language context": 58919,
      "approaches requires": 8327,
      "requires identify": 100277,
      "identify understand": 51571,
      "contextual type": 22506,
      "linguistic meaning": 64507,
      "extensive linguistic": 41942,
      "analysis statistical": 5874,
      "codemixed hindienglish": 18174,
      "hindienglish tweets": 50493,
      "increment stateoftheart": 53720,
      "tweets compare": 126024,
      "improved representation": 52635,
      "detecting offensive": 29844,
      "offensive content": 81541,
      "using stage": 130209,
      "content detection": 21870,
      "importance work": 52094,
      "including using": 53404,
      "sentences online": 106418,
      "flexible allows": 44654,
      "models detect": 73056,
      "detect implicit": 29809,
      "annotations difficult": 6425,
      "competition results": 19626,
      "trained public": 123251,
      "multiple classes": 75516,
      "providing explicit": 94114,
      "explicit labels": 41329,
      "versus models": 131631,
      "online comments": 81746,
      "problem implementing": 91079,
      "new ways": 79247,
      "detection tool": 30083,
      "adding sentiment": 3390,
      "user trying": 129049,
      "set common": 107393,
      "rely entirely": 98690,
      "semantic cues": 105026,
      "target offensive": 117683,
      "offensive posts": 81555,
      "media offensive": 68134,
      "messages previous": 68511,
      "focused detecting": 44852,
      "detecting specific": 29851,
      "target different": 117601,
      "dataset olid": 27068,
      "discuss main": 32264,
      "classification sexism": 17403,
      "result categories": 101363,
      "preliminary work": 89283,
      "categories sexism": 15753,
      "problem time": 91264,
      "technique task": 119817,
      "capture shallow": 15394,
      "captures structure": 15456,
      "revisit problem": 102652,
      "media approach": 68079,
      "multilingual detection": 75238,
      "detection hate": 29960,
      "speech immigrants": 111693,
      "immigrants women": 51850,
      "women twitter": 132895,
      "twitter annotated": 126094,
      "annotated respect": 6225,
      "content target": 21957,
      "performance ranking": 86652,
      "datasets facebook": 27482,
      "comment threads": 18802,
      "interannotator agreements": 56048,
      "presence offensive": 89353,
      "major concern": 66548,
      "content created": 21866,
      "various social": 131199,
      "work english": 134491,
      "detection offensive": 30008,
      "f1score 070": 42705,
      "performing english": 86950,
      "language captures": 58867,
      "survey automated": 116162,
      "consequently past": 21183,
      "years substantial": 135305,
      "proposed date": 93253,
      "highlight challenges": 50257,
      "dialogue safety": 30737,
      "robustness adversarial": 103089,
      "adversarial human": 4126,
      "public forums": 94260,
      "behavior humans": 12660,
      "develop training": 30246,
      "humans models": 51093,
      "approach considerably": 7444,
      "work newly": 134655,
      "analysis typically": 5917,
      "multilingual multitask": 75298,
      "various classification": 131058,
      "bias high": 13809,
      "application developed": 6842,
      "stop words": 113373,
      "annotations enable": 6428,
      "bias study": 13845,
      "guarantee good": 49215,
      "detect abusive": 29794,
      "text poses": 121184,
      "word token": 133604,
      "token based": 122247,
      "character subword": 16470,
      "analyze effectiveness": 5968,
      "studying wikipedia": 114561,
      "sensitivity analysis": 105753,
      "techniques leverage": 119921,
      "models susceptible": 74140,
      "susceptible learning": 116218,
      "time addition": 121979,
      "purpose sentiment": 94439,
      "having sentiment": 49567,
      "biases related": 13883,
      "applied online": 7106,
      "network sites": 77436,
      "facebook twitter": 42744,
      "places people": 87616,
      "opinions sentiments": 82132,
      "different opinions": 31310,
      "model got": 71261,
      "shared online": 107997,
      "share personal": 107959,
      "personal experiences": 87119,
      "experiences collected": 40447,
      "stories extraction": 113399,
      "extraction key": 42361,
      "labels dimensions": 58590,
      "extract key": 42092,
      "present description": 89440,
      "implement deep": 51929,
      "ranking 5th": 95659,
      "content classification": 21863,
      "detection aims": 29872,
      "recall stateoftheart": 96344,
      "50 cases": 1071,
      "training synthetically": 123898,
      "uses characterlevel": 129204,
      "methods outline": 69653,
      "detection comparative": 29902,
      "identify offensive": 51527,
      "focuses detecting": 44895,
      "categorizing offensive": 15776,
      "media offenseval": 68133,
      "detection explore": 29948,
      "detection experiments": 29947,
      "evaluation observed": 39313,
      "measures used": 67901,
      "like facebook": 64041,
      "online user": 81813,
      "networks showed": 77764,
      "comments social": 18826,
      "accuracy traditional": 2304,
      "respectively compare": 101131,
      "capable predicting": 15212,
      "finally comparison": 44156,
      "bias hate": 13807,
      "demographic attributes": 28647,
      "attributes work": 10205,
      "demographic factors": 28648,
      "age country": 4282,
      "italian polish": 57098,
      "classifiers evaluate": 17609,
      "targeted social": 117772,
      "framework analyzing": 45423,
      "processes linguistic": 91607,
      "variation change": 130888,
      "ability analyze": 1484,
      "scale implications": 103725,
      "detecting understanding": 29856,
      "media bias": 68084,
      "process inappropriate": 91509,
      "proposed purpose": 93531,
      "true approaches": 125846,
      "derived structure": 29364,
      "corpora focus": 23489,
      "detection especially": 29941,
      "propose addition": 92529,
      "addition corpus": 3416,
      "scientific works": 104027,
      "communities social": 19063,
      "researchers investigating": 100696,
      "developing systems": 30362,
      "topic far": 122518,
      "dataset offensive": 27064,
      "offensive offensive": 81554,
      "recommendations practitioners": 97094,
      "data arabic": 25617,
      "arabic offensive": 8539,
      "offensive tweet": 81556,
      "building dataset": 14833,
      "arabic dataset": 8502,
      "thoroughly analyze": 121900,
      "dataset determine": 26868,
      "tweets arabic": 126019,
      "speakers use": 111330,
      "results f1": 101789,
      "using sota": 130198,
      "newly annotated": 79264,
      "learning automated": 62376,
      "challenging learning": 16270,
      "instead endtoend": 55658,
      "models crucial": 73006,
      "semisupervised dataset": 105598,
      "media led": 68124,
      "identification provides": 51426,
      "messages limited": 68510,
      "using keywords": 129774,
      "keywords work": 57651,
      "principled manner": 90681,
      "set different": 107417,
      "lower levels": 65440,
      "multimodal dataset": 75423,
      "images text": 51812,
      "study online": 114459,
      "make freely": 66676,
      "apache 20": 6766,
      "corpus italian": 23848,
      "increases probability": 53653,
      "message boards": 68495,
      "preprocessing phase": 89330,
      "aims investigating": 4545,
      "informal written": 54348,
      "written using": 135157,
      "using arabic": 129344,
      "complex preprocessing": 19859,
      "arabic corpora": 8498,
      "subtask offensive": 114948,
      "providing stateoftheart": 94135,
      "accuracy recall": 2258,
      "african american": 4273,
      "english aae": 37062,
      "positive rate": 88339,
      "training mitigate": 123709,
      "able substantially": 1691,
      "investigating deep": 56854,
      "growth internet": 49191,
      "led increase": 63278,
      "needs addressed": 76886,
      "approaches utilizing": 8398,
      "various embeddings": 131089,
      "detecting various": 29859,
      "media detecting": 68105,
      "data presence": 26264,
      "presence various": 89356,
      "detection does": 29925,
      "models published": 73845,
      "investigate assumption": 56720,
      "conditioning context": 20809,
      "labeled posts": 58463,
      "annotators provided": 6498,
      "context surprisingly": 22286,
      "points need": 87865,
      "comments annotated": 18809,
      "annotated context": 6142,
      "available leveraging": 11031,
      "work report": 134776,
      "task organized": 118481,
      "focus developing": 44751,
      "utility training": 130492,
      "vlsp campaign": 131859,
      "facebook task": 42743,
      "build classification": 14750,
      "hate offensive": 49529,
      "attracted large": 10152,
      "number participants": 80938,
      "participants popular": 85323,
      "teams signed": 119733,
      "task 14": 117817,
      "model intent": 71355,
      "language templates": 60156,
      "converge stable": 22919,
      "labelling methodology": 58568,
      "applied linguistic": 7089,
      "better understood": 13759,
      "text authors": 120674,
      "gender study": 46616,
      "study examines": 114374,
      "experiments personality": 41058,
      "author characteristics": 10360,
      "70 cases": 1210,
      "research author": 100422,
      "word list": 133352,
      "list generation": 64692,
      "performance future": 86399,
      "texts crucial": 121489,
      "regular classification": 97722,
      "models ignoring": 73352,
      "work target": 134844,
      "normalization tn": 80351,
      "analysis conducted": 5541,
      "language effectiveness": 58993,
      "seemingly innocuous": 104526,
      "empirically assess": 36217,
      "adaptive pretraining": 3337,
      "2019 significant": 712,
      "stresses need": 113585,
      "vietnamese text": 131696,
      "datasets affect": 27303,
      "data user": 26605,
      "skewed label": 109971,
      "augmentation increases": 10281,
      "new synthetic": 79190,
      "dataset help": 26961,
      "classification fully": 17214,
      "techniques impact": 119904,
      "performance techniques": 86790,
      "classifiers performed": 17625,
      "different constraints": 31056,
      "years previous": 135289,
      "domain like": 33571,
      "new generic": 78937,
      "represent target": 99146,
      "target content": 117591,
      "language quantitative": 60013,
      "having strong": 49570,
      "strong focus": 113677,
      "detection role": 30039,
      "respectively outperforming": 101154,
      "outperforming baselines": 82801,
      "dataset reddit": 27145,
      "offensive abusive": 81540,
      "comparison general": 19546,
      "detection tasks": 30079,
      "battery experiments": 12564,
      "detection evaluation": 29944,
      "dependence specific": 29058,
      "effects social": 35003,
      "similar meaning": 109110,
      "score calculated": 104057,
      "model behaviors": 70756,
      "bias proposed": 13838,
      "model preserving": 71770,
      "preserving overall": 89937,
      "ad hominem": 3162,
      "systems respond": 117119,
      "directly user": 31917,
      "input important": 55349,
      "important study": 52259,
      "classifier analyze": 17521,
      "analyze human": 5976,
      "techniques reduce": 119965,
      "efforts nlp": 35203,
      "research addressing": 100398,
      "process online": 91542,
      "building accurate": 14820,
      "identify categorize": 51483,
      "language general": 59061,
      "furthermore believe": 46149,
      "presents interesting": 89856,
      "black white": 14248,
      "academic community": 1856,
      "study analyse": 114315,
      "effective metric": 34710,
      "model prove": 71818,
      "prove competitive": 93696,
      "comparable superior": 19184,
      "women immigrants": 132893,
      "score using": 104139,
      "tweets provide": 126052,
      "increased rapidly": 53632,
      "sets findings": 107673,
      "problem severe": 91226,
      "labelled datasets": 58560,
      "final annotation": 44095,
      "embedding word2vec": 35530,
      "core contribution": 23319,
      "model weak": 72312,
      "performance increasingly": 86459,
      "gaps biases": 46487,
      "exponential rise": 41679,
      "content identification": 21890,
      "content used": 21968,
      "observed pretrained": 81228,
      "multilingualbert model": 75407,
      "challenge development": 16023,
      "experiments establish": 40938,
      "correction method": 24158,
      "remove existing": 98874,
      "enabled rapid": 36370,
      "language conjunction": 58909,
      "multiple classification": 75517,
      "testing results": 120601,
      "classifiers achieved": 17597,
      "prevalent online": 90359,
      "language varies": 60323,
      "different platforms": 31335,
      "combine existing": 18625,
      "improve detection": 52371,
      "development online": 30409,
      "dataset seen": 27180,
      "apply principles": 7206,
      "compare effects": 19243,
      "main model": 66443,
      "study study": 114525,
      "investigates effects": 56851,
      "learning performance": 62883,
      "detection review": 30038,
      "group individual": 49139,
      "models generalise": 73270,
      "struggle generalise": 114124,
      "research improve": 100532,
      "bias nlp": 13830,
      "models pick": 73744,
      "undesirable biases": 127041,
      "difficult completely": 31614,
      "demonstrate surprising": 28886,
      "refer approach": 97495,
      "biased text": 13861,
      "approach important": 7617,
      "models scoring": 73979,
      "analysis regarded": 5799,
      "according human": 1980,
      "learning detecting": 62496,
      "essays essay": 38534,
      "identify cases": 51482,
      "support human": 115972,
      "built recent": 14937,
      "linguistics particularly": 64621,
      "labeling dataset": 58493,
      "speaker different": 111286,
      "arabic levantine": 8522,
      "systems assist": 116738,
      "annotations proposed": 6455,
      "systems multitask": 117009,
      "results indicated": 101860,
      "users different": 129109,
      "social medias": 110437,
      "creation process": 24722,
      "process annotating": 91420,
      "finally evaluated": 44180,
      "accuracy 55": 2067,
      "text type": 121380,
      "detection exponential": 29949,
      "languages individual": 60643,
      "contents filter": 21989,
      "performed benchmark": 86910,
      "datasets widely": 27797,
      "tweets manually": 126037,
      "message level": 68496,
      "content message": 21905,
      "judged based": 57430,
      "given messages": 48062,
      "portuguese paper": 88200,
      "agreement corpus": 4388,
      "detection compared": 29904,
      "content moderation": 21909,
      "ordinary word": 82451,
      "generation social": 47625,
      "manually expensive": 67061,
      "meaning message": 67648,
      "demonstrate unsupervised": 28894,
      "analyzing words": 6059,
      "art uses": 9100,
      "contextfree word": 22367,
      "embeddings algorithm": 35548,
      "higher detection": 50176,
      "detection accuracies": 29865,
      "dice loss": 30849,
      "identify spans": 51558,
      "spans given": 111203,
      "distribution paper": 32671,
      "issue natural": 57008,
      "manually labeling": 67070,
      "different distributions": 31101,
      "reduced model": 97372,
      "chrome extension": 16944,
      "identify report": 51547,
      "span detection": 111120,
      "outputs paper": 83177,
      "test multiple": 120477,
      "report best": 98981,
      "approach ensemble": 7532,
      "evaluation phase": 39323,
      "media online": 68135,
      "increasingly rely": 53711,
      "order mitigate": 82371,
      "content relevant": 21933,
      "far focused": 43081,
      "method robustly": 69111,
      "detect lexical": 29812,
      "lexical biases": 63733,
      "identifies salient": 51471,
      "reflect human": 97606,
      "understand biases": 126738,
      "detection named": 29996,
      "highest f1score": 50226,
      "corpus russian language": 23979,
      "corpus analysis shows": 23657,
      "study machine learning": 114432,
      "detection using machine": 30096,
      "classifier convolutional neural": 17534,
      "test set consisting": 120500,
      "performance statistically significant": 86754,
      "previous work applied": 90518,
      "based text classification": 12114,
      "existing public datasets": 40261,
      "creating new dataset": 24705,
      "detection paper introduce": 30014,
      "machine learning analyze": 65771,
      "apply methodology english": 7185,
      "methodology english wikipedia": 69256,
      "social media conversations": 110364,
      "conversations social media": 23069,
      "users social media": 129171,
      "language key challenge": 59229,
      "hate speech offensive": 49540,
      "language detection twitter": 58956,
      "online social media": 81804,
      "approach shows promising": 7896,
      "propose approach detecting": 92550,
      "expensive annotation process": 40414,
      "using weakly supervised": 130368,
      "largely overlooked existing": 61346,
      "incorporate context information": 53458,
      "strong baseline f1": 113628,
      "baseline f1 score": 12223,
      "words social media": 134227,
      "hate speech social": 49542,
      "speech social media": 111806,
      "detect hate speech": 29807,
      "recent years increasing": 96617,
      "propose deep neural": 92620,
      "best performing method": 13402,
      "classification paper considers": 17320,
      "social media different": 110374,
      "models prior work": 73801,
      "prior work perform": 90760,
      "previously studied datasets": 90624,
      "train deep learning": 122920,
      "conduct comparative study": 20828,
      "advent social media": 4093,
      "social media recent": 110417,
      "social media post": 110412,
      "classification models dataset": 17279,
      "different deep learning": 31083,
      "future research challenges": 46295,
      "frequency tfidf values": 45849,
      "learning models perform": 62788,
      "comparative analysis models": 19202,
      "social media platform": 110402,
      "makes task challenging": 66810,
      "art results multiple": 9087,
      "detection data sets": 29917,
      "codemixed hindienglish tweets": 18175,
      "tweets using deep": 126073,
      "learning models paper": 62787,
      "speech detection englishhindi": 111672,
      "online discussion forums": 81769,
      "datasets achieve average": 27293,
      "adding sentiment information": 3391,
      "methods significantly outperform": 69753,
      "qualitative analysis model": 94553,
      "target offensive posts": 117685,
      "social media offensive": 110397,
      "identification dataset olid": 51378,
      "task existing approaches": 118166,
      "paper revisit problem": 84414,
      "problem automatically identifying": 90949,
      "social media approach": 110349,
      "approach task using": 7963,
      "multilingual detection hate": 75239,
      "detection hate speech": 29961,
      "hate speech immigrants": 49537,
      "speech immigrants women": 111694,
      "immigrants women twitter": 51851,
      "shared task achieved": 108021,
      "competitive performance ranking": 19665,
      "english paper present": 37234,
      "presence offensive language": 89354,
      "offensive language social": 81550,
      "various social media": 131200,
      "target offensive language": 117684,
      "language detection offensive": 58954,
      "detection offensive language": 30009,
      "offensive language english": 81544,
      "achieves macro averaged": 2816,
      "classification tasks paper": 17452,
      "multitask learning approaches": 75824,
      "need new evaluation": 76839,
      "new evaluation methods": 78908,
      "attention mechanism called": 9888,
      "offensive language detection": 81543,
      "word different words": 132993,
      "language models demonstrate": 59535,
      "data build models": 25712,
      "models susceptible learning": 74141,
      "based idea propose": 11764,
      "social network sites": 110440,
      "vietnamese language speech": 131680,
      "language speech processing": 60116,
      "annotated data set": 6167,
      "media text using": 68172,
      "computational linguistics paper": 20397,
      "learning robust representations": 62979,
      "detection comparative analysis": 29903,
      "identify offensive content": 51528,
      "work focuses detecting": 134538,
      "semeval2019 task identifying": 105544,
      "identifying categorizing offensive": 51586,
      "categorizing offensive language": 15777,
      "social media offenseval": 110396,
      "like facebook twitter": 64042,
      "compare traditional machine": 19306,
      "models capable predicting": 72876,
      "traditional methods deep": 122837,
      "bias hate speech": 13808,
      "italian polish portuguese": 57099,
      "polish portuguese spanish": 87983,
      "information derived structure": 54474,
      "context large corpus": 22160,
      "online communities social": 81749,
      "communities social media": 19064,
      "dataset offensive language": 27065,
      "evaluate computational models": 38812,
      "arabic offensive language": 8540,
      "method building dataset": 68682,
      "social media led": 110389,
      "language identification provides": 59139,
      "test set different": 120504,
      "media platforms like": 68144,
      "make freely available": 66677,
      "apache 20 license": 6767,
      "shared task offensive": 108072,
      "study aims investigating": 114314,
      "classification arabic text": 17122,
      "text arabic language": 120657,
      "language used social": 60304,
      "subtask offensive language": 114949,
      "providing stateoftheart performance": 94136,
      "false positive rate": 43043,
      "adversarial training mitigate": 4178,
      "social media detecting": 110373,
      "make code data": 66633,
      "publicly available leveraging": 94310,
      "shared task organized": 108074,
      "automatic data augmentation": 10508,
      "best models significantly": 13381,
      "vanilla bert model": 130813,
      "multiclass classification task": 75052,
      "classification task based": 17426,
      "attracted large number": 10153,
      "large number participants": 61185,
      "number participants popular": 80939,
      "characteristics language use": 16529,
      "research author profiling": 100423,
      "existing works treat": 40342,
      "text normalization tn": 121150,
      "datasets model outperforms": 27575,
      "text generation algorithms": 120977,
      "large corpus english": 61063,
      "al 2019 significant": 4646,
      "better data selection": 13559,
      "vietnamese text classification": 131697,
      "pretrained transformer network": 90199,
      "recent years previous": 96630,
      "domain paper introduce": 33606,
      "detection english paper": 29935,
      "speech detection tasks": 111674,
      "directly user input": 31918,
      "generated dialogue responses": 47072,
      "introduce new annotated": 56472,
      "challenge nlp community": 16070,
      "nlp models language": 79641,
      "data set 1000": 26398,
      "model improve model": 71307,
      "comparable superior performance": 19185,
      "superior performance monolingual": 115689,
      "performance monolingual models": 86544,
      "datasets paper presents": 27611,
      "word embedding word2vec": 133044,
      "dataset facilitate future": 26925,
      "future research opportunities": 46307,
      "test data using": 120440,
      "systematic gaps biases": 116676,
      "bert multilingual bert": 13185,
      "exponential rise social": 41680,
      "speech offensive content": 111718,
      "2020 shared tasks": 732,
      "training data transfer": 123556,
      "learning approach arabic": 62339,
      "processing nlp field": 91734,
      "model results demonstrate": 71924,
      "transfer learning performance": 124120,
      "future research improve": 46302,
      "does require training": 33398,
      "transformerbased models terms": 124427,
      "datasets paper introduce": 27610,
      "stateoftheart results languages": 112906,
      "solve problem introduce": 110604,
      "data creation process": 25810,
      "detection exponential rise": 29950,
      "large data set": 61069,
      "performed benchmark datasets": 86911,
      "benchmark datasets shows": 12804,
      "stateoftheart performance respect": 112842,
      "datasets widely used": 27798,
      "annotated according different": 6137,
      "results baseline experiments": 101538,
      "generation social media": 47626,
      "state art uses": 112482,
      "word embeddings algorithm": 133048,
      "higher detection accuracies": 50177,
      "task semisupervised learning": 118686,
      "class distribution paper": 17034,
      "various pretrained transformer": 131175,
      "issue natural language": 57009,
      "outputs paper presents": 83178,
      "ensemble models best": 37610,
      "social media online": 110398,
      "detect abusive content": 29795,
      "biases present training": 13881,
      "training data resulting": 123527,
      "model demonstrate method": 70965,
      "better understand biases": 13756,
      "detection named entity": 29997,
      "detection using machine learning": 30097,
      "using machine learning neural": 129848,
      "classifier convolutional neural network": 17535,
      "model trained large corpus": 72198,
      "apply methodology english wikipedia": 7186,
      "using manually annotated data": 129859,
      "hate speech social media": 49543,
      "propose deep neural network": 92621,
      "document frequency tfidf values": 33014,
      "perform comparative analysis models": 85965,
      "state art results multiple": 112471,
      "using deep learning models": 129544,
      "deep learning models paper": 28291,
      "hate speech detection englishhindi": 49534,
      "offensive language identification dataset": 81546,
      "language identification dataset olid": 59130,
      "multilingual detection hate speech": 75240,
      "detection hate speech immigrants": 29963,
      "hate speech immigrants women": 49538,
      "speech immigrants women twitter": 111695,
      "achieved competitive performance ranking": 2611,
      "offensive language social media": 81551,
      "language detection offensive language": 58955,
      "new attention mechanism called": 78806,
      "finetuning large pretrained language": 44473,
      "pretrained language models demonstrate": 90070,
      "statistical natural language processing": 113140,
      "vietnamese language speech processing": 131681,
      "social media text using": 110425,
      "semeval2019 task identifying categorizing": 105545,
      "task identifying categorizing offensive": 118266,
      "identifying categorizing offensive language": 51587,
      "categorizing offensive language social": 15778,
      "language social media offenseval": 60102,
      "approach offensive language detection": 7751,
      "compare traditional machine learning": 19307,
      "models deep neural models": 73025,
      "italian polish portuguese spanish": 57100,
      "online communities social media": 81750,
      "dataset offensive language identification": 27066,
      "social media platforms like": 110405,
      "shared task offensive language": 108073,
      "language used social media": 60305,
      "detection hate speech detection": 29962,
      "experiments publicly available datasets": 41099,
      "make code data publicly": 66634,
      "attracted large number participants": 10154,
      "large number participants popular": 61186,
      "et al 2019 significant": 38734,
      "hate speech detection tasks": 49536,
      "introduce new annotated dataset": 56473,
      "model improve model performance": 71308,
      "paper presents new dataset": 84186,
      "dataset facilitate future research": 26926,
      "facilitate future research opportunities": 42771,
      "bert multilingual bert models": 13187,
      "exponential rise social media": 41681,
      "hate speech offensive content": 49541,
      "pretrained word embeddings glove": 90219,
      "arabic offensive language detection": 8541,
      "language processing nlp field": 59899,
      "issue natural language processing": 57010,
      "downstream tasks work present": 34062,
      "biases present training data": 13882,
      "detection named entity recognition": 29998,
      "frequencyinverse document frequency tfidf values": 45858,
      "offensive language identification dataset olid": 81547,
      "multilingual detection hate speech immigrants": 75241,
      "detection hate speech immigrants women": 29964,
      "hate speech immigrants women twitter": 49539,
      "apply natural language processing techniques": 7195,
      "finetuning large pretrained language models": 44474,
      "statistical natural language processing nlp": 113141,
      "semeval2019 task identifying categorizing offensive": 105546,
      "task identifying categorizing offensive language": 118267,
      "identifying categorizing offensive language social": 51588,
      "categorizing offensive language social media": 15779,
      "offensive language social media offenseval": 81552,
      "make code data publicly available": 66635,
      "attracted large number participants popular": 10155,
      "natural language processing nlp field": 76446,
      "denotes": 29008,
      "handwriting": 49428,
      "coalesce": 18049,
      "numbering": 80998,
      "emulates": 36345,
      "riddled": 102821,
      "metrical": 69914,
      "verse": 131601,
      "poet": 87786,
      "xerox": 135189,
      "font": 44997,
      "quechua": 94908,
      "948": 1417,
      "anova": 6514,
      "7477": 1245,
      "24613": 808,
      "divisions": 32921,
      "tokenbased": 122272,
      "stemmers": 113235,
      "recourse": 97139,
      "adverb": 4096,
      "complications": 19960,
      "disabilities": 31928,
      "masculine": 67281,
      "saudi": 103662,
      "quran": 95419,
      "1115": 236,
      "gulf": 49274,
      "subdialect": 114661,
      "entitled": 37900,
      "dating": 27836,
      "bengal": 13030,
      "district": 32779,
      "foot": 45004,
      "suprasegmental": 116049,
      "intonational": 56332,
      "webservice": 132290,
      "katakana": 57496,
      "arabizi": 8574,
      "remotely": 98868,
      "periodization": 87041,
      "spellchecker": 111896,
      "invalid": 56675,
      "587": 1132,
      "maghrebi": 66373,
      "restoring": 101333,
      "k12": 57484,
      "aesthetics": 4209,
      "fictional": 43927,
      "rejected": 97835,
      "licence": 63968,
      "ict": 51313,
      "keyboards": 57616,
      "dcs": 27850,
      "august": 10352,
      "assists": 9590,
      "covert": 24559,
      "juxtapose": 57483,
      "periodicals": 87040,
      "print": 90686,
      "phonogram": 87306,
      "outofvocab": 82671,
      "uzbek": 130684,
      "cyrillic": 25486,
      "avoidance": 11244,
      "962": 1431,
      "adposition": 3915,
      "snacs": 110299,
      "schneider": 103954,
      "adpositions": 3921,
      "adpositional": 3920,
      "presentday": 89773,
      "bibliographic": 13895,
      "algorithm applies": 4696,
      "words prior": 134139,
      "match results": 67363,
      "input methods": 55366,
      "effective text": 34759,
      "using double": 129608,
      "entirely new": 37734,
      "features texts": 43759,
      "collected processed": 18432,
      "study specifically": 114523,
      "specifically different": 111543,
      "produced standard": 91968,
      "new computational": 78833,
      "automated language": 10448,
      "increasingly available": 53693,
      "rules grammar": 103407,
      "method handwritten": 68869,
      "character text": 16471,
      "works work": 134994,
      "explores application": 41643,
      "algorithm compared": 4706,
      "taking care": 117544,
      "reducing target": 97434,
      "unlike languages": 127436,
      "unique features": 127187,
      "language doesnt": 58983,
      "persian writing": 87088,
      "writing arabic": 135087,
      "additional character": 3498,
      "problem overcome": 91154,
      "world natural": 135038,
      "processing discussed": 91664,
      "morphological database": 74695,
      "language arabic": 58842,
      "scheme represent": 103937,
      "exploited nlp": 41453,
      "scripts used": 104273,
      "script paper": 104258,
      "resolution strategies": 100786,
      "applied arabic": 7046,
      "concepts semantic": 20643,
      "increases rate": 53654,
      "annotation interpretation": 6326,
      "context speech": 22277,
      "resourcelight approach": 100903,
      "learning verb": 63160,
      "resources goal": 100984,
      "language unannotated": 60200,
      "realization automatic": 96113,
      "detection arabic": 29882,
      "especially arabic": 38433,
      "implementation automatic": 51939,
      "properties arabic": 92436,
      "difficult language": 31639,
      "language properties": 59996,
      "agglutinative nature": 4341,
      "generation does": 47375,
      "focused arabic": 44846,
      "algorithm suitable": 4803,
      "maximum accuracy": 67529,
      "achieved compared": 2607,
      "keyphrase extraction": 57618,
      "extracting keyphrases": 42215,
      "enhance efficiency": 37467,
      "annotated arabic": 6139,
      "features document": 43461,
      "candidate keyphrases": 15107,
      "work abstract": 134349,
      "terms abstract": 120263,
      "keyphrases based": 57623,
      "validity selected": 130757,
      "existing arabic": 40066,
      "extractor systems": 42573,
      "corresponding values": 24312,
      "articles arabic": 9170,
      "computer assisted": 20478,
      "assisted language": 9584,
      "number speakers": 80969,
      "text database": 120854,
      "particular highlight": 85419,
      "units work": 127272,
      "paper language": 84032,
      "shows recall": 108621,
      "evaluation arabic": 39122,
      "linguistic coverage": 64456,
      "descriptive components": 29502,
      "based rule": 12000,
      "lmf iso": 64876,
      "iso 24613": 56970,
      "lexical markup": 63780,
      "tei text": 120060,
      "representation design": 99207,
      "based initial": 11771,
      "article focus": 9132,
      "distance paper": 32430,
      "correcting spelling": 24139,
      "spelling errors": 111905,
      "errors arabic": 38357,
      "inserting deleting": 55511,
      "obtained satisfactory": 81408,
      "shows new": 108604,
      "computers requires": 20513,
      "using computer": 129466,
      "scheme using": 103942,
      "generate intermediate": 46959,
      "text significant": 121293,
      "poor language": 88058,
      "paper shown": 84429,
      "results verifying": 102320,
      "punjabi machine": 94401,
      "aims preserve": 4559,
      "words proper": 134144,
      "process extract": 91482,
      "calculating probabilities": 15031,
      "location words": 64970,
      "relative frequency": 98354,
      "researchers natural": 100699,
      "structure tagging": 113967,
      "techniques available": 119841,
      "clusters formed": 17967,
      "inflection word": 54290,
      "algorithms data": 4839,
      "various indian": 131104,
      "data corpora": 25796,
      "annotated treebank": 6253,
      "collection annotated": 18458,
      "requires multiple": 100306,
      "corpora hybrid": 23496,
      "algorithm matching": 4771,
      "granularity level": 48749,
      "level importance": 63457,
      "errors terms": 38412,
      "results classical": 101578,
      "english turkish": 37319,
      "used convert": 128453,
      "word respective": 133483,
      "root form": 103272,
      "suffix prefix": 115298,
      "based suffix": 12084,
      "arabic lexicons": 8523,
      "ways different": 132153,
      "number arabic": 80843,
      "mathematical relations": 67461,
      "entry development": 38178,
      "languages studied": 60900,
      "develop corpus": 30186,
      "framework development": 45496,
      "language providing": 60006,
      "dictionary proposed": 30890,
      "internal data": 56161,
      "insertion deletion": 55513,
      "new records": 79097,
      "framework type": 45725,
      "similarities pairs": 109183,
      "single node": 109776,
      "standard tree": 112333,
      "distance algorithm": 32404,
      "entailment arabic": 37655,
      "presents machine": 89862,
      "morphological generator": 74706,
      "forms given": 45245,
      "morphemes words": 74672,
      "based morphological": 11857,
      "learning morphological": 62803,
      "morphological processing": 74724,
      "form generation": 45090,
      "performance decision": 86274,
      "trees bayesian": 125682,
      "method paper": 69038,
      "partofspeech word": 85568,
      "levels hierarchy": 63542,
      "support special": 115992,
      "method test": 69184,
      "statistical error": 113095,
      "useful alternative": 128857,
      "essential prerequisite": 38563,
      "presents various": 89913,
      "step text": 113304,
      "inflected words": 54285,
      "stemming reduce": 113238,
      "word like": 133351,
      "adjective verb": 3858,
      "reduces size": 97398,
      "inflectional forms": 54292,
      "word common": 132958,
      "common base": 18857,
      "accuracy errors": 2152,
      "challenges persian": 16192,
      "electronic text": 35260,
      "main languages": 66437,
      "briefly introduce": 14624,
      "written corpora": 135119,
      "indoaryan language": 53950,
      "language finite": 59048,
      "written standard": 135150,
      "spread different": 112047,
      "substantial challenge": 114853,
      "order implement": 82338,
      "rules patterns": 103426,
      "words table": 134253,
      "languages important": 60626,
      "applications data": 6911,
      "difficult decide": 31618,
      "semantic entity": 105043,
      "process difficult": 91460,
      "language factors": 59040,
      "spelling pronunciation": 111911,
      "uses dictionary": 129224,
      "special features": 111361,
      "metrics proposed": 69996,
      "compared wellknown": 19488,
      "techniques literature": 119923,
      "verbal morphology": 131535,
      "scope improvement": 104039,
      "subsequently used": 114824,
      "measures combination": 67856,
      "languages studies": 60901,
      "studies applied": 114180,
      "step addition": 113244,
      "step performance": 113293,
      "proposed statistical": 93556,
      "term precision": 120239,
      "poetry text": 87791,
      "reading disabilities": 96018,
      "words change": 133848,
      "search word": 104363,
      "presents algorithm": 89815,
      "word occur": 133372,
      "containing frequent": 21791,
      "words arabic": 133824,
      "words observed": 134086,
      "algorithm select": 4796,
      "proposed al": 93171,
      "approach opens": 7757,
      "automatic natural": 10615,
      "methods studying": 69773,
      "rigorous study": 102856,
      "usually involves": 130431,
      "rules use": 103451,
      "evaluate progress": 38895,
      "rigorous testing": 102857,
      "convert text": 23098,
      "using currently": 129526,
      "sufficient generating": 115279,
      "required word": 100229,
      "new stemming": 79185,
      "documents transformed": 33306,
      "use explore": 128039,
      "results adapting": 101499,
      "comparative work": 19214,
      "morphology free": 74784,
      "significant contributions": 108748,
      "hybrid representation": 51192,
      "improved parsing": 52625,
      "parsing arabic": 85074,
      "transformations resulting": 124280,
      "f1score compared": 42715,
      "various affixes": 131035,
      "automates process": 10487,
      "rules results": 103437,
      "using multilevel": 129913,
      "design patterns": 29558,
      "generation desired": 47364,
      "conflict resolution": 21034,
      "resolution methods": 100769,
      "methods mentioned": 69617,
      "order effectiveness": 82313,
      "proposed schema": 93540,
      "language mt": 59695,
      "ongoing effort": 81735,
      "language arab": 58840,
      "arab world": 8487,
      "growing collection": 49168,
      "110 million": 232,
      "morphological annotation": 74688,
      "respect target": 101106,
      "approach comparable": 7429,
      "spontaneous arabic": 112030,
      "large research": 61237,
      "integration different": 55808,
      "multilingual entity": 75252,
      "work associated": 134382,
      "focuses specific": 44913,
      "classes entities": 17060,
      "strings language": 113605,
      "introduced text": 56600,
      "does know": 33361,
      "based rules": 12001,
      "kannada language": 57491,
      "performance twice": 86816,
      "language rich": 60064,
      "focus modern": 44790,
      "clean corpus": 17699,
      "process morphological": 91531,
      "research human": 100530,
      "phone units": 87276,
      "big corpus": 13990,
      "problem major": 91119,
      "presents statistical": 89902,
      "inverse probability": 56701,
      "uniformly distributed": 127146,
      "ease access": 34406,
      "comprehensive updated": 20282,
      "paper results": 84407,
      "results recent": 102117,
      "conducted identify": 20931,
      "initial list": 55219,
      "direct links": 31790,
      "keeping mind": 57520,
      "rules proposed": 103432,
      "module implemented": 74493,
      "yield accuracy": 135329,
      "accuracy 99": 2089,
      "explored domain": 41623,
      "constituents word": 21556,
      "prosody word": 93651,
      "word prosody": 133423,
      "stress pattern": 113581,
      "syllables sentence": 116309,
      "language aim": 58823,
      "build existing": 14766,
      "correct text": 24124,
      "russian inflection": 103491,
      "share data": 107951,
      "set code": 107392,
      "code public": 18139,
      "discuss attempt": 32235,
      "awadhi bhojpuri": 11252,
      "creation corpora": 24713,
      "used corpora": 128459,
      "study similarity": 114520,
      "language identifier": 59150,
      "currently achieved": 25395,
      "identification languages": 51394,
      "accuracy output": 2227,
      "automatically annotate": 10718,
      "linguistic elements": 64470,
      "identification arabic": 51364,
      "events expressed": 39579,
      "design challenges": 29525,
      "impact development": 51866,
      "transliteration methods": 125524,
      "methods empirically": 69467,
      "arabic japanese": 8518,
      "extract best": 42070,
      "best candidate": 13313,
      "test corpora": 120431,
      "correction paper": 24162,
      "joint morphological": 57298,
      "documents morphological": 33254,
      "partofspeech lemma": 85528,
      "work needed": 134649,
      "possible analyses": 88377,
      "multiple dialects": 75537,
      "history arabic": 50546,
      "rich history": 102746,
      "history language": 50560,
      "written arabic": 135113,
      "sinhala language": 109865,
      "arithmetic problems": 9008,
      "keyword identification": 57637,
      "question identification": 95166,
      "mathematical operation": 67459,
      "combined neural": 18661,
      "correction systems": 24166,
      "language example": 59020,
      "statistical transliteration": 113174,
      "transliterated words": 125519,
      "especially limited": 38465,
      "structures representing": 114102,
      "foreign word": 45044,
      "experiments vietnamese": 41202,
      "outperforms statistical": 83023,
      "50 years": 1079,
      "discussions paper": 32330,
      "participants analyze": 85317,
      "maghrebi arabic": 66374,
      "automatic diacritization": 10512,
      "dependency representation": 29224,
      "treebank paper": 125642,
      "discusses construction": 32307,
      "construction process": 21695,
      "discuss effect": 32249,
      "arabic main": 8524,
      "msa arabic": 74941,
      "roman script": 103245,
      "arabic arabizi": 8490,
      "words include": 133993,
      "given problem": 48092,
      "representation encoding": 99224,
      "specifically present": 111575,
      "capable identifying": 15203,
      "partially fully": 85304,
      "resources created": 100959,
      "provide lexical": 93867,
      "structure provides": 113939,
      "propose algorithmic": 92539,
      "techniques handle": 119899,
      "specific problems": 111479,
      "makes necessary": 66797,
      "capture language": 15322,
      "paper attempted": 83745,
      "make summary": 66735,
      "spelling checkers": 111900,
      "removal techniques": 98871,
      "techniques key": 119914,
      "key importance": 57579,
      "dialects written": 30542,
      "approximately 30": 8464,
      "countries speak": 24449,
      "speak language": 111279,
      "corpus composed": 23705,
      "meet needs": 68251,
      "length different": 63358,
      "solve specified": 110620,
      "algorithms comparing": 4836,
      "algorithms designed": 4841,
      "incorrect input": 53577,
      "analysis common": 5532,
      "differ radically": 30930,
      "english differ": 37112,
      "strong points": 113701,
      "advantage approach": 4036,
      "approach doesnt": 7508,
      "use dictionary": 127999,
      "address important": 3682,
      "application multilingual": 6867,
      "rules english": 103395,
      "focused transfer": 44881,
      "words stems": 134239,
      "results decreasing": 101667,
      "algorithm extracts": 4739,
      "resources hindi": 100987,
      "lemma list": 63329,
      "paper effort": 83887,
      "exhaustive list": 39994,
      "created 12": 24653,
      "various morphological": 131137,
      "assessed using": 9503,
      "tense person": 120182,
      "level forms": 63451,
      "standard pattern": 112290,
      "used set": 128756,
      "applied extract": 7074,
      "identification key": 51392,
      "processing understanding": 91846,
      "disambiguation word": 31996,
      "bhojpuri magahi": 13779,
      "analyze computationally": 5956,
      "expert linguists": 41227,
      "time resources": 122091,
      "relatively highresource": 98394,
      "collected corpora": 18410,
      "extent possible": 41980,
      "text belongs": 120683,
      "corpora character": 23432,
      "partsofspeech pos": 85596,
      "pos chunk": 88207,
      "morphological lexical": 74715,
      "lexical phonological": 63791,
      "results clear": 101579,
      "segmentation fundamental": 104590,
      "fundamental step": 46117,
      "major bottlenecks": 66541,
      "according literature": 1987,
      "used people": 128684,
      "religious texts": 98670,
      "characters used": 16628,
      "project collaborative": 92213,
      "online dictionary": 81766,
      "visual similarity": 131811,
      "russian demonstrate": 103489,
      "yielding results": 135383,
      "results closer": 101581,
      "people public": 85884,
      "problems challenges": 91304,
      "objective study": 81116,
      "requires proper": 100310,
      "segmentation initial": 104592,
      "word initial": 133321,
      "segmentation used": 104648,
      "digital corpus": 31719,
      "validation process": 130742,
      "provide missing": 93874,
      "alignment considering": 4950,
      "alignment process": 4983,
      "workshop computational": 135005,
      "finitestate morphology": 44546,
      "role various": 103223,
      "consisting text": 21463,
      "sentence parallel": 105979,
      "data native": 26152,
      "script language": 104256,
      "language basic": 58858,
      "heldout sentences": 49687,
      "sentences native": 106404,
      "including single": 53374,
      "sentence addition": 105764,
      "parser recently": 84991,
      "parsers proposed": 85041,
      "sentence second": 106057,
      "comparable level": 19149,
      "suggesting word": 115354,
      "prediction furthermore": 89065,
      "presents unique": 89910,
      "language addition": 58813,
      "model utilized": 72288,
      "utilized suggest": 130546,
      "vocabulary various": 131918,
      "persian word": 87087,
      "performed languages": 86927,
      "extracted speech": 42185,
      "dependency corpora": 29137,
      "phenomena languages": 87229,
      "shortterm memorybased": 108351,
      "model benchmark": 70757,
      "words inflectional": 134002,
      "methods automated": 69326,
      "context specific": 22275,
      "digital tools": 31728,
      "onetoone correspondence": 81730,
      "models historical": 73336,
      "disambiguation algorithm": 31954,
      "languages unique": 60936,
      "number characters": 80852,
      "characters languages": 16614,
      "spread multiple": 112050,
      "characters language": 16613,
      "disambiguation ambiguous": 31955,
      "efficiency inference": 35028,
      "accuracy leading": 2195,
      "tree classifier": 125583,
      "learns alignments": 63192,
      "microaveraged f1": 70041,
      "uzbek language": 130685,
      "identification proposed": 51425,
      "adposition case": 3916,
      "case supersenses": 15627,
      "snacs semantic": 110301,
      "network adposition": 77129,
      "supersenses schneider": 115715,
      "schneider et": 103955,
      "snacs inventory": 110300,
      "use adpositions": 127886,
      "adpositions case": 3922,
      "case markers": 15598,
      "function relation": 46041,
      "expertise paper": 41236,
      "phonology morphology": 87319,
      "models vary": 74325,
      "terms complexity": 120294,
      "complexity resource": 19935,
      "indian languages paper": 53824,
      "rule based algorithm": 103346,
      "new computational model": 78834,
      "work explores application": 134521,
      "unlike languages arabic": 127437,
      "processing discussed paper": 91665,
      "morphological analysis generation": 74680,
      "understanding dialogue systems": 126830,
      "experiments suggest approach": 41159,
      "arabic speech recognition": 8557,
      "detection arabic language": 29883,
      "keyphrase extraction using": 57619,
      "using linguistic knowledge": 129818,
      "validity selected features": 130758,
      "computer assisted language": 20479,
      "assisted language learning": 9585,
      "agglutinative language language": 4338,
      "lmf iso 24613": 64877,
      "lexical markup framework": 63781,
      "tei text encoding": 120061,
      "correcting spelling errors": 24140,
      "role improving quality": 103178,
      "researchers natural language": 100700,
      "character level importance": 16442,
      "different writing styles": 31580,
      "nlp applications require": 79573,
      "edit distance algorithm": 34532,
      "paper presents machine": 84177,
      "presents machine learning": 89863,
      "important component natural": 52124,
      "machine learning framework": 65796,
      "learning models learn": 62784,
      "training data models": 123496,
      "word form generation": 133287,
      "different levels hierarchy": 31230,
      "language finite state": 59049,
      "text written standard": 121422,
      "evaluation metrics proposed": 39292,
      "substantial improvements compared": 114861,
      "way improve performance": 132091,
      "automatic natural language": 10616,
      "work presents comprehensive": 134695,
      "using currently available": 129527,
      "classification paper address": 17319,
      "rich morphology free": 102771,
      "morphology free word": 74785,
      "large scale corpus": 61244,
      "language arab world": 58841,
      "spontaneous arabic speech": 112031,
      "language pair model": 59731,
      "paper presents statistical": 84204,
      "nlp research community": 79683,
      "dataset collected multiple": 26797,
      "automatically generate output": 10776,
      "prosody word prosody": 93652,
      "grammatically correct text": 48733,
      "terms accuracy speed": 120265,
      "languages automatic language": 60417,
      "arabic paper introduce": 8543,
      "method automatically identify": 68659,
      "approach evaluated using": 7542,
      "work present largescale": 134679,
      "combined neural network": 18662,
      "error correction systems": 38299,
      "words source language": 134229,
      "linguistic resources available": 64548,
      "framework outperforms statistical": 45645,
      "limited training examples": 64294,
      "linguistic unit language": 64572,
      "available resources available": 11096,
      "little attention given": 64794,
      "analysis arabic language": 5502,
      "applied textual data": 7130,
      "performance proposed algorithm": 86629,
      "various morphological forms": 131138,
      "proposed algorithm based": 93174,
      "disambiguation word sense": 31997,
      "languages various sources": 60966,
      "high error rate": 50070,
      "train model directly": 122956,
      "systems paper describes": 117043,
      "various tasks natural": 131214,
      "dataset paper describes": 27085,
      "provide baseline results": 93761,
      "languages english arabic": 60528,
      "features compared using": 43413,
      "character level neural": 16444,
      "level neural machine": 63479,
      "long shortterm memorybased": 65130,
      "writing systems exist": 135105,
      "report results based": 99035,
      "writing systems language": 135106,
      "decision tree classifier": 27946,
      "microaveraged f1 score": 70042,
      "adposition case supersenses": 3917,
      "snacs semantic network": 110302,
      "semantic network adposition": 105122,
      "network adposition case": 77130,
      "case supersenses schneider": 15628,
      "supersenses schneider et": 115716,
      "schneider et al": 103956,
      "use adpositions case": 127887,
      "adpositions case markers": 3923,
      "task requires modeling": 118644,
      "modeling natural language processing": 72491,
      "computer assisted language learning": 20480,
      "tei text encoding initiative": 120062,
      "important research area field": 52228,
      "researchers natural language processing": 100701,
      "paper presents machine learning": 84178,
      "important component natural language": 52125,
      "component natural language processing": 19991,
      "automatic natural language processing": 10617,
      "rich morphology free word": 102772,
      "morphology free word order": 74786,
      "building natural language processing": 14870,
      "languages automatic language identification": 60418,
      "encoderdecoder recurrent neural network": 36621,
      "proposed framework outperforms statistical": 93296,
      "languages natural language processing": 60754,
      "step natural language processing": 113283,
      "various tasks natural language": 131215,
      "character level neural machine": 16445,
      "level neural machine translation": 63480,
      "snacs semantic network adposition": 110303,
      "semantic network adposition case": 105123,
      "network adposition case supersenses": 77131,
      "adposition case supersenses schneider": 3918,
      "case supersenses schneider et": 15629,
      "supersenses schneider et al": 115717,
      "schneider et al 2018": 103957,
      "use adpositions case markers": 127888,
      "important component natural language processing": 52126,
      "rich morphology free word order": 102773,
      "step natural language processing tasks": 113284,
      "various tasks natural language processing": 131216,
      "character level neural machine translation": 16446,
      "snacs semantic network adposition case": 110304,
      "semantic network adposition case supersenses": 105124,
      "network adposition case supersenses schneider": 77132,
      "adposition case supersenses schneider et": 3919,
      "case supersenses schneider et al": 15630,
      "supersenses schneider et al 2018": 115718,
      "organise": 82463,
      "senticnet": 106558,
      "opiniontarget": 82136,
      "tang": 117576,
      "www": 135185,
      "mse": 74947,
      "aspectrelated": 9369,
      "excessively": 39939,
      "elaborately": 35226,
      "imn": 51853,
      "0914": 103,
      "profoundly": 92110,
      "astonishing": 9671,
      "informality": 54349,
      "folds": 44934,
      "pertinently": 87191,
      "sentimentrelated": 106829,
      "messagepassing": 68501,
      "xsense": 135219,
      "nominate": 80144,
      "newsgroup": 79403,
      "nonadjacent": 80149,
      "downweighting": 34066,
      "harmonized": 49509,
      "acd": 2393,
      "neat": 76743,
      "weakens": 132188,
      "taste": 119648,
      "ptms": 94227,
      "simultaneously identifies": 109678,
      "corresponding sentiment": 24305,
      "efficient variational": 35131,
      "inference test": 54236,
      "labeling accuracy": 58483,
      "aspect phrase": 9341,
      "polysemy context": 88028,
      "review text": 102583,
      "learn feature": 62047,
      "phrase embedding": 87345,
      "experiments review": 41122,
      "stateoftheart strong": 112974,
      "method optimizes": 68999,
      "optimizes objective": 82240,
      "treated sequence": 125559,
      "relations modeled": 98210,
      "information conditional": 54433,
      "results embedding": 101748,
      "classification unlike": 17477,
      "explicitly captures": 41362,
      "lstm attentionbased": 65596,
      "faster lstm": 43183,
      "analysis goal": 5635,
      "comments work": 18829,
      "sentiment target": 106801,
      "particular identify": 85420,
      "text mention": 121111,
      "platforms current": 87663,
      "networks domain": 77575,
      "labels documents": 58593,
      "softly select": 110503,
      "outperforms different": 82880,
      "extraction opinion": 42420,
      "difficult endeavor": 31624,
      "mainly lack": 66491,
      "field specifically": 43983,
      "better generate": 13596,
      "generate detailed": 46925,
      "opinion analysis": 82081,
      "involves identification": 56895,
      "identified terms": 51462,
      "modular flexible": 74467,
      "sequence labeler": 106965,
      "includes lexical": 53256,
      "features polarity": 43658,
      "polarity values": 87931,
      "obtained domain": 81364,
      "best outofdomain": 13386,
      "achieving accuracy": 2926,
      "extraction past": 42428,
      "past domains": 85640,
      "markedly better": 67227,
      "traditional crf": 122808,
      "crf using": 24760,
      "crf training": 24759,
      "having manually": 49561,
      "systems domains": 116859,
      "configuration performs": 21005,
      "classification given": 17218,
      "aspect sentiment": 9347,
      "classification multilingual": 17291,
      "sentiment annotation": 106682,
      "using rhetorical": 130116,
      "paper fills": 83943,
      "reports natural": 99082,
      "set techniques": 107602,
      "derived rhetorical": 29360,
      "structure theory": 113973,
      "textual opinions": 121723,
      "presents prototype": 89892,
      "automatically labelled": 10795,
      "extraction ate": 42266,
      "used efficiently": 128507,
      "construct datasets": 21625,
      "rulebased baseline": 103360,
      "precision values": 88808,
      "beats supervised": 12624,
      "using stacked": 130208,
      "dividing subtasks": 32915,
      "competitively compared": 19701,
      "reviews extract": 102608,
      "recurrent architecture": 97159,
      "component outperforms": 19997,
      "task aspectbased": 117899,
      "subtasks aspect": 114963,
      "approach aspectbased": 7365,
      "analysis improving": 5652,
      "embeddings finegrained": 35723,
      "extracting opinion": 42225,
      "words difficult": 133915,
      "difficult process": 31656,
      "character structure": 16468,
      "labeling using": 58555,
      "performance specifically": 86730,
      "respect baseline": 101074,
      "character patterns": 16457,
      "cover domains": 24484,
      "new creative": 78847,
      "corpora creating": 23447,
      "mechanism select": 68035,
      "aspects use": 9417,
      "supervision evaluating": 115882,
      "evaluating human": 39062,
      "specifically using": 111608,
      "desired sentiment": 29671,
      "consists identifying": 21478,
      "opinionated text": 82114,
      "text detecting": 120870,
      "time energy": 122022,
      "attention deep": 9811,
      "issues remain": 57074,
      "target especially": 117620,
      "utilize interaction": 130510,
      "leftright contexts": 63305,
      "left center": 63296,
      "center right": 15891,
      "target phrase": 117688,
      "target compare": 117589,
      "datasets related": 27666,
      "joint way": 57332,
      "captures interaction": 15448,
      "representations aspects": 99507,
      "architectures recurrent": 8843,
      "task motivated": 118412,
      "memory chains": 68300,
      "update mechanism": 127795,
      "track entities": 122729,
      "demonstrates substantial": 28964,
      "reviews present": 102622,
      "framework tackling": 45709,
      "opinion summary": 82102,
      "analysis problem": 5770,
      "neutral positive": 78756,
      "sentence traditional": 106109,
      "entity considering": 37914,
      "various preprocessing": 131172,
      "task double": 118119,
      "task finegrained": 118214,
      "features users": 43777,
      "highly sophisticated": 50348,
      "sophisticated supervised": 110676,
      "simple cnn": 109384,
      "model employing": 71063,
      "generalpurpose embeddings": 46883,
      "sophisticated existing": 110667,
      "extraction achieve": 42249,
      "gated convolutional": 46504,
      "aspects entities": 9382,
      "summarize previous": 115608,
      "promising task": 92307,
      "idea explicitly": 51324,
      "incorporate representations": 53484,
      "bottomup topdown": 14504,
      "dependency syntactic": 29231,
      "learn treestructured": 62173,
      "methodology highquality": 69263,
      "demonstrate quality": 28855,
      "approaches public": 8301,
      "datasets semeval": 27697,
      "2016 attentionbased": 603,
      "short reviews": 108231,
      "users write": 129185,
      "existing topic": 40319,
      "long reviews": 65085,
      "observe current": 81192,
      "problems end": 91317,
      "reviews viewed": 102638,
      "pairs building": 83495,
      "information clustered": 54419,
      "task vital": 118840,
      "analysis includes": 5653,
      "sentiment provide": 106785,
      "sentence influence": 105906,
      "considered sentiment": 21299,
      "polarity negative": 87916,
      "sentence recurrent": 106018,
      "model deal": 70951,
      "processing rnns": 91787,
      "aspect information": 9336,
      "information considered": 54439,
      "words information": 134003,
      "cnn improves": 18005,
      "representations topic": 99936,
      "data open": 26189,
      "contains high": 21822,
      "quality labels": 94697,
      "lacks data": 58775,
      "extensions recently": 41849,
      "methods target": 69791,
      "classification polarity": 17334,
      "requires finegrained": 100272,
      "semisupervised method": 105614,
      "disentangling latent": 32354,
      "method induces": 68904,
      "methods achieves": 69289,
      "challenging subtasks": 16315,
      "review sentence": 102573,
      "sentence supervised": 106096,
      "perform paper": 86040,
      "restaurant dataset": 101318,
      "constrained attention": 21565,
      "multiaspect sentiment": 75034,
      "studies developed": 114208,
      "methods generating": 69523,
      "mechanism specifically": 68039,
      "multitask settings": 75891,
      "started new": 112403,
      "analysis usergenerated": 5927,
      "usergenerated reviews": 129074,
      "multiple branches": 75508,
      "datasets restaurant": 27684,
      "released semeval": 98494,
      "demonstrates approach": 28941,
      "visualization topic": 131831,
      "implemented simple": 51976,
      "simple set": 109518,
      "languages seven": 60865,
      "features sequence": 43722,
      "expression extraction": 41736,
      "aspects present": 9402,
      "label word": 58420,
      "pattern words": 85721,
      "words particular": 134107,
      "based memory": 11835,
      "memory conditional": 68304,
      "model incorporated": 71329,
      "layer explicitly": 61718,
      "labels experimental": 58600,
      "showed approach": 108368,
      "companies organizations": 19117,
      "account different": 2023,
      "aim overcome": 4465,
      "overcome aforementioned": 83273,
      "created comprehensive": 24658,
      "analysis relevant": 5802,
      "sentimental tendency": 106821,
      "backpropagation time": 11320,
      "patterns address": 85727,
      "encoders modeling": 36658,
      "potential customers": 88544,
      "knowledge exploited": 57919,
      "analysis explore": 5615,
      "datasets comparable": 27359,
      "aims detecting": 4516,
      "datasets effective": 27439,
      "weights calculated": 132374,
      "modeling process": 72517,
      "representations experiment": 99643,
      "automatically mines": 10803,
      "conduct sentiment": 20892,
      "sentiment predictions": 106783,
      "leading substantial": 61914,
      "targets sentiment": 117793,
      "sentence prior": 106004,
      "framework multiple": 45618,
      "approaches framework": 8166,
      "outperforms sequence": 82978,
      "baseline pipeline": 12285,
      "generating long": 47230,
      "coarsetofine decoding": 18061,
      "decoding generating": 28096,
      "semantic slots": 105301,
      "aspect semantics": 9345,
      "model interactive": 71359,
      "terms corresponding": 120300,
      "sentence task": 106102,
      "pipeline manner": 87543,
      "extraction performed": 42432,
      "followed sentiment": 44951,
      "develop approach": 30174,
      "exploit joint": 41423,
      "information helpful": 54653,
      "documentlevel labeled": 33146,
      "labeled sentiment": 58468,
      "network imn": 77278,
      "level unlike": 63516,
      "architecture information": 8673,
      "embeddings attention": 35559,
      "extraction review": 42473,
      "feature attention": 43251,
      "shared sentiment": 108010,
      "caused data": 15829,
      "model crossdomain": 70941,
      "category transfer": 15795,
      "detection essential": 29942,
      "label review": 58409,
      "distributions diverse": 32758,
      "diverse output": 32831,
      "output spaces": 83125,
      "spaces address": 111090,
      "user behavior": 128967,
      "capture informative": 15317,
      "paths using": 85693,
      "distributions output": 32767,
      "series stateoftheart": 107282,
      "followed supervised": 44954,
      "crossdomain evaluation": 24889,
      "evaluation crossdomain": 39165,
      "conduct case": 20825,
      "architecture furthermore": 8665,
      "aspect generated": 9330,
      "accurately generate": 2382,
      "char embeddings": 16415,
      "word characterbased": 132945,
      "lstm word": 65676,
      "coverage train": 24524,
      "fscore results": 45915,
      "unsupervised technique": 127734,
      "networks inherent": 77627,
      "mechanism convolutional": 67962,
      "illustrate proposed": 51743,
      "analysis domain": 5583,
      "comparison ablation": 19523,
      "influence performance": 54315,
      "embedding experimental": 35403,
      "outperform regular": 82737,
      "using parameterized": 130006,
      "capture explicit": 15300,
      "information implicit": 54666,
      "explicit structural": 41349,
      "approaches largely": 8209,
      "modelling semantic": 72620,
      "syntaxaware representation": 116579,
      "architecture enhanced": 8657,
      "task widely": 118845,
      "words accurate": 133803,
      "classify tokens": 17665,
      "stateoftheart bilstm": 112604,
      "labels output": 58627,
      "addition improves": 3436,
      "investigate modeling": 56782,
      "power contextualized": 88631,
      "series simple": 107279,
      "linear classification": 64332,
      "validation dataset": 130739,
      "ignored previous": 51677,
      "methods lack": 69571,
      "adaptation setting": 3265,
      "highly relies": 50341,
      "resolve propose": 100804,
      "latent relations": 61597,
      "weight word": 132332,
      "higher alignment": 50160,
      "alignment weights": 5001,
      "pairwise classification": 83670,
      "fail learn": 42966,
      "polarities different": 87901,
      "models proposes": 73828,
      "novel general": 80581,
      "various sentiment": 131191,
      "complete story": 19770,
      "triplet extraction": 125823,
      "stateoftheart related": 112885,
      "contextbased word": 22351,
      "extract corresponding": 42075,
      "online review": 81798,
      "contain substantial": 21762,
      "design effective": 29534,
      "target mentions": 117673,
      "dominant methods": 33932,
      "leverage label": 63592,
      "information improving": 54675,
      "taskspecific knowledge": 119629,
      "task remains": 118634,
      "modeled paper": 72360,
      "novel dependency": 80542,
      "architecture multitask": 8698,
      "relations types": 98272,
      "types leveraging": 126311,
      "additionally design": 3601,
      "ensure model": 37640,
      "generalpurpose commonsense": 46881,
      "sufficient coverage": 115276,
      "meaningfully improve": 67736,
      "incorporating commonsense": 53519,
      "deep contextual": 28212,
      "opinions entities": 82121,
      "compute sentiment": 20470,
      "better cope": 13551,
      "proposed extensions": 93278,
      "subtle ways": 115002,
      "implicit missing": 52014,
      "method aspect": 68644,
      "requires word": 100332,
      "models needed": 73620,
      "classification identifying": 17224,
      "review summarization": 102581,
      "noisy seed": 80116,
      "aspects experiments": 9383,
      "task lack": 118332,
      "allows generate": 5158,
      "generate diversified": 46934,
      "confirm method": 21015,
      "method alleviates": 68630,
      "alleviates data": 5056,
      "problem significantly": 91230,
      "boosts performances": 14452,
      "performances current": 86876,
      "polarity text": 87930,
      "decay mechanism": 27913,
      "based position": 11938,
      "measured standard": 67839,
      "help pretraining": 49767,
      "used traditional": 128817,
      "constructs sentiment": 21708,
      "word polarity": 133409,
      "classification aiming": 17107,
      "kinds sentiment": 57693,
      "distinct tasks": 32546,
      "added information": 3363,
      "information significantly": 54981,
      "models distinct": 73095,
      "tasks weaklysupervised": 119600,
      "usually involved": 130430,
      "tokenlevel labels": 122290,
      "labels domain": 58594,
      "adaptation strategies": 3269,
      "problems large": 91332,
      "study leveraging": 114426,
      "labels usually": 58655,
      "improve crossdomain": 52365,
      "selfattention recently": 104901,
      "relationship natural": 98303,
      "temporal dependency": 120101,
      "sentiment orientation": 106763,
      "orientation positive": 82483,
      "require lot": 100170,
      "stateoftheart attentionbased": 112589,
      "labeling module": 58513,
      "sentiment tokens": 106807,
      "labeling sentiment": 58539,
      "object detection": 81050,
      "results enhancing": 101758,
      "classification exploiting": 17206,
      "processing analyze": 91614,
      "information local": 54734,
      "features local": 43597,
      "easy adapt": 34473,
      "use given": 128071,
      "words cause": 133843,
      "word semantically": 133514,
      "attention modules": 9938,
      "detection acd": 29867,
      "acd task": 2395,
      "acd aims": 2394,
      "framework opinion": 45635,
      "novel view": 80772,
      "biaffine scorer": 13788,
      "inference phase": 54193,
      "based outputs": 11919,
      "framework semeval": 45677,
      "explaining reason": 41277,
      "reason sentiment": 96205,
      "sentiment existing": 106739,
      "highly related": 50338,
      "motivates build": 74870,
      "analysis predicting": 5760,
      "models emphasize": 73136,
      "extract opinion": 42100,
      "capture opinion": 15354,
      "extract aspectspecific": 42069,
      "evaluate sentiment": 38923,
      "pairs additionally": 83484,
      "usually divided": 130416,
      "pipeline pipeline": 87550,
      "effective inference": 34693,
      "inference strategy": 54229,
      "exploit mutual": 41432,
      "works employ": 134942,
      "target opinion": 117686,
      "word extraction": 133282,
      "syntactic connections": 116382,
      "connections words": 21150,
      "fail benefit": 42951,
      "model gate": 71220,
      "propose mechanism": 92759,
      "scores word": 104221,
      "word sentences": 133544,
      "uses selfattention": 129274,
      "semantics domain": 105412,
      "simultaneously specifically": 109694,
      "analyzing sentiment": 6051,
      "sentiment given": 106742,
      "challenging address": 16222,
      "aspects simultaneously": 9409,
      "sentence experiment": 105861,
      "ranks 2nd": 95701,
      "understand sentiment": 126775,
      "just words": 57475,
      "encode sentiment": 36453,
      "polarity different": 87909,
      "embedding bert": 35381,
      "generic word": 47815,
      "bert improve": 13142,
      "analysis identifying": 5647,
      "polarity certain": 87906,
      "aspect text": 9351,
      "attracting attention": 10164,
      "provides thorough": 94088,
      "accuracy 91": 2082,
      "absolute higher": 1733,
      "including target": 53392,
      "framework enhance": 45522,
      "stage extract": 112146,
      "artificial tags": 9266,
      "field finally": 43953,
      "reviews hotel": 102611,
      "improvement f1score": 52709,
      "result previous": 101391,
      "way processing": 132122,
      "processing compositional": 91643,
      "providing contextual": 94108,
      "annotations domain": 6427,
      "problems likely": 91337,
      "problems present": 91358,
      "sentiment parse": 106765,
      "rules correctly": 103388,
      "method potential": 69058,
      "associations paper": 9634,
      "paper transform": 84477,
      "propose bidirectional": 92576,
      "challenge specifically": 16103,
      "specifically devise": 111542,
      "queries build": 94914,
      "furthermore considering": 46158,
      "approach conduct": 7443,
      "task aspect": 117898,
      "introduce finegrained": 56423,
      "benchmarks propose": 12932,
      "meaning position": 67662,
      "sentiments aspect": 106831,
      "analysis technique": 5890,
      "2016 datasets": 605,
      "percentage point": 85922,
      "2016 dataset": 604,
      "model explaining": 71136,
      "explaining neural": 41275,
      "sentiment value": 106817,
      "propose explanation": 92665,
      "internal dynamics": 56163,
      "classifier simple": 17581,
      "useful word": 128949,
      "does syntax": 33405,
      "models ptms": 73840,
      "contain sufficient": 21763,
      "induced trees": 54009,
      "induced tree": 54008,
      "performances datasets": 86877,
      "model simultaneously identifies": 72050,
      "evaluate performance model": 38884,
      "sentiment analysis challenging": 106575,
      "attentionbased neural network": 10089,
      "review datasets proposed": 102544,
      "outperforms stateoftheart strong": 83015,
      "embedding space specifically": 35500,
      "information conditional random": 54434,
      "experimental results semeval": 40693,
      "sentiment classification deep": 106697,
      "networks domain adaptation": 77576,
      "results demonstrate approach": 101672,
      "relationships different tasks": 98318,
      "platform paper presents": 87656,
      "dataset competitive results": 26809,
      "paper fills gap": 83944,
      "derived rhetorical structure": 29361,
      "rhetorical structure theory": 102715,
      "paper presents prototype": 84199,
      "gold standard dataset": 48440,
      "term extraction ate": 120211,
      "paper introduce architecture": 83989,
      "method automatically construct": 68653,
      "neural architecture sentiment": 77837,
      "approach outperforms standard": 7776,
      "architecture performs competitively": 8727,
      "performs competitively compared": 86994,
      "sentiment analysis proposed": 106638,
      "subtasks aspect term": 114964,
      "sentiment analysis improving": 106599,
      "sequence labeling using": 106991,
      "baseline model experiments": 12258,
      "corpora creating new": 23448,
      "performance compared using": 86242,
      "detection task identifying": 30077,
      "subtask sentiment analysis": 114952,
      "attention deep learning": 9812,
      "sentiment analysis recent": 106642,
      "left center right": 63297,
      "benchmark datasets related": 12800,
      "outperforms stateoftheart techniques": 83018,
      "classification aims identify": 17109,
      "aims identify sentiment": 4538,
      "jointly learns representations": 57360,
      "models machine reading": 73540,
      "substantial improvements stateoftheart": 114863,
      "input sentence conditioned": 55423,
      "stateoftheart methods various": 112716,
      "sentiment analysis problem": 106632,
      "negative neutral positive": 76936,
      "traditional sentiment classification": 122865,
      "various preprocessing steps": 131173,
      "features methods used": 43608,
      "task finegrained sentiment": 118215,
      "sentiment analysis product": 106634,
      "analysis product reviews": 5776,
      "results outperforming stateoftheart": 102019,
      "existing methods knowledge": 40186,
      "gated convolutional networks": 46505,
      "shortterm memory attention": 108317,
      "model based convolutional": 70724,
      "demonstrate efficiency effectiveness": 28733,
      "dependency tree structure": 29236,
      "dependency tree structures": 29237,
      "dependency structure features": 29229,
      "key idea explicitly": 57575,
      "neural models paper": 78188,
      "classification demonstrate effectiveness": 17178,
      "existing topic models": 40320,
      "provides better performance": 94005,
      "words sentence influence": 134194,
      "sentiment polarity negative": 106774,
      "sentence recurrent neural": 106019,
      "sentiment classification using": 106719,
      "sentiment classification performance": 106709,
      "high quality labels": 50120,
      "learning architecture paper": 62369,
      "network conduct experiments": 77208,
      "combination convolutional neural": 18556,
      "shared task 2017": 108016,
      "language understanding requires": 60265,
      "paper proposes semisupervised": 84367,
      "variational autoencoder based": 130910,
      "unlabeled data benefits": 127379,
      "detection important challenging": 29970,
      "supervised methods require": 115805,
      "perform paper propose": 86041,
      "detection task experimental": 30075,
      "proposed unsupervised approach": 93580,
      "multiaspect sentiment analysis": 75035,
      "attention weights shows": 10039,
      "features implemented simple": 43550,
      "features sequence labelling": 43723,
      "memory conditional random": 68305,
      "labels experimental results": 58601,
      "previous approaches model": 90382,
      "patterns address issue": 85728,
      "bert achieve new": 13053,
      "sentiment analysis explore": 106594,
      "sentiment classification paper": 106707,
      "classification paper propose": 17325,
      "significant progress area": 108851,
      "models experiments benchmark": 73204,
      "outperforms sequence tagging": 82979,
      "documentlevel labeled sentiment": 33147,
      "paper propose interactive": 84254,
      "demonstrate superior performance": 28878,
      "embeddings attention mechanism": 35560,
      "sentiment analysis shared": 106656,
      "models proposed address": 73823,
      "propose novel joint": 92932,
      "method outperforms series": 69025,
      "outperforms series stateoftheart": 82982,
      "language model approach": 59291,
      "conduct case study": 20826,
      "previous models typically": 90434,
      "sentence representation generation": 106029,
      "makes use word": 66816,
      "train test sets": 123043,
      "using rhetorical structure": 130117,
      "present unsupervised technique": 89756,
      "attention mechanism convolutional": 9892,
      "mechanism convolutional neural": 67963,
      "dependency tree sentence": 29235,
      "illustrate proposed model": 51744,
      "embedding experimental results": 35404,
      "models achieve excellent": 72665,
      "task jointly predicting": 118326,
      "global structural information": 48274,
      "better performance competitive": 13655,
      "aspects paper propose": 9401,
      "bidirectional lstm architecture": 13950,
      "stateoftheart models learning": 112738,
      "neural attention mechanism": 77843,
      "supervised methods lack": 115804,
      "explore unsupervised domain": 41605,
      "domain adaptation setting": 33453,
      "proposes novel general": 93613,
      "sentences experimental results": 106304,
      "information pos tags": 54848,
      "characterlevel language modeling": 16570,
      "review sentiment classification": 102577,
      "task achieving stateoftheart": 117839,
      "results existing methods": 101775,
      "existing methods consider": 40180,
      "results standard benchmarks": 102202,
      "modeled paper propose": 72361,
      "architecture multitask learning": 8699,
      "exploiting syntactic knowledge": 41483,
      "important role improving": 52242,
      "generalpurpose commonsense knowledge": 46882,
      "commonsense knowledge bases": 18992,
      "sentiment classification question": 106711,
      "domainspecific knowledge bases": 33905,
      "people express opinions": 85875,
      "paper extend stateoftheart": 83940,
      "existing sentiment analysis": 40282,
      "method aspect identification": 68645,
      "new domains languages": 78892,
      "complex neural models": 19847,
      "generalpurpose language models": 46885,
      "domain corpora helps": 33491,
      "challenges task lack": 16216,
      "generation task generating": 47660,
      "results confirm method": 101620,
      "alleviates data scarcity": 5057,
      "task identifying sentiment": 118268,
      "given text segment": 48156,
      "research area nlp": 100415,
      "information significantly improves": 54982,
      "models distinct domains": 73096,
      "propose domain adaptation": 92632,
      "domain adaptation strategies": 33454,
      "domain unlabeled target": 33687,
      "information multiple levels": 54778,
      "work introduce simple": 134579,
      "introduce simple approach": 56536,
      "simple approach based": 109361,
      "relationship natural language": 98304,
      "experiments distinct domains": 40917,
      "orientation positive negative": 82484,
      "application deep learning": 6840,
      "computationally expensive paper": 20452,
      "expensive paper propose": 40428,
      "results datasets different": 101660,
      "generates stateoftheart results": 47177,
      "finegrained sentiment classification": 44383,
      "language processing analyze": 59835,
      "sentiment classification approaches": 106694,
      "context paper proposes": 22213,
      "local context features": 64906,
      "evaluate proposed framework": 38899,
      "explaining reason sentiment": 41278,
      "challenging research question": 16307,
      "using approach leads": 129342,
      "approach leads improved": 7669,
      "performance existing approaches": 86350,
      "analysis demonstrates model": 5564,
      "additionally design effective": 3602,
      "documentlevel sentiment classification": 33165,
      "methods consistently outperform": 69398,
      "novel regularization technique": 80704,
      "proposed model extensively": 93447,
      "propose novel graphbased": 92925,
      "provide important features": 93848,
      "experiment results benchmark": 40490,
      "recently neural networkbased": 96723,
      "different sentiment polarities": 31417,
      "analysis propose novel": 5779,
      "evaluation does bert": 39193,
      "different words sentence": 31577,
      "language model automatically": 59293,
      "persian language present": 87081,
      "language present research": 59822,
      "existing methods short": 40194,
      "representation model bert": 99329,
      "improvement f1score compared": 52710,
      "providing contextual information": 94109,
      "consists multiple subtasks": 21485,
      "detection sentiment classification": 30049,
      "address challenge specifically": 3659,
      "pair paper introduce": 83447,
      "sentence representations experimental": 106035,
      "analysis data augmentation": 5556,
      "used image classification": 128578,
      "percentage point improvement": 85923,
      "paper propose explanation": 84234,
      "useful word information": 128950,
      "pretrained models ptms": 90147,
      "dependency parsing trees": 29206,
      "analysis experiments reveal": 5613,
      "review datasets proposed method": 102545,
      "datasets proposed method outperforms": 27640,
      "information conditional random field": 54435,
      "derived rhetorical structure theory": 29362,
      "aspect term extraction ate": 9350,
      "novel neural architecture sentiment": 80668,
      "subtasks aspect term extraction": 114965,
      "sentiment analysis recent years": 106643,
      "sentiment analysis product reviews": 106635,
      "long shortterm memory attention": 65104,
      "model based convolutional neural": 70725,
      "sentence recurrent neural network": 106020,
      "convolutional neural network paper": 23192,
      "deep learning architecture paper": 28251,
      "considerable performance gains using": 21257,
      "natural language understanding requires": 76581,
      "detection task experimental results": 30076,
      "obtain new stateoftheart results": 81306,
      "bert achieve new stateoftheart": 13054,
      "sentiment classification paper propose": 106708,
      "proposed approach yields better": 93222,
      "languages paper propose novel": 60785,
      "proposed method outperforms series": 93387,
      "method outperforms series stateoftheart": 69026,
      "using rhetorical structure theory": 130118,
      "attention mechanism convolutional neural": 9893,
      "mechanism convolutional neural networks": 67964,
      "aspectbased sentiment analysis paper": 9358,
      "explore unsupervised domain adaptation": 41606,
      "experiments demonstrate effectiveness proposed": 40884,
      "information pos tags dependency": 54849,
      "better performance compared stateoftheart": 13654,
      "task achieving stateoftheart results": 117840,
      "performance experimental results benchmark": 86358,
      "existing sentiment analysis models": 40283,
      "experimental results confirm method": 40574,
      "alleviates data scarcity problem": 5058,
      "popular research area nlp": 88125,
      "source domain unlabeled target": 110750,
      "application deep learning methods": 6841,
      "computationally expensive paper propose": 20453,
      "experimental results datasets different": 40579,
      "results datasets different domains": 101661,
      "natural language processing analyze": 76392,
      "propose multitask learning framework": 92809,
      "approach leads improved performance": 7670,
      "paper propose novel joint": 84299,
      "experiment results benchmark datasets": 40491,
      "outperforms strong baselines tasks": 83027,
      "sentiment analysis propose novel": 106637,
      "persian language present research": 87082,
      "model natural language inference": 71559,
      "language representation model bert": 60040,
      "sentence representations experimental results": 106036,
      "review datasets proposed method outperforms": 102546,
      "model based convolutional neural networks": 70726,
      "demonstrate proposed model outperforms stateoftheart": 28853,
      "sentence recurrent neural network rnn": 106021,
      "classification using convolutional neural network": 17488,
      "bert achieve new stateoftheart results": 13055,
      "datasets proposed approach yields better": 27636,
      "proposed method outperforms series stateoftheart": 93388,
      "attention mechanism convolutional neural networks": 9894,
      "extensive experiments demonstrate effectiveness proposed": 41907,
      "model achieves better performance compared": 70561,
      "achieves better performance compared stateoftheart": 2748,
      "performance experimental results benchmark datasets": 86359,
      "benchmark datasets demonstrate effectiveness approach": 12774,
      "experimental results datasets different domains": 40580,
      "results demonstrate proposed model achieves": 101704,
      "propose multitask learning framework jointly": 92810,
      "datasets demonstrate effectiveness proposed model": 27404,
      "disambiguator": 32002,
      "morphemelike": 74669,
      "cons": 21159,
      "nonconcatenative": 80175,
      "successor": 115202,
      "accentuate": 1884,
      "7959": 1275,
      "zurich": 135555,
      "limitedresource": 64303,
      "syllablelevel": 116306,
      "mexican": 70023,
      "nonexistent": 80200,
      "monolithic": 74635,
      "perlanguage": 87044,
      "recombination": 97082,
      "eyes": 42637,
      "typebased": 126234,
      "talked": 117559,
      "reconstructs": 97128,
      "clms": 17805,
      "bitspercharacter": 14242,
      "crosslinguality": 25051,
      "lemmatize": 63343,
      "9319": 1406,
      "nonhierarchical": 80212,
      "occuring": 81507,
      "chatino": 16682,
      "juan": 57421,
      "quiahije": 95381,
      "mexico": 70024,
      "198": 455,
      "zipf": 135544,
      "climax": 17759,
      "pdp": 85830,
      "lawrence": 61688,
      "island": 56963,
      "kirov": 57702,
      "deadline": 27852,
      "afroasiatic": 4277,
      "humble": 51135,
      "infixation": 54274,
      "charactertocharacter": 16633,
      "109": 200,
      "pawsx": 85808,
      "med": 68075,
      "amplifies": 5375,
      "morphology partofspeech": 74799,
      "output tagger": 83132,
      "tagger lemmatizer": 117358,
      "units model": 127258,
      "mdl principle": 67590,
      "times higher": 122180,
      "used lexicons": 128611,
      "applying morphological": 7257,
      "modelling paper": 72613,
      "scalable method": 103693,
      "morphological representations": 74727,
      "suitably efficient": 115412,
      "presenting results": 89811,
      "obtain improvements": 81295,
      "analysis integrates": 5658,
      "base words": 11493,
      "observed words": 81241,
      "parentchild relations": 84858,
      "set linguistically": 107482,
      "academic commercial": 1855,
      "set parameters": 107523,
      "compositional model": 20119,
      "modeling partofspeech": 72503,
      "thesis investigates": 121854,
      "words rich": 134178,
      "structure subword": 113964,
      "forms approach": 45231,
      "approach encode": 7523,
      "help overcome": 49761,
      "new bayesian": 78810,
      "distributed language": 32613,
      "modelling word": 72631,
      "efficient simple": 35115,
      "form given": 45091,
      "corresponding particular": 24298,
      "solving model": 110644,
      "explicit morphological": 41337,
      "data architecture": 25618,
      "results making": 101905,
      "learning characterlevel": 62425,
      "morphemes word": 74671,
      "predict context": 88878,
      "complex morphology": 19835,
      "morphology large": 74791,
      "large tag": 61287,
      "explore variety": 41613,
      "lstms model": 65713,
      "allow natural": 5086,
      "incorporating morphological": 53548,
      "constructs word": 21709,
      "information unified": 55066,
      "unified probabilistic": 127127,
      "model previously": 71788,
      "analysis continuous": 5547,
      "hard soft": 49479,
      "finnish language": 44558,
      "level characters": 63427,
      "maintaining performance": 66523,
      "prediction novel": 89091,
      "data sample": 26366,
      "task multisource": 118424,
      "version input": 131611,
      "ii multiple": 51711,
      "encoders better": 36635,
      "publish dataset": 94341,
      "analysis derived": 5568,
      "composed smaller": 20073,
      "smaller units": 110245,
      "word does": 133000,
      "directly semantic": 31906,
      "derived meaning": 29354,
      "learns segment": 63233,
      "data jointly": 26058,
      "f1 additionally": 42648,
      "networks yield": 77821,
      "additive models": 3644,
      "morpheme segmentation": 74667,
      "turkish finnish": 125961,
      "processing complex": 91640,
      "correct morphological": 24107,
      "steps natural": 113325,
      "propose uses": 93147,
      "techniques morphological": 119933,
      "disambiguation stateoftheart": 31986,
      "studied work": 114175,
      "results french": 101799,
      "engineering additional": 37043,
      "disambiguation accuracy": 31953,
      "accuracy ambiguous": 2098,
      "present encoderdecoder": 89469,
      "encoderdecoder style": 36625,
      "network produce": 77395,
      "accurate lexicon": 2340,
      "language vocabulary": 60336,
      "objective evaluate": 81075,
      "evaluate tasks": 38934,
      "detection clustering": 29898,
      "gains tasks": 46411,
      "difference results": 30945,
      "focus verb": 44840,
      "morphology neural": 74794,
      "performance maintaining": 86517,
      "thorough investigation": 121889,
      "investigation parameters": 56864,
      "vs characterbased": 131964,
      "encoding layer": 36705,
      "language encoder": 59002,
      "datadriven quantitative": 26675,
      "sources neural": 110911,
      "words capture": 133842,
      "capture morphology": 15346,
      "composing representations": 20083,
      "representations subword": 99911,
      "effective capture": 34639,
      "capture morphological": 15345,
      "words systematically": 134252,
      "morphological typologies": 74753,
      "unit representation": 127210,
      "morphological typology": 74754,
      "language modeled": 59422,
      "character trigram": 16473,
      "predictive accuracy": 89200,
      "morphological analyses": 74677,
      "using loglinear": 129830,
      "log linear": 64982,
      "split points": 111942,
      "generation plays": 47545,
      "morphological generation": 74705,
      "characterbased sequencetosequence": 16500,
      "training characterbased": 123381,
      "distributional vector": 32742,
      "inducing accurate": 54014,
      "distinct lexical": 32537,
      "task dialogue": 118091,
      "morphological embeddings": 74698,
      "embeddings improves": 35751,
      "tagging performance": 117436,
      "attentional sequencetosequence": 10054,
      "training morphological": 123718,
      "analysis outperforms": 5733,
      "best team": 13463,
      "participate task": 85332,
      "supervised morphological": 115812,
      "subtask systems": 114958,
      "submissions subtask": 114753,
      "task submissions": 118750,
      "datasets long": 27551,
      "bias make": 13816,
      "focus neural": 44798,
      "approaches tackle": 8367,
      "limitedresource setting": 64304,
      "characters propose": 16619,
      "approach encoderdecoder": 7524,
      "explicit edit": 41319,
      "experiment character": 40454,
      "method decomposes": 68758,
      "experiments apply": 40775,
      "parsing achieve": 85062,
      "syllables morphemes": 116308,
      "parameters proposed": 84780,
      "existing input": 40144,
      "terms key": 120340,
      "key stroke": 57599,
      "sufficient paper": 115285,
      "paper problem": 84213,
      "smaller general": 110234,
      "analysis errors": 5598,
      "errors suggests": 38410,
      "improve prior": 52504,
      "word incorporated": 133316,
      "model paradigm": 71700,
      "morphology generation": 74787,
      "patterns outperform": 85765,
      "historical lexical": 50527,
      "lexical considerations": 63744,
      "sufficient supervision": 115287,
      "supervision available": 115869,
      "scheme train": 103939,
      "train characterlevel": 122908,
      "ones improving": 81689,
      "language morphological": 59692,
      "provided german": 93968,
      "need external": 76807,
      "corresponding data": 24281,
      "transfer way": 124208,
      "rules work": 103454,
      "quantitatively investigate": 94891,
      "investigate conditions": 56735,
      "language course": 58925,
      "unseen rare": 127536,
      "tagging accuracies": 117371,
      "sequence decoder": 106928,
      "features generating": 43525,
      "addition generating": 3432,
      "set allows": 107353,
      "highresource transfer": 50433,
      "new high": 78945,
      "online support": 81809,
      "morphology semantic": 74802,
      "approach specially": 7927,
      "handle unseen": 49404,
      "srl work": 112089,
      "srl models": 112088,
      "range dependencies": 95556,
      "analyses shed": 5465,
      "trained fully": 123147,
      "generative latentvariable": 47731,
      "analysis partofspeech": 5743,
      "tagging fact": 117391,
      "features known": 43581,
      "actions required": 3093,
      "proposed datadriven": 93250,
      "morphological tags": 74750,
      "learning ner": 62832,
      "performance shown": 86710,
      "case dependency": 15580,
      "beneficial model": 12950,
      "demonstrate explicitly": 28741,
      "morphological case": 74691,
      "approaches training": 8379,
      "sequences results": 107138,
      "mle model": 70443,
      "leads strong": 61968,
      "character language": 16438,
      "learn english": 62042,
      "learn abstract": 61983,
      "model proves": 71821,
      "requires morphological": 100305,
      "context explicit": 22098,
      "decoding paper": 28119,
      "paper documents": 83882,
      "lemma word": 63330,
      "tasks focused": 119137,
      "task introduced": 118308,
      "introduced year": 56603,
      "architecture character": 8637,
      "contributing improvement": 22766,
      "paper submitted": 84449,
      "features individual": 43563,
      "sequencetosequence transduction": 107211,
      "corpora combined": 23435,
      "generating missing": 47231,
      "tagging disambiguation": 117386,
      "disambiguation models": 31972,
      "models estonian": 73171,
      "tagging standard": 117449,
      "standard multiclass": 112268,
      "single unit": 109810,
      "nonneural baselines": 80254,
      "featured data": 43340,
      "task featured": 118205,
      "task asked": 117894,
      "asked participants": 9301,
      "submissions task": 114754,
      "propose view": 93157,
      "baselines establishing": 12388,
      "establishing stateoftheart": 38621,
      "2016 lexical": 609,
      "propose multi": 92792,
      "learning wordlevel": 63170,
      "wordlevel tag": 133757,
      "tags words": 117490,
      "framework effectiveness": 45514,
      "effectiveness training": 34967,
      "multiple loss": 75607,
      "features phonological": 43656,
      "bring information": 14630,
      "representation dense": 99206,
      "especially beneficial": 38434,
      "beneficial low": 12948,
      "different treebanks": 31520,
      "architecture characterlevel": 8638,
      "languages perspective": 60800,
      "individual hidden": 53912,
      "morphological properties": 74725,
      "backward hidden": 11350,
      "languages improving": 60631,
      "standard languages": 112251,
      "dictionary headword": 30881,
      "task encoderdecoder": 118143,
      "available historical": 11006,
      "historical corpora": 50517,
      "test proposed": 120483,
      "languages underlying": 60935,
      "data problem": 26279,
      "form using": 45144,
      "scarce unavailable": 103797,
      "unavailable lowresource": 126597,
      "comparison baseline": 19529,
      "context simple": 22269,
      "seeks map": 104524,
      "forms canonical": 45233,
      "results 20": 101486,
      "addition training": 3481,
      "information characters": 54414,
      "crucial components": 25139,
      "integration word": 55817,
      "models segmentation": 73983,
      "obtain final": 81283,
      "final word": 44138,
      "different segmentation": 31403,
      "components including": 20027,
      "configurations based": 21007,
      "comparable outperform": 19154,
      "features inferred": 43565,
      "context develop": 22056,
      "neural hybrid": 77928,
      "showing utility": 108436,
      "better character": 13533,
      "models clms": 72910,
      "bitspercharacter bpc": 14243,
      "embeddings identify": 35743,
      "language extract": 59039,
      "lexical network": 63786,
      "words perform": 134112,
      "clustering resulting": 17955,
      "semantic categories": 104998,
      "sigmorphon 2019": 108670,
      "analysis lemmatization": 5678,
      "produce lemma": 91908,
      "transfer training": 124198,
      "transfer multiple": 124158,
      "lemmatization finding": 63339,
      "morphological form": 74703,
      "tagger accuracy": 117354,
      "pushing limits": 94466,
      "hard come": 49455,
      "novel twostep": 80760,
      "twostep attention": 126169,
      "crucial factors": 25142,
      "multilingual parsing": 75327,
      "parsing raw": 85207,
      "text universal": 121387,
      "task extrinsic": 118197,
      "extrinsic parser": 42624,
      "finally merge": 44204,
      "margin lemmatization": 67196,
      "russian languages": 103494,
      "improvements word": 52942,
      "baselines present": 12443,
      "mechanisms work": 68074,
      "twostep generation": 126171,
      "generation setup": 47618,
      "second languages": 104424,
      "contextual analysis": 22443,
      "100 language": 162,
      "tasks examining": 119110,
      "task improved": 118280,
      "models morphologically": 73584,
      "enhancing word": 37540,
      "far explored": 43077,
      "perplexity gains": 87065,
      "introduce discriminative": 56414,
      "segmentation orthographic": 104606,
      "occur word": 81506,
      "trained currently": 123105,
      "naturally occuring": 76645,
      "subsequently train": 114823,
      "evaluate segmentation": 38919,
      "modeling recently": 72522,
      "learning generative": 62615,
      "like information": 64052,
      "languages lacking": 60669,
      "morphology english": 74783,
      "successfully generate": 115188,
      "available tasks": 11124,
      "machinelearning approaches": 66321,
      "conllu format": 21110,
      "need supervised": 76863,
      "relatively easy": 98389,
      "researched languages": 100670,
      "problems occurring": 91353,
      "different intrinsic": 31186,
      "structure trained": 113975,
      "trained word2vec": 123339,
      "compared embeddings": 19362,
      "san juan": 103622,
      "juan quiahije": 57422,
      "quiahije chatino": 95382,
      "tags based": 117466,
      "results core": 101639,
      "learn morphological": 62093,
      "metalearning problem": 68557,
      "suggested approach": 115341,
      "proposed crosslingual": 93245,
      "transfer model": 124152,
      "ud treebank": 126534,
      "bootstrapping techniques": 14463,
      "sparse vocabularies": 111242,
      "word complexity": 132961,
      "general scarcity": 46708,
      "data poses": 26248,
      "language generate": 59063,
      "train encoderdecoder": 122930,
      "data resampling": 26351,
      "natural distribution": 76252,
      "demonstrates feasibility": 28945,
      "task generates": 118239,
      "ii additional": 51695,
      "14 typologically": 321,
      "multiple research": 75671,
      "similar challenges": 109076,
      "incorporating symbolic": 53561,
      "nowadays neural": 80796,
      "bootstrap neural": 14454,
      "data predict": 26256,
      "approach treat": 7986,
      "completely independent": 19782,
      "independent word": 53783,
      "complexity challenge": 19902,
      "closer average": 17872,
      "text prediction": 121190,
      "st lawrence": 112099,
      "lawrence island": 61689,
      "island yupik": 56964,
      "networks cognitive": 77536,
      "kirov cotterell": 57703,
      "cotterell 2018": 24411,
      "modern encoderdecoder": 74386,
      "encoderdecoder ed": 36596,
      "marcus et": 67179,
      "novel nouns": 80677,
      "designed avoid": 29581,
      "ed model": 34518,
      "simulate realistic": 109633,
      "data development": 25836,
      "systems surprisingly": 117176,
      "largely unsolved": 61350,
      "broad goal": 14672,
      "aims investigate": 4544,
      "typologically distinct": 126495,
      "distinct languages": 32536,
      "90 languages": 1368,
      "languages total": 60923,
      "submitted task": 114779,
      "transformers massively": 124454,
      "problem disambiguating": 91015,
      "disambiguating ambiguous": 31948,
      "produces multiple": 92004,
      "unambiguous words": 126579,
      "surpasses state": 116110,
      "art including": 9052,
      "theory computational": 121815,
      "changes research": 16391,
      "stages language": 112160,
      "single layer": 109753,
      "model creates": 70939,
      "predicting stem": 89013,
      "method simplicity": 69153,
      "best languages": 13360,
      "just single": 57471,
      "use contextualized": 127963,
      "embeddings revisiting": 35921,
      "segmentation typically": 104645,
      "use rulebased": 128256,
      "discuss limitations": 32262,
      "analysis disambiguation": 5578,
      "crucial preprocessing": 25159,
      "forms new": 45253,
      "classifier achieve": 17519,
      "association rules": 9627,
      "form meaning": 45104,
      "errors systems": 38411,
      "patterns tasks": 85785,
      "use modified": 128150,
      "scores possible": 104191,
      "score used": 104138,
      "decoder input": 28024,
      "tasks relate": 119447,
      "external lexicon": 42028,
      "characters external": 16607,
      "ud languages": 126530,
      "better simple": 13721,
      "inflection neural": 54289,
      "global optimum": 48257,
      "observations suggest": 81183,
      "corpus demonstrated": 23748,
      "completely fail": 19781,
      "shortcoming introducing": 108275,
      "report significant": 99043,
      "previous data": 90403,
      "german paper present": 47906,
      "statistical partofspeech tagger": 113151,
      "using output tagger": 129996,
      "length mdl principle": 63367,
      "language resources used": 60058,
      "word representations language": 133457,
      "representations language modelling": 99714,
      "perform intrinsic extrinsic": 86016,
      "set linguistically motivated": 107483,
      "vectors word type": 131489,
      "language modeling partofspeech": 59453,
      "modeling partofspeech tagging": 72504,
      "language words rich": 60345,
      "models automatically learn": 72796,
      "learning problem present": 62911,
      "model language independent": 71407,
      "rich languages achieve": 102758,
      "results existing stateoftheart": 101776,
      "morphology paper presents": 74798,
      "languages complex morphology": 60459,
      "complex morphology large": 19836,
      "unlabeled data experiments": 127384,
      "allow natural language": 5087,
      "constructs word embeddings": 21710,
      "prior distribution latent": 90699,
      "analysis continuous representations": 5548,
      "hard soft attention": 49480,
      "models propose new": 73819,
      "provide complementary information": 93783,
      "probabilistic model word": 90820,
      "jointly learns segment": 57361,
      "neural networks yield": 78611,
      "steps natural language": 113326,
      "learning techniques morphological": 63094,
      "proposed architecture achieves": 93226,
      "feature engineering additional": 43272,
      "encoderdecoder style neural": 36626,
      "neural network produce": 78377,
      "demonstrate model able": 28788,
      "consistent gains tasks": 21378,
      "morphology neural machine": 74795,
      "models various levels": 74322,
      "conduct thorough investigation": 20903,
      "thorough investigation parameters": 121890,
      "datadriven quantitative evaluation": 26676,
      "bayesian model unsupervised": 12583,
      "morphological segmentation paper": 74733,
      "information different sources": 54493,
      "use neural word": 128170,
      "different information sources": 31180,
      "sources neural word": 110912,
      "different morphological typologies": 31270,
      "characterbased sequencetosequence models": 16501,
      "distributional vector space": 32743,
      "language understanding systems": 60278,
      "task dialogue state": 118092,
      "embeddings named entity": 35821,
      "languages work present": 60978,
      "attentional sequencetosequence models": 10055,
      "network model using": 77339,
      "using long shortterm": 129835,
      "use additional unlabeled": 127884,
      "model copy mechanism": 70926,
      "strong results languages": 113708,
      "problem outofvocabulary words": 91153,
      "terms key stroke": 120341,
      "context prediction model": 22222,
      "sufficient supervision available": 115288,
      "lowresource languages learning": 65516,
      "neural model train": 78154,
      "train single multilingual": 123023,
      "single multilingual model": 109769,
      "crosslingual morphological tagging": 24980,
      "improves performance lowresource": 53014,
      "power neural network": 88644,
      "surface forms model": 116062,
      "morphological analysis using": 74683,
      "relevant information word": 98558,
      "languages different morphological": 60504,
      "models use word": 74285,
      "long range dependencies": 65081,
      "range dependencies model": 95557,
      "analyses shed light": 5466,
      "trained fully supervised": 123148,
      "end introduce novel": 36812,
      "generative latentvariable model": 47732,
      "efficient variational inference": 35132,
      "analysis partofspeech tagging": 5744,
      "language previous work": 59826,
      "model improving named": 71322,
      "shown linguistic features": 108493,
      "demonstrate explicitly modeling": 28742,
      "previous approaches training": 90383,
      "character language models": 16439,
      "wide range languages": 132506,
      "propose new methods": 92867,
      "representation input neural": 99265,
      "shared task featured": 108050,
      "task asked participants": 117895,
      "words sentential context": 134201,
      "multi task deep": 75028,
      "propose multi task": 92793,
      "multiple loss functions": 75608,
      "model establishes new": 71097,
      "new stateoftheart accuracy": 79155,
      "stateoftheart accuracy score": 112573,
      "models used extensively": 74288,
      "forward backward hidden": 45324,
      "test proposed model": 120484,
      "sentence representations previous": 106043,
      "scarce unavailable lowresource": 103798,
      "unavailable lowresource languages": 126598,
      "lowresource languages addition": 65508,
      "corpus data using": 23742,
      "morphological tagging lemmatization": 74743,
      "word representations use": 133478,
      "systematic comparative analysis": 116663,
      "general framework learning": 46661,
      "language models clms": 59529,
      "word embeddings identify": 133126,
      "sigmorphon 2019 shared": 108671,
      "neural conditional random": 77869,
      "neural models propose": 78192,
      "pos tagger accuracy": 88220,
      "success crosslingual transfer": 115066,
      "shared task multilingual": 108067,
      "task multilingual parsing": 118419,
      "multilingual parsing raw": 75328,
      "parsing raw text": 85208,
      "raw text universal": 95862,
      "text universal dependencies": 121388,
      "shared task extrinsic": 108047,
      "task extrinsic parser": 118198,
      "extrinsic parser evaluation": 42625,
      "pretrained contextualized embeddings": 90011,
      "wide margin lemmatization": 132495,
      "learning nlp models": 62849,
      "models evaluated word": 73175,
      "evaluated word sense": 39027,
      "english russian languages": 37267,
      "baseline models using": 12266,
      "present different approaches": 89454,
      "different approaches problem": 31000,
      "analysis model outputs": 5700,
      "100 language pairs": 163,
      "neural nonneural baselines": 78620,
      "models morphologically rich": 73585,
      "using subword units": 130236,
      "compared competitive baseline": 19346,
      "baseline models morphological": 12263,
      "available nlp resources": 11056,
      "accuracy partofspeech tagging": 2233,
      "generation using neural": 47694,
      "newly introduced task": 79275,
      "language model architecture": 59292,
      "languages paper describes": 60777,
      "corpora available tasks": 23420,
      "nlp tasks training": 79774,
      "language model tasks": 59406,
      "san juan quiahije": 103623,
      "juan quiahije chatino": 57423,
      "baseline results core": 12298,
      "results core nlp": 101640,
      "proposed crosslingual transfer": 93246,
      "transfer model outperforms": 124153,
      "word embeddings approach": 133051,
      "train encoderdecoder model": 122931,
      "given raw text": 48101,
      "task consists generating": 118012,
      "14 typologically diverse": 322,
      "report empirical results": 98991,
      "st lawrence island": 112100,
      "lawrence island yupik": 61690,
      "neural networks cognitive": 78455,
      "networks cognitive models": 77537,
      "kirov cotterell 2018": 57704,
      "modern encoderdecoder ed": 74387,
      "marcus et al": 67180,
      "collect new dataset": 18388,
      "shared task unsupervised": 108102,
      "task novel task": 118465,
      "typologically distinct languages": 126496,
      "transformers massively multilingual": 124455,
      "training lowresource languages": 123694,
      "collected different sources": 18418,
      "rich languages propose": 102762,
      "disambiguating ambiguous words": 31949,
      "surpasses state art": 116111,
      "state art including": 112442,
      "models using manually": 74304,
      "question neural network": 95191,
      "neural network capable": 78272,
      "works best languages": 134927,
      "best languages tasks": 13361,
      "nlp tasks experimental": 79719,
      "short context windows": 108205,
      "local context model": 64908,
      "context sentence level": 22260,
      "just single word": 57472,
      "context target word": 22293,
      "word subword character": 133594,
      "morphological analysis disambiguation": 74679,
      "dataset collected crowdsourcing": 26795,
      "data case study": 25719,
      "novel hybrid approach": 80595,
      "improvements compared baseline": 52827,
      "language generation case": 59068,
      "learning sequence models": 63012,
      "models successfully applied": 74125,
      "description length mdl principle": 29466,
      "perform intrinsic extrinsic evaluations": 86017,
      "language modeling partofspeech tagging": 59454,
      "morphologically rich languages achieve": 74770,
      "languages complex morphology large": 60460,
      "steps natural language processing": 113327,
      "language paper propose new": 59795,
      "machine learning techniques morphological": 65835,
      "morphology neural machine translation": 74796,
      "models obtain stateoftheart performance": 73656,
      "analyze representations learned neural": 6001,
      "models various levels granularity": 74323,
      "conduct thorough investigation parameters": 20904,
      "use neural word embeddings": 128171,
      "task dialogue state tracking": 118093,
      "present new stateoftheart results": 89592,
      "neural network model using": 78340,
      "using long shortterm memory": 129836,
      "use additional unlabeled data": 127885,
      "long range dependencies model": 65082,
      "model improving named entity": 71323,
      "2018 shared task paper": 682,
      "morphologically rich languages mrls": 74771,
      "model establishes new stateoftheart": 71098,
      "scarce unavailable lowresource languages": 103799,
      "sigmorphon 2019 shared task": 108672,
      "deep neural models propose": 28353,
      "2018 shared task multilingual": 680,
      "shared task multilingual parsing": 108068,
      "task multilingual parsing raw": 118420,
      "multilingual parsing raw text": 75329,
      "parsing raw text universal": 85209,
      "raw text universal dependencies": 95863,
      "2018 shared task extrinsic": 678,
      "shared task extrinsic parser": 108048,
      "task extrinsic parser evaluation": 118199,
      "evaluated word sense disambiguation": 39028,
      "artificial neural network models": 9262,
      "models morphologically rich languages": 73586,
      "compared competitive baseline models": 19347,
      "embeddings trained large corpora": 35986,
      "san juan quiahije chatino": 103624,
      "provide baseline results core": 93762,
      "baseline results core nlp": 12299,
      "results core nlp tasks": 101641,
      "st lawrence island yupik": 112101,
      "encoderdecoder neural networks cognitive": 36615,
      "neural networks cognitive models": 78456,
      "morphologically rich languages propose": 74774,
      "works best languages tasks": 134928,
      "nlp tasks experimental results": 79720,
      "propose novel hybrid approach": 92929,
      "statistically significant improvements compared": 113188,
      "deep learning sequence models": 28315,
      "minimum description length mdl principle": 70208,
      "using long shortterm memory lstm": 129837,
      "model improving named entity recognition": 71324,
      "conll 2018 shared task multilingual": 21086,
      "2018 shared task multilingual parsing": 681,
      "shared task multilingual parsing raw": 108069,
      "task multilingual parsing raw text": 118421,
      "multilingual parsing raw text universal": 75330,
      "parsing raw text universal dependencies": 85210,
      "2018 shared task extrinsic parser": 679,
      "shared task extrinsic parser evaluation": 108049,
      "evaluated word sense disambiguation task": 39029,
      "provide baseline results core nlp": 93763,
      "baseline results core nlp tasks": 12300,
      "encoderdecoder neural networks cognitive models": 36616,
      "dsambigusation": 34204,
      "narrowing": 76196,
      "translatable": 124521,
      "lagrangian": 58783,
      "redirect": 97308,
      "fofe": 44932,
      "generalising": 46745,
      "monetary": 74540,
      "traindevelopmenttest": 123058,
      "currency": 25254,
      "wordshape": 134333,
      "brownclusterbased": 14722,
      "wordembeddingbased": 133709,
      "innately": 55273,
      "unfair": 127064,
      "poleval": 87936,
      "224": 781,
      "prioritizes": 90774,
      "crfsuite": 24772,
      "truecasing": 125856,
      "nesting": 77111,
      "plugging": 87764,
      "ace2004": 2401,
      "typo": 126478,
      "linguisticallyinspired": 64600,
      "breakdown": 14577,
      "perpetuate": 87057,
      "lining": 64631,
      "promised": 92260,
      "enforcement": 37005,
      "gillick": 47965,
      "ji": 57230,
      "multidataset": 75062,
      "httpsgithubcomneulabinterpreteval": 50704,
      "fixes": 44625,
      "856": 1326,
      "german evaluation": 47891,
      "classification usually": 17497,
      "context does": 22070,
      "getting right": 47956,
      "performance penn": 86604,
      "branch natural": 14558,
      "main methodological": 66439,
      "methodological approaches": 69236,
      "stateoftheart conditional": 112616,
      "form linear": 45102,
      "solve resulting": 110616,
      "lagrangian relaxation": 58784,
      "resolution stateoftheart": 100785,
      "networkbased language": 77484,
      "contributions new": 22790,
      "networks key": 77633,
      "yield good": 135337,
      "semantic generalization": 105064,
      "crossdomain setting": 24902,
      "reduction neural": 97453,
      "recognition stateoftheart": 97013,
      "features domainspecific": 43466,
      "lstms conditional": 65708,
      "segments using": 104679,
      "using transitionbased": 130318,
      "transitionbased approach": 124501,
      "shiftreduce parsers": 108188,
      "parsers models": 85032,
      "rely sources": 98742,
      "words characterbased": 133852,
      "supervised corpus": 115743,
      "learned unannotated": 62271,
      "recognition essential": 96873,
      "called deep": 15055,
      "forms entity": 45243,
      "matching given": 67406,
      "redirect pages": 97309,
      "provides automatic": 93999,
      "polish wikipedia": 87986,
      "recognition fails": 96881,
      "approach easy": 7512,
      "easy train": 34484,
      "language news": 59717,
      "knowledge report": 58142,
      "large indomain": 61114,
      "domain similar": 33639,
      "labelling problem": 58572,
      "encoding fofe": 36693,
      "sentence fragment": 105878,
      "predict entity": 88885,
      "label individual": 58400,
      "tasks local": 119272,
      "labelling methods": 58569,
      "changing language": 16399,
      "feature sparsity": 43318,
      "approaches struggle": 8356,
      "entire web": 37724,
      "method obtaining": 68996,
      "serving input": 107332,
      "tasks ner": 119337,
      "better capacity": 13523,
      "context structured": 22282,
      "document accurate": 32947,
      "dates times": 27835,
      "problem sparseness": 91244,
      "difficult maintain": 31644,
      "features smaller": 43734,
      "residual lstm": 100740,
      "languages standard": 60894,
      "particular given": 85418,
      "baseline ner": 12269,
      "direct output": 31796,
      "output information": 83082,
      "specific named": 111466,
      "learning combined": 62442,
      "lightweight architecture": 64010,
      "nearly stateoftheart": 76736,
      "task computationally": 117997,
      "performance just": 86476,
      "modular framework": 74468,
      "train conditional": 122913,
      "challenge named": 16058,
      "leverages unsupervised": 63663,
      "using gazetteer": 129694,
      "embeddings include": 35753,
      "embeddingbased sentence": 35537,
      "locations organizations": 64975,
      "solutions work": 110585,
      "starting vanilla": 112414,
      "networks finally": 77602,
      "finally adding": 44143,
      "adding external": 3376,
      "external word": 42050,
      "significantly increased": 108968,
      "gcn perform": 46568,
      "improvements requiring": 52905,
      "additional languagespecific": 3538,
      "crowd annotations": 25096,
      "crowd annotation": 25095,
      "labels multiple": 58621,
      "uses common": 129206,
      "sets chinese": 107655,
      "domains experimental": 33778,
      "scores strong": 104210,
      "words effective": 133928,
      "obtain large": 81297,
      "campaign report": 15089,
      "vlsp 2018": 131857,
      "scheme applied": 103913,
      "word wordshape": 133662,
      "wordshape features": 134334,
      "brownclusterbased features": 14723,
      "features wordembeddingbased": 43788,
      "wordembeddingbased features": 133710,
      "entities levels": 37812,
      "labels utilized": 58656,
      "text domains": 120892,
      "learn entity": 62043,
      "crf paper": 24757,
      "designing specific": 29649,
      "highorder dependency": 50357,
      "ner named": 77057,
      "speech named": 111711,
      "pipeline process": 87551,
      "approach disadvantages": 7494,
      "disadvantages error": 31932,
      "final task": 44132,
      "outperform sequential": 82742,
      "directly extracts": 31878,
      "unique neural": 127190,
      "french data": 45804,
      "data classical": 25729,
      "carefully handcrafted": 15526,
      "systems lexical": 116978,
      "features actually": 43354,
      "stateoftheart f1": 112661,
      "matching stateoftheart": 67433,
      "conll2003 dataset": 21096,
      "test claim": 120427,
      "hybrid learning": 51179,
      "including manual": 53322,
      "manual features": 67000,
      "english shared": 37276,
      "training requirements": 123811,
      "domainspecific dictionary": 33897,
      "handcrafting features": 49359,
      "better ner": 13634,
      "processing currently": 91653,
      "architecture consists": 8641,
      "adjacent layers": 3850,
      "crossdomain named": 24894,
      "architectures effective": 8802,
      "newswire datasets": 79420,
      "significant resources": 108860,
      "typically degrades": 126420,
      "emerging domains": 36066,
      "domains social": 33858,
      "media significant": 68163,
      "methods efficient": 69464,
      "text serves": 121289,
      "indispensable component": 53892,
      "trees exploited": 125692,
      "based bilstmcrf": 11558,
      "vary respect": 131247,
      "datasets special": 27727,
      "conll03 dataset": 21090,
      "knowledgebased approaches": 58250,
      "sources propose": 110918,
      "propose named": 92813,
      "including contextual": 53277,
      "embeddings long": 35791,
      "use entity": 128031,
      "poleval 2018": 87937,
      "linking model": 64662,
      "identify mentions": 51522,
      "text belonging": 120682,
      "foundation natural": 45349,
      "corpora offtheshelf": 23548,
      "survey representative": 116189,
      "area transfer": 8885,
      "data solution": 26465,
      "difference source": 30946,
      "transferred target": 124236,
      "manual efforts": 66991,
      "classification ner": 17305,
      "internal datasets": 56162,
      "size furthermore": 109924,
      "result addition": 101359,
      "improvements ner": 52879,
      "obtains 22": 81452,
      "outputs work": 83190,
      "time effective": 122018,
      "model high": 71283,
      "text shown": 121292,
      "detection noisy": 30007,
      "ner word": 77097,
      "data dramatically": 25867,
      "similarity source": 109314,
      "good predictors": 48493,
      "usefulness pretrained": 128962,
      "30 data": 900,
      "data pairs": 26205,
      "better pretraining": 13679,
      "exists large": 40345,
      "datasets share": 27703,
      "explore question": 41579,
      "improving ner": 53125,
      "performance investigate": 86470,
      "applied method": 7092,
      "results result": 102134,
      "context fragment": 22109,
      "improvement overall": 52735,
      "fullyconnected layers": 45985,
      "layers finally": 61780,
      "tags provided": 117482,
      "tags combination": 117467,
      "dataset wall": 27276,
      "types layers": 126310,
      "hope public": 50635,
      "techniques nested": 119940,
      "way perform": 132117,
      "end formulate": 36809,
      "task prove": 118582,
      "key feature": 57569,
      "label entity": 58395,
      "sentence does": 105829,
      "method generalize": 68851,
      "challenging inherent": 16263,
      "3rd workshop": 1000,
      "german recent": 47909,
      "improve classical": 52347,
      "crfbased methods": 24762,
      "entities addition": 37742,
      "mechanism utilized": 68047,
      "utilized framework": 130542,
      "dominant approaches": 33929,
      "limited recurrent": 64269,
      "feedforward architectures": 43843,
      "explore local": 41558,
      "fuse local": 46224,
      "layers process": 61800,
      "sequential manner": 107235,
      "compatible different": 19597,
      "domains inspired": 33794,
      "choose different": 16926,
      "demonstrate good": 28750,
      "performance framework": 86391,
      "work heavily": 134549,
      "information leaving": 54727,
      "noisy annotations": 80089,
      "based weakly": 12157,
      "focusing large": 44926,
      "modules combined": 74521,
      "identify label": 51513,
      "manually correct": 67049,
      "correct label": 24105,
      "training specifically": 123874,
      "names difficult": 76142,
      "method recognize": 69094,
      "design objective": 29555,
      "sequence nested": 107021,
      "best path": 13389,
      "entities iteratively": 37800,
      "ones inner": 81690,
      "handling nested": 49416,
      "court decision": 24470,
      "order recognize": 82395,
      "similarities word": 109189,
      "entity candidates": 37911,
      "court decisions": 24471,
      "obtained competitive": 81354,
      "core tasks": 23334,
      "task years": 118859,
      "explicitly account": 41360,
      "recognize entities": 97051,
      "based syntax": 12104,
      "using treelstm": 130322,
      "field classifier": 43934,
      "confirm findings": 21014,
      "longdistance syntactic": 65152,
      "trees capture": 125685,
      "types dependency": 126273,
      "better better": 13520,
      "points stanford": 87876,
      "annotation training": 6398,
      "models quality": 73847,
      "new diagnostic": 78880,
      "supplementary materials": 115944,
      "method domain": 68786,
      "methods acquire": 69293,
      "function better": 46005,
      "possible spans": 88434,
      "lower performance": 65443,
      "systems data": 116829,
      "combined embeddings": 18653,
      "network layer": 77301,
      "improvements come": 52825,
      "techniques substantially": 119988,
      "tagger trained": 117361,
      "using crossentropy": 129517,
      "training tagger": 123901,
      "datasets analyze": 27309,
      "popular deep": 88090,
      "long sequence": 65090,
      "examine traditional": 39757,
      "token probability": 122261,
      "bottleneck existing": 14494,
      "directions improvement": 31843,
      "project page": 92222,
      "comprehensive summary": 20273,
      "biases originating": 13879,
      "depending corpus": 29270,
      "associated specific": 9608,
      "model answers": 70668,
      "silver lining": 109065,
      "corpora mitigate": 23529,
      "datasets f1": 27481,
      "comparable large": 19148,
      "datasets comprising": 27369,
      "news given": 79339,
      "systems allow": 116728,
      "interpretability analysis": 56218,
      "understand predictions": 126769,
      "domains english": 33774,
      "learning achieve": 62304,
      "context examine": 22086,
      "bert context": 13090,
      "annotators evaluate": 6491,
      "feasibility inferring": 43237,
      "able infer": 1659,
      "majority errors": 66599,
      "indicate current": 53835,
      "typing finegrained": 126474,
      "typing challenging": 126471,
      "finegrained types": 44391,
      "structures typical": 114113,
      "improving chinese": 53075,
      "study named": 114448,
      "sufficient context": 115275,
      "open situations": 81932,
      "facing challenges": 42814,
      "order explore": 82322,
      "explore impact": 41549,
      "models verify": 74329,
      "model discontinuous": 71002,
      "tagging techniques": 117456,
      "effectively recognize": 34841,
      "sacrificing accuracy": 103526,
      "entity candidate": 37910,
      "upstream task": 127830,
      "needs downstream": 76890,
      "approaches ignore": 8184,
      "network adaptive": 77128,
      "adaptive context": 3330,
      "contexts encode": 22389,
      "approaches experimental": 8152,
      "traditional named": 122845,
      "features strong": 43744,
      "exist languages": 40025,
      "crosslingual entity": 24951,
      "languages average": 60421,
      "manning 2009": 66967,
      "corpora achieving": 23401,
      "labeling systems": 58545,
      "perform reliably": 86055,
      "ideal conditions": 51345,
      "corrupted inputs": 24327,
      "perturbed input": 87199,
      "stability training": 112106,
      "noise model": 80059,
      "improved robustness": 52638,
      "models preserving": 73778,
      "embeddings label": 35771,
      "lowfrequency labels": 65469,
      "classes entity": 17061,
      "gillick et": 47966,
      "approach labeling": 7660,
      "ji et": 57231,
      "inspired fact": 55568,
      "f1 absolute": 42646,
      "improvement english": 52703,
      "work alleviate": 134368,
      "propose local": 92753,
      "virtual samples": 131747,
      "sampled training": 103573,
      "learning ensemble": 62549,
      "approach employed": 7519,
      "emnlp wnut2020": 36086,
      "learning ensembling": 62550,
      "final submission": 44128,
      "micro f1score": 70037,
      "effective fewshot": 34676,
      "based nearest": 11882,
      "learning structured": 63064,
      "uses supervised": 129286,
      "metalearning approaches": 68547,
      "cheap effective": 16687,
      "entity tags": 38109,
      "structured decoding": 113998,
      "special structure": 111367,
      "status classification": 113215,
      "depend handcrafted": 29048,
      "improvement 105": 52663,
      "designed capturing": 29586,
      "texts semantic": 121601,
      "semantic augmentation": 104997,
      "sparsity problems": 111260,
      "propose neuralbased": 92829,
      "text augmented": 120672,
      "respectively extensive": 101140,
      "studies datasets": 114202,
      "properties entities": 92451,
      "leveraging different": 63672,
      "information attentive": 54383,
      "dependencies parsing": 29105,
      "task held": 118253,
      "domain total": 33677,
      "seven teams": 107873,
      "does tell": 33406,
      "perform differently": 85981,
      "choices paper": 16918,
      "weaknesses current": 132207,
      "make easy": 66662,
      "run similar": 103464,
      "documentlevel features": 33142,
      "features named": 43616,
      "typically consider": 126418,
      "sentencelevel model": 106164,
      "ner offers": 77061,
      "features standard": 43742,
      "answering dialogue": 6624,
      "development training": 30434,
      "large labelled": 61125,
      "entity problem": 37987,
      "problem entities": 91030,
      "idea use": 51341,
      "datasets realworld": 27660,
      "surpasses prior": 116109,
      "understanding finegrained": 126845,
      "report introduces": 99011,
      "fulfill requirements": 45925,
      "requirements different": 100235,
      "academic industrial": 1858,
      "methods single": 69759,
      "multiple annotation": 75489,
      "acquisition bottleneck": 3046,
      "order measure": 82370,
      "identify training": 51568,
      "different thresholds": 31497,
      "designed special": 29624,
      "traditional ner": 122850,
      "training adding": 123348,
      "different circumstances": 31037,
      "architecture natural": 8704,
      "particularly task": 85500,
      "significant learning": 108821,
      "methods past": 69669,
      "downstream ner": 34012,
      "linking processes": 64668,
      "supervision lowresource": 115895,
      "supervision allows": 115865,
      "settings limited": 107819,
      "handannotated data": 49330,
      "tuning step": 125944,
      "annotation linguistic": 6331,
      "18 points": 420,
      "incomplete syntax": 53425,
      "finegrained annotation": 44334,
      "context implicit": 22127,
      "names long": 76147,
      "documents conduct": 33206,
      "annotation fundamental": 6317,
      "noisy clean": 80094,
      "approach selftraining": 7874,
      "framework boosting": 45444,
      "experiments general": 40952,
      "recognition instead": 96896,
      "rules challenging": 103386,
      "requires considerable": 100254,
      "new labeling": 78972,
      "nodes representing": 80039,
      "rules finally": 103403,
      "apply augmented": 7162,
      "improvement 20": 52672,
      "english german evaluation": 37151,
      "performance penn treebank": 86605,
      "branch natural language": 14559,
      "main methodological approaches": 66440,
      "stateoftheart conditional random": 112617,
      "neural networkbased language": 78424,
      "words known word": 134022,
      "neural networks key": 78496,
      "key challenges natural": 57547,
      "yield good performance": 135338,
      "languages work investigate": 60977,
      "handcrafted features domainspecific": 49340,
      "bidirectional lstms conditional": 13961,
      "lstms conditional random": 65709,
      "rely sources information": 98743,
      "sources information words": 110901,
      "information words characterbased": 55102,
      "words characterbased word": 133853,
      "representations learned supervised": 99730,
      "learned supervised corpus": 62263,
      "learned unannotated corpora": 62272,
      "novel technique called": 80748,
      "model trained labeled": 72196,
      "f1 scores range": 42695,
      "large indomain training": 61115,
      "directly paper propose": 31897,
      "small indomain training": 110161,
      "sequence labelling problem": 106995,
      "achieves competitive accuracy": 2768,
      "competitive accuracy stateoftheart": 19634,
      "problem deep neural": 90999,
      "specific named entities": 111467,
      "reported results standard": 99072,
      "typically requires large": 126458,
      "labeled data work": 58451,
      "deep learning combined": 28263,
      "computationally expensive requires": 20454,
      "nearly stateoftheart performance": 76737,
      "performance standard datasets": 86734,
      "stateofthe art systems": 112564,
      "best knowledge use": 13355,
      "external word embeddings": 42051,
      "models evaluated datasets": 73174,
      "encoding input tokens": 36702,
      "data sets chinese": 26419,
      "domains experimental results": 33779,
      "evaluation campaign report": 39140,
      "labeling problem using": 58527,
      "word wordshape features": 133663,
      "brownclusterbased features wordembeddingbased": 14724,
      "features wordembeddingbased features": 43789,
      "unified neural network": 127125,
      "training corpus paper": 123403,
      "propose novel effective": 92916,
      "entity type information": 38117,
      "proposed model experimental": 93446,
      "recognition ner named": 96935,
      "speech named entity": 111712,
      "approach disadvantages error": 7495,
      "disadvantages error propagation": 31933,
      "present study endtoend": 89724,
      "experimental results endtoend": 40613,
      "improved neural network": 52622,
      "feature vector representing": 43327,
      "new stateoftheart f1": 79159,
      "neural models allow": 78165,
      "additional human effort": 3522,
      "language processing currently": 59858,
      "crossdomain named entity": 24895,
      "efforts shown neural": 35207,
      "stateoftheart results standard": 112920,
      "models performance model": 73733,
      "domains social media": 33859,
      "neural ner model": 78227,
      "model new domain": 71579,
      "extensive empirical studies": 41872,
      "nlp tasks relation": 79761,
      "use external knowledge": 128044,
      "word embeddings long": 133152,
      "embeddings long shortterm": 35792,
      "use entity linking": 128032,
      "model entity linking": 71088,
      "entity linking model": 37957,
      "model knowledge base": 71395,
      "foundation natural language": 45350,
      "achieving good performance": 2949,
      "existing deep learning": 40108,
      "learned source domain": 62260,
      "model trained target": 72212,
      "difference source target": 30947,
      "competitive result compared": 19678,
      "model size furthermore": 72056,
      "machine translation outputs": 66138,
      "train test data": 123038,
      "word vectors language": 133639,
      "target task data": 117722,
      "vectors pretrained word": 131464,
      "annotated datasets expensive": 6179,
      "new approach named": 78792,
      "different annotation schemes": 30993,
      "dataset propose use": 27115,
      "learn neural network": 62100,
      "models model based": 73575,
      "entity recognition english": 38012,
      "downstream tasks ner": 34046,
      "dataset wall street": 27277,
      "hope public release": 50636,
      "multitask neural network": 75884,
      "german recent advances": 47910,
      "language models improve": 59567,
      "improve ner performance": 52434,
      "ner performance experimental": 77065,
      "current stateoftheart baselines": 25353,
      "complex recurrent neural": 19869,
      "relying external resources": 98758,
      "model performance different": 71726,
      "results demonstrate good": 101683,
      "demonstrate good performance": 28751,
      "design objective function": 29556,
      "objective function training": 81085,
      "random field based": 95476,
      "tree structures model": 125616,
      "random field classifier": 95477,
      "entity types based": 38119,
      "sentence syntactic relations": 106098,
      "entity types dependency": 38120,
      "proposed model improving": 93452,
      "errors hard impossible": 38374,
      "flair elmo bert": 44633,
      "automatically annotated data": 10720,
      "datasets perform poorly": 27620,
      "models conditional random": 72954,
      "techniques substantially improve": 119989,
      "pretrained embeddings word": 90029,
      "using crossentropy loss": 129518,
      "multiple tasks datasets": 75724,
      "entity typing finegrained": 38129,
      "typing finegrained entity": 126475,
      "entity typing challenging": 38126,
      "paper introduce corpus": 83990,
      "study named entity": 114449,
      "pretrained model achieved": 90129,
      "achieved promising performance": 2671,
      "ner data sets": 77035,
      "detection task paper": 30078,
      "approaches experimental results": 8153,
      "traditional named entity": 122846,
      "performance recent work": 86658,
      "work demonstrated utility": 134463,
      "lowresource languages challenging": 65510,
      "exist languages address": 40026,
      "languages address problem": 60393,
      "crosslingual entity linking": 24952,
      "experiments lowresource languages": 40992,
      "recognition dependency parsing": 96856,
      "evaluation corpora achieving": 39162,
      "sequence labeling systems": 106983,
      "experiments english german": 40934,
      "gillick et al": 47967,
      "al 2018 paper": 4635,
      "2018 paper present": 667,
      "ji et al": 57232,
      "based data augmentation": 11621,
      "ner named entity": 77058,
      "data experiments conducted": 25925,
      "benchmarks demonstrate effectiveness": 12898,
      "publicly released code": 94333,
      "contextualised word embeddings": 22535,
      "handcrafted linguistic features": 49352,
      "overall accuracy improvement": 83212,
      "data sparsity problems": 26487,
      "datasets collected english": 27352,
      "collected english chinese": 18420,
      "english chinese social": 37092,
      "previous studies datasets": 90492,
      "universal dependencies parsing": 127288,
      "shared task held": 108052,
      "choices paper present": 16919,
      "features named entity": 43617,
      "paper perform comparative": 84067,
      "perform comparative evaluation": 85966,
      "document context present": 32980,
      "question answering dialogue": 95041,
      "entity types second": 38123,
      "knowledge acquisition bottleneck": 57743,
      "architecture natural language": 8705,
      "distant supervision allows": 32461,
      "propose finegrained annotation": 92678,
      "finegrained annotation scheme": 44335,
      "deep learning shown": 28316,
      "effectiveness method code": 34904,
      "entity recognition instead": 38020,
      "unlabeled data create": 127380,
      "branch natural language processing": 14560,
      "words known word embeddings": 134023,
      "key challenges natural language": 57548,
      "based bidirectional lstms conditional": 11554,
      "bidirectional lstms conditional random": 13962,
      "lstms conditional random fields": 65710,
      "rely sources information words": 98744,
      "sources information words characterbased": 110902,
      "information words characterbased word": 55103,
      "words characterbased word representations": 133854,
      "characterbased word representations learned": 16506,
      "word representations learned supervised": 133460,
      "representations learned supervised corpus": 99731,
      "small indomain training data": 110162,
      "best reported results standard": 13428,
      "reported results standard benchmark": 99073,
      "requires large amounts labeled": 100286,
      "stateoftheart performance standard datasets": 112847,
      "recognition named entity recognition": 96915,
      "sequence labeling problem using": 106979,
      "brownclusterbased features wordembeddingbased features": 14725,
      "paper propose novel effective": 84294,
      "entity recognition ner named": 38043,
      "approach disadvantages error propagation": 7496,
      "paper present study endtoend": 84139,
      "performance extensive experiments benchmark": 86368,
      "natural language processing currently": 76414,
      "crossdomain named entity recognition": 24896,
      "nlp tasks relation extraction": 79762,
      "recurrent neural networks stateoftheart": 97249,
      "word embeddings long shortterm": 133153,
      "embeddings long shortterm memory": 35793,
      "shortterm memory lstm layers": 108331,
      "manually annotated datasets expensive": 67030,
      "named entity recognition english": 76066,
      "dataset wall street journal": 27278,
      "entity recognition paper presents": 38056,
      "improve ner performance experimental": 52435,
      "ner performance experimental results": 77066,
      "complex recurrent neural networks": 19870,
      "experimental results demonstrate good": 40590,
      "results demonstrate good performance": 101684,
      "conditional random field based": 20776,
      "conditional random field classifier": 20777,
      "named entity recognition dependency": 76063,
      "work propose simple effective": 134742,
      "finegrained entity typing finegrained": 44354,
      "entity typing finegrained entity": 38130,
      "typing finegrained entity typing": 126476,
      "finegrained entity typing challenging": 44353,
      "traditional named entity recognition": 122847,
      "entity recognition dependency parsing": 38010,
      "al 2018 paper present": 4636,
      "recognition social media texts": 97002,
      "datasets collected english chinese": 27353,
      "collected english chinese social": 18421,
      "english chinese social media": 37093,
      "chinese social media platforms": 16822,
      "features named entity recognition": 43618,
      "language processing information extraction": 59872,
      "propose finegrained annotation scheme": 92679,
      "demonstrate effectiveness method code": 28713,
      "named entity recognition instead": 76071,
      "component natural language processing nlp": 19992,
      "key challenges natural language processing": 57549,
      "based bidirectional lstms conditional random": 11555,
      "bidirectional lstms conditional random fields": 13963,
      "rely sources information words characterbased": 98745,
      "sources information words characterbased word": 110903,
      "information words characterbased word representations": 55104,
      "words characterbased word representations learned": 133855,
      "characterbased word representations learned supervised": 16507,
      "word representations learned supervised corpus": 133461,
      "best reported results standard benchmark": 13429,
      "entity recognition named entity recognition": 38027,
      "recognition named entity recognition ner": 96916,
      "chinese named entity recognition ner": 16791,
      "task sequence labeling problem using": 118699,
      "named entity recognition ner named": 76089,
      "performance extensive experiments benchmark datasets": 86369,
      "bidirectional recurrent neural network rnn": 13972,
      "approach significantly outperforms stateoftheart methods": 7915,
      "word embeddings long shortterm memory": 133154,
      "long shortterm memory lstm layers": 65117,
      "named entity recognition paper presents": 76099,
      "improve ner performance experimental results": 52436,
      "experimental results demonstrate good performance": 40591,
      "experiments benchmark datasets demonstrate method": 40797,
      "finegrained entity typing finegrained entity": 44355,
      "entity typing finegrained entity typing": 38131,
      "named entity recognition dependency parsing": 76064,
      "datasets collected english chinese social": 27354,
      "collected english chinese social media": 18422,
      "english chinese social media platforms": 37094,
      "models natural language processing tasks": 73612,
      "natural language processing information extraction": 76426,
      "graphics": 48936,
      "wordsasclassifiers": 134326,
      "initialise": 55232,
      "mscoco": 74946,
      "conquer": 21158,
      "automaticallygenerated": 10841,
      "videototext": 131666,
      "modulates": 74473,
      "attenuate": 10130,
      "amplifying": 5378,
      "abc": 1468,
      "doubly": 33970,
      "lightens": 64002,
      "173": 406,
      "visionandlanguage": 131771,
      "vln": 131854,
      "genome": 47825,
      "moderatesized": 74377,
      "vectorial": 131402,
      "heat": 49646,
      "pan": 83688,
      "olive": 81647,
      "fitness": 44584,
      "denotational": 29004,
      "orthogonality": 82591,
      "beneath": 12944,
      "balls": 11408,
      "kstep": 58354,
      "visuallygrounded": 131846,
      "reshaping": 100733,
      "gendermarked": 46623,
      "cmcl": 17974,
      "nonpaired": 80261,
      "unambiguously": 126580,
      "commonsenserelated": 19015,
      "reversible": 102530,
      "combining use": 18756,
      "cohesive coherent": 18335,
      "models user": 74295,
      "task interactive": 118304,
      "document intended": 33026,
      "intended support": 55871,
      "framework multimodal": 45617,
      "vision natural": 131763,
      "descriptions train": 29498,
      "bilinear model": 14018,
      "descriptions given": 29483,
      "considerably simpler": 21279,
      "grounded representations": 49110,
      "visual input": 131797,
      "wordsasclassifiers model": 134327,
      "model common": 70847,
      "language refer": 60028,
      "phrase composition": 87341,
      "words recently": 134159,
      "improve statistical": 52551,
      "visual space": 131812,
      "point strong": 87815,
      "text technique": 121363,
      "algorithm finds": 4741,
      "information vector": 55086,
      "representation present": 99381,
      "common modeling": 18894,
      "outperform recent": 82735,
      "learning benchmarks": 62403,
      "translation latent": 124895,
      "variable capture": 130837,
      "underlying semantic": 126698,
      "conducted englishgerman": 20920,
      "baseline incorporating": 12233,
      "sentence ii": 105899,
      "impact adding": 51856,
      "brings additional": 14642,
      "set multilingual": 107501,
      "data meaning": 26121,
      "meaning model": 67649,
      "quality end": 94643,
      "nmt task": 79985,
      "multilingual signals": 75370,
      "used reranking": 128728,
      "reranking nbest": 100375,
      "efficiently exploit": 35143,
      "corpora report": 23574,
      "decoding enhance": 28090,
      "decoder uses": 28059,
      "recently effectiveness": 96681,
      "usually include": 130427,
      "concatenation paper": 20542,
      "approach multimodal": 7719,
      "improvement automatic": 52682,
      "descriptions generated": 29482,
      "free errors": 45756,
      "gender identification": 46599,
      "observing gains": 81245,
      "set identify": 107460,
      "report different": 98988,
      "machine translating": 65878,
      "improving multimodal": 53118,
      "translation multimodal": 124991,
      "cnn extract": 18003,
      "linguistic visual": 64581,
      "features convolutional": 43428,
      "visual context": 131788,
      "output instead": 83083,
      "text investigating": 121064,
      "unsupervised speech": 127722,
      "speech pattern": 111720,
      "english available": 37074,
      "application image": 6855,
      "lack fluency": 58708,
      "module utilize": 74517,
      "fluency scores": 44700,
      "improves fluency": 52984,
      "submissions shared": 114752,
      "key difference": 57564,
      "evaluating accuracy": 39034,
      "difficult given": 31630,
      "given video": 48171,
      "video clip": 131656,
      "answer measure": 6542,
      "captioning task": 15265,
      "task 2016": 117819,
      "2016 shown": 619,
      "multimodal representations": 75452,
      "paper sheds": 84425,
      "larger units": 61386,
      "goal information": 48358,
      "types unstructured": 126381,
      "data images": 26016,
      "images video": 51815,
      "shortcoming propose": 108277,
      "attributevalue pairs": 10207,
      "baselines demonstrate": 12384,
      "relative effectiveness": 98350,
      "grounding problem": 49122,
      "results transfer": 102281,
      "visual grounding": 131795,
      "exploit specific": 41443,
      "words strong": 134242,
      "attending source": 9784,
      "model visual": 72309,
      "analysis pipeline": 5751,
      "pipeline finally": 87539,
      "different test": 31486,
      "multimodal model": 75442,
      "association pairs": 9626,
      "inconsistent incomplete": 53442,
      "syntax lexical": 116540,
      "textual contexts": 121681,
      "annotated named": 6211,
      "incorporates relevant": 53513,
      "information generic": 54639,
      "informative ones": 55144,
      "attention significantly": 10009,
      "provided visual": 93991,
      "visual contexts": 131789,
      "combine neural": 18634,
      "objective produce": 81107,
      "poses problem": 88272,
      "sampling small": 103610,
      "subset possible": 114834,
      "decisions finally": 27959,
      "method testing": 69186,
      "pragmatic speaker": 88757,
      "baseline wordlevel": 12340,
      "networks long": 77646,
      "significantly various": 109055,
      "descriptions target": 29497,
      "scenario data": 103823,
      "locations dataset": 64974,
      "attentive transformer": 10129,
      "results englishgerman": 101757,
      "visual attention": 131785,
      "model multimodal": 71540,
      "translator model": 125511,
      "collected new": 18428,
      "online shopping": 81802,
      "dataset visual": 27275,
      "semantic attention": 104996,
      "attention time": 10023,
      "according generated": 1977,
      "datasets reaches": 27659,
      "reaches stateoftheart": 95920,
      "wmt 2018": 132774,
      "learning lead": 62683,
      "adapting transformer": 3326,
      "architecture multimodal": 8697,
      "features small": 43733,
      "year propose": 135239,
      "convolutional features": 23157,
      "refine using": 97590,
      "information final": 54615,
      "submissions ranked": 114751,
      "semantics using": 105479,
      "linked language": 64645,
      "comparing sparse": 19519,
      "interpretable linguistic": 56241,
      "descriptions language": 29485,
      "improvements additional": 52802,
      "recent findings": 96462,
      "presented following": 89783,
      "attention years": 10043,
      "automated metric": 10455,
      "study architecture": 114325,
      "tasks multimodal": 119313,
      "instructional videos": 55707,
      "obtain deeper": 81274,
      "seen multimodal": 104535,
      "available test": 11125,
      "including multitask": 53327,
      "comparison assessing": 19528,
      "techniques present": 119952,
      "main feature": 66419,
      "baselines models": 12431,
      "word endings": 133262,
      "alignments distributed": 5012,
      "models attempt": 72781,
      "attempt incorporate": 9747,
      "features yielding": 43796,
      "25 relative": 818,
      "general case": 46638,
      "quality image": 94682,
      "way integrated": 132098,
      "text captions": 120693,
      "common embedding": 18868,
      "comparable current": 19141,
      "retrieval benchmark": 102384,
      "content resulting": 21936,
      "importance visual": 52090,
      "translation embedding": 124780,
      "deeply understand": 28465,
      "relations natural": 98217,
      "translating rare": 124602,
      "combine approaches": 18616,
      "improvements 124": 52788,
      "bleu achieve": 14276,
      "novel crowdsourced": 80521,
      "crowdsourced resource": 25113,
      "relations like": 98202,
      "penn discourse": 85858,
      "discourse treebank": 32095,
      "treebank new": 125641,
      "natural communication": 76242,
      "findings implications": 44302,
      "debiasing word": 27904,
      "embeddings explored": 35714,
      "debiasing techniques": 27903,
      "seldom studied": 104685,
      "multimodal transformer": 75462,
      "language sequences": 60083,
      "mixture natural": 70426,
      "timeseries data": 122196,
      "modalities paper": 70480,
      "attention attends": 9795,
      "distinct time": 32547,
      "timeseries model": 122197,
      "visionandlanguage navigation": 131772,
      "navigation vln": 76684,
      "propose decompose": 92614,
      "procedure set": 91399,
      "access different": 1906,
      "vln task": 131855,
      "inference text": 54237,
      "images present": 51810,
      "image tasks": 51797,
      "symbolic information": 116314,
      "based inputs": 11773,
      "specific cases": 111414,
      "example presence": 39791,
      "consequence models": 21173,
      "draft translation": 34082,
      "englishhindi machine": 37409,
      "wat 2019": 132042,
      "translations address": 125451,
      "using state": 130213,
      "used multimodal": 128643,
      "assess utility": 9496,
      "text does": 120890,
      "provides excellent": 94026,
      "provides stronger": 94083,
      "possible computational": 88393,
      "far approaches": 43069,
      "embeddings spoken": 35950,
      "work furthermore": 134542,
      "encoding word": 36735,
      "translation growing": 124847,
      "language grounding": 59113,
      "image remains": 51791,
      "works partially": 134968,
      "recent promising": 96503,
      "adversarial architectures": 4100,
      "architectures evaluate": 8805,
      "reconstruction auxiliary": 97115,
      "task helps": 118255,
      "based visual": 12155,
      "tokens input": 122313,
      "furthermore jointly": 46185,
      "features finegrained": 43518,
      "olive oil": 81648,
      "information guided": 54650,
      "benchmark report": 12852,
      "features exploited": 43501,
      "explored visual": 41639,
      "used denote": 128481,
      "approaches prior": 8288,
      "context capture": 22023,
      "bi directional": 13782,
      "directional lstm": 31833,
      "sentences certain": 106235,
      "descriptions derived": 29478,
      "derived structural": 29363,
      "insights performance": 55542,
      "better characterization": 13534,
      "works related": 134972,
      "reviews major": 102617,
      "resources tasks": 101054,
      "endtoend pipeline": 36953,
      "targeted evaluations": 117767,
      "task drawn": 118122,
      "years evaluation": 135260,
      "surface similarity": 116079,
      "assessing models": 9515,
      "set synthetic": 107597,
      "construct diagnostic": 21626,
      "diagnostic datasets": 30511,
      "generation image": 47431,
      "paper contributes": 83788,
      "extending dataset": 41816,
      "dataset vietnamese": 27274,
      "active field": 3108,
      "representations focus": 99658,
      "module consider": 74482,
      "unlike approaches": 127426,
      "decoding sequence": 28131,
      "varying difficulties": 131258,
      "simpler datasets": 109554,
      "translation improvement": 124860,
      "suggests difficult": 115371,
      "estimate model": 38631,
      "proposed lookahead": 93325,
      "combination visual": 18594,
      "largely untested": 61351,
      "representations produces": 99828,
      "inference compared": 54125,
      "representations improvement": 99684,
      "computing similarity": 20530,
      "complex multimodal": 19838,
      "information decoding": 54467,
      "analyzed effect": 6025,
      "media blogs": 68085,
      "generation translation": 47686,
      "end neural": 36818,
      "using encoderdecoder": 129625,
      "score generated": 104074,
      "data efficacy": 25879,
      "learning unfortunately": 63132,
      "value measuring": 130782,
      "contemporary nlp": 21851,
      "analysis guide": 5636,
      "guide future": 49236,
      "grounded neural": 49109,
      "specific contribution": 111422,
      "shi et": 108171,
      "recent approach": 96428,
      "components various": 20053,
      "task boost": 117943,
      "trained pairs": 123231,
      "using gating": 129691,
      "perform statistical": 86077,
      "analysis known": 5667,
      "effects human": 34987,
      "prior segmentation": 90729,
      "build lexicon": 14782,
      "topdown approach": 122488,
      "learn reliable": 62124,
      "mapping provided": 67144,
      "networks architecture": 77510,
      "segments used": 104678,
      "lowlevel highlevel": 65476,
      "challenge 2020": 15988,
      "highquality text": 50409,
      "video text": 131661,
      "semantic correspondences": 105025,
      "multimodal semantic": 75456,
      "learn node": 62102,
      "product attribute": 92026,
      "usually incomplete": 130428,
      "product attributes": 92027,
      "attributes values": 10203,
      "easier extract": 34414,
      "attributes given": 10197,
      "attribute value": 10186,
      "text read": 121227,
      "seek understand": 104519,
      "context helpful": 22121,
      "improvement low": 52720,
      "datasets train": 27763,
      "world population": 135043,
      "utilizing sentence": 130579,
      "scene graph": 103877,
      "graph encoding": 48803,
      "recognition quality": 96976,
      "grounding speech": 49124,
      "entities masked": 37815,
      "rely global": 98698,
      "possible recover": 88422,
      "automatic object": 10620,
      "context shown": 22267,
      "useful automatic": 128864,
      "corpus multimodal": 23889,
      "noisy scenarios": 80115,
      "proposed incorporate": 93315,
      "multimodal interactions": 75436,
      "datasets popular": 27623,
      "translation tackle": 125325,
      "image essential": 51783,
      "corpus benchmarks": 23681,
      "architecture new": 8712,
      "uses existing": 129228,
      "effectively adapt": 34785,
      "multimodal understanding": 75464,
      "improve capability": 52345,
      "unified semantic": 127128,
      "space corpus": 110986,
      "representations textual": 99933,
      "problem simultaneous": 91233,
      "available textual": 11127,
      "information agent": 54370,
      "construct largest": 21637,
      "10 billion": 123,
      "100 billion": 159,
      "image generation": 51786,
      "subtitles dataset": 114993,
      "dataset valuable": 27268,
      "possible text": 88437,
      "text fragment": 120964,
      "fragment sentence": 45373,
      "humans evaluate": 51078,
      "image selection": 51795,
      "automatic tasks": 10687,
      "useful complement": 128869,
      "quality introducing": 94692,
      "irrelevant text": 56952,
      "text remaining": 121239,
      "processing visual": 91851,
      "robust accurate": 103014,
      "accurate natural": 2345,
      "modeling texts": 72566,
      "input second": 55420,
      "textonly baselines": 121445,
      "learning zeroshot": 63176,
      "outside world": 83201,
      "highly beneficial": 50294,
      "range benchmarks": 95547,
      "computer vision natural": 20497,
      "vision natural language": 131764,
      "present simple model": 89709,
      "language model produce": 59383,
      "models achieves comparable": 72699,
      "representation word sentence": 99466,
      "human language learning": 50899,
      "use language refer": 128112,
      "pretrained convolutional neural": 90017,
      "neural network extract": 78299,
      "reference resolution task": 97530,
      "approach does depend": 7504,
      "improvements bleu point": 52817,
      "bleu point strong": 14301,
      "visual textual representations": 131816,
      "yields high quality": 135412,
      "high quality models": 50121,
      "models outperform recent": 73681,
      "machine translation latent": 66026,
      "generates target word": 47186,
      "parts source sentence": 85585,
      "latent variable capture": 61626,
      "experiments conducted englishgerman": 40841,
      "task model outperforms": 118397,
      "nmt models incorporate": 79915,
      "features different parts": 43458,
      "image features extracted": 51785,
      "decoder hidden state": 28019,
      "neural model significantly": 78153,
      "data set multilingual": 26412,
      "function training objective": 46049,
      "reranking nbest lists": 100376,
      "machine translation attention": 65900,
      "translation attention mechanism": 124652,
      "error analysis recent": 38283,
      "translation task english": 125332,
      "network cnn extract": 77193,
      "according automatic evaluation": 1967,
      "data set paper": 26413,
      "texttospeech tts synthesis": 121663,
      "conventional machine translation": 22884,
      "paper introduce simple": 84002,
      "correct answer measure": 24092,
      "paper sheds light": 84426,
      "address shortcoming propose": 3765,
      "contextual information entity": 22471,
      "representations different modalities": 99596,
      "novel dynamic fusion": 80552,
      "word association pairs": 132928,
      "sampling small subset": 103611,
      "neural networks long": 78507,
      "novel annotation scheme": 80465,
      "multihead attention layer": 75124,
      "achieves competitive stateoftheart": 2777,
      "product description dataset": 92032,
      "using different architectures": 129568,
      "transformer neural machine": 124364,
      "wmt18 shared task": 132851,
      "presented following sections": 89784,
      "natural language emerging": 76292,
      "obtain deeper understanding": 81275,
      "model latent variable": 71420,
      "visual textual inputs": 131815,
      "training does require": 123591,
      "including multitask learning": 53328,
      "translation task present": 125341,
      "task present submission": 118550,
      "attention neural attention": 9954,
      "network rnn language": 77418,
      "language model features": 59322,
      "datasets used prior": 27778,
      "analysis provides insights": 5786,
      "generate better translations": 46908,
      "sentence encoder trained": 105850,
      "sentence embeddings deep": 105839,
      "common embedding space": 18869,
      "results comparable current": 101591,
      "comparable current stateoftheart": 19142,
      "semantic content resulting": 105019,
      "demonstrate importance visual": 28757,
      "importance visual information": 52091,
      "data resulting poor": 26358,
      "resulting poor performance": 101465,
      "translating rare words": 124603,
      "rare word translation": 95748,
      "discourse relations paper": 32079,
      "penn discourse treebank": 85859,
      "recent years pretrained": 96629,
      "word embeddings explored": 133112,
      "word embeddings high": 133123,
      "embeddings high dimensional": 35740,
      "techniques proposed address": 119959,
      "proposed method gives": 93370,
      "high quality image": 50118,
      "mixture natural language": 70427,
      "empirical analysis suggests": 36152,
      "visionandlanguage navigation vln": 131773,
      "sentence representations paper": 106042,
      "representations paper use": 99798,
      "models tend learn": 74173,
      "englishhindi machine translation": 37410,
      "using state art": 130214,
      "noise input text": 80056,
      "sentence representations obtained": 106040,
      "representations obtained using": 99788,
      "textual similarity experiments": 121739,
      "transformer models achieve": 124347,
      "previous work furthermore": 90531,
      "image remains challenging": 51792,
      "previous works partially": 90576,
      "reconstruction auxiliary task": 97116,
      "auxiliary task helps": 10887,
      "conducting extensive experiments": 20964,
      "translation performance transformer": 125117,
      "words used denote": 134292,
      "paper explore model": 83919,
      "dataset results demonstrate": 27168,
      "bi directional lstm": 13783,
      "study different types": 114358,
      "recognition image captioning": 96890,
      "recent years evaluation": 96611,
      "task remains challenging": 118635,
      "image captioning task": 51772,
      "task automatic generation": 117910,
      "performance simpler datasets": 86718,
      "machine translation improvement": 66005,
      "combination visual textual": 18595,
      "present novel task": 89625,
      "task significantly challenging": 118716,
      "social media blogs": 110354,
      "given input sentence": 48044,
      "learning methods results": 62746,
      "guide future work": 49237,
      "shi et al": 108172,
      "supervision target task": 115925,
      "model model trained": 71522,
      "perform statistical analysis": 86078,
      "information present simple": 54860,
      "present simple way": 89710,
      "machine translation challenge": 65919,
      "learn node representations": 62103,
      "experimental results indepth": 40621,
      "results indepth analysis": 101853,
      "practical applications paper": 88702,
      "code dataset released": 18093,
      "method consists phases": 68733,
      "speech recognition quality": 111770,
      "propose model uses": 92788,
      "previous work demonstrated": 90526,
      "analysis shows models": 5853,
      "poorly understood work": 88073,
      "context information obtained": 22144,
      "document understanding tasks": 33102,
      "selfattention mechanism transformer": 104887,
      "mechanism transformer architecture": 68043,
      "margin achieves new": 67186,
      "free text corpus": 45766,
      "unified semantic space": 127129,
      "space experimental results": 111002,
      "tasks code pretrained": 118982,
      "translation models demonstrate": 124953,
      "paper introduces largescale": 84008,
      "dataset aims facilitate": 26729,
      "dataset valuable resource": 27269,
      "evaluate quality automatic": 38905,
      "human evaluation automatic": 50802,
      "crossmodal attention module": 25063,
      "translation quality paper": 125182,
      "quality paper proposes": 94746,
      "accurate natural language": 2346,
      "language modeling classification": 59427,
      "wide range benchmarks": 132500,
      "computer vision natural language": 20498,
      "vision natural language processing": 131765,
      "paper present simple model": 84137,
      "pretrained convolutional neural network": 90018,
      "convolutional neural network extract": 23189,
      "improvements bleu point strong": 52818,
      "neural machine translation latent": 78015,
      "relevant parts source sentence": 98569,
      "neural machine translation attention": 77972,
      "machine translation attention mechanism": 65901,
      "neural network cnn extract": 78283,
      "paper introduce simple effective": 84003,
      "propose novel dynamic fusion": 92915,
      "machine translation introduce novel": 66014,
      "transformer neural machine translation": 124365,
      "neural network rnn language": 78390,
      "network rnn language model": 77419,
      "embeddings deep neural networks": 35634,
      "demonstrate importance visual information": 28758,
      "training data resulting poor": 123528,
      "word embeddings high dimensional": 133124,
      "sentence representations obtained using": 106041,
      "semantic textual similarity experiments": 105331,
      "models perform significantly better": 73729,
      "image remains challenging task": 51793,
      "speech recognition image captioning": 111755,
      "yields better results using": 135403,
      "experimental results indepth analysis": 40622,
      "large margin achieves new": 61140,
      "margin achieves new stateoftheart": 67187,
      "tasks code pretrained models": 118983,
      "accurate natural language processing": 2347,
      "computer vision natural language processing": 20499,
      "attentionbased neural machine translation nmt": 10088,
      "neural machine translation attention mechanism": 77973,
      "convolutional neural network cnn extract": 23184,
      "recurrent neural network rnn language": 97217,
      "neural network rnn language model": 78391,
      "large margin achieves new stateoftheart": 61141,
      "margin achieves new stateoftheart results": 67188,
      "markedup": 67228,
      "repairing": 98905,
      "rfc": 102704,
      "respected": 101112,
      "presumed": 89964,
      "contextdriven": 22358,
      "componentwise": 20055,
      "matlab": 67466,
      "configured": 21009,
      "connective": 21151,
      "headroom": 49602,
      "labellers": 58564,
      "concession": 20697,
      "unsupported": 127760,
      "drss": 34192,
      "drs": 34191,
      "signaling": 108698,
      "signalling": 108699,
      "rnngs": 102951,
      "leftbranching": 63301,
      "ckystyle": 17004,
      "identifiability": 51357,
      "fruitfully": 45906,
      "tapping": 117579,
      "ensembled": 37624,
      "nuclearity": 80826,
      "sentimentannotated": 106824,
      "method discovering": 68780,
      "aid identifying": 4434,
      "earlier approaches": 34377,
      "theoretically sound": 121806,
      "fragment size": 45374,
      "linear text": 64379,
      "paper aimed": 83724,
      "segmentation procedure": 104623,
      "applied corpus": 7053,
      "genres results": 47838,
      "obtained application": 81342,
      "anaphora discourse": 6062,
      "structure argue": 113810,
      "structure conclude": 113824,
      "discourse interpretation": 32062,
      "reference encoding": 97515,
      "access multiple": 1919,
      "tools perform": 122465,
      "textbased machine": 121426,
      "needs improvement": 76893,
      "primitive discourse": 90663,
      "technique similarity": 119815,
      "linear sequence": 64371,
      "effort create": 35164,
      "achieves encouraging": 2781,
      "encouraging performance": 36779,
      "elements discourse": 35278,
      "ml approaches": 70433,
      "empirical support": 36210,
      "doubly annotated": 33971,
      "taxonomy text": 119669,
      "problems proposed": 91362,
      "multilevel language": 75191,
      "dynamic patterns": 34320,
      "patterns data": 85736,
      "series tools": 107284,
      "features centered": 43390,
      "single token": 109803,
      "token predict": 122259,
      "tokens propose": 122326,
      "reduction stateoftheart": 97458,
      "performance approaching": 86148,
      "frameworks using": 45747,
      "universal linguistic": 127311,
      "linguistic events": 64472,
      "classification annotation": 17115,
      "context translation": 22306,
      "markers languages": 67232,
      "discourse marker": 32067,
      "language rhetorical": 60063,
      "variety research": 131013,
      "parsing focused": 85114,
      "adapts models": 3345,
      "sets various": 107725,
      "work described": 134467,
      "articles essays": 9185,
      "segmentation language": 104596,
      "tasks problem": 119402,
      "better supervised": 13734,
      "approaches special": 8346,
      "focus given": 44770,
      "approaches supervised": 8365,
      "difficult text": 31665,
      "compiled small": 19718,
      "induction approach": 54020,
      "meanings relations": 67748,
      "relation analysis": 97938,
      "encoder generate": 36529,
      "discourse semantic": 32088,
      "attention semantic": 10002,
      "generates discourse": 47159,
      "models approximate": 72764,
      "standard stochastic": 112309,
      "performance readily": 86654,
      "small label": 110164,
      "set work": 107640,
      "neural chinese": 77858,
      "results hold": 101831,
      "shallow discourse": 107917,
      "optimization paper": 82202,
      "shortterm memories": 108314,
      "combine representations": 18639,
      "driven language": 34154,
      "previous proposals": 90444,
      "knowledge modeled": 58064,
      "predicates entities": 88864,
      "standard ngram": 112283,
      "quality semantic": 94786,
      "microblog texts": 70048,
      "understanding information": 126861,
      "argumentative structure": 8972,
      "annotated paper": 6218,
      "present best": 89395,
      "crosslingual discourse": 24939,
      "task discourse": 118102,
      "discourse connective": 32045,
      "task human": 118259,
      "languages tool": 60921,
      "step develop": 113258,
      "theory rst": 121825,
      "manually segmented": 67075,
      "texts results": 121595,
      "domains typically": 33879,
      "segment boundaries": 104559,
      "assuming gold": 9658,
      "token segmentation": 122264,
      "highquality syntactic": 50407,
      "heuristics generally": 49884,
      "rely gold": 98699,
      "language fully": 59056,
      "f1 english": 42653,
      "automatic disambiguation": 10516,
      "used convey": 128454,
      "disambiguation english": 31963,
      "networks discourse": 77570,
      "bagofwords approach": 11369,
      "measuring sentiment": 67931,
      "according hierarchical": 1979,
      "scores individual": 104181,
      "learn resulting": 62131,
      "salient passages": 103552,
      "different frameworks": 31156,
      "frameworks makes": 45746,
      "corpora unified": 23608,
      "labels different": 58589,
      "identify sources": 51557,
      "future annotation": 46257,
      "generated statistical": 47126,
      "achieves recall": 2843,
      "classifier identify": 17548,
      "argument labeling": 8946,
      "networks argument": 77511,
      "model argument": 70686,
      "multiple configurations": 75524,
      "task shallow": 118705,
      "complementary approaches": 19738,
      "standard machine": 112253,
      "achieves f1score": 2788,
      "framework develop": 45492,
      "f1 identification": 42657,
      "parsing long": 85146,
      "long treated": 65146,
      "units use": 127268,
      "external parsers": 42036,
      "discourse levels": 32066,
      "like words": 64115,
      "spans challenging": 111197,
      "meaning text": 67700,
      "methods exploited": 69495,
      "exploited syntactic": 41454,
      "treelstm model": 125677,
      "arguments relation": 8994,
      "relation leverage": 98052,
      "composition process": 20101,
      "analyzing behavior": 6036,
      "like time": 64104,
      "seen kind": 104533,
      "measures extracted": 67864,
      "measures natural": 67882,
      "complex dynamic": 19810,
      "improving implicit": 53104,
      "classification modeling": 17277,
      "modeling interdependencies": 72451,
      "argue semantic": 8928,
      "clause interpreted": 17690,
      "structure goal": 113871,
      "relations benefits": 98114,
      "benefits nlp": 13016,
      "articles different": 9183,
      "structural integrity": 113774,
      "annotation workflow": 6404,
      "evaluating discourse": 39050,
      "level semantic": 63500,
      "contribution analysis": 22768,
      "analysis improvements": 5651,
      "features providing": 43680,
      "neural segmenter": 78660,
      "challenging important": 16261,
      "tasks discourse": 119066,
      "parsing lack": 85137,
      "relations implicit": 98186,
      "implicit source": 52020,
      "language explicitly": 59032,
      "multiple translations": 75737,
      "quality additional": 94588,
      "parsing face": 85112,
      "based formal": 11724,
      "facilitate learning": 42778,
      "silver training": 109067,
      "classification difficult": 17186,
      "difficulty stems": 31707,
      "stems fact": 113240,
      "refined representations": 97593,
      "benchmark multiple": 12844,
      "study information": 114408,
      "readability text": 95949,
      "text formatting": 120963,
      "information number": 54808,
      "communicative function": 19051,
      "information commonly": 54425,
      "fusion sentence": 46247,
      "fusion task": 46252,
      "task joining": 118323,
      "sentences single": 106497,
      "rules identifying": 103410,
      "identifying diverse": 51594,
      "various discourse": 131082,
      "recent dataset": 96443,
      "10k examples": 209,
      "learning transferable": 63123,
      "transferable sentence": 124223,
      "learning prediction": 62901,
      "yields state": 135438,
      "different transfer": 31510,
      "clear models": 17727,
      "room improvements": 103267,
      "perform domain": 85982,
      "problems addressed": 91295,
      "model stacking": 72079,
      "discourse unit": 32098,
      "2018 propose": 670,
      "examine model": 39752,
      "order assess": 82280,
      "structure compared": 113819,
      "parser finally": 84958,
      "provides little": 94048,
      "hurting performance": 51168,
      "performance implicit": 86438,
      "challenge identifying": 16041,
      "context opendomain": 22203,
      "text inherently": 121055,
      "topics present": 122650,
      "identification pipeline": 51421,
      "specifically tuned": 111605,
      "argument pairs": 8948,
      "dialogic turns": 30619,
      "resulting novel": 101460,
      "leverage dialogue": 63582,
      "classification automatically": 17130,
      "challenging fact": 16254,
      "text evaluated": 120923,
      "corpus obtains": 23914,
      "suite tasks": 115420,
      "propose variety": 93154,
      "annotations wikipedia": 6481,
      "document structures": 33083,
      "textual signals": 121736,
      "al 1993": 4589,
      "including summarization": 53385,
      "classification previous": 17339,
      "parsing develop": 85100,
      "branching bias": 14563,
      "outperforms discriminative": 82881,
      "transform syntactically": 124257,
      "simple regular": 109503,
      "regular structure": 97730,
      "structure easier": 113845,
      "process downstream": 91463,
      "applications purpose": 6995,
      "semantic hierarchy": 105070,
      "hierarchy form": 50026,
      "form core": 45077,
      "core facts": 23320,
      "facts accompanying": 42905,
      "accompanying contexts": 1947,
      "contexts identifying": 22399,
      "identifying rhetorical": 51620,
      "rhetorical relations": 102712,
      "interpretability downstream": 56223,
      "detecting implicit": 29839,
      "inadequate capturing": 53197,
      "effective semantic": 34743,
      "representations arguments": 99506,
      "arguments automatically": 8981,
      "superiority model": 115705,
      "tree generation": 125595,
      "parser structure": 84998,
      "parser does": 84938,
      "performance parser": 86601,
      "propose scheme": 93027,
      "parsing firstly": 85113,
      "dialogues machine": 30837,
      "parsing crosslingual": 85091,
      "crosslingual mapping": 24972,
      "results corresponding": 101645,
      "parsing need": 85173,
      "previous datadriven": 90404,
      "phrases result": 87450,
      "signal words": 108696,
      "identification context": 51374,
      "type annotations": 126186,
      "specific tokens": 111503,
      "need order": 76841,
      "methods preliminary": 69677,
      "based resources": 11994,
      "evaluations systems": 39486,
      "reference results": 97531,
      "unsupervised adversarial": 127598,
      "counterparts tackle": 24441,
      "2015 present": 592,
      "works adversarial": 134920,
      "work contributes": 134439,
      "contributes new": 22764,
      "properties chinese": 92439,
      "english resource": 37262,
      "achieve highly": 2482,
      "reliable results": 98620,
      "considerably recent": 21277,
      "current strong": 25372,
      "20 performance": 519,
      "parsing relation": 85217,
      "representation module": 99344,
      "fusion module": 46244,
      "particular error": 85411,
      "keeping good": 57519,
      "structure drs": 113844,
      "drawing attention": 34126,
      "hierarchical tree": 50018,
      "parsing pretraining": 85198,
      "pretraining contrastive": 90240,
      "13 average": 281,
      "26 absolute": 838,
      "markers sentence": 67234,
      "markers semantic": 67233,
      "relations annotated": 98111,
      "number highquality": 80894,
      "online data": 81758,
      "smaller manually": 110236,
      "nlp benchmark": 79581,
      "current paper": 25320,
      "efficacy pretrained": 35013,
      "global neural": 48253,
      "parsing largely": 85141,
      "greedy parsers": 49073,
      "label assigned": 58383,
      "performance global": 86423,
      "accuracy classifying": 2116,
      "features produced": 43670,
      "benchmark furthermore": 12826,
      "demonstrated impressive": 28921,
      "recent papers": 96487,
      "method quantitatively": 69088,
      "knowledge intermediate": 58022,
      "introduce types": 56562,
      "sentence paragraphlevel": 105978,
      "approach augment": 7370,
      "scoring task": 104233,
      "methods serve": 69743,
      "nontrivial baseline": 80311,
      "reflect different": 97604,
      "english define": 37108,
      "research finally": 100515,
      "parsing work": 85282,
      "novel scalable": 80709,
      "scalable methodology": 103694,
      "corpora sentiment": 23583,
      "augmentation sentiment": 10301,
      "requires methods": 100297,
      "methods capturing": 69362,
      "capturing complex": 15461,
      "framework exploit": 45536,
      "structure aware": 113811,
      "aware approach": 11257,
      "task numerous": 118467,
      "role understanding": 103221,
      "treebank parsing": 125644,
      "crosslingual documentlevel": 24945,
      "representations documentlevel": 99606,
      "accordance rhetorical": 1961,
      "corpora address": 23405,
      "longrange dependency": 65181,
      "modeling documentlevel": 72421,
      "semiautomatic methods": 105571,
      "research leverage": 100546,
      "classification small": 17411,
      "deeplearning architectures": 28456,
      "additional techniques": 3582,
      "accuracy setting": 2274,
      "labelling introduce": 58567,
      "problem goal": 91071,
      "document individual": 33023,
      "explore traditional": 41597,
      "traditional recurrent": 122860,
      "metric proposed": 69895,
      "finally present results": 44219,
      "present results evaluation": 89669,
      "different genres results": 31165,
      "results obtained application": 102005,
      "textbased machine translation": 121427,
      "learning ml approaches": 62751,
      "global features previous": 48239,
      "tokens propose novel": 122327,
      "reduction stateoftheart performance": 97459,
      "documents news articles": 33261,
      "language compare approaches": 58896,
      "deep semantic meaning": 28421,
      "semantic meaning representation": 105107,
      "experiments benchmark data": 40792,
      "standard stochastic gradient": 112310,
      "stochastic gradient methods": 113358,
      "relations natural language": 98218,
      "neural models require": 78196,
      "small label set": 110165,
      "network models based": 77345,
      "models perform consistently": 73724,
      "long shortterm memories": 65101,
      "representations surface features": 99920,
      "trained neural language": 123218,
      "structure documents previous": 113843,
      "present best knowledge": 89396,
      "best knowledge experiments": 13349,
      "structure theory rst": 113974,
      "languages domains paper": 60517,
      "sentiment analysis based": 106573,
      "neural networks discourse": 78467,
      "approach measuring sentiment": 7709,
      "generated statistical machine": 47127,
      "neural networks argument": 78440,
      "achieved f1 measure": 2622,
      "standard machine learning": 112254,
      "recent years research": 96631,
      "text spans challenging": 121318,
      "task recent studies": 118609,
      "syntactic parse tree": 116436,
      "measures natural language": 67883,
      "text segmentation task": 121277,
      "benefits nlp tasks": 13017,
      "discourse analysis previous": 32041,
      "use paper propose": 128188,
      "neural semantic parser": 78662,
      "silver training data": 109068,
      "relation classification difficult": 97946,
      "difficulty stems fact": 31708,
      "case study information": 15619,
      "features used improve": 43774,
      "extraction case study": 42275,
      "knowledge used improve": 58226,
      "sentence fusion sentence": 105884,
      "sentence fusion task": 105885,
      "heavily rely manually": 49659,
      "rely manually annotated": 98721,
      "methods present work": 69679,
      "massive amounts data": 67324,
      "representation learning prediction": 99308,
      "yields state art": 135439,
      "dividing text segments": 32917,
      "detection paper present": 30015,
      "nlp tasks learning": 79731,
      "2018 propose structured": 671,
      "propose structured attention": 93091,
      "dialogue previous work": 30725,
      "using zeroshot transfer": 130394,
      "english training data": 37312,
      "pretrained sentence embeddings": 90184,
      "makes use natural": 66815,
      "et al 1993": 38681,
      "tasks including summarization": 119190,
      "summarization document classification": 115499,
      "paper present generative": 84104,
      "parsing develop novel": 85101,
      "transform syntactically complex": 124258,
      "simple regular structure": 109504,
      "regular structure easier": 97731,
      "structure easier process": 113846,
      "easier process downstream": 34423,
      "downstream semantic applications": 34025,
      "semantic applications purpose": 104991,
      "hierarchy form core": 50027,
      "form core facts": 45078,
      "core facts accompanying": 23321,
      "facts accompanying contexts": 42906,
      "accompanying contexts identifying": 1948,
      "contexts identifying rhetorical": 22400,
      "identifying rhetorical relations": 51621,
      "rhetorical relations hold": 102713,
      "semantic information paper": 105076,
      "demonstrate superiority model": 28882,
      "superiority model previous": 115706,
      "model previous stateoftheart": 71785,
      "task sentiment classification": 118693,
      "paper propose scheme": 84316,
      "dependency parsing crosslingual": 29175,
      "simple efficient method": 109420,
      "supervised neural network": 115817,
      "english corpus annotated": 37099,
      "features neural models": 43625,
      "corpus manually annotated": 23878,
      "development language technology": 30394,
      "work contributes new": 134440,
      "considerably recent years": 21278,
      "retrieval text summarization": 102436,
      "text summarization work": 121355,
      "approaches standard datasets": 8349,
      "particular error rates": 85412,
      "classified different levels": 17510,
      "sentences recent work": 106468,
      "sentence representations useful": 106045,
      "classification tasks work": 17462,
      "language models finetuned": 59555,
      "parser does require": 84939,
      "proposed pretraining approach": 93524,
      "language modeling pretraining": 59456,
      "identify different types": 51492,
      "present novel scalable": 89621,
      "augmentation sentiment analysis": 10302,
      "sentiment analysis especially": 106593,
      "attention model experiments": 9931,
      "documents previous approaches": 33271,
      "nlp task numerous": 79702,
      "paper demonstrate simple": 83803,
      "important role understanding": 52247,
      "annotated data work": 6172,
      "data achieve stateoftheart": 25563,
      "stateoftheart performance crosslingual": 112817,
      "accordance rhetorical structure": 1962,
      "improve performance multitask": 52471,
      "highlevel semantic tasks": 50251,
      "introduce novel dynamic": 56511,
      "texts different genres results": 121500,
      "machine learning ml approaches": 65804,
      "standard stochastic gradient methods": 112311,
      "neural network models based": 78345,
      "rhetorical structure theory rst": 102716,
      "generated statistical machine translation": 47128,
      "outperforms previous stateoftheart systems": 82953,
      "method outperforms previous state": 69020,
      "work propose method automatically": 134726,
      "yields state art results": 135440,
      "2018 propose structured attention": 672,
      "propose structured attention mechanism": 93092,
      "using zeroshot transfer learning": 130395,
      "marcus et al 1993": 67181,
      "simple regular structure easier": 109505,
      "regular structure easier process": 97732,
      "structure easier process downstream": 113847,
      "downstream semantic applications purpose": 34026,
      "form core facts accompanying": 45079,
      "core facts accompanying contexts": 23322,
      "facts accompanying contexts identifying": 42907,
      "accompanying contexts identifying rhetorical": 1949,
      "contexts identifying rhetorical relations": 22401,
      "identifying rhetorical relations hold": 51622,
      "semantic information paper propose": 105077,
      "superiority model previous stateoftheart": 115707,
      "tasks information retrieval text": 119206,
      "information retrieval text summarization": 54952,
      "deep natural language understanding": 28344,
      "attention recent years previous": 9987,
      "pretrained language models finetuned": 90076,
      "plays important role understanding": 87735,
      "accordance rhetorical structure theory": 1963,
      "pretrained contextual language model": 90007,
      "improve performance multitask learning": 52472,
      "propose neural network models based": 92828,
      "experimental results model outperforms previous": 40650,
      "model outperforms previous stateoftheart systems": 71660,
      "method outperforms previous state art": 69021,
      "2018 propose structured attention mechanism": 673,
      "simple regular structure easier process": 109506,
      "regular structure easier process downstream": 97733,
      "form core facts accompanying contexts": 45080,
      "core facts accompanying contexts identifying": 23323,
      "facts accompanying contexts identifying rhetorical": 42908,
      "accompanying contexts identifying rhetorical relations": 1950,
      "contexts identifying rhetorical relations hold": 22402,
      "extensive experiments proposed model outperforms": 41924,
      "tasks information retrieval text summarization": 119207,
      "accordance rhetorical structure theory rst": 1964,
      "strokes": 113615,
      "microtexts": 70065,
      "granted": 48738,
      "961": 1430,
      "ctb5": 25191,
      "gaussianbased": 46554,
      "bakeoff": 11394,
      "dataindependent": 26689,
      "wubi": 135184,
      "microtext": 70063,
      "grant": 48737,
      "switcher": 116292,
      "optimistic": 82185,
      "dualchannel": 34244,
      "spatially": 111272,
      "longlength": 65175,
      "applied multiple": 7095,
      "subtasks need": 114974,
      "usually learned": 130434,
      "learned isolation": 62220,
      "algorithm evaluated": 4729,
      "learn joint": 62073,
      "particles characters": 85374,
      "characters form": 16608,
      "meaning does": 67633,
      "basic semantic": 12534,
      "understanding way": 127009,
      "binary tree": 14138,
      "segmentation chinese": 104578,
      "problem main": 91117,
      "granularity problem": 48751,
      "work chinese": 134413,
      "pruning function": 94167,
      "effect number": 34604,
      "general lexicon": 46673,
      "wordbased features": 133698,
      "employ wordbased": 36290,
      "penn chinese": 85856,
      "generation comparison": 47336,
      "language strings": 60126,
      "predict character": 88875,
      "text network": 121143,
      "usefulness learned": 128958,
      "supervised character": 115735,
      "paper gap": 83965,
      "existing embedding": 40122,
      "use character": 127935,
      "complex hierarchical": 19815,
      "use structure": 128301,
      "graph weights": 48882,
      "efficiency text": 35048,
      "task corpora": 118027,
      "classifier possible": 17571,
      "2015 shared": 594,
      "paper overview": 84064,
      "overview shared": 83375,
      "conference natural": 20978,
      "2015 chinese": 583,
      "different popular": 31337,
      "systems report": 117110,
      "results online": 102013,
      "useful capturing": 128866,
      "work innovatively": 134571,
      "develop chinese": 30183,
      "features traditional": 43761,
      "segmentation use": 104647,
      "media experimental": 68111,
      "ner new": 77059,
      "segment representations": 104565,
      "different composition": 31050,
      "results conll03": 101624,
      "local windows": 64942,
      "simple interactions": 109447,
      "combination neural": 18572,
      "help feature": 49724,
      "classification applied": 17117,
      "corpus article": 23668,
      "provides interesting": 94043,
      "corpus corresponding": 23731,
      "ability alleviating": 1483,
      "feature templates": 43321,
      "traditional discrete": 122813,
      "distance dependency": 32411,
      "information recurrent": 54904,
      "accurate word": 2374,
      "pragmatic approach": 88748,
      "train characterbased": 122907,
      "treebank experiments": 125636,
      "existing domain": 40119,
      "varieties dialects": 130955,
      "use general": 128064,
      "ngram new": 79449,
      "features series": 43724,
      "lowresource chinese": 65487,
      "work significantly": 134806,
      "furthermore paper": 46198,
      "extensively evaluated": 41966,
      "stateoftheart tagger": 112994,
      "work modeled": 134636,
      "compositionality words": 20146,
      "words creating": 133895,
      "model effect": 71040,
      "characters based": 16596,
      "rare characters": 95737,
      "datasets adversarial": 27302,
      "multiple heterogeneous": 75573,
      "criteria experiments": 24792,
      "segmentation research": 104628,
      "hand statistical": 49328,
      "using rich": 130119,
      "chinese news": 16796,
      "18 classes": 415,
      "effort feature": 35172,
      "utilize external": 130507,
      "learn chinese": 62000,
      "directly learned": 31883,
      "improve chinese": 52346,
      "public crosslingual": 94242,
      "tasks crosslingual": 119022,
      "improved ensemble": 52602,
      "evaluated official": 38997,
      "official data": 81596,
      "outstanding accuracies": 83203,
      "compared participating": 19407,
      "rely local": 98718,
      "words subwords": 134247,
      "label decoding": 58388,
      "step simple": 113301,
      "result significant": 101404,
      "networks conventional": 77550,
      "datasets lack": 27536,
      "tasks insufficient": 119216,
      "learn knowledge": 62075,
      "design various": 29573,
      "second make": 104427,
      "neural solution": 78683,
      "simple elegant": 109422,
      "solution train": 110567,
      "single joint": 109744,
      "remains unchanged": 98823,
      "stateoftheart learning": 112690,
      "input method": 55365,
      "rely predefined": 98729,
      "online algorithm": 81742,
      "dual long": 34238,
      "characters commonly": 16602,
      "smaller parts": 110242,
      "example languages": 39787,
      "components called": 20019,
      "networks employ": 77581,
      "conversion highly": 23086,
      "knowledge shared": 58171,
      "framework chinese": 45454,
      "problem directly": 91014,
      "complex paper": 19855,
      "exploited experiments": 41451,
      "design unsupervised": 29570,
      "results corpora": 101642,
      "using lattice": 129803,
      "characters potential": 16618,
      "does suffer": 33404,
      "recurrent cells": 97163,
      "character features": 16433,
      "challenge chinese": 16005,
      "semantic unit": 105344,
      "generic way": 47814,
      "maps word": 67169,
      "inputs resulting": 55495,
      "believed crucial": 12708,
      "understanding widely": 127013,
      "corpus conducted": 23713,
      "evaluated linguistic": 38985,
      "linguistic experts": 64475,
      "experts model": 41243,
      "segmentation wide": 104652,
      "accuracy experimental": 2157,
      "stateoftheart accuracies": 112571,
      "works achieved": 134917,
      "sharing lstm": 108134,
      "procedure extensive": 91392,
      "respectively leveraging": 101147,
      "endtoend text": 36979,
      "types paper": 126333,
      "similar characters": 109078,
      "task analyse": 117875,
      "performance benefit": 86175,
      "conducted compare": 20914,
      "lstm structure": 65669,
      "effective subword": 34753,
      "comprehension inference": 20185,
      "models characterlevel": 72899,
      "ignoring linguistic": 51691,
      "series unsupervised": 107285,
      "methods subword": 69776,
      "improves baselines": 52950,
      "provide comparable": 93779,
      "ground use": 49102,
      "ability discover": 1508,
      "discovery learning": 32139,
      "words fit": 133959,
      "conditioning model": 20812,
      "bayesian word": 12588,
      "need largescale": 76829,
      "realworld tasks": 96182,
      "frequencies corpus": 45834,
      "analysis comparing": 5534,
      "comparing neural": 19509,
      "ngrambased language": 79458,
      "token time": 122270,
      "information gathered": 54631,
      "microtext normalization": 70064,
      "available word": 11148,
      "propose flexible": 92682,
      "new criteria": 78848,
      "robust chinese": 103023,
      "dealing outofvocabulary": 27871,
      "rate used": 95806,
      "shown superior": 108535,
      "significant progresses": 108853,
      "embeddings crossdomain": 35626,
      "challenge despite": 16019,
      "proposed derive": 93258,
      "embeddings optimized": 35842,
      "special domains": 111359,
      "entities word": 37897,
      "unsupervised crossdomain": 127616,
      "annotation criteria": 6294,
      "layers proposed": 61802,
      "errors occur": 38394,
      "different auxiliary": 31021,
      "prediction help": 89068,
      "help task": 49783,
      "usually considered": 130411,
      "characterbased convolutional": 16485,
      "selfattention layer": 104879,
      "models depending": 73040,
      "task construct": 118015,
      "path length": 85677,
      "best proposed": 13422,
      "method 60": 68578,
      "crossdomain semantic": 24899,
      "domain obtain": 33599,
      "toolkit supports": 122420,
      "using bilstmcrf": 129415,
      "predictions recently": 89186,
      "attentionbased sequence": 10099,
      "competitive alternative": 19635,
      "alternative lstms": 5233,
      "gives highly": 48197,
      "complex model": 19833,
      "representation layer": 99279,
      "achieves inference": 2810,
      "faster stateoftheart": 43190,
      "incorporated pretrained": 53500,
      "detection challenging": 29893,
      "function proposed": 46039,
      "capturing sequential": 15491,
      "training encoder": 123606,
      "transformer variant": 124392,
      "unigram features": 127157,
      "features scoring": 43710,
      "results highest": 101824,
      "terms strict": 120386,
      "enhancing pretrained": 37538,
      "semantics expressed": 105415,
      "chinese propose": 16809,
      "potential issue": 88567,
      "chinese spelling": 16825,
      "framework detect": 45490,
      "process dynamically": 91465,
      "dynamically integrates": 34348,
      "sequence translation": 107092,
      "recently chinese": 96669,
      "novel sequencetosequence": 80724,
      "tackle nlp": 117300,
      "jointly endtoend": 57341,
      "mode experiments": 70490,
      "recognition augmented": 96819,
      "combined generate": 18655,
      "possible candidates": 88387,
      "datasets hierarchical": 27503,
      "hierarchical character": 49941,
      "literature suggests": 64775,
      "structures learn": 114085,
      "verify claim": 131579,
      "segmentation article": 104576,
      "article proposed": 9152,
      "problem instead": 91087,
      "clustering model": 17949,
      "85 percent": 1323,
      "text chinese": 120703,
      "ner incorporating": 77050,
      "computation gpus": 20348,
      "lexiconbased models": 63918,
      "similarities language": 109180,
      "spelling check": 111899,
      "check chinese": 16692,
      "detect correct": 29797,
      "graph set": 48868,
      "classifiers classifiers": 17603,
      "subword sequences": 115037,
      "based subword": 12082,
      "exploiting secondorder": 41476,
      "lexical words": 63843,
      "researches proposed": 100722,
      "performed simple": 86933,
      "knowledge provided": 58129,
      "based propose": 11956,
      "strategy integrate": 113523,
      "approaching neural": 8409,
      "learning era": 62553,
      "characterlevel translation": 16587,
      "simplified model": 109603,
      "translation supervised": 125303,
      "including regularization": 53360,
      "translationbased method": 125445,
      "problem achieved": 90914,
      "information multilevel": 54774,
      "detect word": 29821,
      "applied character": 7049,
      "recent researches": 96510,
      "usually adopt": 130405,
      "architecture incorporates": 8671,
      "solved task": 110631,
      "weaknesses existing": 132208,
      "available release": 11090,
      "quickly evaluate": 95391,
      "endtoend chinese": 36880,
      "character inputs": 16437,
      "structures particular": 114096,
      "word knowledge": 133328,
      "unit analysis": 127201,
      "framework general": 45551,
      "time efficient": 122020,
      "chunking named": 16960,
      "notably improved": 80381,
      "shortterm dependencies": 108313,
      "algorithms described": 4840,
      "segmentation helps": 104591,
      "stateoftheart different": 112641,
      "potential capture": 88541,
      "framework incrementally": 45574,
      "construct phrase": 21647,
      "baselines achieved": 12348,
      "sentences verify": 106543,
      "effectiveness modeling": 34915,
      "deeper understanding way": 28454,
      "problem previous work": 91171,
      "previous work chinese": 90524,
      "appear training data": 6805,
      "penn chinese treebank": 85857,
      "embeddings learning word": 35786,
      "natural language strings": 76538,
      "propose learn text": 92744,
      "simple recurrent network": 109501,
      "programming language code": 92130,
      "2015 shared task": 595,
      "overview shared task": 83376,
      "conference natural language": 20979,
      "information successfully applied": 55015,
      "media paper propose": 68140,
      "social media experimental": 110378,
      "use linguistic features": 128128,
      "new stateoftheart word": 79180,
      "distributed representations word": 32621,
      "help feature engineering": 49725,
      "text classification applied": 120710,
      "neural network applied": 78241,
      "neural models extract": 78175,
      "feature based models": 43254,
      "models use different": 74275,
      "model complex compositional": 70866,
      "existing domain adaptation": 40120,
      "performance domain adaptation": 86313,
      "studies shown effectiveness": 114281,
      "shown effectiveness using": 108466,
      "data propose transfer": 26292,
      "work significantly improves": 134807,
      "rich contextual information": 102736,
      "model extensively evaluated": 71157,
      "model allows better": 70660,
      "minimal feature engineering": 70160,
      "current neural models": 25311,
      "obtains significant improvement": 81475,
      "codes paper available": 18208,
      "paper available github": 83753,
      "effort feature engineering": 35173,
      "word representations character": 133448,
      "evaluated official data": 38998,
      "previous studies chinese": 90491,
      "order address problem": 82274,
      "experiment shows method": 40511,
      "tasks insufficient data": 119217,
      "datasets diverse domains": 27436,
      "framework consists parts": 45468,
      "conduct experiments lowresource": 20852,
      "improvements datasets compared": 52840,
      "use word information": 128363,
      "neural model able": 78144,
      "able capture rich": 1618,
      "word embeddings evaluate": 133101,
      "evaluate model benchmark": 38864,
      "model benchmark datasets": 70758,
      "present simple elegant": 89705,
      "high performance large": 50096,
      "dual long shortterm": 34239,
      "complex paper propose": 19856,
      "approach outperforms best": 7767,
      "achieve word segmentation": 2588,
      "lstm model chinese": 65642,
      "word embeddings ensemble": 133100,
      "learning approaches based": 62357,
      "neural networks word": 78606,
      "like pretrained word": 64082,
      "embeddings proposed model": 35891,
      "believed crucial step": 12709,
      "crucial step natural": 25169,
      "language understanding widely": 60293,
      "understanding widely studied": 127014,
      "widely studied recent": 132557,
      "tasks trained model": 119570,
      "sequence tagging framework": 107077,
      "languages different writing": 60506,
      "accuracy experimental results": 2158,
      "training procedure extensive": 123790,
      "procedure extensive experiments": 91393,
      "based morphological analysis": 11858,
      "types paper propose": 126334,
      "endtoend learning model": 36913,
      "results previous stateoftheart": 102053,
      "conducted compare performance": 20915,
      "reading comprehension inference": 95985,
      "significantly improves baselines": 108944,
      "models achieved competitive": 72678,
      "language model combines": 59308,
      "generalization power neural": 46792,
      "context improves performance": 22132,
      "corpus paper introduce": 23921,
      "article propose approach": 9151,
      "paper propose flexible": 84241,
      "error rate used": 38334,
      "simple bidirectional lstm": 109379,
      "compared traditional supervised": 19480,
      "traditional supervised learning": 122876,
      "supervised learning methods": 115771,
      "word similarity sentiment": 133568,
      "sentiment analysis named": 106613,
      "various datasets model": 131072,
      "capture domainspecific knowledge": 15296,
      "different auxiliary tasks": 31022,
      "characterbased convolutional neural": 16486,
      "external resources like": 42040,
      "stateoftheart methods word": 112717,
      "stateoftheart method 60": 112700,
      "enriching word embeddings": 37570,
      "joint learning sequential": 57288,
      "models different domains": 73069,
      "self attention network": 104871,
      "gives highly competitive": 48198,
      "highly competitive results": 50302,
      "models best results": 72844,
      "proposed method easily": 93363,
      "capturing sequential information": 15492,
      "greedy decoding algorithm": 49070,
      "features scoring model": 43711,
      "word level propose": 133346,
      "tasks demonstrate model": 119038,
      "sequence translation problem": 107093,
      "methods successfully applied": 69778,
      "entity recognition augmented": 37992,
      "hierarchical character embeddings": 49942,
      "potentially lead better": 88616,
      "benefit downstream tasks": 12965,
      "neural network chinese": 78275,
      "word segmentation article": 133491,
      "language model introduced": 59336,
      "methods achieve new": 69279,
      "recognition ner incorporating": 96932,
      "parallel computation gpus": 84592,
      "chinese spelling check": 16826,
      "correct spelling errors": 24122,
      "deep learning era": 28271,
      "machine translation supervised": 66245,
      "used previous works": 128698,
      "knowledge different existing": 57861,
      "promising direction future": 92270,
      "stateoftheart results endtoend": 112899,
      "extensive experiments tasks": 41932,
      "syntactic chunking named": 116377,
      "chunking named entity": 16961,
      "results f1 score": 101790,
      "model learning objective": 71438,
      "long shortterm dependencies": 65100,
      "different evaluation settings": 31134,
      "construct phrase representations": 21648,
      "experiments syntactic chunking": 41166,
      "methods significantly outperformed": 69754,
      "qualitative analysis study": 94556,
      "conference natural language processing": 20980,
      "social media paper propose": 110401,
      "convolutional neural network applied": 23177,
      "recent studies shown effectiveness": 96526,
      "propose transfer learning method": 93119,
      "proposed model extensively evaluated": 93448,
      "source codes paper available": 110731,
      "codes paper available github": 18209,
      "dual long shortterm memory": 34240,
      "neural networks word embeddings": 78607,
      "like pretrained word embeddings": 64083,
      "model outperforms baseline models": 71637,
      "believed crucial step natural": 12710,
      "crucial step natural language": 25170,
      "step natural language understanding": 113285,
      "natural language understanding widely": 76592,
      "language understanding widely studied": 60294,
      "understanding widely studied recent": 127015,
      "widely studied recent years": 132558,
      "studied recent years endtoend": 114172,
      "languages different writing systems": 60507,
      "training procedure extensive experiments": 123791,
      "language models based deep": 59509,
      "models based deep learning": 72805,
      "characterlevel convolutional neural networks": 16562,
      "machine reading comprehension inference": 65853,
      "compared traditional supervised learning": 19481,
      "word similarity sentiment analysis": 133569,
      "sentiment analysis named entity": 106614,
      "annotated data target domain": 6170,
      "characterbased convolutional neural network": 16487,
      "results approach outperforms stateoftheart": 101517,
      "similarity word analogy tasks": 109334,
      "gives highly competitive results": 48199,
      "absolute improvement previous stateoftheart": 1737,
      "bert language model introduced": 13151,
      "methods achieve new stateoftheart": 69280,
      "entity recognition ner incorporating": 38041,
      "lowresource machine translation task": 65534,
      "neural machine translation supervised": 78108,
      "syntactic chunking named entity": 116378,
      "chunking named entity recognition": 16962,
      "methods significantly outperformed previous": 69755,
      "source codes paper available github": 110732,
      "model achieves stateoftheart performance datasets": 70600,
      "believed crucial step natural language": 12711,
      "crucial step natural language understanding": 25171,
      "step natural language understanding widely": 113286,
      "natural language understanding widely studied": 76593,
      "language understanding widely studied recent": 60295,
      "understanding widely studied recent years": 127016,
      "widely studied recent years endtoend": 132559,
      "sentiment analysis named entity recognition": 106615,
      "word similarity word analogy tasks": 133575,
      "work propose simple effective method": 134743,
      "methods achieve new stateoftheart results": 69281,
      "named entity recognition ner incorporating": 76087,
      "syntactic chunking named entity recognition": 116379,
      "chunking named entity recognition ner": 16963,
      "displaying": 32380,
      "specializes": 111396,
      "scws": 104275,
      "churn": 16973,
      "resourcefree": 100895,
      "210": 764,
      "macroscopic": 66366,
      "practitioner": 88743,
      "phylogenies": 87474,
      "vecmap": 131292,
      "anchorbased": 6085,
      "problems providing": 91365,
      "analysis time": 5899,
      "representations multilingual": 99759,
      "representations semantically": 99873,
      "distributed semantics": 32625,
      "hypothesis multilingual": 51267,
      "embeddings semantically": 35927,
      "number diverse": 80875,
      "languages extend": 60576,
      "marginalizing alignments": 67216,
      "work relying": 134774,
      "prior published": 90726,
      "born equal": 14470,
      "models singlelanguage": 74058,
      "syntactic role": 116459,
      "concepts related": 20638,
      "multilingual topics": 75384,
      "topics method": 122638,
      "unlike previously": 127446,
      "scalable large": 103692,
      "approach refer": 7849,
      "method reaches": 69092,
      "reaches level": 95914,
      "accuracy absolute": 2090,
      "heavily relied": 49654,
      "translation resources": 125219,
      "resources dictionaries": 100964,
      "resources syntactic": 101050,
      "comparison approach": 19526,
      "probabilistic topic": 90832,
      "tested language": 120571,
      "parsing web": 85280,
      "tasks systematic": 119541,
      "including intrinsic": 53305,
      "experience applying": 40442,
      "applying distributional": 7238,
      "optimal linear": 82163,
      "presented method": 89788,
      "translation levenshtein": 124904,
      "clustering algorithms": 17934,
      "requiring linguistic": 100346,
      "graphs study": 48973,
      "embeddings monolingual": 35813,
      "employed task": 36309,
      "method lexical": 68943,
      "bilingual projection": 14054,
      "algorithms feature": 4854,
      "traditional alignment": 122791,
      "stateoftheart embedding": 112649,
      "embeddings future": 35729,
      "accuracy syntactic": 2297,
      "similarity compared": 109209,
      "nonparallel data": 80266,
      "applying topic": 7275,
      "new bilingual": 78816,
      "problem noise": 91146,
      "translation selection": 125235,
      "measure evaluate": 67797,
      "languages aim": 60395,
      "sentences provide": 106460,
      "structure syntax": 113966,
      "classes based": 17059,
      "limited languages": 64249,
      "uses crosslingual": 129216,
      "crosslingual translation": 25041,
      "standard clustering": 112211,
      "dimensions features": 31766,
      "monolingual documents": 74586,
      "model 93": 70503,
      "data included": 26031,
      "systems errors": 116873,
      "argue evaluating": 8915,
      "learning image": 62636,
      "translation selected": 125234,
      "learned way": 62284,
      "indicate previous": 53847,
      "scale problem": 103750,
      "pairs obtained": 83593,
      "method lowresource": 68946,
      "considerably outperform": 21275,
      "setting uses": 107785,
      "approach mapping": 7703,
      "pair hindi": 83434,
      "applying multiple": 7258,
      "models solving": 74068,
      "structural similarities": 113795,
      "seed dictionaries": 104500,
      "structures monolingual": 114088,
      "performance bilingual": 86183,
      "comparable using": 19192,
      "language vector": 60328,
      "language linear": 59254,
      "different unsupervised": 31542,
      "languages exists": 60566,
      "highly parallel": 50334,
      "single common": 109716,
      "emerging language": 36068,
      "considered semantic": 21298,
      "used disambiguate": 128494,
      "linguistic signal": 64554,
      "grounded translation": 49115,
      "amounts human": 5338,
      "use highly": 128088,
      "setup learning": 107853,
      "language typology": 60198,
      "common baseline": 18858,
      "techniques typically": 120004,
      "concatenation different": 20541,
      "gap stateoftheart": 46481,
      "monolingual baseline": 74549,
      "leverages shared": 63660,
      "shared crosslingual": 107970,
      "queries documents": 94918,
      "spaces induced": 111095,
      "embeddings induced": 35761,
      "framework step": 45700,
      "scarce nonexistent": 103792,
      "maps learned": 67164,
      "maps using": 67168,
      "perform predictions": 86047,
      "predictions languages": 89179,
      "highly predictable": 50336,
      "transfer word": 124209,
      "translation combine": 124696,
      "strongly supervised": 113751,
      "dictionaries propose": 30866,
      "leads promising": 61952,
      "embeddings increasingly": 35758,
      "important multilingual": 52185,
      "nlp recently": 79678,
      "embeddings effectively": 35675,
      "effectively learned": 34827,
      "spaces linear": 111099,
      "crosslingual synonyms": 25019,
      "alignment typically": 4998,
      "typically learned": 126439,
      "crosslingual evaluation": 24953,
      "mwes represent": 75996,
      "space unsupervised": 111073,
      "art learning": 9054,
      "unsupervised bilingual": 127609,
      "directly exploits": 31876,
      "key roles": 57596,
      "methods involve": 69564,
      "steps including": 113322,
      "building idea": 14847,
      "alignment english": 4955,
      "english embeddings": 37123,
      "challenge community": 16009,
      "induced using": 54010,
      "believe understanding": 12704,
      "shows cases": 108559,
      "differences similar": 30972,
      "similar effects": 109088,
      "easily obtained": 34462,
      "varying hyperparameters": 131259,
      "aims establish": 4524,
      "semantic mappings": 105103,
      "mappings words": 67160,
      "languages supervised": 60906,
      "require crosslingual": 100125,
      "losses use": 65310,
      "optimize objective": 82225,
      "knowledge wordnet": 58236,
      "distributional space": 32737,
      "examples learning": 39852,
      "adversarial component": 4107,
      "realistic output": 96102,
      "output vectors": 83148,
      "method languages": 68930,
      "learns bilingual": 63197,
      "embeddings align": 35549,
      "space crosslingual": 110987,
      "similarity scws": 109303,
      "scws dataset": 104276,
      "superior quality": 115694,
      "quality sense": 94787,
      "evaluated monolingual": 38993,
      "capture common": 15274,
      "task adopts": 117852,
      "distribution differences": 32648,
      "process form": 91489,
      "embedding target": 35510,
      "contexts concepts": 22383,
      "line work": 64328,
      "best evaluation": 13329,
      "structure common": 113818,
      "entities design": 37765,
      "knowledge attention": 57758,
      "linking results": 64671,
      "properly evaluate": 92431,
      "embeddings strong": 35954,
      "baselines comparative": 12371,
      "modeling meaning": 72472,
      "evaluated single": 39005,
      "task bilingual": 117937,
      "correctly interpret": 24178,
      "properties different": 92445,
      "evaluate supervised": 38931,
      "nlp empirically": 79610,
      "depends task": 29294,
      "entries bilingual": 38156,
      "dictionary source": 30894,
      "documentlevel representations": 33162,
      "better reflects": 13692,
      "translation retrieval": 125223,
      "used wordlevel": 128854,
      "expanding text": 40364,
      "digital divide": 31721,
      "categorizing texts": 15780,
      "churn intent": 16974,
      "embeddings architectures": 35558,
      "multilingual joint": 75262,
      "nlp current": 79595,
      "cleaning process": 17717,
      "model compatible": 70862,
      "semantic shifts": 105274,
      "shifts words": 108193,
      "words documents": 133924,
      "language single": 60097,
      "word retrieval": 133486,
      "words map": 134046,
      "information single": 54985,
      "concept level": 20577,
      "encode meaning": 36438,
      "similarity independent": 109241,
      "moderate strong": 74375,
      "lowresource crosslingual": 65490,
      "extra features": 42060,
      "pair training": 83461,
      "shared vector": 108115,
      "rich resources": 102780,
      "separately language": 106873,
      "dictionaries pretrained": 30865,
      "pairs little": 83576,
      "research perform": 100581,
      "frequency dictionary": 45839,
      "language construction": 58914,
      "learning case": 62421,
      "scenario evaluation": 103824,
      "multiple crosslingual": 75530,
      "independently paper": 53789,
      "lack supervision": 58755,
      "makes approaches": 66756,
      "multisense word": 75777,
      "processing like": 91698,
      "embeddings define": 35635,
      "counterparts work": 24443,
      "aims bring": 4505,
      "translations closer": 125456,
      "integration monolingual": 55811,
      "gains current": 46390,
      "distances used": 32441,
      "retrieval crosslingual": 102389,
      "problems argue": 91298,
      "bilingual lexical": 14042,
      "induction experimental": 54023,
      "using wasserstein": 130364,
      "english recent": 37259,
      "impact downstream": 51870,
      "evaluating established": 39052,
      "embeddings baselines": 35568,
      "method resulting": 69108,
      "original embeddings": 82515,
      "tagging english": 117389,
      "performance observed": 86584,
      "easier interpret": 34415,
      "combination measures": 18567,
      "recent embeddingbased": 96459,
      "results generally": 101808,
      "spelling information": 111907,
      "work space": 134810,
      "scripts demonstrate": 104268,
      "challenges lowresource": 16173,
      "represented linear": 100031,
      "embeddings necessary": 35826,
      "languages carry": 60433,
      "provide direct": 93808,
      "additional insight": 3530,
      "typological differences": 126484,
      "present series": 89691,
      "sparse language": 111228,
      "like multilingual": 64067,
      "benefit language": 12976,
      "language phylogenies": 59812,
      "accuracy tasks": 2299,
      "expensive retraining": 40432,
      "related approaches": 97844,
      "language manually": 59273,
      "timeconsuming costly": 122144,
      "languages increasing": 60638,
      "adaptable different": 3204,
      "closelyrelated languages": 17868,
      "augmentation unsupervised": 10314,
      "hold true": 50587,
      "general paper": 46694,
      "using pseudoparallel": 130064,
      "pseudoparallel corpus": 94191,
      "language helps": 59119,
      "typologically similar": 126500,
      "embeddings examine": 35703,
      "semisupervised bilingual": 105593,
      "annotations previous": 6454,
      "methods fully": 69519,
      "data hinders": 26006,
      "hinders improvement": 50456,
      "study proves": 114492,
      "incorrect assumption": 53576,
      "especially distant": 38451,
      "piecewise linear": 87502,
      "space low": 111018,
      "resource target": 100881,
      "experiments simple": 41143,
      "japanese vietnamese": 57201,
      "translations typically": 125503,
      "quality gains": 94671,
      "resulting lexical": 101450,
      "providing interpretable": 94120,
      "comparable quality": 19168,
      "presents methods": 89868,
      "pairs study": 83642,
      "trained unpaired": 123320,
      "model xlmr": 72332,
      "reaches state": 95918,
      "semantics present novel": 105448,
      "bilingual word representations": 14074,
      "tasks requiring knowledge": 119466,
      "data previously used": 26276,
      "unlike previously proposed": 127447,
      "previously proposed methods": 90611,
      "improve previous stateoftheart": 52503,
      "improvement previous state": 52746,
      "data propose new": 26290,
      "model learns dense": 71442,
      "probabilistic topic modeling": 90833,
      "tested language pairs": 120572,
      "embeddings introduce new": 35766,
      "new methods estimating": 79017,
      "corpus data collection": 23741,
      "linguistic resources language": 64549,
      "word embeddings monolingual": 133164,
      "feature set used": 43311,
      "word embeddings future": 133118,
      "applying topic models": 7276,
      "better capture semantic": 13526,
      "words different languages": 133913,
      "machine translation learn": 66027,
      "capture underlying semantics": 15413,
      "representations words based": 99989,
      "previous work does": 90527,
      "problem learning crosslingual": 91103,
      "extracting bilingual lexicon": 42203,
      "word embeddings source": 133209,
      "target language evaluate": 117637,
      "improves current state": 52967,
      "embeddings existing methods": 35706,
      "proposed unsupervised method": 93581,
      "model align words": 70653,
      "word translation pairs": 133608,
      "sufficiently large corpus": 115295,
      "performance comparable complex": 86220,
      "crosslingual sentence representations": 25010,
      "embedding spaces induced": 35505,
      "data scarce nonexistent": 26371,
      "languages word embeddings": 60971,
      "transfer word embeddings": 124210,
      "machine translation combine": 65924,
      "vector spaces linear": 131393,
      "vector space unsupervised": 131387,
      "unsupervised bilingual word": 127610,
      "model substantially outperforms": 72103,
      "play key roles": 87703,
      "results performance comparable": 102034,
      "hard paper presents": 49475,
      "paper presents challenge": 84156,
      "presents challenge community": 89830,
      "word embeddings induced": 133135,
      "based distributional information": 11654,
      "corresponding word embedding": 24315,
      "unsupervised learning approach": 127654,
      "evaluation benchmark datasets": 39134,
      "competitive performance proposed": 19664,
      "method compared stateoftheart": 68711,
      "baseline methods language": 12249,
      "distributional word vectors": 32749,
      "unsupervised sense embeddings": 127714,
      "sense embeddings paper": 105677,
      "learning model jointly": 62762,
      "jointly learns bilingual": 57359,
      "word similarity scws": 133566,
      "similarity scws dataset": 109304,
      "shows superior quality": 108637,
      "capture common linguistic": 15275,
      "previous work task": 90556,
      "work task adopts": 134846,
      "require parallel corpora": 100184,
      "entity linking results": 37962,
      "evaluated single task": 39006,
      "comprehensive evaluation crosslingual": 20259,
      "crosslingual sentence embeddings": 25009,
      "texts written different": 121655,
      "churn intent detection": 16975,
      "text classification explore": 120722,
      "multilingual joint training": 75263,
      "words time periods": 134269,
      "methods use pretrained": 69830,
      "languages improves performance": 60630,
      "documents source target": 33295,
      "using small number": 130188,
      "embedding space model": 35498,
      "applied language pair": 7083,
      "suggest models trained": 115320,
      "hierarchical models robust": 49978,
      "shared vector space": 108116,
      "capturing semantic similarities": 15489,
      "available language pairs": 11024,
      "performance training data": 86809,
      "significantly outperforms competitive": 109007,
      "language pairs fully": 59760,
      "multisense word embeddings": 75778,
      "used supervised learning": 128795,
      "challenging tasks natural": 16336,
      "language processing like": 59879,
      "embeddings encode information": 35688,
      "compared traditional evaluation": 19478,
      "induction experimental results": 54024,
      "par stateoftheart models": 84526,
      "english recent work": 37260,
      "word embeddings created": 133073,
      "empirical evaluation demonstrates": 36159,
      "evaluation demonstrates effectiveness": 39177,
      "pos tagging english": 88231,
      "crosslingual nlp tasks": 24988,
      "capture complementary information": 15277,
      "good word vector": 48511,
      "present series experiments": 89692,
      "word embeddings existing": 133105,
      "different context windows": 31059,
      "translation achieve competitive": 124620,
      "word embeddings transfer": 133230,
      "annotated data scarce": 6166,
      "different domains including": 31113,
      "data augmentation unsupervised": 25656,
      "outperforms alternative approaches": 82837,
      "data contains information": 25790,
      "vast majority languages": 131277,
      "majority languages world": 66607,
      "typologically similar languages": 126501,
      "datasets significant improvement": 27711,
      "languages good quality": 60613,
      "similar language pairs": 109104,
      "chinese japanese vietnamese": 16778,
      "word meaning context": 133360,
      "learning crosslingual sentence": 62473,
      "unsupervised language model": 127649,
      "reaches state art": 95919,
      "state art unsupervised": 112481,
      "tasks recent work shown": 119441,
      "improvement previous state art": 52747,
      "word embeddings introduce new": 133139,
      "better capture semantic information": 13527,
      "neural machine translation learn": 78016,
      "problem learning crosslingual representations": 91104,
      "improves current state art": 52968,
      "word vectors word embeddings": 133653,
      "results datasets different languages": 101662,
      "model substantially outperforms previous": 72104,
      "paper presents challenge community": 84157,
      "resourcerich languages english lowresource": 100928,
      "unsupervised sense embeddings paper": 127715,
      "sense embeddings paper proposes": 105678,
      "contextual word similarity scws": 22525,
      "word similarity scws dataset": 133567,
      "outperforms competitive baselines large": 82869,
      "et al 2018 paper": 38724,
      "does require parallel corpora": 33397,
      "method significantly outperforms competitive": 69145,
      "significantly outperforms competitive baselines": 109008,
      "methods significantly improve performance": 69752,
      "used supervised learning approach": 128796,
      "challenging tasks natural language": 16337,
      "natural language processing like": 76430,
      "embeddings word embeddings standard": 36030,
      "crosslingual natural language inference": 24984,
      "compared traditional evaluation methods": 19479,
      "learning crosslingual sentence representations": 62474,
      "resourcerich languages english lowresource languages": 100929,
      "unsupervised sense embeddings paper proposes": 127716,
      "stanford contextual word similarity scws": 112363,
      "contextual word similarity scws dataset": 22526,
      "results proposed model outperforms competitive": 102082,
      "outperforms competitive baselines large margin": 82870,
      "proposed method significantly outperforms competitive": 93402,
      "method significantly outperforms competitive baselines": 69146,
      "word embeddings word embeddings standard": 133250,
      "counted": 24421,
      "undetermined": 127043,
      "bee": 12628,
      "mg": 70026,
      "dogs": 33416,
      "greeks": 49078,
      "consciousness": 21161,
      "traitement": 123971,
      "textes": 121436,
      "avec": 11162,
      "unavoidable": 126602,
      "crosswords": 25093,
      "gesture": 47948,
      "crossing": 24918,
      "christiansen": 16942,
      "puzzles": 94472,
      "dlm": 32925,
      "gini": 47970,
      "pressures": 89960,
      "biblical": 13894,
      "extinction": 42056,
      "unsolvable": 127561,
      "illustrations": 51753,
      "dryer": 34202,
      "acronym": 3053,
      "taylors": 119672,
      "fluctuation": 44689,
      "exponent": 41674,
      "artifact": 9239,
      "heaps": 49639,
      "obligatory": 81154,
      "106": 193,
      "reexamination": 97489,
      "neighbours": 77004,
      "ideogram": 51641,
      "dynasty": 34354,
      "tentative": 120199,
      "unmarked": 127462,
      "corpusdriven": 24082,
      "eighteen": 35220,
      "ana": 5407,
      "2019b": 719,
      "processes language": 91606,
      "words noun": 134083,
      "historical background": 50515,
      "form modern": 45106,
      "substantial linguistic": 114870,
      "distinguish text": 32575,
      "considered reliable": 21297,
      "data compression": 25761,
      "memory properties": 68369,
      "quality current": 94625,
      "texts possible": 121581,
      "number components": 80857,
      "provide good": 93839,
      "definite knowledge": 28516,
      "processing specifically": 91797,
      "completely random": 19783,
      "useful predicting": 128917,
      "text direct": 120876,
      "theory presented": 121824,
      "obtained based": 81349,
      "main linguistic": 66438,
      "avec des": 11163,
      "correlation word": 24247,
      "functions apply": 46069,
      "efficient communication": 35064,
      "article studies": 9160,
      "leads precise": 61951,
      "computations required": 20462,
      "evolve time": 39698,
      "cultural economic": 25229,
      "economic political": 34512,
      "political issues": 87995,
      "test differences": 120444,
      "differences features": 30958,
      "usage different": 127857,
      "different characters": 31035,
      "distributions given": 32762,
      "network network": 77370,
      "reveal new": 102502,
      "networks exhibit": 77591,
      "small world": 110224,
      "furthermore compared": 46154,
      "similar properties": 109125,
      "seemingly obvious": 104527,
      "psycholinguistic evidence": 94203,
      "variety examples": 130978,
      "appear frequently": 6797,
      "value given": 130778,
      "patterns written": 85796,
      "streams using": 113560,
      "statistics information": 113201,
      "linguistics especially": 64610,
      "field digital": 43949,
      "qualitative information": 94566,
      "novel metrics": 80641,
      "behavior computational": 12649,
      "linguistics statistical": 64625,
      "studies languages": 114246,
      "published european": 94354,
      "consist words": 21351,
      "words general": 133968,
      "time diversity": 122017,
      "computer simulations": 20491,
      "sequence nodes": 107023,
      "present fundamental": 89498,
      "linguistics information": 64611,
      "finally paper": 44212,
      "words distinct": 133918,
      "gini index": 47971,
      "beginning word": 12637,
      "word roots": 133487,
      "unique aspects": 127181,
      "2015 work": 598,
      "work corpora": 134445,
      "order preferences": 82385,
      "appropriate control": 8413,
      "alternative theories": 5248,
      "meaning relations": 67667,
      "languages differ": 60499,
      "terms dependency": 120305,
      "languages change": 60438,
      "hierarchical taxonomy": 50012,
      "number type": 80989,
      "observed pattern": 81227,
      "type token": 126230,
      "results reinforce": 102120,
      "vocabulary acquisition": 131865,
      "method calculating": 68685,
      "old method": 81642,
      "measure average": 67780,
      "average information": 11197,
      "content use": 21967,
      "tokens word": 122352,
      "help establish": 49717,
      "understand performance": 126767,
      "languages employ": 60523,
      "grammatical dependency": 48695,
      "chinese vietnamese": 16833,
      "books bible": 14422,
      "differences way": 30979,
      "known information": 58306,
      "statistics different": 113198,
      "distribution pattern": 32672,
      "fluent speech": 44720,
      "signal phonological": 108689,
      "class models": 17045,
      "change study": 16370,
      "effect change": 34587,
      "model sound": 72069,
      "hypothesis tested": 51278,
      "word surprisal": 133596,
      "using bigram": 129407,
      "model calculation": 70798,
      "occurrence word": 81513,
      "new referents": 79100,
      "information structural": 55007,
      "distance new": 32429,
      "report complex": 98983,
      "specific issues": 111452,
      "understanding evolution": 126839,
      "brain processes": 14553,
      "fundamental property": 46110,
      "algorithm estimate": 4728,
      "properties texts": 92486,
      "text attempted": 120666,
      "main text": 66470,
      "text transcribed": 121373,
      "typology order": 126505,
      "poisson regression": 87897,
      "systems derived": 116839,
      "instances concept": 55621,
      "world trade": 135052,
      "language users": 60310,
      "different concepts": 31051,
      "kernel method": 57533,
      "correlation entropy": 24229,
      "taylors law": 119673,
      "event time": 39549,
      "application language": 6856,
      "texts exhibit": 121512,
      "code results": 18153,
      "known zipfs": 58334,
      "cross language": 24863,
      "finding indicates": 44276,
      "languages statistical": 60895,
      "cognitive mechanisms": 18269,
      "mechanisms results": 68068,
      "time corpus": 122003,
      "heaps law": 49640,
      "law heaps": 61684,
      "prove superior": 93703,
      "accurate estimates": 2333,
      "errors text": 38413,
      "eyetracking study": 42640,
      "correct words": 24129,
      "increases difficulty": 53640,
      "paper proves": 84373,
      "classifier language": 17554,
      "specific classifiers": 111418,
      "used nouns": 128666,
      "2007 use": 555,
      "corresponding semantics": 24303,
      "languages statistically": 60896,
      "approach examine": 7546,
      "followed series": 44953,
      "closed class": 17841,
      "class words": 17054,
      "spreading activation": 112054,
      "post hoc": 88456,
      "family names": 43057,
      "predictions regarding": 89187,
      "natural communicative": 76243,
      "remarkably similar": 98854,
      "verbs results": 131560,
      "constraints described": 21595,
      "sets furthermore": 107675,
      "explored role": 41635,
      "process result": 91563,
      "patterns remarkably": 85774,
      "empirical data": 36157,
      "yields excellent": 135409,
      "word island": 133327,
      "words results": 134177,
      "methodology using": 69269,
      "potential overcome": 88577,
      "difficulties existing": 31682,
      "human word": 50988,
      "consistency results": 21367,
      "studies experiments": 114223,
      "method demonstrate": 68761,
      "canonical word": 15157,
      "largescale experiments": 61438,
      "role linguistic": 103199,
      "evidence shown": 39670,
      "alphabetic writing": 5208,
      "usage frequency": 127860,
      "tang dynasty": 117577,
      "showed different": 108373,
      "distribution consistent": 32644,
      "small case": 110140,
      "english native": 37214,
      "chinese classical": 16744,
      "language syntactic": 60138,
      "metrics evaluate": 69957,
      "adjective ordering": 3857,
      "ordering rules": 82440,
      "speakers corpus": 111310,
      "word indicate": 133318,
      "typically exhibit": 126428,
      "evolution word": 39693,
      "attempts explain": 9768,
      "regarding word": 97662,
      "perspective particular": 87170,
      "genetic algorithms": 47820,
      "psycholinguistic studies": 94207,
      "al 2019b": 4648,
      "information earlier": 54510,
      "methods establish": 69479,
      "later word": 61650,
      "word present": 133418,
      "grammatical classes": 48691,
      "role comparing": 103162,
      "networks paper presents": 77691,
      "novel metrics proposed": 80642,
      "unify different areas": 127149,
      "word order preferences": 133380,
      "offer simple alternative": 81573,
      "different positions sentence": 31342,
      "selecting best candidate": 104746,
      "algorithm takes advantage": 4805,
      "takes advantage fact": 117527,
      "effect different writing": 34593,
      "chinese texts using": 16829,
      "analysis syntactic structures": 5882,
      "spoken written language": 112026,
      "known zipfs law": 58335,
      "word frequency distributions": 133294,
      "law heaps law": 61685,
      "prove superior accuracy": 93704,
      "languages statistically significant": 60897,
      "closed class words": 17842,
      "data sets furthermore": 26426,
      "patterns remarkably similar": 85775,
      "usage frequency distributions": 127861,
      "language models leads": 59579,
      "native speakers corpus": 76234,
      "evolutionary model language": 39696,
      "using genetic algorithms": 129707,
      "et al 2019b": 38736,
      "existing methods comparing": 40179,
      "statistically significant differences": 113183,
      "linguistic features associated": 64480,
      "networks paper presents preliminary": 77692,
      "sociology": 110480,
      "worthwhile": 135075,
      "userspecified": 129188,
      "featureengineering": 43342,
      "presume": 89963,
      "agenda": 4291,
      "omnipresent": 81660,
      "manipulative": 66934,
      "helpfulness": 49803,
      "thats": 121773,
      "gun": 49280,
      "bipolar": 14225,
      "crosstopic": 25085,
      "subreddit": 114804,
      "217": 767,
      "topicdependent": 122607,
      "facet": 42754,
      "14k": 336,
      "theoreticallymotivated": 121807,
      "incurred": 53752,
      "counterarguments": 24424,
      "18k": 429,
      "instruct": 55702,
      "supreme": 116051,
      "47000": 1048,
      "parliaments": 84874,
      "accommodates": 1941,
      "laypeople": 61823,
      "bootstraps": 14465,
      "alludes": 5200,
      "maze": 67559,
      "negate": 76907,
      "accommodation": 1942,
      "diction": 30855,
      "generalised": 46743,
      "understanding ways": 127010,
      "frame arguments": 45382,
      "understanding public": 126932,
      "paper adopt": 83719,
      "problems involving": 91329,
      "used understanding": 128833,
      "corpora facilitate": 23487,
      "study preliminary": 114473,
      "preliminary analyses": 89262,
      "analyses suggest": 5469,
      "popular text": 88137,
      "text finding": 120953,
      "promising line": 92278,
      "challenges given": 16165,
      "data adapting": 25571,
      "extensive annotation": 41858,
      "structures proposed": 114099,
      "performance base": 86163,
      "essays annotated": 38533,
      "scheme annotation": 103912,
      "problems understanding": 91379,
      "prediction score": 89121,
      "classifiers ablation": 17596,
      "study observe": 114457,
      "systems final": 116905,
      "tasks extracting": 119129,
      "based pointer": 11929,
      "following reasons": 44982,
      "argument relations": 8954,
      "attempts learn": 9773,
      "far superior": 43101,
      "performance regular": 86666,
      "component type": 20009,
      "research tends": 100638,
      "generation design": 47362,
      "models subtask": 74116,
      "model public": 71828,
      "public corpora": 94240,
      "evidence graph": 39647,
      "different argument": 31009,
      "technique evaluate": 119784,
      "articles social": 9217,
      "analysis traditional": 5904,
      "carry task": 15560,
      "task relies": 118632,
      "levels paper": 63556,
      "compared proposed": 19427,
      "embeddings formed": 35728,
      "articles results": 9215,
      "comparison handcrafted": 19547,
      "neural endtoend": 77910,
      "argumentative relation": 8970,
      "identification argument": 51365,
      "nlp typically": 79788,
      "typically includes": 126437,
      "arguments diverse": 8983,
      "problem sentencelevel": 91214,
      "argument detection": 8939,
      "collected online": 18430,
      "types explore": 126287,
      "explore leveraging": 41555,
      "reviews work": 102640,
      "annotations annotations": 6408,
      "components usually": 20052,
      "superior style": 115697,
      "effects linguistic": 34989,
      "features allows": 43360,
      "allows predict": 5179,
      "74 accuracy": 1236,
      "arguments allows": 8977,
      "identify linguistic": 51517,
      "review helpfulness": 102552,
      "identification problem": 51423,
      "relations known": 98195,
      "time online": 122062,
      "online conversations": 81755,
      "variables interact": 130857,
      "processing belief": 91630,
      "rapidly expanding": 95728,
      "expanding application": 40362,
      "support new": 115982,
      "casual conversation": 15683,
      "initial working": 55230,
      "online dialogue": 81764,
      "apply bootstrapping": 7165,
      "representing linguistic": 100055,
      "describes english": 29402,
      "speech recording": 111792,
      "controversial topics": 22866,
      "production data": 92053,
      "summaries dialogues": 115434,
      "focused summarizing": 44878,
      "aimed identifying": 4484,
      "represented small": 100035,
      "value judgment": 130781,
      "patterns work": 85795,
      "detailed annotation": 29760,
      "reliable corpus": 98612,
      "patterns test": 85786,
      "core technology": 23335,
      "argument search": 8958,
      "despite usefulness": 29739,
      "heterogeneous texts": 49867,
      "new sentential": 79128,
      "opponent ad": 82138,
      "hominem argument": 50611,
      "research lacks": 100542,
      "performing largescale": 86955,
      "providing linguistic": 94122,
      "investigate multitask": 56786,
      "identification results": 51433,
      "results mtl": 101963,
      "particularly better": 85472,
      "effective argument": 34634,
      "argument construction": 8938,
      "machines work": 66346,
      "given statement": 48136,
      "propose encoderdecoder": 92648,
      "model enriched": 71082,
      "producing final": 92018,
      "keyphrases experiments": 57624,
      "method identifying": 68879,
      "new users": 79237,
      "users domains": 129110,
      "study large": 114422,
      "mining machine": 70239,
      "assessing crosslingual": 9508,
      "works equally": 134943,
      "using costly": 129513,
      "studies discussion": 114214,
      "sufficiently paper": 115296,
      "accuracy terms": 2300,
      "boosting model": 14444,
      "score 85": 104049,
      "review fact": 102550,
      "costly need": 24387,
      "scientific writing": 104028,
      "assess efficiency": 9476,
      "essential element": 38550,
      "errors making": 38383,
      "arguments work": 8999,
      "sources annotated": 110877,
      "argument annotations": 8935,
      "crowdsourcing high": 25119,
      "achieving close": 2938,
      "robust sentence": 103070,
      "errors publicly": 38406,
      "contrast learning": 22695,
      "performance respective": 86675,
      "hurt performance": 51165,
      "dataset properties": 27111,
      "finally develop": 44167,
      "aid future": 4432,
      "improves claim": 52959,
      "problem fine": 91060,
      "reddit corpus": 97299,
      "relations documents": 98142,
      "generation retrieval": 47605,
      "planning realization": 87642,
      "generation appealing": 47308,
      "study specific": 114522,
      "realization decoder": 96114,
      "media provides": 68159,
      "content diversity": 21875,
      "diversity automatic": 32867,
      "evaluation largescale": 39252,
      "comparisons human": 19592,
      "achieving impressive": 2955,
      "argument classification": 8936,
      "classification improve": 17228,
      "debates work": 27895,
      "arguments given": 8986,
      "identify arguments": 51476,
      "know time": 57735,
      "assessment argument": 9520,
      "factor compared": 42855,
      "annotated independently": 6203,
      "subjective nature": 114696,
      "task annotation": 117880,
      "suggest neural": 115321,
      "discussions goal": 32329,
      "classification developed": 17182,
      "online dialogues": 81765,
      "introduce annotation": 56379,
      "produce reliable": 91920,
      "aim propose": 4471,
      "small nonexistent": 110175,
      "argument representation": 8955,
      "information incorporating": 54680,
      "prove usefulness": 93707,
      "quality larger": 94702,
      "dataset previous": 27107,
      "implicit knowledge": 52012,
      "clause types": 17691,
      "knowledge terms": 58205,
      "guide process": 49244,
      "revealing implicit": 102510,
      "topics discourse": 122622,
      "progress far": 92150,
      "far focus": 43080,
      "ignoring dynamic": 51688,
      "study automatically": 114327,
      "supreme court": 116052,
      "topic discourse": 122507,
      "especially social": 38510,
      "impact research": 51887,
      "claims propose": 17018,
      "digital formats": 31722,
      "performance stance": 86731,
      "publish datasets": 94342,
      "concise summary": 20704,
      "predict key": 88896,
      "advance study": 3934,
      "audio transcripts": 10241,
      "analysis covering": 5552,
      "argumentative writing": 8973,
      "debate forums": 27888,
      "paid little": 83415,
      "extracting propositions": 42229,
      "2016 presidential": 612,
      "presidential debates": 89947,
      "building findings": 14842,
      "forums provide": 45317,
      "arguments online": 8990,
      "features play": 43657,
      "summarization crossdomain": 115489,
      "develop method": 30211,
      "intrinsic quality": 56369,
      "dimensions natural": 31770,
      "language arguments": 58845,
      "relevance topic": 98519,
      "topic knowledge": 122534,
      "corpus observe": 23910,
      "features turn": 43768,
      "mining computational": 70229,
      "work presenting": 134693,
      "consider aspects": 21190,
      "important downstream": 52152,
      "judgments crowdsourcing": 57442,
      "include text": 53238,
      "text irrelevant": 121066,
      "seed examples": 104502,
      "examples approach": 39809,
      "approach adapted": 7329,
      "corpora providing": 23565,
      "corpus quality": 23954,
      "dataset prior": 27108,
      "exist apply": 40017,
      "engine available": 37027,
      "identifying extracting": 51598,
      "evaluation corpus": 39163,
      "structure grammar": 113872,
      "spirit reproducible": 111932,
      "language according": 58807,
      "novel perspective": 80683,
      "social status": 110457,
      "agreement present": 4400,
      "networks promising": 77708,
      "results suggesting": 102236,
      "scripts needed": 104269,
      "lies core": 63976,
      "framework aims": 45418,
      "partofspeech dependency": 85525,
      "finetuned language": 44429,
      "testing ground": 120594,
      "annotation phase": 6351,
      "literature empirical": 64753,
      "study measuring": 114436,
      "learning finetuned": 62584,
      "sentences taken": 106516,
      "performance fully": 86395,
      "unit trained": 127216,
      "trained accomplish": 123064,
      "accomplish tasks": 1953,
      "contextaware models": 22339,
      "types contextual": 126265,
      "context incorporated": 22136,
      "model classifying": 70832,
      "types improve": 126297,
      "classifiers perform": 17623,
      "units conducted": 127237,
      "apply tokenlevel": 7224,
      "tokenlevel classification": 122288,
      "argumentative essays": 8969,
      "middle school": 70072,
      "compare variety": 19312,
      "models discrete": 73083,
      "bertbased architectures": 13269,
      "level classification": 63428,
      "research field computational": 100506,
      "new gold standard": 78939,
      "article present novel": 9146,
      "annotation scheme annotation": 6369,
      "encourage future research": 36756,
      "language propose new": 59998,
      "different training strategies": 31509,
      "model based pointer": 70740,
      "based pointer network": 11930,
      "pointer network architecture": 87825,
      "joint model achieves": 57290,
      "individual models subtask": 53921,
      "experiment results model": 40499,
      "stateoftheart supervised learning": 112979,
      "articles social media": 9218,
      "word embeddings formed": 133117,
      "comparison handcrafted features": 19548,
      "improves performance argument": 53010,
      "rnns propose novel": 102978,
      "rnn based methods": 102905,
      "rapidly expanding application": 95729,
      "application dialogue systems": 6845,
      "paper describes english": 83821,
      "rich source information": 102788,
      "training data consisting": 123445,
      "opponent ad hominem": 82139,
      "ad hominem argument": 3163,
      "architectures multitask learning": 8830,
      "investigate multitask learning": 56787,
      "learning method applied": 62718,
      "french spanish chinese": 45825,
      "sentencelevel classification task": 106147,
      "improve performance respective": 52481,
      "learning improve performance": 62642,
      "compared singletask learning": 19444,
      "different domains data": 31107,
      "problem fine tuning": 91061,
      "empirical results using": 36194,
      "using approach improves": 129341,
      "diverse domains social": 32810,
      "twostep generation model": 126172,
      "automatic evaluation largescale": 10531,
      "yields significantly higher": 135434,
      "significantly higher bleu": 108918,
      "word embeddings experiment": 133106,
      "achieving impressive results": 2957,
      "far know time": 43088,
      "subjective nature task": 114697,
      "tackle problem predicting": 117306,
      "network models able": 77342,
      "character word level": 16478,
      "prior work largely": 90750,
      "argumentative relation classification": 8971,
      "framework significantly outperform": 45685,
      "addition analyze quality": 3400,
      "commonsense knowledge relations": 19001,
      "especially social media": 38511,
      "able generate highquality": 1647,
      "performance stance detection": 86732,
      "stance detection models": 112169,
      "publish datasets code": 94343,
      "generating concise summary": 47206,
      "serve strong baselines": 107304,
      "baselines future work": 12400,
      "paid little attention": 83416,
      "present largescale analysis": 89532,
      "online debate forums": 81760,
      "substantially outperform previous": 114905,
      "natural language arguments": 76270,
      "important downstream tasks": 52153,
      "present experiments using": 89484,
      "demonstrate performance subtasks": 28817,
      "language processing complex": 59854,
      "models bert xlnet": 72841,
      "spirit reproducible research": 111933,
      "provide novel perspective": 93883,
      "neural networks promising": 78540,
      "finetuned language model": 44430,
      "literature empirical study": 64754,
      "learning labeled data": 62668,
      "different domains approach": 31106,
      "performance fully supervised": 86396,
      "types contextual information": 126266,
      "models discrete features": 73084,
      "research field computational linguistics": 100507,
      "new gold standard corpus": 78940,
      "network rnn based methods": 77415,
      "rnn based methods use": 102906,
      "opponent ad hominem argument": 82140,
      "little training data available": 64832,
      "diverse domains social media": 32811,
      "paper tackle problem predicting": 84466,
      "neural network models able": 78343,
      "serve strong baselines future": 107305,
      "strong baselines future work": 113648,
      "natural language processing complex": 76410,
      "neural network rnn based methods": 78388,
      "network rnn based methods use": 77416,
      "serve strong baselines future work": 107306,
      "259": 832,
      "backed": 11295,
      "damaged": 25526,
      "treetosequence": 125732,
      "dropping": 34185,
      "groundhog": 49119,
      "memorizes": 68282,
      "394": 988,
      "unfolding": 127074,
      "jhu": 57229,
      "44k": 1031,
      "englishtogermen": 37452,
      "unexploited": 127059,
      "enfi": 37000,
      "fellow": 43868,
      "outlets": 82624,
      "sparsemax": 111245,
      "5x": 1145,
      "delays": 28591,
      "counterintuitively": 24432,
      "additionsubtraction": 3639,
      "interrogating": 56298,
      "bothside": 14485,
      "6layer": 1206,
      "chunked": 16955,
      "welltuned": 132443,
      "ffn": 43920,
      "transformerbase": 124398,
      "transformerbig": 124441,
      "sequenceaware": 107102,
      "wmt15 english": 132826,
      "existing best": 40084,
      "flexible attention": 44656,
      "coverage neural": 24518,
      "tends ignore": 120173,
      "nmt consider": 79849,
      "standard attentionbased": 112195,
      "nmt adding": 79829,
      "models alleviate": 72741,
      "keeps updating": 57524,
      "largescale chinesetoenglish": 61406,
      "enhanced model": 37512,
      "quality significantly": 94791,
      "submission achieves": 114727,
      "sentence memory": 105947,
      "step empirical": 113262,
      "set supervised": 107593,
      "generates candidate": 47157,
      "approach baseline": 7386,
      "pairs approach": 83489,
      "gains baseline": 46384,
      "smt baselines": 110280,
      "better alignments": 13505,
      "based nmt": 11900,
      "nmt interactive": 79879,
      "track interaction": 122731,
      "interaction history": 55947,
      "2016 neural": 610,
      "iterative refinement": 57134,
      "later stage": 61648,
      "improve output": 52448,
      "layers allows": 61760,
      "englishgerman task": 37383,
      "lstm baseline": 65600,
      "ensembling improves": 37630,
      "reducing dimensionality": 97413,
      "japaneseenglish report": 57212,
      "requirements output": 100236,
      "codes experiments": 18203,
      "scores approach": 104153,
      "vector target": 131395,
      "states paper": 113048,
      "state history": 112499,
      "history experiments": 50555,
      "nmt beam": 79841,
      "best achieves": 13303,
      "accuracyspeed tradeoff": 2323,
      "nmt learning": 79884,
      "translation past": 125103,
      "capture words": 15423,
      "factor decoding": 42857,
      "final softmax": 44127,
      "candidate lists": 15110,
      "layer nmt": 61735,
      "current nmt": 25317,
      "candidate pool": 15117,
      "losing translation": 65243,
      "memory structure": 68387,
      "nmt inference": 79878,
      "architectures neural": 8831,
      "proposed far": 93280,
      "carried english": 15536,
      "machine training": 65875,
      "best improvements": 13339,
      "sentence simple": 106075,
      "respectively neural": 101152,
      "models hybrid": 73347,
      "words translated": 134279,
      "translation probabilities": 125141,
      "models nmt": 73646,
      "sentence predicting": 105995,
      "difficult model": 31646,
      "relationship target": 98309,
      "achieved stateofart": 2702,
      "rnn units": 102939,
      "introduced nmt": 56594,
      "relieve issue": 98666,
      "incorporate relation": 53483,
      "network standard": 77444,
      "enhanced neural": 37513,
      "translation learning": 124901,
      "decoding conducted": 28081,
      "stage novel": 112150,
      "refine translation": 97589,
      "using chineseenglish": 129440,
      "nmt improvements": 79877,
      "work specifically": 134812,
      "studies attention": 114185,
      "provides analysis": 93998,
      "attention similar": 10010,
      "including global": 53296,
      "role neural": 103209,
      "attention selectively": 10001,
      "aligned source": 4922,
      "predicted target": 88968,
      "propose double": 92633,
      "vector global": 131308,
      "chinesetoenglish englishtogermen": 16868,
      "englishtogermen translation": 37453,
      "source encoding": 110753,
      "target decoding": 117599,
      "decoding phases": 28123,
      "counterparts target": 24442,
      "distance source": 32433,
      "ones experiments": 81686,
      "translation individual": 124868,
      "future contents": 46262,
      "effective architecture": 34633,
      "encoder yields": 36587,
      "structure example": 113856,
      "raw form": 95854,
      "motivated demand": 74846,
      "addition hidden": 3434,
      "decoding empirical": 28089,
      "translation dominant": 124771,
      "perform translation": 86096,
      "right left": 102834,
      "decoder order": 28038,
      "decoding nmt": 28114,
      "sequence produced": 107039,
      "generate targetside": 47030,
      "decoder performs": 28043,
      "translation prediction": 125128,
      "contexts improve": 22403,
      "chineseenglish wmt": 16864,
      "use temporal": 128322,
      "score experiments": 104071,
      "enfi translation": 37001,
      "score addition": 104052,
      "structure context": 113827,
      "model article": 70687,
      "affect final": 4216,
      "warmup steps": 132031,
      "maximum sentence": 67550,
      "hope observations": 50632,
      "data constraints": 25786,
      "fully parallelize": 45969,
      "pruning algorithms": 94164,
      "decoder results": 28053,
      "hyperparameter optimization": 51227,
      "generated news": 47099,
      "news outlets": 79353,
      "compare nmt": 19265,
      "proposed leverage": 93323,
      "modeling lead": 72459,
      "lead performance": 61865,
      "experiments wmt14": 41209,
      "single point": 109783,
      "point neural": 87808,
      "propose variations": 93153,
      "learned sentence": 62257,
      "dropped source": 34184,
      "provided languages": 93973,
      "faster transformer": 43197,
      "stateoftheart bleu": 112606,
      "networks encoderdecoder": 77585,
      "improve learning": 52405,
      "attention connections": 9807,
      "dense connection": 29011,
      "structure competitive": 113821,
      "attention propose": 9981,
      "agent decides": 4300,
      "segmentation strategy": 104637,
      "suffers issues": 115259,
      "novel keyvalue": 80609,
      "decoder focuses": 28010,
      "appropriate source": 8440,
      "stacking layers": 112134,
      "number recurrent": 80952,
      "feedforward layers": 43844,
      "new layer": 78989,
      "parameters paper": 84775,
      "pseudoparallel corpora": 94190,
      "effectively increasing": 34824,
      "convergence poor": 22922,
      "challenges paper": 16189,
      "character input": 16436,
      "training begins": 123372,
      "model ngrams": 71584,
      "attention entities": 9832,
      "tasks wmt": 119603,
      "investigates alternative": 56849,
      "translation modeling": 124950,
      "twodimensional 2d": 126136,
      "case word": 15633,
      "disambiguation neural": 31978,
      "models hypothesize": 73348,
      "patterns translating": 85789,
      "tokens layers": 122314,
      "units existing": 127241,
      "interpretable experiments": 56238,
      "quality speed": 94796,
      "nmt far": 79867,
      "introduce recurrent": 56526,
      "constant memory": 21516,
      "sentences linear": 106383,
      "network decode": 77218,
      "representation unlike": 99457,
      "sentence passive": 105981,
      "routing policy": 103324,
      "core properties": 23331,
      "work capsule": 134406,
      "networks empirically": 77580,
      "shown potential": 108509,
      "translation named": 124998,
      "model enhances": 71081,
      "including transformer": 53400,
      "variants transformer": 130883,
      "translation simplification": 125259,
      "sums word": 115664,
      "conventional models": 22891,
      "big performance": 13997,
      "nmt uses": 80005,
      "network explicit": 77249,
      "translation unit": 125400,
      "unit words": 127223,
      "suggesting possibility": 115349,
      "performance explicit": 86361,
      "combines multihead": 18692,
      "syntactic formalism": 116407,
      "phrases enhance": 87421,
      "power attention": 88629,
      "explicitly used": 41398,
      "generation function": 47413,
      "process compared": 91439,
      "germanenglish dataset": 47932,
      "demonstrate attention": 28674,
      "models reliably": 73918,
      "characterbased translation": 16502,
      "baseline paper": 12281,
      "use known": 128107,
      "deeper transformer": 28452,
      "stateoftheart transformerbased": 113014,
      "systems follow": 116911,
      "encoder selfattention": 36572,
      "encoder effectively": 36518,
      "capture general": 15305,
      "information resulting": 54927,
      "representation backbone": 99177,
      "backbone information": 11287,
      "focused paper": 44869,
      "compression method": 20311,
      "representation nmt": 99358,
      "ways including": 132162,
      "backbone sourceside": 11289,
      "sourceside fusion": 110936,
      "fusion targetside": 46248,
      "targetside fusion": 117797,
      "fusion bothside": 46233,
      "bothside fusion": 14486,
      "fusion integrate": 46237,
      "fusion neural": 46246,
      "models deeper": 73027,
      "deeper neural": 28450,
      "framework ease": 45509,
      "desired representation": 29670,
      "finegrained representation": 44375,
      "evidence showing": 39669,
      "difficult slow": 31661,
      "works character": 134930,
      "6layer transformer": 1207,
      "transformer base": 124300,
      "morphological phenomena": 74718,
      "models extremely": 73225,
      "commonly applied": 18957,
      "approach focuses": 7577,
      "transformerbased approach": 124400,
      "character levels": 16448,
      "developed characterlevel": 30258,
      "translation objective": 125068,
      "given continuous": 48006,
      "quality propose": 94764,
      "vector neural": 131331,
      "translation effective": 124775,
      "introduced sequence": 56597,
      "nmt based": 79837,
      "sentences improving": 106350,
      "predict language": 88897,
      "language depending": 58943,
      "model changes": 70817,
      "universal model": 127312,
      "translation transformer": 125393,
      "significantly pushed": 109043,
      "pushed forward": 94460,
      "attentional network": 10049,
      "connect distant": 21117,
      "words shorter": 134216,
      "shorter network": 108292,
      "network paths": 77385,
      "paths rnns": 85691,
      "difficulty fully": 31697,
      "smt approach": 110278,
      "representations transformer": 99943,
      "wmt 14": 132765,
      "14 englishgerman": 313,
      "tasks strong": 119524,
      "transformer big": 124312,
      "sentences substantially": 106508,
      "valuable contribution": 130760,
      "deep transformers": 28434,
      "stabilizes training": 112110,
      "25 bleu": 811,
      "layers perform": 61797,
      "shallow models": 107923,
      "14 times": 320,
      "units improves": 127249,
      "chinesetoenglish wmt14": 16877,
      "model 54": 70502,
      "parameters results": 84781,
      "transformers vaswani": 124467,
      "vulnerable noise": 131977,
      "models problem": 73804,
      "training idea": 123653,
      "entire test": 37719,
      "networks construct": 77545,
      "networks capturing": 77531,
      "sentence performed": 105983,
      "ngram feature": 79432,
      "normalized output": 80357,
      "given sequential": 48121,
      "input recurrent": 55413,
      "use attentionbased": 127902,
      "approach higher": 7603,
      "little work exploring": 64838,
      "method achieved improvement": 68589,
      "standard attentionbased nmt": 112196,
      "attentional neural machine": 10051,
      "dataset demonstrate proposed": 26859,
      "translation nmt adding": 125027,
      "experiments largescale chinesetoenglish": 40980,
      "translation quality significantly": 125184,
      "shared task news": 108070,
      "task news translation": 118458,
      "machine translation efficient": 65965,
      "propose novel mechanism": 92936,
      "approach achieves better": 7302,
      "source sentence generating": 110811,
      "propose supervised attention": 93097,
      "nmt paper propose": 79940,
      "nmt baseline stateoftheart": 79839,
      "al 2016 neural": 4613,
      "2016 neural machine": 611,
      "existing machine translation": 40169,
      "decoding algorithms generate": 28074,
      "existing translation based": 40326,
      "bidirectional lstm baseline": 13951,
      "translation nmt improve": 125036,
      "improves quality generated": 53034,
      "prediction paper propose": 89099,
      "layer neural machine": 61732,
      "binary codes experiments": 14127,
      "reducing memory usage": 97427,
      "vector target word": 131396,
      "states paper propose": 113049,
      "models achieve significant": 72671,
      "standard machine translation": 112255,
      "nmt beam search": 79842,
      "deep recurrent model": 28403,
      "machine translation past": 66153,
      "attend source words": 9779,
      "decoding neural machine": 28111,
      "target language vocabulary": 117661,
      "language vocabulary size": 60337,
      "target words address": 117753,
      "words address problem": 133810,
      "source sentence usually": 110818,
      "losing translation quality": 65244,
      "rare words propose": 95751,
      "architectures neural machine": 8832,
      "strong nmt baseline": 113696,
      "improvements stateoftheart baselines": 52915,
      "problem existing nmt": 91041,
      "largescale chinesetoenglish englishtogermen": 61407,
      "chinesetoenglish englishtogermen translation": 16869,
      "englishtogermen translation tasks": 37454,
      "tasks proposed approach": 119412,
      "distance source target": 32434,
      "experiments analysis presented": 40768,
      "analysis presented paper": 5763,
      "presented paper demonstrate": 89792,
      "paper demonstrate proposed": 83802,
      "models able significantly": 72648,
      "modeling past future": 72506,
      "past future contents": 85643,
      "tasks specifically proposed": 119514,
      "model outperforms conventional": 71642,
      "english french task": 37146,
      "machine translation dominant": 65961,
      "chineseenglish wmt englishgerman": 16865,
      "source word paper": 110867,
      "word paper propose": 133398,
      "final translation quality": 44135,
      "propose new recurrent": 92882,
      "results significantly lower": 102184,
      "attention networks recently": 9952,
      "wmt14 englishgerman translation": 132817,
      "neural networks encoderdecoder": 78474,
      "networks encoderdecoder framework": 77586,
      "outperforms previously proposed": 82961,
      "remarkable progress recent": 98842,
      "increase number parameters": 53607,
      "number parameters paper": 80936,
      "parameters paper propose": 84776,
      "comparable translation quality": 19191,
      "model evaluate performance": 71108,
      "translation tasks wmt": 125369,
      "analysis attention mechanisms": 5506,
      "sense disambiguation neural": 105663,
      "disambiguation neural machine": 31979,
      "learn extract features": 62046,
      "competitive performance english": 19663,
      "translation quality speed": 125185,
      "introduce recurrent neural": 56527,
      "dynamic routing policy": 34324,
      "results stateoftheart nmt": 102207,
      "machine translation named": 66076,
      "text classification applications": 120709,
      "nmt models better": 79910,
      "understanding neural machine": 126903,
      "machine translation simplification": 66220,
      "sums word embeddings": 115665,
      "standard transformer recurrent": 112330,
      "embeddings attention mechanisms": 35561,
      "big performance drop": 13998,
      "multihead selfattention network": 75131,
      "information prior work": 54869,
      "combines multihead selfattention": 18693,
      "encoder effectively capture": 36519,
      "effectively capture general": 34795,
      "capture general information": 15306,
      "general information resulting": 46664,
      "specifically focused paper": 111553,
      "focused paper propose": 44870,
      "backbone sourceside fusion": 11290,
      "sourceside fusion targetside": 110937,
      "fusion targetside fusion": 46249,
      "targetside fusion bothside": 117798,
      "fusion bothside fusion": 46234,
      "bothside fusion integrate": 14487,
      "translation systems require": 125315,
      "result new stateoftheart": 101386,
      "deeper neural networks": 28451,
      "deep nmt models": 28389,
      "iwslt translation tasks": 57167,
      "translation tasks translation": 125365,
      "tasks translation directions": 119576,
      "requires deep architectures": 100259,
      "works character level": 134931,
      "better capture morphological": 13525,
      "machine translation objective": 66130,
      "machine translation effective": 65963,
      "sequence sequence based": 107054,
      "improving overall accuracy": 53134,
      "accuracy paper propose": 2229,
      "machine translation transformer": 66278,
      "translation model vaswani": 124947,
      "significantly pushed forward": 109044,
      "connect distant words": 21118,
      "distant words shorter": 32483,
      "words shorter network": 134217,
      "shorter network paths": 108293,
      "network paths rnns": 77386,
      "translation smt approach": 125269,
      "ability capture longdistance": 1492,
      "improvements wmt 14": 52940,
      "wmt 14 englishgerman": 132766,
      "transformer base models": 124302,
      "transformer big models": 124313,
      "substantially fewer parameters": 114890,
      "nmt using simple": 80007,
      "open question paper": 81925,
      "achieve remarkable success": 2522,
      "experimental results machine": 40633,
      "nist chinesetoenglish wmt14": 79504,
      "chinesetoenglish wmt14 englishtogerman": 16878,
      "robust neural machine": 103052,
      "transformers vaswani et": 124468,
      "approach higher bleu": 7604,
      "attentional neural machine translation": 10052,
      "machine translation nmt adding": 66091,
      "shared task news translation": 108071,
      "paper propose new attention": 84277,
      "achieve significant improvements previous": 2539,
      "et al 2016 neural": 38704,
      "al 2016 neural machine": 4614,
      "model convolutional neural network": 70923,
      "machine translation nmt improve": 66100,
      "paper propose new method": 84280,
      "layer neural machine translation": 61733,
      "model significantly improves translation": 72024,
      "decoding neural machine translation": 28112,
      "target language vocabulary size": 117662,
      "target words address problem": 117754,
      "architectures neural machine translation": 8833,
      "neural machine translation word": 78132,
      "paper propose use word": 84341,
      "englishtogerman translation tasks proposed": 37451,
      "achieves substantial improvements stateoftheart": 2914,
      "substantial improvements stateoftheart baselines": 114864,
      "neural machine translation provides": 78084,
      "experiments largescale chinesetoenglish englishtogermen": 40981,
      "largescale chinesetoenglish englishtogermen translation": 61408,
      "chinesetoenglish englishtogermen translation tasks": 16870,
      "englishtogermen translation tasks proposed": 37455,
      "translation tasks proposed approach": 125361,
      "source target word embeddings": 110846,
      "experiments analysis presented paper": 40769,
      "analysis presented paper demonstrate": 5764,
      "presented paper demonstrate proposed": 89793,
      "proposed model outperforms conventional": 93462,
      "tasks demonstrate model achieves": 119039,
      "neural networks encoderdecoder framework": 78475,
      "achieved remarkable progress recent": 2686,
      "remarkable progress recent years": 98843,
      "number parameters paper propose": 80937,
      "statistical machine translation paper": 113109,
      "sequence sequence model neural": 107059,
      "sequence model neural machine": 107010,
      "word sense disambiguation neural": 133519,
      "sense disambiguation neural machine": 105664,
      "disambiguation neural machine translation": 31980,
      "bahdanau et al 2014": 11389,
      "results stateoftheart nmt systems": 102208,
      "achieve best results reported": 2423,
      "wmt14 englishtogerman englishtofrench translation": 132823,
      "understanding neural machine translation": 126904,
      "tasks proposed approach consistently": 119413,
      "proposed approach consistently improves": 93189,
      "encoder effectively capture general": 36520,
      "effectively capture general information": 34796,
      "capture general information resulting": 15307,
      "specifically focused paper propose": 111554,
      "focused paper propose explicit": 44871,
      "backbone sourceside fusion targetside": 11291,
      "sourceside fusion targetside fusion": 110938,
      "fusion targetside fusion bothside": 46250,
      "targetside fusion bothside fusion": 117799,
      "fusion bothside fusion integrate": 46235,
      "machine translation systems require": 66253,
      "iwslt translation tasks translation": 57168,
      "translation tasks translation directions": 125366,
      "neural machine translation transformer": 78119,
      "transformer translation model vaswani": 124390,
      "translation model vaswani et": 124948,
      "based multihead attention mechanism": 11863,
      "connect distant words shorter": 21119,
      "distant words shorter network": 32484,
      "words shorter network paths": 134218,
      "shorter network paths rnns": 108294,
      "machine translation smt approach": 66226,
      "improvements wmt 14 englishgerman": 52941,
      "experimental results machine translation": 40634,
      "results machine translation tasks": 101901,
      "nist chinesetoenglish wmt14 englishtogerman": 79505,
      "robust neural machine translation": 103053,
      "transformers vaswani et al": 124469,
      "improve neural machine translation nmt": 52440,
      "neural machine translation nmt adding": 78042,
      "paper propose new attention mechanism": 84278,
      "et al 2016 neural machine": 38705,
      "technique neural machine translation nmt": 119805,
      "neural machine translation nmt improve": 78050,
      "experiments proposed model significantly improves": 41085,
      "decoding neural machine translation nmt": 28113,
      "achieves substantial improvements stateoftheart baselines": 2915,
      "experiments largescale chinesetoenglish englishtogermen translation": 40982,
      "largescale chinesetoenglish englishtogermen translation tasks": 61409,
      "chinesetoenglish englishtogermen translation tasks proposed": 16871,
      "experiments analysis presented paper demonstrate": 40770,
      "analysis presented paper demonstrate proposed": 5765,
      "architectures neural machine translation nmt": 8834,
      "achieved remarkable progress recent years": 2687,
      "sequence sequence model neural machine": 107060,
      "sequence model neural machine translation": 107011,
      "word sense disambiguation neural machine": 133520,
      "sense disambiguation neural machine translation": 105665,
      "translation tasks proposed approach consistently": 125362,
      "encoder effectively capture general information": 36521,
      "effectively capture general information resulting": 34797,
      "specifically focused paper propose explicit": 111555,
      "backbone sourceside fusion targetside fusion": 11292,
      "sourceside fusion targetside fusion bothside": 110939,
      "fusion targetside fusion bothside fusion": 46251,
      "targetside fusion bothside fusion integrate": 117800,
      "neural machine translation systems require": 78112,
      "iwslt translation tasks translation directions": 57169,
      "transformer translation model vaswani et": 124391,
      "translation model vaswani et al": 124949,
      "connect distant words shorter network": 21120,
      "distant words shorter network paths": 32485,
      "words shorter network paths rnns": 134219,
      "statistical machine translation smt approach": 113113,
      "experimental results machine translation tasks": 40635,
      "transformers vaswani et al 2017": 124470,
      "lexicalfunctional": 63846,
      "pitmanyor": 87577,
      "om": 81649,
      "stackpropagation": 112140,
      "o1": 81040,
      "dragnn": 34085,
      "347": 958,
      "stacklstms": 112137,
      "nonbinary": 80166,
      "lee": 63292,
      "on4": 81663,
      "ims": 53188,
      "blending": 14272,
      "bilexical": 14016,
      "workinprogress": 134912,
      "casebased": 15634,
      "successors": 115203,
      "transitional": 124497,
      "displacement": 32377,
      "ics": 51312,
      "july": 57455,
      "blow": 14379,
      "linearizations": 64398,
      "results penn": 102031,
      "increases increasing": 53645,
      "parse selection": 84884,
      "text tokens": 121369,
      "stanford typed": 112383,
      "typed dependencies": 126236,
      "parsing algorithms": 85068,
      "tradeoff accuracy": 122774,
      "spmrl shared": 111959,
      "models highorder": 73335,
      "secondorder features": 104479,
      "scanned books": 103782,
      "gains parsing": 46403,
      "domains best": 33735,
      "trees encoding": 125689,
      "experiments resulting": 41119,
      "surpass current": 116090,
      "model generative": 71245,
      "overcomes limitations": 83304,
      "previous generative": 90416,
      "accurate inference": 2336,
      "adapt beam": 3171,
      "predicting pos": 89000,
      "door applications": 33950,
      "present structured": 89721,
      "parsing learn": 85143,
      "parser reaches": 84990,
      "knowledge best": 57817,
      "ablative analysis": 1588,
      "analysis determine": 5574,
      "aspects model": 9395,
      "strengths datadriven": 113572,
      "correct parsing": 24114,
      "complexity stateoftheart": 19938,
      "resulting parse": 101461,
      "solely number": 110532,
      "firstorder models": 44570,
      "models encodes": 73149,
      "similarly original": 109347,
      "parsing hierarchical": 85121,
      "compositional vector": 20138,
      "parser achieving": 84918,
      "implementation available": 51940,
      "parser globally": 84961,
      "graphbased parser": 48907,
      "simple architectures": 109366,
      "match surpass": 67369,
      "typically leverage": 126440,
      "templates demonstrate": 120084,
      "demonstrate better": 28682,
      "19 languages": 435,
      "trees training": 125717,
      "obtain impressive": 81292,
      "minimal features": 70161,
      "model parser": 71711,
      "automatically identifies": 10783,
      "parsing does": 85103,
      "parsers different": 85019,
      "minimum bayes": 70200,
      "bayes risk": 12573,
      "risk decoding": 102875,
      "ambiguity second": 5286,
      "applying standard": 7270,
      "richer features": 102804,
      "hand incremental": 49320,
      "right hand": 102833,
      "shiftreduce parsing": 108189,
      "leverages sentence": 63659,
      "constituent parser": 21543,
      "parsing context": 85087,
      "provably optimal": 93694,
      "networks despite": 77565,
      "approaches constituency": 8108,
      "best f1": 13333,
      "stateoftheart dependency": 112636,
      "parser yields": 85007,
      "apply reinforcement": 7212,
      "unlabeled dependencies": 127393,
      "absolute labeled": 1743,
      "joint pos": 57310,
      "classifiers designed": 17607,
      "differ terms": 30934,
      "terms order": 120353,
      "topdown parsers": 122491,
      "rely strong": 98747,
      "encoder input": 36534,
      "traversal syntactic": 125544,
      "set transition": 107615,
      "transition actions": 124480,
      "combining generative": 18726,
      "expectationmaximization algorithm": 40386,
      "using discriminative": 129585,
      "use rich": 128252,
      "features universal": 43769,
      "linguistic priors": 64525,
      "propose exact": 92664,
      "algorithm parsing": 4780,
      "achieved comparable": 2605,
      "score penn": 104102,
      "present minimal": 89558,
      "score reported": 104115,
      "stacklstms present": 112138,
      "constituent nodes": 21541,
      "gold tree": 48447,
      "exhibit ambiguity": 39996,
      "used cross": 128464,
      "explore training": 41598,
      "parsing latent": 85142,
      "learning syntactic": 63080,
      "linear complexity": 64338,
      "compete complex": 19605,
      "increasingly wide": 53716,
      "used syntactic": 128800,
      "mind propose": 70146,
      "article novel": 9142,
      "outperforms static": 83022,
      "gradient method": 48602,
      "potential reduce": 88582,
      "practical implementation": 88704,
      "treebanks make": 125662,
      "empirically effective": 36230,
      "languages improved": 60628,
      "tagging component": 117379,
      "benchmark english": 12818,
      "stateoftheart pos": 112867,
      "average pos": 11206,
      "extended compared": 41804,
      "number important": 80895,
      "reduction parameter": 97454,
      "parameter redundancy": 84719,
      "biaffine classifier": 13787,
      "mechanism modeling": 68013,
      "parsing deep": 85096,
      "deep biaffine": 28201,
      "standard version": 112335,
      "features removed": 43699,
      "embeddings notably": 35835,
      "multilayer lstm": 75171,
      "syntax paper": 116550,
      "task dynamic": 118125,
      "accuracy comparison": 2121,
      "obtained performing": 81390,
      "task subtask": 118754,
      "second place": 104442,
      "teams paper": 119727,
      "paper components": 83768,
      "goal subtask": 48388,
      "analyze relation": 5997,
      "internal structural": 56174,
      "action decisions": 3076,
      "childsum treelstm": 16732,
      "encoder does": 36517,
      "neural variational": 78728,
      "learning dependency": 62494,
      "syntax annotated": 116530,
      "data traditional": 26564,
      "models global": 73303,
      "performance lags": 86479,
      "treebank empirical": 125634,
      "substantially increases": 114900,
      "solve ambiguities": 110587,
      "unlexicalized models": 127424,
      "necessary achieve": 76750,
      "modelling open": 72612,
      "open issue": 81912,
      "issue neural": 57012,
      "perform model": 86032,
      "composition helps": 20097,
      "lstm performance": 65659,
      "features crucial": 43433,
      "lstm better": 65603,
      "schema dependency": 103896,
      "casebased reasoning": 15635,
      "reasoning order": 96282,
      "based pos": 11937,
      "syntactic analyses": 116367,
      "predefined templates": 88837,
      "analyses directly": 5452,
      "ngram matches": 79443,
      "comes expense": 18778,
      "information coming": 54423,
      "information comes": 54422,
      "attempt formulate": 9744,
      "syntactic input": 116425,
      "high inference": 50077,
      "parsing order": 85179,
      "novel parser": 80682,
      "linguistics nlp": 64619,
      "advances dependency": 3986,
      "networks excel": 77590,
      "sentence kind": 105915,
      "networks able": 77499,
      "work wide": 134878,
      "time complexities": 121996,
      "spaces time": 111106,
      "parsers evaluate": 85022,
      "varies different": 130949,
      "reverse true": 102527,
      "paper shed": 84423,
      "meaningful correlations": 67715,
      "additional loss": 3546,
      "graphs use": 48974,
      "parser directly": 84937,
      "loss paper": 65290,
      "time presents": 122074,
      "structural learning": 113777,
      "modeling useful": 72584,
      "relations examples": 98161,
      "syntactic heads": 116415,
      "outside context": 83195,
      "used identifying": 128575,
      "compare common": 19230,
      "strategies experimental": 113460,
      "english dependency": 37109,
      "parsing sequencetosequence": 85234,
      "topdown tree": 122494,
      "date paper": 27829,
      "instead based": 55655,
      "costs inference": 24402,
      "efficiency models": 35033,
      "increasing recent": 53683,
      "faster baseline": 43163,
      "small increase": 110157,
      "annotation correct": 6291,
      "capability utilizing": 15187,
      "hope survey": 50642,
      "rate observe": 95798,
      "enforce constraint": 37003,
      "parsers based": 85015,
      "parsing proposed": 85203,
      "size making": 109932,
      "train lexicalized": 122950,
      "achieves 10": 2728,
      "input parameters": 55389,
      "chartbased model": 16649,
      "model constituent": 70899,
      "model gains": 71219,
      "use structural": 128300,
      "stateoftheart statistical parsers": 112973,
      "results penn treebank": 102032,
      "stanford typed dependencies": 112384,
      "tradeoff accuracy speed": 122775,
      "parsing paper describes": 85181,
      "spmrl shared task": 111960,
      "gains parsing accuracy": 46404,
      "high accuracy model": 50033,
      "previous generative models": 90417,
      "decoding algorithm based": 28071,
      "predicting pos tags": 89001,
      "model obtains better": 71604,
      "opening door applications": 81996,
      "achieving stateoftheart accuracies": 2985,
      "models typically leverage": 74256,
      "training set training": 123854,
      "minimum bayes risk": 70201,
      "bayes risk decoding": 12574,
      "approaches constituency parsing": 8109,
      "high overall accuracy": 50093,
      "stateoftheart dependency parser": 112637,
      "apply reinforcement learning": 7213,
      "learning improves accuracy": 62645,
      "conll 2017 shared": 21078,
      "shared task baseline": 108027,
      "joint pos tagging": 57311,
      "variety natural languages": 131006,
      "parsing present novel": 85197,
      "network model learns": 77334,
      "uses bidirectional lstms": 129199,
      "outperforms stateoftheart neural": 83012,
      "stateoftheart neural networkbased": 112786,
      "mitigate issues propose": 70372,
      "set transition actions": 107616,
      "simple general effective": 109440,
      "achieved comparable performance": 2606,
      "performance stateoftheart approaches": 86743,
      "network architecture propose": 77143,
      "score penn treebank": 104103,
      "used cross entropy": 128465,
      "entropy loss function": 38166,
      "parsers trained using": 85050,
      "based bidirectional recurrent": 11556,
      "novel approach outperforms": 80485,
      "policy gradient method": 87952,
      "pos tagging accuracy": 88225,
      "average pos tagging": 11207,
      "annotated training corpora": 6248,
      "performance drops significantly": 86326,
      "reduce parameter redundancy": 97350,
      "multilayer lstm language": 75172,
      "achieved second place": 2692,
      "internal structural information": 56175,
      "model improve stateoftheart": 71310,
      "framework outperforms previous": 45642,
      "paper makes attempt": 84040,
      "makes attempt formulate": 66758,
      "model jointly perform": 71387,
      "task achieves best": 117837,
      "dependency parsing models": 29183,
      "linear time complexity": 64381,
      "inference paper propose": 54186,
      "universal dependency treebanks": 127300,
      "computational linguistics nlp": 20396,
      "advances dependency parsing": 3987,
      "outperforming previous work": 82822,
      "previous work wide": 90564,
      "stateoftheart results fully": 112903,
      "paper shed light": 84424,
      "semisupervised neural tagging": 105619,
      "dependency parse trees": 29158,
      "strategies experimental results": 113461,
      "knowledge large model": 58036,
      "large model smaller": 61153,
      "small increase performance": 110158,
      "survey existing approaches": 116172,
      "future research topic": 46310,
      "making difficult train": 66835,
      "stateoftheart methods joint": 112709,
      "experiments penn treebank": 41055,
      "make use unlabeled data": 66747,
      "recently neural network approaches": 96721,
      "minimum bayes risk decoding": 70202,
      "conll 2017 shared task": 21079,
      "neural network model learns": 78335,
      "model outperforms stateoftheart neural": 71676,
      "outperforms stateoftheart neural networkbased": 83013,
      "mitigate issues propose novel": 70373,
      "cross entropy loss function": 24862,
      "dyer et al 2016": 34293,
      "multilayer lstm language model": 75173,
      "significantly better results compared": 108890,
      "paper makes attempt formulate": 84041,
      "proposed approach achieves stateoftheart": 93184,
      "knowledge large model smaller": 58037,
      "facilitate future research topic": 42772,
      "agreeing": 4385,
      "sp": 110967,
      "cumulative": 25237,
      "amplitude": 5379,
      "chung": 16949,
      "fertile": 43878,
      "possessive": 88363,
      "conspicuously": 21514,
      "21st": 769,
      "rejecting": 97836,
      "testable": 120554,
      "controller": 22849,
      "unnecessarily": 127477,
      "paradox": 84560,
      "strand": 113440,
      "processing typically": 91845,
      "statistical regularities": 113156,
      "addressing question": 3816,
      "using number": 129984,
      "agreement english": 4390,
      "furthermore language": 46186,
      "augmenting model": 10342,
      "important differences": 52141,
      "understanding deep": 126825,
      "measuring ability": 67905,
      "ability combine": 1495,
      "work helps": 134551,
      "errors inherent": 38378,
      "single rnn": 109789,
      "rnn perform": 102929,
      "nets learn": 77117,
      "statistical laws": 113102,
      "task processing": 118567,
      "arithmetic expressions": 9006,
      "visualisation techniques": 131820,
      "predictions features": 89171,
      "features hidden": 43537,
      "state representations": 112512,
      "representations time": 99934,
      "network novel": 77375,
      "expressions seen": 41763,
      "techniques does": 119873,
      "larger deeper": 61362,
      "networks dealing": 77559,
      "understanding internal": 126867,
      "input utterance": 55467,
      "typically need": 126448,
      "need make": 76832,
      "simple examples": 109429,
      "hypothesis space": 51276,
      "rnns constrained": 102965,
      "way simulate": 132130,
      "formation hierarchical": 45214,
      "rnns achieved": 102959,
      "languages italian": 60652,
      "rely semantic": 98737,
      "models scaling": 73975,
      "probabilistic contextfree": 90801,
      "line linguistic": 64326,
      "goto architecture": 48537,
      "sequence language": 106999,
      "distribution inputs": 32659,
      "rewriting task": 102699,
      "clearly defined": 17737,
      "training distribution": 123588,
      "evaluating standard": 39102,
      "deep rnns": 28419,
      "encode significant": 36454,
      "linguistic attributes": 64430,
      "general models": 46680,
      "data affect": 25584,
      "tokens longer": 122315,
      "explicitly using": 41400,
      "grammatical constructions": 48692,
      "complexity metrics": 19924,
      "metrics word": 70021,
      "hierarchical multiscale": 49980,
      "lstm hierarchical": 65627,
      "chung et": 16950,
      "structure characterlevel": 113816,
      "does correlate": 33338,
      "nlp help": 79620,
      "networks current": 77558,
      "correct form": 24101,
      "representations acquired": 99492,
      "evaluating grammaticality": 39061,
      "model assign": 70692,
      "accuracy large": 2193,
      "rapidly adapt": 95726,
      "abstract syntactic": 1781,
      "understand ability": 126735,
      "items paper": 57112,
      "results proven": 102085,
      "suite nlp": 115416,
      "relationship multiple": 98302,
      "multiple syntactic": 75717,
      "syntactic positions": 116451,
      "demonstrates stateoftheart": 28962,
      "behavior reflects": 12668,
      "broadly test": 14707,
      "test new": 120478,
      "learn appropriate": 61989,
      "reflexive pronouns": 97630,
      "learning compositionality": 62450,
      "trained squad": 123291,
      "did change": 30912,
      "structure lstm": 113904,
      "models powerful": 73762,
      "tasks sequential": 119495,
      "raises questions": 95463,
      "task agreement": 117859,
      "requires implicit": 100278,
      "regularities human": 97737,
      "language recurrent": 60027,
      "calls greater": 15074,
      "interaction neural": 55957,
      "constraints language": 21603,
      "bias exhibited": 13805,
      "experiment directly": 40464,
      "directly manipulating": 31885,
      "observe model": 81205,
      "contextfree rules": 22366,
      "investigate concepts": 56734,
      "improves learning": 52992,
      "supervision hierarchical": 115889,
      "structure enhances": 113852,
      "enhances learning": 37530,
      "represent hierarchical": 99111,
      "results types": 102289,
      "dependencies neural": 29100,
      "structures test": 114109,
      "represent syntactic": 99145,
      "way comparable": 132066,
      "languages typological": 60933,
      "order morphological": 82374,
      "affect ability": 4213,
      "propose paradigm": 92984,
      "synthetic versions": 116653,
      "findings performance": 44306,
      "bias predicting": 13832,
      "prediction significantly": 89124,
      "significantly easier": 108908,
      "account present": 2033,
      "number information": 80896,
      "units importantly": 127248,
      "models possess": 73756,
      "effective generalization": 34682,
      "handle linguistic": 49390,
      "rules deep": 103390,
      "structure investigating": 113886,
      "constructions particular": 21705,
      "environment paper": 38191,
      "pronoun gender": 92339,
      "block information": 14359,
      "linguistics cognitive": 64607,
      "ed models": 34519,
      "demonstrate humanlike": 28755,
      "human participant": 50923,
      "model findings": 71186,
      "architectures better": 8787,
      "evidence claim": 39639,
      "ability implicitly": 1531,
      "temporal structure": 120129,
      "suggest temporal": 115333,
      "representing natural": 100060,
      "input work": 55476,
      "input shown": 55436,
      "data structures": 26511,
      "present suite": 89728,
      "suite experiments": 115415,
      "order properly": 82391,
      "languages build": 60431,
      "network natural": 77368,
      "entities exhibit": 37778,
      "morphological aspects": 74690,
      "capable distinguishing": 15196,
      "evaluating computational": 39043,
      "analyses revealed": 5464,
      "structure vocabulary": 113982,
      "models generative": 73294,
      "gans text": 46437,
      "good indicator": 48480,
      "techniques exploit": 119884,
      "exploit cooccurrence": 41408,
      "cooccurrence semantic": 23232,
      "useful work": 128951,
      "additional module": 3552,
      "state current": 112488,
      "achieves lower": 2814,
      "lower perplexities": 65444,
      "lms perform": 64887,
      "organized hierarchically": 82477,
      "linguistically interpretable": 64590,
      "pragmatic inferences": 88752,
      "learns use": 63241,
      "methodology based": 69249,
      "based minimal": 11844,
      "lstm achieve": 65589,
      "structure types": 113978,
      "data generalize": 25972,
      "biases neural": 13878,
      "architecture investigate": 8678,
      "linear order": 64359,
      "investigated qualitatively": 56842,
      "use treestructured": 128342,
      "treestructured model": 125726,
      "syntactic constructions": 116385,
      "examples particular": 39864,
      "word provide": 133424,
      "analysis trained": 5905,
      "learning signals": 63024,
      "models years": 74366,
      "strings generated": 113604,
      "similar ones": 109115,
      "different syntax": 31471,
      "learning modern": 62798,
      "require extensive": 100134,
      "linguistic skills": 64556,
      "probing models": 90895,
      "knowledge raw": 58132,
      "data exposed": 25935,
      "work theoretical": 134852,
      "encoding grammatical": 36696,
      "data latent": 26079,
      "models encoding": 73150,
      "properties results": 92479,
      "relative clause": 98344,
      "assign probabilities": 9545,
      "cases multiple": 15658,
      "fail acquire": 42948,
      "lstmbased recurrent": 65687,
      "approach refines": 7850,
      "elements subject": 35284,
      "complete view": 19775,
      "hallmark human": 49300,
      "processing storing": 91804,
      "gender information": 46601,
      "agreement capturing": 4387,
      "architecture long": 8687,
      "status multiple": 113217,
      "process sentences": 91568,
      "linguistic performance": 64519,
      "structural bias": 113759,
      "evaluate bert": 38803,
      "linguistic universals": 64578,
      "exploring context": 41661,
      "vectors outputs": 131459,
      "2019 recently": 707,
      "multiple factors": 75564,
      "evaluate recent": 38912,
      "lstms comparable": 65707,
      "information grammatical": 54645,
      "limited input": 64241,
      "corpora reflect": 23572,
      "training long": 123690,
      "demonstrating model": 28977,
      "showing humans": 108407,
      "compared transformer": 19484,
      "contrary humans": 22681,
      "despite model": 29709,
      "modeling structural": 72553,
      "sentence test": 106103,
      "learned base": 62203,
      "work characterize": 134412,
      "generative process": 47754,
      "time compare": 121994,
      "times longer": 122183,
      "rnns generate": 102970,
      "provide theoretical": 93940,
      "languages reflect": 60838,
      "known results": 58321,
      "finally algorithm": 44144,
      "accuracy grammatical": 2175,
      "nouns systematically": 80445,
      "robust grammatical": 103037,
      "performance grammatical": 86424,
      "nouns grammatical": 80442,
      "variation grammatical": 130891,
      "contextual variability": 22513,
      "mechanisms different": 68060,
      "occur training": 81504,
      "perform natural": 86035,
      "applied tools": 7131,
      "units long": 127255,
      "functions sets": 46078,
      "finally tested": 44239,
      "advances recurrent": 4014,
      "study recursive": 114500,
      "length sentence": 63377,
      "bias learning": 13813,
      "model processing": 71795,
      "structures complex": 114065,
      "words nouns": 134084,
      "models miss": 73569,
      "models argue": 72770,
      "argue current": 8912,
      "argue different": 8914,
      "handle different": 49379,
      "types meaning": 126318,
      "analysis modern": 5704,
      "language processing typically": 59980,
      "using language models": 129783,
      "overall accuracy errors": 83211,
      "formal language theory": 45165,
      "models neural nets": 73631,
      "neural nets learn": 78232,
      "paper provides empirical": 84385,
      "hidden state representations": 49909,
      "networks rnns achieved": 77744,
      "languages italian english": 60653,
      "probabilistic contextfree grammar": 90802,
      "seq2seq based neural": 106890,
      "present set experiments": 89694,
      "models trained natural": 74218,
      "chung et al": 16951,
      "overall performance model": 83246,
      "language modeling recurrent": 59460,
      "modeling recurrent neural": 72524,
      "network architectures used": 77150,
      "predictions language model": 89178,
      "model automatically construct": 70713,
      "suite nlp tasks": 115417,
      "sequence modeling natural": 107013,
      "models implicitly learn": 73360,
      "structure sentence propose": 113952,
      "learn understand language": 62176,
      "extensive training data": 41961,
      "training data demonstrate": 123449,
      "performance similar human": 86715,
      "trained large datasets": 123180,
      "word order morphological": 133378,
      "natural language syntactic": 76539,
      "linguistics cognitive science": 64608,
      "network models proposed": 77352,
      "model task model": 72144,
      "ability implicitly learn": 1532,
      "processing tasks remains": 91827,
      "representing natural language": 100061,
      "probing neural language": 90897,
      "results provide evidence": 102087,
      "natural languages build": 76607,
      "neural network natural": 78363,
      "language models generative": 59559,
      "models generative adversarial": 73295,
      "gans text generation": 46438,
      "analysis reveals language": 5819,
      "information sequence words": 54971,
      "final word prediction": 44139,
      "model learns use": 71453,
      "structure neural language": 113916,
      "training data generalize": 123469,
      "recent studies suggest": 96527,
      "studies suggest models": 114286,
      "tests models ability": 120611,
      "lstm model trained": 65643,
      "deep learning modern": 28297,
      "modern deep neural": 74384,
      "data evaluate performance": 25908,
      "natural language acquisition": 76264,
      "language model use": 59417,
      "lstmbased recurrent neural": 65688,
      "multiple factors including": 75565,
      "evaluate recent neural": 38913,
      "shortterm memory network": 108343,
      "models generated output": 73291,
      "trained large quantities": 123181,
      "large quantities text": 61230,
      "standard language modeling": 112250,
      "models able leverage": 72645,
      "best known results": 13358,
      "natural languages work": 76610,
      "types training data": 126378,
      "training data results": 123529,
      "attention mechanisms different": 9924,
      "perform natural language": 86036,
      "recent advances recurrent": 96420,
      "advances recurrent neural": 4015,
      "representations neural network": 99777,
      "language models argue": 59505,
      "models neural nets learn": 73632,
      "neural networks rnns achieved": 78562,
      "neural language models neural": 77950,
      "models trained natural language": 74219,
      "trained natural language data": 123213,
      "language modeling recurrent neural": 59461,
      "modeling recurrent neural networks": 72525,
      "recurrent neural networks shown": 97248,
      "language models achieved stateoftheart": 59501,
      "sequence modeling natural language": 107014,
      "results natural language tasks": 101978,
      "language models trained large": 59671,
      "neural network models proposed": 78350,
      "language processing tasks remains": 59968,
      "probing neural language models": 90898,
      "language models generative adversarial": 59560,
      "models generative adversarial networks": 73296,
      "structure neural language models": 113917,
      "recent studies suggest models": 96528,
      "modern deep neural networks": 74385,
      "lstmbased recurrent neural networks": 65689,
      "long shortterm memory network": 65124,
      "neural language models language": 77945,
      "models language models lms": 73453,
      "recent advances recurrent neural": 96421,
      "advances recurrent neural networks": 4016,
      "recurrent neural networks rnns achieved": 97242,
      "deep recurrent neural networks rnns": 28407,
      "models trained natural language data": 74220,
      "language modeling recurrent neural networks": 59462,
      "variety natural language processing nlp": 131002,
      "language models achieved stateoftheart results": 59502,
      "natural language processing tasks remains": 76501,
      "language models generative adversarial networks": 59561,
      "models generative adversarial networks gans": 73297,
      "neural language models language models": 77946,
      "language models language models lms": 59575,
      "recent advances recurrent neural networks": 96422,
      "700k": 1215,
      "tactical": 117332,
      "wa": 131981,
      "autoregression": 10856,
      "advocates": 4203,
      "treegeneration": 125672,
      "guider": 49267,
      "curious": 25252,
      "welllabeled": 132425,
      "allinone": 5066,
      "basketball": 12550,
      "266": 847,
      "cooccurs": 23249,
      "easiest": 34426,
      "fooling": 45002,
      "fosters": 45341,
      "jensenshannon": 57225,
      "ldc2017t10": 61838,
      "heuristicbased": 49880,
      "restful": 101329,
      "montecarlo": 74654,
      "faithfully": 43006,
      "parenting": 84859,
      "lasertagger": 61553,
      "adhere": 3841,
      "adhering": 3844,
      "dub": 34249,
      "unlikelihood": 127456,
      "welleck": 132405,
      "reminiscent": 98866,
      "delve": 28630,
      "commongen": 18955,
      "highprobability": 50366,
      "intelligence natural": 55831,
      "follow idea": 44938,
      "suggest avenues": 115304,
      "introduces neural": 56619,
      "large rich": 61238,
      "domains experiment": 33777,
      "vocabulary compared": 131869,
      "work conditional": 134425,
      "task nearly": 118444,
      "datatotext systems": 27819,
      "describes general": 29406,
      "provides recommendations": 94066,
      "obtain baseline": 81259,
      "improvements incorporating": 52860,
      "fluent sentence": 44718,
      "sentence desired": 105819,
      "variants predict": 130878,
      "constrained sentences": 21573,
      "updated iteratively": 127802,
      "words replacing": 134171,
      "generate plausible": 46988,
      "text identify": 121039,
      "obstacle training": 81249,
      "techniques make": 119928,
      "structure contained": 113825,
      "english portion": 37247,
      "diseases like": 32341,
      "tree structured": 125614,
      "models performs": 73741,
      "approach treats": 7987,
      "future contexts": 46264,
      "results generation": 101812,
      "producing fluent": 92019,
      "informative texts": 55158,
      "framework various": 45739,
      "planning recent": 87643,
      "task stages": 118738,
      "stages given": 112158,
      "dataset sequencetosequence": 27189,
      "generation word": 47700,
      "structured inputs": 114005,
      "controlled experiment": 22834,
      "generated templates": 47136,
      "rl training": 102894,
      "sparse reward": 111235,
      "rl approach": 102887,
      "approach overcome": 7783,
      "nextword prediction": 79423,
      "intermediate rewards": 56142,
      "generate neural": 46974,
      "sentence present": 105999,
      "estimate quality": 38633,
      "curious case": 25253,
      "effect quality": 34608,
      "attracted significant": 10161,
      "controlling form": 22853,
      "text style": 121336,
      "work controlling": 134444,
      "generation intention": 47442,
      "reducing frequency": 97416,
      "text infilling": 121052,
      "search text": 104357,
      "generation scenarios": 47608,
      "scenarios given": 103851,
      "context challenging": 22026,
      "existing greedy": 40141,
      "tasks consistently": 119007,
      "abundant paper": 1837,
      "autoencoder dae": 10401,
      "entity modeling": 37975,
      "approaches datatotext": 8116,
      "thanks use": 121772,
      "rely representation": 98733,
      "vocabulary tokens": 131909,
      "propose entitycentric": 92659,
      "entitycentric neural": 38137,
      "data input": 26043,
      "writing process": 135094,
      "summary addition": 115632,
      "generation limited": 47461,
      "prediction needs": 89085,
      "fewer annotated": 43884,
      "score neural": 104095,
      "designed using": 29631,
      "using modular": 129896,
      "nonlinguistic input": 80230,
      "transformer automatic": 124298,
      "better texts": 13744,
      "ones generated": 81688,
      "abstract generation": 1764,
      "generation effective": 47381,
      "row column": 103327,
      "stock market": 113364,
      "market data": 67241,
      "data means": 26123,
      "historical time": 50538,
      "described text": 29381,
      "selection human": 104786,
      "marginal likelihood": 67206,
      "model controllable": 70913,
      "generation largescale": 47458,
      "text preserving": 121197,
      "data likely": 26088,
      "faster original": 43187,
      "incorporate simple": 53486,
      "process faster": 91486,
      "source propose": 110805,
      "propose confidence": 92593,
      "variational bayes": 130921,
      "score data": 104065,
      "conditioned structured": 20807,
      "leverage data": 63581,
      "task participated": 118518,
      "performance improve": 86440,
      "importance using": 52089,
      "missing sentences": 70332,
      "leveraging power": 63695,
      "analyze influence": 5979,
      "environment finally": 38184,
      "problem structured": 91251,
      "papers contribution": 84503,
      "examples generated": 39840,
      "generation fact": 47407,
      "game summaries": 46420,
      "task focus": 118221,
      "computational creativity": 20374,
      "encouraging preliminary": 36780,
      "open set": 81931,
      "learning templates": 63101,
      "large paired": 61197,
      "manipulation paper": 66931,
      "focus new": 44799,
      "practical task": 88717,
      "transfer aims": 124035,
      "style reference": 114591,
      "basketball game": 12551,
      "report corpus": 98986,
      "texts achieve": 121450,
      "transfer better": 124039,
      "better style": 13731,
      "pairs empirical": 83524,
      "direct control": 31786,
      "closer human": 17873,
      "new metrics": 79021,
      "score jensenshannon": 104087,
      "jensenshannon divergence": 57226,
      "semantically consistent": 105366,
      "assessed human": 9500,
      "measures generated": 67866,
      "significant advancement": 108719,
      "years models": 135273,
      "dynamically select": 34349,
      "novel objective": 80678,
      "dataset terms": 27234,
      "bleu metrics": 14294,
      "texts generated": 121523,
      "descriptions existing": 29481,
      "text description": 120868,
      "stateoftheart large": 112688,
      "stylized text": 114638,
      "shared codebase": 107966,
      "restful api": 101330,
      "users build": 129098,
      "outperforms finetuning": 82905,
      "selected topic": 104737,
      "approach leveraging": 7683,
      "corpus finally": 23802,
      "models tends": 74174,
      "generation procedure": 47563,
      "traditional generation": 122820,
      "generation phases": 47543,
      "sentences wikipedia": 106545,
      "extracting corresponding": 42208,
      "content plans": 21923,
      "generator topic": 47782,
      "controlling sentiment": 22860,
      "separately model": 106875,
      "topic relevance": 122567,
      "applications case": 6903,
      "phrases output": 87441,
      "consequently models": 21182,
      "text contribution": 120822,
      "modifying model": 74460,
      "growing field": 49172,
      "scenarios require": 103866,
      "training maximum": 123703,
      "represented form": 100028,
      "content tend": 21959,
      "networks guided": 77617,
      "inference techniques": 54234,
      "text lead": 121082,
      "mechanism propose": 68029,
      "mechanism variational": 68050,
      "controlled generation": 22837,
      "editing present": 34550,
      "based iterative": 11781,
      "text leveraging": 121089,
      "using trivial": 130324,
      "model filtered": 71179,
      "opens possibility": 82012,
      "accuracy datatotext": 2125,
      "inference major": 54164,
      "data output": 26201,
      "words long": 134040,
      "based finetuning": 11722,
      "gpt2 produce": 48566,
      "performance extensively": 86370,
      "proposed existing": 93275,
      "grounded theory": 49114,
      "traits language": 123974,
      "finally based": 44149,
      "taskspecific vector": 119645,
      "setting outperforms": 107766,
      "examples topics": 39890,
      "lms generate": 64882,
      "relies variety": 98663,
      "context addition": 22001,
      "unlikelihood training": 127457,
      "using powerful": 130035,
      "propose finetuning": 92681,
      "generation apply": 47309,
      "welleck et": 132406,
      "improving generation": 53099,
      "perform quite": 86053,
      "does order": 33375,
      "form neural": 45109,
      "systems changing": 116777,
      "framework selects": 45675,
      "metric ranking": 69897,
      "time reduced": 122086,
      "bleu neural": 14296,
      "samples available": 103580,
      "samples text": 103593,
      "model bleu": 70782,
      "constrained text": 21574,
      "studies constrained": 114200,
      "focus commongen": 44746,
      "set concepts": 107398,
      "rely supervised": 98749,
      "sense coverage": 105656,
      "sense concept": 105655,
      "concept coverage": 20564,
      "guided decoding": 49255,
      "significantly increases": 108969,
      "increases concept": 53638,
      "diversity neural": 32887,
      "weighting neural": 132365,
      "words inspired": 134007,
      "distribution use": 32687,
      "generated samples": 47117,
      "hypotheses proposed": 51258,
      "captioning datatotext": 15259,
      "corresponds higher": 24319,
      "interactive writing": 56038,
      "writing assistants": 135089,
      "topics user": 122662,
      "training components": 123391,
      "demonstrate topic": 28889,
      "options better": 82250,
      "artificial intelligence natural": 9252,
      "intelligence natural language": 55832,
      "framework framework consists": 45550,
      "language model text": 59408,
      "model text generation": 72165,
      "significantly outperforms classical": 109006,
      "datatotext systems use": 27820,
      "new largescale corpus": 78987,
      "experiments effectiveness proposed": 40926,
      "training data major": 123490,
      "generating large amounts": 47229,
      "models models ranked": 73578,
      "topdown tree structured": 122495,
      "text generation text": 121019,
      "generation achieved remarkable": 47292,
      "data address issue": 25581,
      "automatic evaluation results": 10539,
      "significant attention recently": 108727,
      "complex sentence simple": 19875,
      "generation recent work": 47585,
      "controlling form text": 22854,
      "data significant improvements": 26449,
      "improvements competitive baselines": 52834,
      "inference algorithm based": 54112,
      "generative models text": 47745,
      "great success text": 49031,
      "abundant paper propose": 1838,
      "method make use": 68951,
      "denoising autoencoder dae": 28996,
      "different text generation": 31491,
      "rely representation learning": 98734,
      "propose entitycentric neural": 92660,
      "baselines automatic human": 12357,
      "fewer annotated data": 43885,
      "corpus evaluate model": 23781,
      "bleu score neural": 14328,
      "nonlinguistic input data": 80231,
      "models generalize better": 73275,
      "effective text generation": 34760,
      "paper tackles problem": 84469,
      "headline generation tasks": 49598,
      "language models promising": 59635,
      "improve models ability": 52420,
      "recently neural models": 96719,
      "recent advancements neural": 96391,
      "advancements neural language": 3972,
      "systems human automatic": 116938,
      "models bert gpt2": 72837,
      "empirically demonstrate effectiveness": 36225,
      "understanding information extraction": 126862,
      "unlike previous methods": 127441,
      "public data set": 94244,
      "methods language modeling": 69573,
      "area computational creativity": 8867,
      "text generation based": 120980,
      "encouraging preliminary results": 36781,
      "paper focus new": 83949,
      "text style transfer": 121337,
      "style transfer aims": 114602,
      "pairs empirical results": 83525,
      "text generation propose": 121002,
      "text generation quality": 121003,
      "allows direct control": 5142,
      "achieving impressive performance": 2956,
      "order evaluate model": 82320,
      "propose new metrics": 92869,
      "recent years models": 96619,
      "novel objective function": 80679,
      "superior performance stateoftheart": 115691,
      "text generation knowledge": 120986,
      "triples natural language": 125816,
      "stateoftheart large margin": 112689,
      "stylized text generation": 114639,
      "advantages different models": 4076,
      "model large corpus": 71414,
      "consistently significantly outperforms": 21443,
      "approach lowresource setting": 7697,
      "new task propose": 79203,
      "results framework outperforms": 101798,
      "content work present": 21980,
      "words challenging task": 133847,
      "unlike existing models": 127434,
      "outperform stateoftheart approach": 82750,
      "modifying model architecture": 74461,
      "research text generation": 100640,
      "learning better representation": 62407,
      "generation aims automatically": 47301,
      "stateoftheart performance automatic": 112808,
      "evaluations model outperforms": 39473,
      "making language generation": 66850,
      "models proposed models": 73825,
      "language quality generated": 60009,
      "domain adaptation using": 33460,
      "generation task based": 47659,
      "based finetuning pretrained": 11723,
      "model parameters frozen": 71705,
      "lms generate text": 64883,
      "improving neural text": 53131,
      "welleck et al": 132407,
      "approach outperforms competitive": 7768,
      "structured data text": 113996,
      "constrained text generation": 21575,
      "text generation generate": 120985,
      "increase diversity generated": 53595,
      "diversity generated samples": 32877,
      "language generation despite": 59069,
      "neural models prone": 78191,
      "image captioning datatotext": 51766,
      "captioning datatotext generation": 15260,
      "artificial intelligence natural language": 9253,
      "intelligence natural language processing": 55833,
      "experiments effectiveness proposed method": 40927,
      "text generation text generation": 121020,
      "generation achieved remarkable progress": 47293,
      "network models trained endtoend": 77357,
      "received significant attention recently": 96377,
      "baselines automatic human evaluation": 12358,
      "generation experimental results model": 47400,
      "results model outperforms existing": 101942,
      "recent advancements neural language": 96392,
      "text style transfer aims": 121338,
      "require large amounts labeled": 100158,
      "new task propose novel": 79204,
      "content work present novel": 21981,
      "achieve stateoftheart performance automatic": 2559,
      "text generation tasks including": 121015,
      "using policy gradient reinforcement": 130030,
      "approach outperforms competitive baselines": 7769,
      "image captioning datatotext generation": 51767,
      "artificial intelligence natural language processing": 9254,
      "use recurrent neural networks rnns": 128233,
      "neural network models trained endtoend": 78355,
      "using policy gradient reinforcement learning": 130031,
      "335": 948,
      "contrarily": 22675,
      "appeals": 6793,
      "philosophers": 87264,
      "3dimensional": 993,
      "highranking": 50418,
      "representatives": 100025,
      "literals": 64732,
      "blei": 14268,
      "worddocument": 133707,
      "ldabased": 61833,
      "5m": 1143,
      "20newsgroup": 752,
      "neology": 77012,
      "parallelly": 84707,
      "clusterings": 17962,
      "dtm": 34221,
      "migration": 70077,
      "flooding": 44677,
      "subtopics": 115005,
      "graphofwords": 48938,
      "task addressed": 117848,
      "method providing": 69085,
      "estimation techniques": 38668,
      "dependencies experiments": 29084,
      "approaches topic": 8377,
      "annotation technique": 6386,
      "concept hierarchy": 20575,
      "content new": 21912,
      "approach topic": 7975,
      "topic selection": 122573,
      "selection instead": 104788,
      "conceptual hierarchy": 20662,
      "concepts represent": 20640,
      "discussed document": 32295,
      "generic concepts": 47793,
      "semantic topics": 105336,
      "distributions specific": 32771,
      "learned topics": 62268,
      "unsupervised partofspeech": 127692,
      "task make": 118372,
      "improves interpretability": 52987,
      "discovering temporal": 32127,
      "temporal trends": 120130,
      "lowdimensional representation": 65411,
      "features techniques": 43755,
      "embedded semantic": 35360,
      "distribution documents": 32650,
      "cluster assignments": 17924,
      "main topics": 66472,
      "highdimensional distributed": 50153,
      "generates list": 47164,
      "labeled lda": 58461,
      "lda topic": 61830,
      "promote interpretability": 92318,
      "language topic": 60175,
      "latent documentlevel": 61585,
      "vectors contrast": 131421,
      "simultaneously learning": 109683,
      "vectors linear": 131452,
      "approaches sentencelevel": 8333,
      "number challenges": 80851,
      "adapt approach": 3170,
      "structure improves": 113877,
      "studied effect": 114157,
      "alignment based": 4947,
      "problem estimating": 91036,
      "relevant tasks": 98592,
      "tune experiments": 125928,
      "idea using": 51342,
      "document titles": 33097,
      "compute neural": 20468,
      "shared documents": 107974,
      "topics lowdimensional": 122637,
      "lowdimensional dense": 65409,
      "models assume": 72779,
      "result trained": 101415,
      "trained external": 123144,
      "corpus help": 23822,
      "distribution defined": 32647,
      "representative vectors": 100023,
      "context outside": 22205,
      "outside current": 83196,
      "experiments range": 41102,
      "sentences topic": 106523,
      "modeling topics": 72571,
      "meanings given": 67744,
      "approach documentlevel": 7502,
      "analysis highly": 5640,
      "predicting topic": 89018,
      "topicbased methods": 122606,
      "indicate methods": 53840,
      "algorithm identify": 4749,
      "group documents": 49136,
      "belonging topic": 12723,
      "knowledge probabilistic": 58122,
      "leads larger": 61944,
      "thesaurus wordnet": 121843,
      "wordnet used": 133781,
      "relations combined": 98124,
      "identify text": 51567,
      "cooccurrence data": 23224,
      "interpretability components": 56222,
      "components learn": 20029,
      "observe improvement": 81196,
      "words algorithm": 133813,
      "highdimensional space": 50154,
      "words classifying": 133859,
      "words topic": 134271,
      "extract topics": 42129,
      "detection topic": 30085,
      "computerassisted text": 20504,
      "generation datadriven": 47352,
      "model literature": 71466,
      "applications importantly": 6935,
      "measure topic": 67827,
      "meaningful topics": 67731,
      "topics past": 122646,
      "related time": 97907,
      "topic structures": 122577,
      "case descriptions": 15581,
      "ldabased models": 61834,
      "properties document": 92446,
      "models defining": 73030,
      "models creating": 73000,
      "dictionary resources": 30892,
      "resources topic": 101057,
      "corpora vocabulary": 23618,
      "depends corpus": 29282,
      "corpus characteristics": 23691,
      "discover topics": 32120,
      "idea generate": 51325,
      "topics discovered": 122623,
      "extensive qualitative": 41949,
      "approach linear": 7688,
      "topics clusters": 122616,
      "words clusters": 133863,
      "topic memory": 122538,
      "sparsity address": 111250,
      "topic representations": 122568,
      "topical words": 122595,
      "results jointly": 101870,
      "modeled discourse": 72356,
      "representations effectively": 99615,
      "provide supplementary": 93937,
      "commenting requires": 18807,
      "train automatic": 122902,
      "commenting model": 18806,
      "model relying": 71899,
      "comments based": 18810,
      "corpora achieves": 23400,
      "performance semisupervised": 86696,
      "assumptions training": 9669,
      "different multilingual": 31273,
      "empirical insights": 36171,
      "model yield": 72334,
      "problem conventional": 90983,
      "learning clustering": 62440,
      "methods introduced": 69561,
      "optimization target": 82213,
      "works context": 134933,
      "wiki dataset": 132625,
      "hidden topics": 49916,
      "inadequate document": 53198,
      "highlight benefits": 50255,
      "duplicate detection": 34267,
      "assigning correct": 9554,
      "fashion use": 43123,
      "detection study": 30066,
      "syntagmatic paradigmatic": 116527,
      "analysis far": 5619,
      "performance enable": 86336,
      "ways using": 132175,
      "mining data": 70230,
      "scalability large": 103683,
      "information microtexts": 54764,
      "work novel": 134658,
      "expression propose": 41743,
      "clustering words": 17960,
      "vectors concatenated": 131416,
      "final text": 44133,
      "evaluation classification": 39145,
      "deployed model": 29306,
      "learning hard": 62627,
      "manner use": 66962,
      "gumbelsoftmax distribution": 49278,
      "comparison word": 19587,
      "words acquire": 133805,
      "semiautomatic way": 105572,
      "clues particular": 17921,
      "lexicon use": 63910,
      "indicate word": 53854,
      "way better": 132061,
      "quantify number": 94854,
      "improvements overall": 52891,
      "topics traditional": 122660,
      "crosslingual contextualized": 24934,
      "german portuguese": 47908,
      "clusters pretrained": 17968,
      "story paper": 113417,
      "provide benchmarks": 93767,
      "zeroshot topic": 135531,
      "identification experiments": 51387,
      "outperforms systems": 83034,
      "zeroshot settings": 135527,
      "propose robust": 93022,
      "scheme extending": 103926,
      "potentially benefit": 88605,
      "encoding contextual": 36685,
      "addresses issue": 3796,
      "wordlevel embedding": 133732,
      "topics word": 122665,
      "discovering coherent": 32125,
      "documents utilize": 33311,
      "document labels": 33029,
      "modeling aims": 72371,
      "method realworld": 69093,
      "produces output": 92006,
      "layers popular": 61798,
      "roberta does": 102997,
      "discourse level": 32065,
      "representations captures": 99533,
      "tokens measure": 122317,
      "model attend": 70698,
      "topic related": 122566,
      "inference perform": 54190,
      "keyphrase generation": 57620,
      "clustering performance": 17951,
      "detection automatic": 29886,
      "reader characteristics": 95953,
      "information detection": 54478,
      "modeling design": 72415,
      "bernoulli distribution": 13049,
      "issues discussed": 57044,
      "outline methods": 82632,
      "application using": 6893,
      "environmental issues": 38195,
      "cognitive social": 18283,
      "topic granularity": 122526,
      "term ranking": 120240,
      "ranking use": 95693,
      "interpretation demonstrate": 56261,
      "general news": 46685,
      "topics various": 122664,
      "data showcase": 26445,
      "like unsupervised": 64108,
      "methods finetuning": 69512,
      "constructing new": 21681,
      "new topic": 79224,
      "task integrating": 118303,
      "quality finetuning": 94669,
      "source natural": 110797,
      "data employ": 25888,
      "used generate synthetic": 128565,
      "word distributions specific": 132998,
      "unsupervised partofspeech tagging": 127693,
      "contained large corpora": 21775,
      "discovering temporal trends": 32128,
      "approaches skipgram model": 8344,
      "model provide good": 71823,
      "learning work propose": 63173,
      "propose semisupervised method": 93035,
      "word vectors jointly": 133638,
      "topic modeling approaches": 122544,
      "information sentence boundaries": 54966,
      "significantly improves interpretability": 108947,
      "document clustering text": 32973,
      "context outside current": 22206,
      "outside current sentence": 83197,
      "word representations combined": 133449,
      "results indicate methods": 101856,
      "large collections documents": 61053,
      "documents belonging topic": 33193,
      "probabilistic topic models": 90834,
      "problem given document": 91070,
      "shared task provided": 108087,
      "large collections text": 61054,
      "collections text documents": 18505,
      "social science research": 110452,
      "standard metrics fail": 112264,
      "paper aim analyze": 83721,
      "method effectively learns": 68793,
      "depends corpus characteristics": 29283,
      "idea generate multiple": 51326,
      "qualitative quantitative evaluation": 94573,
      "shows proposed approach": 108616,
      "lda topic modeling": 61831,
      "prior works focus": 90767,
      "short texts data": 108262,
      "data sparsity address": 26481,
      "sparsity address issue": 111251,
      "topic memory mechanism": 122539,
      "latent topic representations": 61619,
      "propose unsupervised model": 93137,
      "unsupervised model jointly": 127671,
      "representation learning clustering": 99284,
      "real world datasets": 96092,
      "using word vectors": 130386,
      "approximate posterior distribution": 8460,
      "propose new text": 92888,
      "text representation scheme": 121251,
      "compared existing baselines": 19366,
      "words acquire new": 133806,
      "neural topic models": 78713,
      "models recent neural": 73873,
      "models results indicate": 73954,
      "model model learns": 71521,
      "different languages italian": 31205,
      "french german portuguese": 45810,
      "largescale neural networks": 61487,
      "evaluation word sense": 39444,
      "topics word embeddings": 122666,
      "downstream tasks sentiment": 34056,
      "capture semantic patterns": 15380,
      "topic modeling text": 122546,
      "case study focuses": 15618,
      "text classification jointly": 120728,
      "classification jointly learning": 17241,
      "proposed method jointly": 93377,
      "attention networks neural": 9951,
      "existing neural models": 40231,
      "embeddings transfer learning": 35989,
      "performance different classifiers": 86297,
      "comment generation approaches": 18800,
      "methods terms automatic": 69798,
      "source natural language": 110798,
      "twitter data employ": 126104,
      "context outside current sentence": 22207,
      "datasets demonstrate model outperforms": 27410,
      "large collections text documents": 61055,
      "data sparsity address issue": 26482,
      "sparsity address issue propose": 111252,
      "parallel data propose novel": 84648,
      "evaluation word sense disambiguation": 39445,
      "data sparsity address issue propose": 26483,
      "nell": 77006,
      "william": 132703,
      "explosively": 41673,
      "ke": 57515,
      "deceive": 27916,
      "tough": 122705,
      "ring": 102859,
      "knowledgebases": 58262,
      "consider learning": 21212,
      "variety embedding": 130975,
      "models link": 73512,
      "composition relations": 20102,
      "matrix multiplication": 67480,
      "allows explicitly": 5153,
      "information manually": 54749,
      "jointly embedding": 57339,
      "relations mentions": 98207,
      "population paper": 88166,
      "graph word": 48883,
      "models relationships": 73911,
      "yields accurate": 135395,
      "head entity": 49575,
      "tail entity": 117499,
      "method remarkably": 69101,
      "recently knowledge": 96701,
      "associated corresponding": 9594,
      "method experiments": 68825,
      "construction information": 21690,
      "text patterns": 121174,
      "entities recently": 37857,
      "versions universal": 131625,
      "schema used": 103902,
      "generalize text": 46820,
      "explicit entity": 41320,
      "accuracy entity": 2151,
      "expressed textual": 41726,
      "prediction typically": 89147,
      "aggregation functions": 4360,
      "language database": 58934,
      "importantly make": 52297,
      "sequential text": 107258,
      "considered significant": 21301,
      "challenges constructing": 16142,
      "plain texts": 87626,
      "entities plain": 37837,
      "corresponding entities": 24285,
      "build knowledge": 14776,
      "utilize recurrent": 130525,
      "network encode": 77238,
      "measure informativeness": 67805,
      "knowledge unified": 58220,
      "entity prediction": 37986,
      "bases far": 12499,
      "bases different": 12496,
      "extended traditional": 41812,
      "facts paper": 42921,
      "utilize structural": 130528,
      "prediction triplet": 89146,
      "william shakespeare": 132704,
      "considering local": 21327,
      "provided test": 93987,
      "composing distributed": 20081,
      "multiple similarity": 75694,
      "additive composition": 3641,
      "mechanism experiments": 67983,
      "does enable": 33343,
      "advances stateoftheart": 4022,
      "multiple filters": 75567,
      "problem inducing": 91083,
      "kg embeddings": 57656,
      "method induce": 68903,
      "methods dont": 69459,
      "extraction report": 42468,
      "report evaluation": 98993,
      "ones introduce": 81691,
      "demonstrate knowledge": 28765,
      "generally effective": 46855,
      "stateofart neural": 112555,
      "datasets knowledge": 27534,
      "problems learning": 91334,
      "dependent specific": 29266,
      "knowledge bayesian": 57815,
      "set learning": 107478,
      "knowledge demand": 57854,
      "graph entity": 48804,
      "relational learning": 98094,
      "coverage previous": 24520,
      "studies knowledge": 114243,
      "training instance": 123664,
      "using gradient": 129713,
      "embeddings systematically": 35970,
      "leads sizable": 61960,
      "extraction conventional": 42290,
      "extraction usually": 42535,
      "improve approach": 52335,
      "benchmarks proposed": 12933,
      "knowledge heterogeneous": 58000,
      "reflect changes": 97601,
      "world represent": 135045,
      "problems prediction": 91357,
      "model problems": 71792,
      "task fact": 118201,
      "fact propose": 42831,
      "assessing difficulty": 9510,
      "important challenges": 52112,
      "challenges address": 16124,
      "relations defined": 98136,
      "create largest": 24621,
      "relations supervision": 98263,
      "interactive way": 56037,
      "set obtained": 107517,
      "using query": 130073,
      "demonstrated vector": 28937,
      "classifier classify": 17529,
      "documents context": 33212,
      "entities occur": 37830,
      "graph relations": 48861,
      "relations model": 98209,
      "attractive properties": 10172,
      "graphs entity": 48950,
      "linking entities": 64657,
      "improvement paper": 52736,
      "better entity": 13573,
      "kg completion": 57654,
      "attention pruning": 9984,
      "knowledge completing": 57833,
      "structural differences": 113765,
      "alignments extensive": 5013,
      "model structural": 72091,
      "explicitly utilize": 41401,
      "utilize useful": 130536,
      "useful relation": 128921,
      "improving entity": 53095,
      "iteratively learn": 57141,
      "studies example": 114222,
      "better extract": 13584,
      "stage experiments": 112145,
      "entities source": 37872,
      "approaches encode": 8143,
      "entities including": 37795,
      "edges represent": 34525,
      "like cnns": 64029,
      "aggregation scheme": 4366,
      "approaches handle": 8176,
      "missing links": 70330,
      "number stateoftheart": 80971,
      "papers report": 84512,
      "employ simple": 36280,
      "robust features": 103034,
      "designing new": 29643,
      "expressive features": 41771,
      "adding just": 3384,
      "addition compared": 3410,
      "advantage abundant": 4035,
      "knowledge plms": 58108,
      "works remarkably": 134973,
      "kg link": 57659,
      "benchmark stateoftheart": 12860,
      "large kg": 61119,
      "connection text": 21136,
      "potentially applied": 88603,
      "simulated annealing": 109635,
      "initialization approach": 55235,
      "quadratic time": 94542,
      "approach easily": 7511,
      "case results": 15607,
      "reasoning crosslingual": 96240,
      "encoding knowledge": 36703,
      "source entity": 110754,
      "decisions paper": 27963,
      "reasoning methods": 96271,
      "knowledge resolve": 58151,
      "propose enhanced": 92657,
      "baseline addition": 12183,
      "performance reasoning": 86656,
      "generative network": 47747,
      "generates vector": 47190,
      "classifier unlike": 17589,
      "generate unseen": 47041,
      "kgs contain": 57665,
      "fact descriptions": 42818,
      "learn structured": 62159,
      "hardly generalizable": 49494,
      "major drawbacks": 66556,
      "aim alleviate": 4444,
      "alleviate drawbacks": 5033,
      "scheme improve": 103928,
      "magnitude compared": 66378,
      "entity given": 37942,
      "approaches operate": 8268,
      "making inference": 66846,
      "newly emerging": 79272,
      "framework able": 45404,
      "representations previously": 99822,
      "entities method": 37820,
      "bases present": 12506,
      "approach reasoning": 7839,
      "crisp logical": 24789,
      "proposed metalearning": 93333,
      "satisfactory results": 103653,
      "methods 15": 69271,
      "learning automl": 62379,
      "task embeddings": 118134,
      "score function": 104072,
      "efficiently using": 35159,
      "using gradientbased": 129714,
      "results knowledgeaware": 101872,
      "building unified": 14896,
      "methods overestimated": 69662,
      "crosslingual monolingual": 24978,
      "monolingual datasets": 74584,
      "datasets ablation": 27289,
      "cases similar": 15667,
      "scalability paper": 103684,
      "knowledgebases kbs": 58263,
      "kbs approach": 57510,
      "current supervised": 25376,
      "objective achieve": 81058,
      "fewshot knowledge": 43902,
      "focus current": 44749,
      "relation given": 98038,
      "representations predictive": 99812,
      "data privacy": 26278,
      "descriptive information": 29503,
      "improve understandability": 52578,
      "ignored paper": 51675,
      "types achieve": 126244,
      "promising improvements": 92276,
      "topics paper": 122644,
      "fundamental theory": 46128,
      "graph considered": 48771,
      "embedding embedding": 35400,
      "progresses achieved": 92187,
      "probabilities estimated": 90841,
      "learning unseen": 63142,
      "furthermore inference": 46182,
      "process easily": 91467,
      "instances possible": 55631,
      "15 18": 338,
      "inductive transductive": 54050,
      "contain multiple": 21748,
      "adversarial knowledge": 4128,
      "dual adversarial": 34226,
      "address structural": 3769,
      "information unsupervised": 55072,
      "space second": 111057,
      "produce multiple": 91910,
      "context multiple": 22187,
      "models use syntactic": 74283,
      "evaluate performance approach": 38882,
      "performance approach compared": 86146,
      "vector representations natural": 131356,
      "corpus training data": 24045,
      "improve quality word": 52512,
      "knowledge graph embedding": 57963,
      "remarkably improve performance": 98853,
      "recently knowledge graph": 96702,
      "knowledge generative model": 57952,
      "conducted benchmark datasets": 20913,
      "natural language database": 76279,
      "magnitude fewer parameters": 66383,
      "utilize recurrent neural": 130526,
      "semantic information sentence": 105078,
      "knowledge bases far": 57803,
      "knowledge bases different": 57800,
      "results experiments proposed": 101783,
      "knowledge graph structure": 57975,
      "introduce neural models": 56467,
      "neural models encode": 78173,
      "gating mechanism applied": 46536,
      "architecture experiments models": 8663,
      "learning methods assume": 62737,
      "models knowledge graph": 73442,
      "composing distributed representations": 20082,
      "gating mechanism experiments": 46537,
      "propose novel embedding": 92917,
      "head entity relation": 49576,
      "models benchmark datasets": 72827,
      "models generally effective": 73282,
      "limitations existing methods": 64174,
      "augmented neural networks": 10331,
      "prior knowledge bayesian": 90706,
      "knowledge graph entity": 57966,
      "previous studies knowledge": 90496,
      "studies knowledge graph": 114244,
      "performance improvements existing": 86450,
      "set predefined relations": 107536,
      "new data new": 78854,
      "relations expressed text": 98170,
      "results framework achieves": 101797,
      "analysis demonstrates proposed": 5565,
      "knowledge graphs entity": 57984,
      "improvement paper propose": 52737,
      "better entity representations": 13574,
      "entity relation representations": 38094,
      "paper simple effective": 84432,
      "substantially outperforms stateoftheart": 114910,
      "develop novel generative": 30223,
      "generative model framework": 47737,
      "entities source language": 37873,
      "different languages propose": 31210,
      "outperforms existing systems": 82904,
      "traditional neural networks": 122853,
      "neural networks like": 78505,
      "like cnns rnns": 64030,
      "prediction models employ": 89083,
      "vector representations entities": 131351,
      "encoding knowledge graph": 36704,
      "significantly improve existing": 108927,
      "generates vector representation": 47191,
      "based generative adversarial": 11736,
      "classification tasks semantic": 17456,
      "different languages provide": 31211,
      "representations previously unseen": 99823,
      "source target entities": 110834,
      "knowledge bases present": 57810,
      "reasoning knowledge graphs": 96260,
      "demonstrate model robust": 28797,
      "prediction task predicting": 89135,
      "use automated machine": 127905,
      "automated machine learning": 10450,
      "machine learning automl": 65779,
      "effectiveness method compared": 34905,
      "context propose novel": 22235,
      "nlp tasks proposed": 79754,
      "knowledge language understanding": 58033,
      "significantly better standard": 108891,
      "objective achieve stateoftheart": 81059,
      "fewshot knowledge graph": 43903,
      "graph kg completion": 48825,
      "knowledge graphs paper": 57990,
      "benchmark datasets results": 12801,
      "ignored paper propose": 51676,
      "knowledge graph reasoning": 57973,
      "downstream tasks provide": 34050,
      "research topics paper": 100647,
      "topics paper present": 122645,
      "paper present fundamental": 84102,
      "knowledge graph considered": 57959,
      "variable model approach": 130843,
      "given knowledge graph": 48051,
      "prior knowledge furthermore": 90707,
      "uses neural model": 129255,
      "dual adversarial learning": 34227,
      "models use syntactic information": 74284,
      "experiments conducted benchmark datasets": 40839,
      "benchmark datasets proposed method": 12796,
      "order magnitude fewer parameters": 82360,
      "prediction experimental results demonstrate": 89060,
      "experiment results model significantly": 40500,
      "results experiments proposed model": 101784,
      "consistent improvements compared stateoftheart": 21386,
      "paper propose novel embedding": 84295,
      "problem propose novel framework": 91179,
      "experimental results framework achieves": 40619,
      "approach substantially outperforms stateoftheart": 7942,
      "use automated machine learning": 127906,
      "automated machine learning automl": 10451,
      "benchmark datasets approach significantly": 12767,
      "objective achieve stateoftheart results": 81060,
      "knowledge graph kg completion": 57969,
      "research topics paper present": 100648,
      "latent variable model approach": 61630,
      "model achieves substantial improvements stateoftheart": 70612,
      "significant consistent improvements compared stateoftheart": 108745,
      "use automated machine learning automl": 127907,
      "paper propose novel neural architecture": 84305,
      "benchmark datasets approach significantly outperforms": 12768,
      "disregards": 32387,
      "occupation": 81490,
      "liberal": 63958,
      "coevolution": 18249,
      "stereotypic": 113341,
      "pig": 87503,
      "unequally": 127052,
      "winobias": 132724,
      "classics": 17095,
      "weat": 132214,
      "genderbalanced": 46619,
      "makers": 66751,
      "occupied": 81493,
      "feminine": 43875,
      "rhyme": 102718,
      "shouldnt": 108359,
      "differentials": 31587,
      "mirror": 70283,
      "operationalize": 82063,
      "glean": 48222,
      "irish": 56935,
      "portrayal": 88189,
      "mindful": 70147,
      "unmasking": 127464,
      "tropes": 125841,
      "trope": 125840,
      "creator": 24736,
      "seat": 104380,
      "sociological": 110479,
      "ethnic": 38752,
      "worrying": 135061,
      "unmasked": 127463,
      "approach validated": 8011,
      "gender gender": 46598,
      "behavior great": 12659,
      "perform study": 86079,
      "affects way": 4251,
      "focus especially": 44760,
      "associations present": 9635,
      "study types": 114538,
      "associations finally": 9633,
      "contemporary theories": 21853,
      "liberal conservative": 63959,
      "models motivates": 73587,
      "prediction gender": 89066,
      "focused lexical": 44863,
      "features allow": 43359,
      "models examining": 73185,
      "human biases": 50770,
      "carefully chosen": 15517,
      "systems took": 117199,
      "task semeval2018": 118680,
      "gender stereotypes": 46615,
      "gender roles": 46612,
      "gender inequality": 46600,
      "ways detect": 132152,
      "race ethnicity": 95423,
      "reviews considered": 102600,
      "customers valuable": 25453,
      "targeted advertising": 117763,
      "learn human": 62065,
      "systematic gender": 116677,
      "vast range": 131280,
      "amplified models": 5374,
      "regularization loss": 97750,
      "model unstable": 72248,
      "paper quantify": 84393,
      "encode gender": 36429,
      "stateoftheart coreference": 112623,
      "useful wide": 128948,
      "lack interpretability": 58719,
      "loss work": 65307,
      "methods lexicon": 69596,
      "qualitative insights": 94567,
      "societal biases": 110467,
      "bias existing": 13806,
      "furthermore analyze": 46143,
      "bias reduction": 13840,
      "used illustrate": 128576,
      "spaces built": 111092,
      "tool paper": 122397,
      "paper series": 84420,
      "way used": 132139,
      "term loss": 120225,
      "bias evaluation": 13804,
      "debiasing strategies": 27902,
      "biased training": 13862,
      "examine differences": 39740,
      "mitigating gender": 70383,
      "languages consequently": 60463,
      "level tags": 63510,
      "study phenomenon": 114468,
      "embedding association": 35377,
      "association test": 9629,
      "amplify social": 5377,
      "results capturing": 101561,
      "evaluating gender": 39055,
      "measures word": 67903,
      "capturing latent": 15473,
      "attempt quantify": 9751,
      "biases better": 13864,
      "metric corpus": 69874,
      "properties specific": 92484,
      "processing literature": 91700,
      "rise popularity": 102869,
      "success modeling": 115099,
      "furthermore discuss": 46164,
      "gender debiasing": 46594,
      "investigate presence": 56799,
      "genderbalanced dataset": 46620,
      "embedding associations": 35378,
      "model implicitly": 71302,
      "common association": 18856,
      "test word": 120551,
      "datasets novel": 27600,
      "people write": 85906,
      "ways people": 132169,
      "evaluations methods": 39468,
      "languages grammatical": 60614,
      "grammatical gender": 48706,
      "gender recent": 46610,
      "english finally": 37136,
      "utility embeddings": 130479,
      "automatic categorization": 10498,
      "limitations terms": 64194,
      "gender nouns": 46605,
      "suggested answer": 115340,
      "spanish speakers": 111174,
      "languages finding": 60588,
      "quantifying semantic": 94859,
      "gender systems": 46617,
      "use canonical": 127928,
      "application methods": 6864,
      "reflect bias": 97600,
      "data examine": 25911,
      "chitchat dialogue": 16882,
      "highly imbalanced": 50324,
      "bias light": 13814,
      "social bias": 110325,
      "reasoning social": 96311,
      "content offensive": 21914,
      "pragmatic implications": 88750,
      "introduce social": 56542,
      "inference corpus": 54129,
      "modelling evaluation": 72604,
      "explanations terms": 41299,
      "terms social": 120383,
      "inference commonsense": 54123,
      "tasks point": 119382,
      "existence biases": 40036,
      "representation open": 99361,
      "study gender": 114390,
      "implicit gender": 52009,
      "despite prevalence": 29715,
      "identifying gender": 51599,
      "artifacts data": 9241,
      "involves reducing": 56901,
      "reducing influence": 97423,
      "way capture": 132063,
      "capture subtle": 15404,
      "unique opportunity": 127191,
      "opportunity study": 82146,
      "embeddings average": 35564,
      "dimensions social": 31772,
      "highly confident": 50304,
      "measures reflect": 67892,
      "output various": 83147,
      "bias pretrained": 13835,
      "datasets designed": 27421,
      "directly input": 31882,
      "language gender": 59060,
      "biases language": 13874,
      "input prompts": 55407,
      "groups analyze": 49151,
      "analyze scenarios": 6004,
      "types biases": 126257,
      "pragmatic semantic": 88756,
      "crowdsourced evaluation": 25110,
      "enables train": 36404,
      "meaning provide": 67665,
      "control gender": 22804,
      "languages czech": 60479,
      "information gender": 54632,
      "support classic": 115961,
      "relevant classification": 98530,
      "deeper investigation": 28445,
      "presents research": 89893,
      "human culture": 50789,
      "effect grammatical": 34596,
      "attribute transfer": 10184,
      "change words": 16376,
      "network grammatical": 77271,
      "provided word": 93992,
      "linguistic hypotheses": 64486,
      "analyze errors": 5971,
      "gender based": 46582,
      "features articles": 43368,
      "performance dramatically": 86323,
      "experiments removing": 41114,
      "gender explore": 46597,
      "unintended correlations": 127172,
      "metrics reveal": 70005,
      "gap corpus": 46446,
      "2019 method": 704,
      "popular media": 88099,
      "media reflects": 68161,
      "automatically score": 10825,
      "enables analysis": 36373,
      "decrease performance": 28171,
      "dutch datasets": 34277,
      "topical preferences": 122593,
      "spontaneous linguistic": 112033,
      "linguistic productions": 64530,
      "largescale empirical": 61431,
      "shown distributional": 108461,
      "dialects egyptian": 30537,
      "bias social": 13844,
      "technical methods": 119748,
      "race sex": 95425,
      "variety metrics": 130993,
      "models covering": 72996,
      "experimental conditions": 40527,
      "conditions results": 20820,
      "extrinsic measures": 42620,
      "sets annotated": 107647,
      "effort release": 35182,
      "methodology analyzing": 69244,
      "gender racial": 46608,
      "biases results": 13884,
      "different conclusions": 31052,
      "performance interestingly": 86467,
      "resources knowledge": 100993,
      "space spanned": 111063,
      "desktop application": 29676,
      "commonsense models": 19004,
      "wed like": 132312,
      "like encourage": 64038,
      "evaluating social": 39101,
      "metrics use": 70017,
      "importance sentence": 52079,
      "orders magnitude larger": 82447,
      "systems automatic machine": 116747,
      "automatic sentiment analysis": 10653,
      "reviews using word": 102637,
      "word embedding approach": 133005,
      "models text corpora": 74184,
      "text corpora exhibit": 120825,
      "efficacy proposed method": 35018,
      "encode gender information": 36430,
      "natural language like": 76368,
      "term loss function": 120226,
      "languages present novel": 60804,
      "word embedding association": 133006,
      "evaluating gender bias": 39056,
      "mitigating gender bias": 70384,
      "gender bias natural": 46587,
      "language processing literature": 59880,
      "nlp models shown": 79643,
      "furthermore discuss advantages": 46165,
      "sentiment analysis work": 106680,
      "analysis models task": 5703,
      "downstream tasks release": 34053,
      "models using approach": 74298,
      "test word embeddings": 120552,
      "human evaluations methods": 50848,
      "languages grammatical gender": 60615,
      "grammatical gender recent": 48708,
      "new metrics evaluating": 79022,
      "outperforms deep learning": 82879,
      "deep learning traditional": 28323,
      "use canonical correlation": 127929,
      "gender bias existing": 46586,
      "existing dialogue datasets": 40111,
      "targeted data collection": 117765,
      "using various evaluation": 130354,
      "inference commonsense reasoning": 54124,
      "commonsense reasoning social": 19008,
      "gender bias effects": 46585,
      "present general approach": 89501,
      "languages czech german": 60480,
      "paper presents research": 84200,
      "systematic gender bias": 116678,
      "space models semantics": 111026,
      "neural network grammatical": 78305,
      "words grammatical gender": 133977,
      "grammatical gender based": 48707,
      "bert contextual language": 13092,
      "semantic space word": 105306,
      "al 2019 method": 4645,
      "english language based": 37184,
      "novel large diverse": 80617,
      "spontaneous linguistic productions": 112034,
      "perform largescale empirical": 86024,
      "different time periods": 31499,
      "research nlp research": 100571,
      "systems case study": 116773,
      "experimental conditions results": 40528,
      "release code new": 98439,
      "proposed method applied": 93348,
      "benchmark datasets evaluate": 12781,
      "best performance interestingly": 13393,
      "embedding space experimental": 35495,
      "research efforts make": 100484,
      "word embeddings given": 133119,
      "commonsense knowledge resources": 19002,
      "performance different tasks": 86301,
      "room future work": 103261,
      "nlp systems used": 79699,
      "nlp tasks taken": 79772,
      "different types biases": 31525,
      "languages present novel approach": 60805,
      "contextual word embeddings bert": 22520,
      "gender bias natural language": 46588,
      "bias natural language processing": 13824,
      "natural language processing literature": 76431,
      "use canonical correlation analysis": 127930,
      "vector space models semantics": 131375,
      "et al 2019 method": 38733,
      "embedding space experimental results": 35496,
      "gender bias natural language processing": 46589,
      "taught": 119652,
      "mostafazadeh": 74825,
      "textgeneration": 121437,
      "fables": 42724,
      "yearly": 135240,
      "fanfiction": 43061,
      "fullyspecified": 45987,
      "gscore": 49205,
      "backpropagated": 11317,
      "humansubject": 51121,
      "intervene": 56322,
      "stolen": 113368,
      "meticulous": 69862,
      "theatre": 121777,
      "went": 132447,
      "novelists": 80783,
      "tf": 121757,
      "discuss methodology": 32267,
      "ported easily": 88180,
      "requiring language": 100343,
      "algebra word": 4677,
      "science fiction": 103976,
      "changing underlying": 16402,
      "math word": 67451,
      "problems help": 91324,
      "help students": 49782,
      "thematic coherence": 121779,
      "represent new": 99123,
      "work states": 134817,
      "story cloze": 113406,
      "task mostafazadeh": 118408,
      "mostafazadeh et": 74826,
      "strategies try": 113493,
      "given story": 48137,
      "ranking 3rd": 95658,
      "generation sequence": 47614,
      "creativity originality": 24735,
      "modern nlg": 74406,
      "creative tasks": 24730,
      "suggested approaches": 115342,
      "require semantic": 100196,
      "input required": 55417,
      "results measured": 101909,
      "representation provide": 99388,
      "variations generated": 130935,
      "nlg component": 79514,
      "automatically produce": 10813,
      "produce large": 91906,
      "given content": 47997,
      "content propose": 21926,
      "methods developing": 69444,
      "quantitatively evaluate": 94889,
      "view finally": 131705,
      "easily create": 34444,
      "graph use": 48879,
      "representation generate": 99246,
      "affect readers": 4224,
      "graphs present": 48968,
      "graphs deep": 48946,
      "tool new": 122395,
      "pairs events": 83536,
      "methods high": 69532,
      "causal relation": 15810,
      "researchers nlp": 100702,
      "reasoning approach": 96228,
      "work implement": 134560,
      "pairs learn": 83573,
      "collecting human": 18447,
      "indicate use": 53852,
      "increases average": 53637,
      "form communication": 45073,
      "communication human": 19033,
      "building automatic": 14823,
      "automatic tagging": 10686,
      "finegrained set": 44384,
      "corpus useful": 24059,
      "determine potential": 30130,
      "ending story": 36856,
      "set achieve": 107346,
      "task stylistic": 118748,
      "stylistic differences": 114627,
      "validation test": 130746,
      "considering just": 21325,
      "new characters": 78828,
      "forum dataset": 45303,
      "novel form": 80571,
      "new gated": 78932,
      "model factor": 71169,
      "semantic supervision": 105313,
      "story comprehension": 113410,
      "test propose": 120482,
      "neural memory": 78139,
      "story ending": 113414,
      "demonstrates superior": 28965,
      "characters simple": 16621,
      "simple commonsense": 109387,
      "reading lines": 96024,
      "addressing challenge": 3807,
      "new commonsense": 78832,
      "events situations": 39599,
      "unseen events": 127525,
      "participants addition": 85316,
      "description events": 29462,
      "generation computer": 47339,
      "present analyze": 89368,
      "ending given": 36852,
      "story context": 113411,
      "understand context": 126743,
      "controllable neural": 22828,
      "reward shaping": 102676,
      "lm predict": 64869,
      "sentence add": 105763,
      "coherence present": 18307,
      "order guide": 82332,
      "studies generated": 114233,
      "event ordering": 39531,
      "background commonsense": 11298,
      "knowledge significant": 58173,
      "language describes": 58944,
      "work restricted": 134785,
      "hierarchical generation": 49964,
      "topic generated": 122523,
      "problem covers": 90986,
      "separately consider": 106870,
      "information course": 54459,
      "help nli": 49758,
      "story narrative": 113416,
      "distillation module": 32513,
      "interaction features": 55945,
      "crucial semantic": 25166,
      "module different": 74485,
      "ways evaluate": 132156,
      "deal outofvocabulary": 27864,
      "generator produce": 47777,
      "texts words": 121652,
      "extracts text": 42582,
      "possibilities interpretable": 88365,
      "interpretable controllable": 56237,
      "human collaboration": 50778,
      "finally humans": 44192,
      "implications future": 51992,
      "work date": 134453,
      "work finally": 134529,
      "persona based": 87107,
      "explores idea": 41648,
      "various persona": 131165,
      "based story": 12076,
      "inspect various": 55551,
      "style using": 114608,
      "multiview model": 75936,
      "attention shows": 10008,
      "improvement methods": 52724,
      "turning points": 125988,
      "storytelling previous": 113425,
      "human eye": 50860,
      "functions designed": 46071,
      "knowledge tackle": 58200,
      "temporal order": 120113,
      "utilize commonsense": 130496,
      "finetuning automatic": 44447,
      "selection sentences": 104827,
      "general unsupervised": 46726,
      "need handcrafted": 76815,
      "set phrases": 107529,
      "rough sketch": 103309,
      "story present": 113418,
      "learn different": 62021,
      "tracking important": 122757,
      "propose sentencelevel": 93037,
      "set fluent": 107447,
      "method treats": 69202,
      "candidates sentence": 15146,
      "integrate domainspecific": 55750,
      "modeling domain": 72422,
      "baselines work": 12492,
      "dialogues dialogue": 30827,
      "dialogue plan": 30720,
      "different role": 31392,
      "want model": 132017,
      "incrementally generate": 53745,
      "chains events": 15982,
      "time scale": 122095,
      "propose design": 92623,
      "automatic textual": 10697,
      "sentence important": 105901,
      "characters account": 16593,
      "task demands": 118056,
      "performance story": 86755,
      "observe issues": 81199,
      "coherency generated": 18312,
      "consistency controlled": 21356,
      "errors improve": 38376,
      "improve coreference": 52359,
      "stories challenging": 113398,
      "approaches domain": 8135,
      "phrases generation": 87428,
      "contexts general": 22395,
      "text examine": 120925,
      "social situations": 110456,
      "investigation human": 56862,
      "models accuracies": 72653,
      "lower human": 65432,
      "modeling event": 72426,
      "tool text": 122402,
      "annotation proposed": 6361,
      "include information": 53228,
      "formalize task": 45199,
      "highlevel understanding": 50253,
      "generally fail": 46857,
      "acquire knowledge": 3032,
      "guaranteed text": 49217,
      "guide generator": 49240,
      "built automatically": 14912,
      "explore challenges": 41528,
      "generation addressed": 47297,
      "directions new": 31849,
      "predicts semantic": 89220,
      "sentences represent": 106477,
      "baselines indicating": 12415,
      "using frame": 129685,
      "dataset comprised": 26813,
      "content does": 21878,
      "models unable": 74260,
      "theoretically grounded": 121804,
      "using different machine": 129573,
      "algebra word problems": 4678,
      "math word problems": 67453,
      "new words target": 79255,
      "baselines release new": 12453,
      "work automatic detection": 134391,
      "story cloze test": 113408,
      "task mostafazadeh et": 118409,
      "mostafazadeh et al": 74827,
      "generation nlg systems": 47517,
      "network rnn architecture": 77413,
      "systems require semantic": 117115,
      "paper present automatic": 84085,
      "semantic representation input": 105228,
      "results measured terms": 101910,
      "generation nlg component": 47512,
      "automatically produce large": 10814,
      "produce large number": 91907,
      "propose solution problem": 93074,
      "methods high quality": 69533,
      "unsupervised methods learning": 127669,
      "collecting human judgments": 18448,
      "validation test sets": 130747,
      "close stateoftheart performance": 17836,
      "automated human evaluations": 10446,
      "inspired previous studies": 55576,
      "roc story cloze": 103141,
      "demonstrates superior performance": 28966,
      "introduce new annotation": 56474,
      "commonsense inference task": 18988,
      "previously unseen events": 90630,
      "able automatically generate": 1610,
      "existing generative models": 40140,
      "according human evaluation": 1981,
      "evaluation automatic evaluation": 39125,
      "ending given story": 36853,
      "given story context": 48138,
      "story context addition": 113412,
      "character word sentence": 16480,
      "background commonsense knowledge": 11299,
      "according automatic human": 1968,
      "test set respectively": 120518,
      "guide generation process": 49239,
      "paper considers problem": 83785,
      "traditional metrics bleu": 122842,
      "advantage external knowledge": 4043,
      "better stateoftheart systems": 13729,
      "model learn generate": 71432,
      "systems large margin": 116971,
      "utilize commonsense knowledge": 130497,
      "state tracking important": 112528,
      "does need model": 33374,
      "effectiveness approach stateoftheart": 34871,
      "story cloze task": 113407,
      "task promising results": 118571,
      "domainspecific features common": 33901,
      "provides new approach": 94057,
      "works focus using": 134948,
      "addition commonly used": 3409,
      "previous works shown": 90578,
      "challenging problem requires": 16298,
      "interface human users": 56101,
      "implications understanding human": 51997,
      "significantly lower human": 108975,
      "lower human performance": 65433,
      "proposed approach shown": 93210,
      "research directions new": 100474,
      "work focused predicting": 134535,
      "provides challenging testbed": 94009,
      "including neural models": 53339,
      "presents machine learning approach": 89864,
      "using different machine learning": 129574,
      "task mostafazadeh et al": 118410,
      "mostafazadeh et al 2016a": 74828,
      "language generation nlg systems": 59087,
      "neural network rnn architecture": 78386,
      "language generation nlg component": 59082,
      "automatically produce large number": 10815,
      "roc story cloze test": 103142,
      "human evaluation automatic evaluation": 50803,
      "ending given story context": 36854,
      "traditional metrics bleu rouge": 122843,
      "work propose neural architecture": 134729,
      "significantly lower human performance": 108976,
      "data paper propose new": 26218,
      "future research directions new": 46299,
      "human evaluation shows proposed": 50836,
      "paper presents machine learning approach": 84179,
      "task mostafazadeh et al 2016a": 118411,
      "natural language generation nlg systems": 76314,
      "recurrent neural network rnn architecture": 97214,
      "natural language generation nlg component": 76310,
      "clarified": 17024,
      "concierge": 20698,
      "electronics": 35263,
      "trackers": 122745,
      "calendar": 15041,
      "ghazvininejad": 47957,
      "exaequo": 39728,
      "submodules": 114786,
      "naik": 76005,
      "serialized": 107266,
      "encountering": 36746,
      "smartly": 110257,
      "duconv": 34255,
      "dm": 32926,
      "userbased": 129060,
      "sharedprivate": 108120,
      "expandable": 40358,
      "multiterm": 75902,
      "thingtalk": 121862,
      "woz": 135077,
      "reannotation": 96193,
      "temporal reference": 120117,
      "explicit temporal": 41355,
      "temporal references": 120118,
      "annotations fully": 6435,
      "detailed manual": 29779,
      "anaphoric relations": 6075,
      "ambiguity data": 5280,
      "strategies make": 113475,
      "systems try": 117209,
      "performance adapting": 86123,
      "corpus 80": 23637,
      "resolving pronouns": 100818,
      "knowledge conversation": 57842,
      "manual work": 67021,
      "systems learning": 116977,
      "traditional dialog": 122812,
      "domains endtoend": 33773,
      "strengths shortcomings": 113576,
      "systems goaloriented": 116924,
      "results comparing": 101609,
      "search browse": 104303,
      "multiturn spoken": 75927,
      "baselines sequencetosequence": 12462,
      "suffer drawbacks": 115217,
      "user behaviour": 128968,
      "acts corresponding": 3141,
      "corresponding user": 24311,
      "process reinforcement": 91557,
      "assistants siri": 9581,
      "learning article": 62371,
      "steps necessary": 113328,
      "dialogue domains": 30668,
      "users important": 129131,
      "random forests": 95497,
      "furthermore used": 46220,
      "applying approaches": 7231,
      "taskoriented nontaskoriented": 118900,
      "research activity": 100397,
      "using tweets": 130325,
      "utterances characterize": 130627,
      "use tweets": 128344,
      "queries improves": 94923,
      "utterance understanding": 130617,
      "systems state": 117158,
      "proposed hybrid": 93308,
      "finance domain": 44251,
      "systems struggle": 117163,
      "work seek": 134790,
      "architecture simultaneously": 8750,
      "dialog research": 30582,
      "humantohuman dialog": 51123,
      "systems furthermore": 116913,
      "ghazvininejad et": 47958,
      "addition unclear": 3482,
      "role human": 103174,
      "agent generates": 4301,
      "natural informative": 76260,
      "comments given": 18815,
      "context combining": 22030,
      "engaging user": 37023,
      "relevant semantic": 98577,
      "framework iterative": 45587,
      "deep rl": 28418,
      "policies deep": 87940,
      "conversations dialog": 23054,
      "dialog flow": 30562,
      "responses automatically": 101255,
      "prior conversations": 90696,
      "agents paper": 4325,
      "algorithm extends": 4735,
      "algorithm data": 4711,
      "advances new": 4006,
      "new frontiers": 78930,
      "systems dialogue": 116850,
      "techniques help": 119900,
      "algorithms finally": 4855,
      "dialogue taskoriented": 30799,
      "current user": 25386,
      "agent model": 4307,
      "model integration": 71354,
      "users experiments": 129115,
      "performance tracking": 86803,
      "reasonable responses": 96213,
      "benefits training": 13024,
      "humanmachine conversation": 51060,
      "collection training": 18494,
      "recent statistical": 96520,
      "addressed extending": 3782,
      "models reflect": 73900,
      "simulated user": 109643,
      "banking finance": 11421,
      "state conversation": 112487,
      "provides basis": 94000,
      "sense domain": 105674,
      "models efficiency": 73120,
      "method suffer": 69171,
      "suffer mismatch": 115231,
      "state distribution": 112491,
      "model map": 71501,
      "utterances interaction": 130651,
      "utterances evaluate": 130637,
      "model atis": 70695,
      "benefits modeling": 13013,
      "make systems": 66737,
      "prize contest": 90789,
      "coarsegrained entity": 18055,
      "conduct conversation": 20833,
      "dialogue contextaware": 30650,
      "contextaware question": 22346,
      "reward signal": 102677,
      "furthermore online": 46196,
      "rl typically": 102895,
      "users suffering": 129176,
      "methods discuss": 69453,
      "master thesis": 67349,
      "questions difficulty": 95302,
      "difficulty maintaining": 31705,
      "challenge goal": 16038,
      "research better": 100432,
      "understand current": 126744,
      "agent provide": 4311,
      "dataset presented": 27103,
      "selection mechanisms": 104799,
      "resolving referring": 100820,
      "approaches addressed": 8051,
      "diverse schemas": 32840,
      "learning candidate": 62415,
      "candidate transformations": 15127,
      "naik et": 76006,
      "instead model": 55669,
      "query reformulation": 94978,
      "reformulation task": 97638,
      "state serialized": 112514,
      "language query": 60015,
      "methodology creating": 69254,
      "support research": 115988,
      "possible user": 88442,
      "account design": 2022,
      "consequently existing": 21179,
      "dialogue process": 30727,
      "user actions": 128966,
      "following given": 44969,
      "natural engaging": 76253,
      "learn grammar": 62058,
      "amounts taskspecific": 5360,
      "generation modules": 47491,
      "complex multidomain": 19837,
      "strong taskspecific": 113723,
      "approach holds": 7609,
      "agents deep": 4321,
      "efficient dialogue": 35069,
      "sequencetosequence methods": 107164,
      "make dialogue": 66657,
      "method datadriven": 68757,
      "induction deep": 54021,
      "discovering user": 32131,
      "motivated intuition": 74853,
      "expressed user": 41728,
      "analysis key": 5666,
      "policy taskoriented": 87969,
      "dialogue existing": 30671,
      "interactions challenge": 55981,
      "challenge users": 16115,
      "predicting multiple": 88996,
      "novel policy": 80685,
      "approaches code": 8097,
      "approaches addressing": 8052,
      "rules lack": 103413,
      "problem remedy": 91204,
      "framework builds": 45448,
      "dataset user": 27259,
      "need support": 76864,
      "everincreasing number": 39623,
      "ontology domain": 81843,
      "spanning 16": 111186,
      "16 domains": 378,
      "building largescale": 14857,
      "largescale virtual": 61519,
      "schemaguided paradigm": 103906,
      "dynamic set": 34326,
      "intents slots": 55926,
      "building proposed": 14878,
      "capable zeroshot": 15223,
      "new apis": 78780,
      "goaloriented conversation": 48406,
      "transfer networks": 124164,
      "data annotations": 25607,
      "unsupervised dialogue": 127622,
      "sources different": 110883,
      "dataset stanford": 27218,
      "literature identifying": 64757,
      "multiple components": 75520,
      "ai solutions": 4425,
      "evaluation real": 39351,
      "popular ones": 88115,
      "performance integrating": 86466,
      "recent progresses": 96502,
      "research taskoriented": 100636,
      "retraining models": 102375,
      "models underlying": 74263,
      "concerning best": 20690,
      "track 8th": 122723,
      "8th dialogue": 1361,
      "dstc track": 34212,
      "new elements": 78896,
      "evaluated subtasks": 39009,
      "network multiturn": 77363,
      "models multiturn": 73596,
      "reasoning problems": 96294,
      "common human": 18884,
      "results really": 102114,
      "strong knowledge": 113684,
      "key desiderata": 57563,
      "end investigate": 36813,
      "explicit use": 41357,
      "systems ignore": 116943,
      "user reactions": 129032,
      "knowledge multitask": 58071,
      "rl problem": 102893,
      "learning enhancing": 62548,
      "survey shed": 116195,
      "supplementary input": 115942,
      "information add": 54362,
      "like talk": 64100,
      "appropriate conversation": 8414,
      "approaches solve": 8345,
      "problem modelbased": 91130,
      "collaborative filtering": 18359,
      "suggest promising": 115328,
      "explain models": 41256,
      "models recommendation": 73891,
      "offline metrics": 81613,
      "stored structured": 113393,
      "collecting datasets": 18446,
      "timeconsuming work": 122150,
      "define dialogue": 28483,
      "enabling better": 36409,
      "addressed dialogue": 3781,
      "domains realworld": 33845,
      "fewshot semantic": 43913,
      "representation previous": 99384,
      "thingtalk language": 121863,
      "graphbased state": 48914,
      "model match": 71505,
      "replicating experiments": 98965,
      "multimodal dialog": 75424,
      "continuously learning": 22654,
      "used production": 128703,
      "learning built": 62414,
      "built evaluated": 14921,
      "defined paper": 28501,
      "paper reports results": 84405,
      "implementation evaluation model": 51944,
      "new domains endtoend": 78891,
      "baselines sequencetosequence model": 12463,
      "sequence dialogue acts": 106930,
      "dialogue acts corresponding": 30635,
      "sequencetosequence model outperforms": 107172,
      "using gaussian process": 129693,
      "process reinforcement learning": 91558,
      "used support vector": 128799,
      "model endtoend differentiable": 71076,
      "model dialogue state": 70990,
      "networks promising approach": 77709,
      "humantohuman dialog corpus": 51124,
      "agent generates natural": 4302,
      "engaging user experience": 37024,
      "present deep reinforcement": 89433,
      "policies deep rl": 87941,
      "results produced using": 102058,
      "10 absolute improvement": 122,
      "recent advances new": 96416,
      "advances new frontiers": 4007,
      "systems deep learning": 116835,
      "taskoriented dialogue taskoriented": 118896,
      "dialogue taskoriented dialogue": 30800,
      "model deep reinforcement": 70959,
      "tracking dialogue state": 122754,
      "significant improvement task": 108778,
      "need large amounts": 76828,
      "collection training data": 18495,
      "proposed framework used": 93301,
      "dialogue state distribution": 30745,
      "evaluate model atis": 38862,
      "training efficiency model": 123600,
      "dialogue systems dialogue": 30776,
      "dialogue systems tasks": 30791,
      "adversarial learning method": 4138,
      "using ensemble model": 129636,
      "training data significantly": 123536,
      "resolving referring expressions": 100821,
      "previous approaches addressed": 90381,
      "learning candidate transformations": 62416,
      "naik et al": 76007,
      "instead model reference": 55670,
      "user query reformulation": 129027,
      "query reformulation task": 94979,
      "reformulation task dialog": 97639,
      "task dialog state": 118088,
      "dialog state serialized": 30586,
      "natural language query": 76517,
      "consequently existing systems": 21180,
      "recent progress language": 96499,
      "evaluations proposed model": 39482,
      "approach holds promise": 7610,
      "reinforcement learning methods": 97815,
      "performance models task": 86540,
      "data enhance performance": 25895,
      "schemaguided dialogue dataset": 103905,
      "spanning 16 domains": 111187,
      "largescale virtual assistants": 61520,
      "training data building": 123440,
      "capable zeroshot generalization": 15224,
      "zeroshot generalization new": 135504,
      "generalization new apis": 46788,
      "research problem paper": 100588,
      "information dialogue act": 54483,
      "majority existing methods": 66601,
      "understanding user utterances": 127006,
      "building large scale": 14856,
      "paper survey recent": 84456,
      "survey recent advances": 116186,
      "8th dialogue technology": 1362,
      "systems evaluated subtasks": 116879,
      "study problem employing": 114481,
      "problem employing pretrained": 91025,
      "adapt pretrained language": 3191,
      "language models multiturn": 59605,
      "target domain domain": 117608,
      "learning rl problem": 62977,
      "survey shed light": 116196,
      "task dialogue systems": 118094,
      "different people different": 31326,
      "new research topic": 79108,
      "appropriate conversation context": 8415,
      "prediction proposed model": 89112,
      "using neural latent": 129950,
      "neural latent variable": 77959,
      "gives better performance": 48184,
      "dialogue state representations": 30747,
      "state tracking model": 112530,
      "code replicating experiments": 18147,
      "replicating experiments available": 98966,
      "taskoriented spoken dialogue systems": 118906,
      "neural networks promising approach": 78541,
      "paper present deep reinforcement": 84093,
      "present deep reinforcement learning": 89434,
      "recent advances new frontiers": 96417,
      "taskoriented dialogue taskoriented dialogue": 118897,
      "dialogue taskoriented dialogue systems": 30801,
      "model deep reinforcement learning": 70960,
      "naik et al 2018": 76008,
      "instead model reference resolution": 55671,
      "user query reformulation task": 129028,
      "query reformulation task dialog": 94980,
      "reformulation task dialog state": 97640,
      "task dialog state serialized": 118089,
      "recent progress language modeling": 96500,
      "automatic human evaluations proposed": 10560,
      "using natural language descriptions": 129935,
      "model dialogue state tracking": 70991,
      "capable zeroshot generalization new": 15225,
      "zeroshot generalization new apis": 135505,
      "paper survey recent advances": 84457,
      "study problem employing pretrained": 114482,
      "pretrained language models multiturn": 90088,
      "large training data available": 61300,
      "reinforcement learning rl problem": 97827,
      "survey shed light future": 116197,
      "users natural language questions": 129147,
      "using neural latent variable": 129951,
      "neural latent variable models": 77960,
      "gives better performance compared": 48185,
      "code replicating experiments available": 18148,
      "paper present deep reinforcement learning": 84094,
      "user query reformulation task dialog": 129029,
      "query reformulation task dialog state": 94981,
      "reformulation task dialog state serialized": 97641,
      "capable zeroshot generalization new apis": 15226,
      "survey shed light future research": 116198,
      "collected amazon alexa prize 2018": 18399,
      "using neural latent variable models": 129952,
      "interfaced": 56105,
      "speculate": 111635,
      "globallocal": 48285,
      "fce": 43228,
      "conll14": 21094,
      "f05": 42643,
      "129": 275,
      "lifting": 63984,
      "015": 7,
      "vertical": 131633,
      "longformer": 65172,
      "windowed": 132715,
      "mae": 66368,
      "responsibilities": 101308,
      "proximal": 94152,
      "destined": 29747,
      "transformerstyle": 124474,
      "smp": 110272,
      "interspersed": 56316,
      "wallclock": 132005,
      "318": 925,
      "linearcomplexity": 64395,
      "277": 860,
      "employed solve": 36307,
      "solve natural": 110601,
      "processing problems": 91776,
      "especially large": 38464,
      "novel realistic": 80698,
      "understanding source": 126967,
      "decoder targetside": 28056,
      "targetside word": 117803,
      "embeddings making": 35801,
      "using fixedsize": 129684,
      "compact set": 19112,
      "attention contexts": 9809,
      "memory approach": 68285,
      "sentences contrast": 106263,
      "subset tokens": 114839,
      "inefficient train": 54075,
      "process soft": 91569,
      "reward signals": 102678,
      "facilitate training": 42796,
      "training hard": 123646,
      "neural transformer": 78717,
      "fast train": 43153,
      "replace original": 98930,
      "robust consistent": 103026,
      "networks san": 77752,
      "multilevel attention": 75190,
      "chinese poem": 16804,
      "ability existing": 1515,
      "pervasive attention": 87203,
      "2d convolutional": 876,
      "networks sequencetosequence": 77761,
      "prediction current": 89047,
      "sequence generate": 106944,
      "relies single": 98660,
      "translation architectures": 124647,
      "features source": 43737,
      "cnns outperform": 18043,
      "attention autoencoder": 9797,
      "learn universal": 62177,
      "sequence encoding": 106939,
      "range 10": 95538,
      "models decoding": 73020,
      "values generated": 130795,
      "information current": 54463,
      "multisource transformer": 75792,
      "multisource sequencetosequence": 75790,
      "contextaware selfattention": 22348,
      "information proven": 54885,
      "simplicity flexibility": 109569,
      "layers used": 61814,
      "furthermore conducted": 46156,
      "successes nlp": 115146,
      "resource constraints": 100834,
      "tasks led": 119251,
      "questions ways": 95375,
      "positional embedding": 88299,
      "clean test": 17703,
      "lm text": 64871,
      "roles played": 103234,
      "played important": 87711,
      "shown strong": 108529,
      "model slow": 72060,
      "policy jointly": 87957,
      "results yields": 102351,
      "yields average": 135396,
      "bleu stateoftheart": 14337,
      "speedup works": 111889,
      "dropout methods": 34179,
      "layer recurrent": 61743,
      "normalization propose": 80346,
      "normalization layers": 80336,
      "wmt iwslt": 132793,
      "ability encoding": 1514,
      "strategies encode": 113458,
      "tree experimental": 125590,
      "information past": 54831,
      "token identity": 122256,
      "syntax knowledge": 116538,
      "leading erroneous": 61891,
      "erroneous translations": 38273,
      "module trained": 74512,
      "introduced training": 56601,
      "relations directly": 98139,
      "rnn structure": 102936,
      "addition help": 3433,
      "designed multiple": 29611,
      "tradeoffs compared": 122784,
      "source syntax": 110828,
      "past papers": 85647,
      "sourceside syntax": 110943,
      "syntax using": 116569,
      "data manipulation": 26116,
      "particular train": 85459,
      "heads produce": 49609,
      "multitask setups": 75893,
      "trees instead": 125696,
      "final decoder": 44101,
      "furthermore make": 46188,
      "replacement standard": 98942,
      "selfattention combines": 104875,
      "aggregation strategy": 4369,
      "algorithm taking": 4806,
      "low layer": 65371,
      "layer capsules": 61703,
      "high layer": 50081,
      "translation test": 125375,
      "wmt datasets": 132784,
      "scores different": 104167,
      "architectures achieved": 8776,
      "tasks evidence": 119108,
      "experts propose": 41248,
      "block coordinate": 14357,
      "coordinate descent": 23260,
      "parameters experiments": 84748,
      "experts different": 41239,
      "asr work": 9455,
      "present multiple": 89563,
      "multiple strategies": 75705,
      "layers effectively": 61771,
      "pure transformer": 94411,
      "layers number": 61796,
      "function obtain": 46037,
      "context jointly": 22154,
      "aggregate semantic": 4350,
      "adaptively combine": 3340,
      "simultaneously preserving": 109692,
      "model facilitating": 71168,
      "masking method": 67312,
      "empirical analyses": 36148,
      "better relative": 13693,
      "order preserve": 82386,
      "notion word": 80412,
      "quality wide": 94827,
      "methods transformer": 69814,
      "nonlinear mappings": 80225,
      "store history": 113380,
      "dynamic memory": 34315,
      "significantly reduced": 109046,
      "pattern shows": 85720,
      "equal attention": 38221,
      "feed forward": 43816,
      "forward neural": 45327,
      "change course": 16358,
      "course training": 24466,
      "training attention": 123365,
      "allowed train": 5105,
      "component language": 19985,
      "transformer paper": 124368,
      "knowledge lm": 58057,
      "better conventional": 13550,
      "increase computational": 53591,
      "finetuning specifically": 44515,
      "different downstream": 31116,
      "metalearning paradigm": 68556,
      "inference experimental": 54140,
      "released future": 98485,
      "revolution natural": 102660,
      "mechanisms based": 68055,
      "existing new": 40234,
      "multiple structural": 75708,
      "different structural": 31456,
      "priors different": 90778,
      "vocabulary experiments": 131876,
      "ones values": 81709,
      "layers residual": 61806,
      "time convergence": 122002,
      "baseline loss": 12243,
      "significant word": 108874,
      "consistent perplexity": 21397,
      "encoding text": 36731,
      "information meaning": 54759,
      "enhance transformer": 37498,
      "approach typical": 7989,
      "dedicated models": 28181,
      "compression approach": 20306,
      "conclude comparing": 20710,
      "efficiency gains": 35025,
      "gains especially": 46392,
      "heuristic feature": 49873,
      "yield suboptimal": 135362,
      "named dynamic": 76027,
      "propose cascade": 92580,
      "wikitext2 corpora": 132700,
      "corpora proposed": 23563,
      "solve natural language": 110602,
      "language processing problems": 59935,
      "corpora used nlp": 23612,
      "propose use recurrent": 93143,
      "word embeddings making": 133157,
      "promising results machine": 92296,
      "seq2seq model efficient": 106900,
      "encoder decoder states": 36514,
      "training time translation": 123930,
      "different network architectures": 31286,
      "use convolutional layers": 127968,
      "improve performance sequence": 52486,
      "text generation machine": 120987,
      "2d convolutional neural": 877,
      "neural networks sequencetosequence": 78573,
      "encode input sequence": 36435,
      "input sequence generate": 55428,
      "sequence generate output": 106945,
      "machine translation architectures": 65898,
      "training model large": 123711,
      "traditional recurrent neural": 122861,
      "facilitate information flow": 42776,
      "achieves superior results": 2920,
      "multisource transformer decoder": 75793,
      "successes nlp tasks": 115147,
      "grammatical error detection": 48699,
      "network model pretrained": 77336,
      "work investigate effect": 134590,
      "nlp tasks led": 79732,
      "experiments wide range": 41205,
      "advanced stateoftheart various": 3959,
      "sentence propose strategies": 106008,
      "tree experimental results": 125591,
      "context information experiments": 22143,
      "achieves comparable translation": 2765,
      "transformer attention heads": 124296,
      "response generation task": 101213,
      "impact translation quality": 51898,
      "transformer base model": 124301,
      "simple linear transformation": 109458,
      "machine translation test": 66267,
      "languages french spanish": 60601,
      "architectures achieved stateoftheart": 8777,
      "mixture experts propose": 70420,
      "block coordinate descent": 14358,
      "performance paper present": 86596,
      "deep learning field": 28272,
      "complexity grows quadratically": 19914,
      "context document classification": 22065,
      "results compared state": 101604,
      "language model achieves": 59288,
      "significant improvements semantic": 108802,
      "translation recent studies": 125200,
      "problem propose simple": 91183,
      "handle long sequences": 49392,
      "widely used stateoftheart": 132591,
      "feed forward neural": 43817,
      "forward neural network": 45328,
      "based sequencetosequence seq2seq": 12034,
      "large paired data": 61198,
      "transformer paper proposes": 124369,
      "propose method called": 92767,
      "different downstream tasks": 31117,
      "inference experimental results": 54141,
      "revolution natural language": 102661,
      "attention mechanisms based": 9921,
      "future nlp research": 46287,
      "translation using different": 125418,
      "using different datasets": 129570,
      "performance various machine": 86843,
      "masked language modelling": 67295,
      "widely used standard": 132590,
      "training data order": 123505,
      "downstream tasks evaluation": 34037,
      "achieves similar better": 2868,
      "machine translation experiment": 65982,
      "datasets analysis shows": 27308,
      "finetuning pretrained transformers": 44505,
      "treebank wikitext2 corpora": 125653,
      "natural language processing problems": 76476,
      "propose use recurrent neural": 93144,
      "promising results machine translation": 92297,
      "experiments proposed method significantly": 41077,
      "method significantly improve performance": 69136,
      "text generation machine translation": 120988,
      "2d convolutional neural networks": 879,
      "input sequence generate output": 55429,
      "sequence generate output sequence": 106946,
      "2d convolutional neural network": 878,
      "neural machine translation architectures": 77971,
      "alleviate problems propose novel": 5049,
      "translation tasks demonstrate effectiveness": 125352,
      "neural network model pretrained": 78337,
      "pretrained language representation model": 90108,
      "tasks proposed method significantly": 119416,
      "achieves comparable translation quality": 2766,
      "results compared state art": 101605,
      "machine translation recent studies": 66189,
      "feed forward neural network": 43818,
      "based sequencetosequence seq2seq model": 12035,
      "revolution natural language processing": 102662,
      "machine translation using different": 66294,
      "achieves similar better performance": 2869,
      "penn treebank wikitext2 corpora": 85865,
      "input sequence generate output sequence": 55430,
      "neural machine translation recent studies": 78088,
      "bp": 14535,
      "796": 1276,
      "conll2005": 21098,
      "conll2012": 21100,
      "extractable": 42133,
      "notwithstanding": 80420,
      "interlanguage": 56115,
      "parserbased": 85008,
      "pps": 88694,
      "unanalyzed": 126581,
      "syntacticallyinformed": 116521,
      "peter": 87206,
      "866": 1333,
      "humanvalidated": 51126,
      "denser": 29029,
      "involved creating": 56880,
      "evaluation englishgerman": 39203,
      "manually automatically": 67039,
      "explore usefulness": 41609,
      "model evaluations": 71115,
      "changing model": 16400,
      "resulting small": 101474,
      "techniques approach": 119839,
      "structures related": 114100,
      "approaches utilize": 8397,
      "identify position": 51534,
      "structures use": 114114,
      "roles based": 103229,
      "stateofart models": 112554,
      "verbs semantic": 131561,
      "proposed theoretical": 93571,
      "chinese semantic": 16816,
      "single corpus": 109720,
      "usually exists": 130419,
      "generation robust": 47607,
      "robust semantic": 103069,
      "robust nlp": 103054,
      "process train": 91579,
      "art standard": 9093,
      "task linguistic": 118356,
      "task comes": 117974,
      "propose syntax": 93101,
      "sequence semantic": 107047,
      "resources automatic": 100948,
      "automatic semantic": 10649,
      "enables detection": 36381,
      "sentences participants": 106441,
      "portuguese bp": 88194,
      "corpus syntactic": 24031,
      "represent real": 99125,
      "balanced corpus": 11403,
      "architecture predicting": 8729,
      "sentences extended": 106309,
      "building semantic": 14884,
      "systems wellstudied": 117233,
      "wellstudied languages": 132434,
      "usually used": 130465,
      "addition common": 3407,
      "value important": 130779,
      "exploration concept": 41497,
      "target syntactic": 117718,
      "bootstrapping method": 14459,
      "fscore improvement": 45912,
      "techniques related": 119968,
      "mtl model": 75017,
      "subtask semantic": 114950,
      "present effort": 89462,
      "texts particular": 121575,
      "statistical classifiers": 113087,
      "experiments achieves": 40758,
      "research believe": 100429,
      "conll2012 shared": 21101,
      "tokens second": 122331,
      "documents need": 33257,
      "restrict search": 101337,
      "types identified": 126296,
      "grammatical functions": 48705,
      "extraction domain": 42312,
      "lstm encoding": 65619,
      "encoding strategy": 36728,
      "increased accuracy": 53625,
      "sequence simultaneously": 107070,
      "attend syntactic": 9780,
      "outofdomain text": 82664,
      "independent decisions": 53764,
      "polyglot semantic": 88014,
      "approaches multilingual": 8240,
      "role labeler": 103181,
      "baseline analysis": 12187,
      "polyglot model": 88012,
      "shows advantageous": 108547,
      "experiments multitask": 41029,
      "role labels": 103197,
      "decoding constraints": 28082,
      "endtoend semantic": 36962,
      "importance syntactic": 52085,
      "case task": 15631,
      "standard automatic": 112197,
      "l2 data": 58374,
      "data l2": 26063,
      "indicating importance": 53871,
      "explore semantic": 41584,
      "baseline semisupervised": 12306,
      "based explicitly": 11705,
      "instances train": 55637,
      "data varying": 26620,
      "major focus": 66561,
      "processing characteristics": 91636,
      "task purpose": 118590,
      "reported previous": 99069,
      "predicates corresponding": 88863,
      "embeddings enhanced": 35693,
      "prepositional phrases": 89314,
      "pp attachment": 88683,
      "results especially": 101763,
      "classification pps": 17335,
      "improving sentence": 53162,
      "2012 datasets": 566,
      "roles evaluation": 103231,
      "approach event": 7545,
      "obviate need": 81483,
      "elmo outperforms": 35319,
      "gap f1": 46454,
      "golden age": 48451,
      "dependency span": 29226,
      "sentence endtoend": 105856,
      "successful paper": 115168,
      "representation deal": 99202,
      "identification subtask": 51447,
      "argument identification": 8945,
      "variants results": 130882,
      "combines strengths": 18700,
      "performance conll": 86250,
      "neural nli": 78613,
      "datasets capture": 27341,
      "issues make": 57059,
      "labelling model": 58570,
      "models indomain": 73397,
      "nlp motivated": 79648,
      "thorough study": 121897,
      "representations beneficial": 99518,
      "representations approaches": 99504,
      "10 respectively": 148,
      "integrate syntax": 55768,
      "effectiveness deep": 34880,
      "structure refinement": 113942,
      "multilayer lstms": 75174,
      "encode prior": 36445,
      "nonlocal interactions": 80237,
      "modeling choice": 72390,
      "involves extracting": 56891,
      "sentences stateoftheart": 106504,
      "model nonlocal": 71591,
      "maintaining efficient": 66513,
      "use capsule": 127931,
      "types mistakes": 126320,
      "different quality": 31366,
      "experiments newly": 41040,
      "style paper": 114587,
      "providing solution": 94134,
      "syntaxaware multitask": 116576,
      "attention improve": 9858,
      "performance exploiting": 86362,
      "knowledge achieving": 57740,
      "research orientations": 100576,
      "adopt simple": 3896,
      "simple unified": 109537,
      "representations dependency": 99588,
      "scores test": 104212,
      "translated training": 124575,
      "efforts research": 35205,
      "performances crosslingual": 86875,
      "sentences reliable": 106472,
      "reliable linguistic": 98616,
      "annotated version": 6261,
      "include annotations": 53221,
      "adapted languages": 3285,
      "features contextualized": 43424,
      "crucial crosslingual": 25140,
      "representations bring": 99526,
      "important neural": 52194,
      "models firstorder": 73256,
      "annotation styles": 6380,
      "humanvalidated test": 51127,
      "sets use": 107722,
      "polyglot training": 88015,
      "idea training": 51339,
      "baselines especially": 12387,
      "fact simple": 42833,
      "embeddings coupled": 35621,
      "contrary traditional": 22686,
      "transitionbased framework": 124506,
      "transformers transfer": 124465,
      "propbank semantic": 92406,
      "proposed english": 93268,
      "indonesian news": 53993,
      "article paper": 9144,
      "200 words": 536,
      "shown incorporating": 108487,
      "soft bias": 110488,
      "manually automatically annotated": 67040,
      "approach motivated observation": 7717,
      "semantic roles based": 105261,
      "role labeling task": 103195,
      "annotated corpus training": 6155,
      "release new corpus": 98461,
      "state art standard": 112476,
      "paper propose syntax": 84330,
      "brazilian portuguese bp": 14573,
      "mining applications paper": 70226,
      "test data approach": 120434,
      "important step understanding": 52258,
      "syntactic information paper": 116421,
      "approaches outperform stateoftheart": 8271,
      "model makes predictions": 71495,
      "semantic parsing natural": 105153,
      "conll2012 shared task": 21102,
      "information extraction open": 54586,
      "dependency parsing partofspeech": 29191,
      "parsing partofspeech tagging": 85190,
      "semantic role labeler": 105245,
      "model shows advantageous": 72013,
      "active learning neural": 3115,
      "learning neural semantic": 62846,
      "encourage research area": 36763,
      "semantic role labels": 105259,
      "studies semantic parsing": 114277,
      "corpora method based": 23527,
      "reported previous work": 99070,
      "stateoftheart strong baselines": 112975,
      "using human judgments": 129745,
      "frame semantic parsing": 45387,
      "improve crosslingual transfer": 52367,
      "multilingual representation learning": 75349,
      "use dependency parsing": 127994,
      "model combines strengths": 70845,
      "tags dependency parse": 117470,
      "stateoftheart method uses": 112701,
      "neural nli models": 78614,
      "nli models trained": 79547,
      "systems better understanding": 116760,
      "sequence labelling model": 106994,
      "encode prior knowledge": 36446,
      "use capsule networks": 127932,
      "syntactic information derived": 116418,
      "model learns general": 71447,
      "dependency span representations": 29227,
      "proposed methods effective": 93413,
      "syntaxaware multitask learning": 116577,
      "performance exploiting syntactic": 86363,
      "scores test data": 104213,
      "competitive performances crosslingual": 19673,
      "constructing highquality training": 21676,
      "syntactic information neural": 116420,
      "use multilingual bert": 128155,
      "performance base model": 86164,
      "embeddings derived bert": 35641,
      "generation network generates": 47503,
      "confirm proposed model": 21019,
      "systems english chinese": 116869,
      "propbank semantic role": 92407,
      "achieved better performance": 2601,
      "allows model use": 5174,
      "semantic role labeling task": 105258,
      "semantic parsing natural language": 105154,
      "parsing natural language sentences": 85172,
      "conll2012 shared task dataset": 21103,
      "open information extraction open": 81908,
      "dependency parsing partofspeech tagging": 29192,
      "improve performance strong baselines": 52489,
      "crucial natural language understanding": 25155,
      "propose method automatically construct": 92765,
      "propbank semantic role labeling": 92408,
      "diffusion": 31711,
      "volatility": 131934,
      "powerlaw": 88681,
      "prison": 90780,
      "metaphoric": 68563,
      "haha": 49294,
      "tomorrows": 122360,
      "randomness": 95536,
      "hurricane": 51163,
      "revolutionised": 102663,
      "timerelated": 122164,
      "obsolete": 81246,
      "kim": 57671,
      "tons": 122365,
      "slowly": 110104,
      "75k": 1258,
      "romantic": 103257,
      "mirroring": 70284,
      "languageinternal": 60370,
      "timespecific": 122198,
      "uob": 127779,
      "fastchanging": 43157,
      "qmulsds": 94535,
      "earliest": 34382,
      "cosines": 24342,
      "backandforth": 11284,
      "rdt": 95885,
      "networks human": 77621,
      "introduces human": 56615,
      "models past": 73719,
      "change language": 16363,
      "language emergence": 58997,
      "types social": 126360,
      "networks modeling": 77662,
      "diachronic corpora": 30488,
      "return volatility": 102480,
      "token partofspeech": 122258,
      "levels variation": 63568,
      "temporal analysis": 120090,
      "model google": 71260,
      "words changed": 133849,
      "changed significantly": 16378,
      "develop test": 30243,
      "known historical": 58304,
      "using historical": 129741,
      "reasons including": 96328,
      "different distributional": 31100,
      "words distributional": 133922,
      "methods allows": 69305,
      "model topic": 72174,
      "style similarity": 114594,
      "classify texts": 17663,
      "texts respect": 121594,
      "word unigram": 133617,
      "machines classifier": 66333,
      "date documents": 27824,
      "half century": 49296,
      "performed investigate": 86926,
      "task linked": 118357,
      "sense given": 105679,
      "figurative meaning": 44015,
      "dynamic graph": 34311,
      "previous time": 90511,
      "senses firstly": 105715,
      "different periods": 31327,
      "languages presented": 60807,
      "ngram compared": 79429,
      "hypernym detection": 51215,
      "change build": 16357,
      "contextual variations": 22515,
      "detecting semantic": 29850,
      "terms contextual": 120298,
      "models degree": 73032,
      "define baseline": 28482,
      "order overcome": 82378,
      "issues data": 57042,
      "started pay": 112404,
      "words approaches": 133823,
      "specific metadata": 111463,
      "compare meaning": 19253,
      "space measure": 111021,
      "period time": 87038,
      "ideology detection": 51647,
      "words likely": 134037,
      "words senses": 134192,
      "computation paper": 20352,
      "despite potential": 29714,
      "shows current": 108569,
      "english need": 37218,
      "online writing": 81819,
      "growth online": 49192,
      "learns languages": 63216,
      "varies time": 130952,
      "time change": 121993,
      "train random": 123002,
      "annotation lexical": 6330,
      "framework extends": 45538,
      "tools currently": 122445,
      "realtime analysis": 96136,
      "techniques enables": 119879,
      "loss precision": 65292,
      "fast linear": 43143,
      "language availability": 58852,
      "taken reflect": 117513,
      "fiction texts": 43926,
      "used quantify": 128710,
      "extracted massive": 42163,
      "widely believed": 132540,
      "interactive tools": 56034,
      "survey current": 116165,
      "related semantic": 97895,
      "phenomena propose": 87244,
      "produced approach": 91948,
      "interpreting results": 56291,
      "available applying": 10946,
      "data fundamental": 25963,
      "fundamental differences": 46095,
      "approaches lexical": 8218,
      "senses new": 105717,
      "lexical replacement": 63796,
      "review article": 102535,
      "semantic changes": 105004,
      "research automatically": 100426,
      "systematically control": 116694,
      "possible biases": 88385,
      "evaluating lexical": 39067,
      "extending benchmark": 41815,
      "common stateoftheart": 18930,
      "detection domainspecific": 29926,
      "kim et": 57672,
      "particular study": 85452,
      "articles generated": 9190,
      "allows understand": 5193,
      "techniques specifically": 119987,
      "news different": 79333,
      "possible identify": 88408,
      "consists 75k": 21467,
      "detection generating": 29957,
      "senseannotated data": 105709,
      "procedure simulate": 91400,
      "dataset represents": 27158,
      "compared common": 19343,
      "weaknesses propose": 132211,
      "linguistic application": 64425,
      "domains terms": 33872,
      "german based": 47880,
      "times present": 122186,
      "published new": 94363,
      "realworld environment": 96168,
      "showcase importance": 108362,
      "detecting words": 29861,
      "quantitative manner": 94871,
      "semeval2020 shared": 105548,
      "ranking words": 95695,
      "time analyse": 121981,
      "detection algorithms": 29874,
      "task interestingly": 118305,
      "particular algorithm": 85378,
      "corpora differing": 23460,
      "belong topic": 12718,
      "task addresses": 117849,
      "german latin": 47902,
      "low dimensionality": 65359,
      "results semeval2020": 102150,
      "used lexical": 128610,
      "vector initialization": 131309,
      "time location": 122047,
      "changes taking": 16392,
      "include temporal": 53237,
      "different connotations": 31055,
      "russian nouns": 103498,
      "bert architectures": 13066,
      "modeling russian": 72536,
      "contexts extracted": 22393,
      "room researchers": 103270,
      "uob semeval2020": 127780,
      "approach lexical": 7684,
      "based bayesian": 11545,
      "identification approach": 51363,
      "used submission": 128789,
      "propose semantic": 93031,
      "results main": 101902,
      "near perfect": 76712,
      "approaches yield": 8406,
      "tune parameters": 125930,
      "approach historical": 7608,
      "linguistics word": 64627,
      "approach field": 7572,
      "contexts using": 22433,
      "extracted annotated": 42138,
      "change information": 16361,
      "task ranked": 118598,
      "detection ranked": 30031,
      "research far": 100503,
      "methods modelling": 69623,
      "aspects problem": 9403,
      "twitter word": 126133,
      "models semeval2020": 73993,
      "field investigate": 43958,
      "shift combining": 108174,
      "false discovery": 43037,
      "discovery rate": 32148,
      "data semeval2020": 26391,
      "change logistic": 16365,
      "linguistic evidence": 64473,
      "propose consider": 92594,
      "cases jointly": 15649,
      "paper introduces human": 84007,
      "words changed significantly": 133850,
      "word partofspeech tag": 133402,
      "text classification algorithms": 120708,
      "vector machines classifier": 131322,
      "researchers started pay": 100711,
      "started pay attention": 112405,
      "emerging research field": 36072,
      "train random forest": 123003,
      "annotated dataset use": 6177,
      "years seen surge": 135299,
      "distributional word embedding": 32746,
      "paper compare models": 83766,
      "kim et al": 57673,
      "able capture semantic": 1619,
      "new high german": 78946,
      "new method leverages": 79013,
      "performance comparable current": 86221,
      "detection task detecting": 30074,
      "nlp field recently": 79615,
      "transformerbased bert model": 124404,
      "compare performance systems": 19278,
      "systems present results": 117072,
      "present results shared": 89677,
      "semeval2020 shared task": 105549,
      "scores test set": 104214,
      "shared task addresses": 108022,
      "present results semeval2020": 89675,
      "results semeval2020 task": 102151,
      "model based skipgram": 70745,
      "quantitative qualitative evaluations": 94879,
      "model compares favorably": 70855,
      "test sets consist": 120528,
      "test set task": 120522,
      "uob semeval2020 task": 127781,
      "novel word senses": 80780,
      "approach lexical semantic": 7685,
      "novel word sense": 80779,
      "analysis method based": 5694,
      "present results main": 89671,
      "results main findings": 101903,
      "model ranked 3rd": 71845,
      "historical linguistics word": 50529,
      "using computational linguistics": 129463,
      "task ranked 1st": 118599,
      "embedding models different": 35454,
      "model language based": 71404,
      "models semeval2020 task": 73994,
      "nlp tasks does": 79715,
      "target word encoded": 117748,
      "false discovery rate": 43038,
      "demonstrate performance approach": 28815,
      "data semeval2020 task": 26392,
      "logistic regression models": 65038,
      "support vector machines classifier": 116011,
      "researchers started pay attention": 100712,
      "corpora recent studies shown": 23570,
      "recent years seen surge": 96635,
      "propose new method leverages": 92866,
      "present results shared task": 89678,
      "present results semeval2020 task": 89676,
      "model compares favorably stateoftheart": 70856,
      "present results main findings": 89672,
      "word embedding models different": 133025,
      "embased": 35346,
      "msh": 74948,
      "polysemic": 88022,
      "dna": 32927,
      "senselevel": 105712,
      "594": 1135,
      "immediacy": 51842,
      "fullcoverage": 45929,
      "conflation": 21032,
      "glossbert": 48294,
      "affordances": 4261,
      "knowledge especially": 57904,
      "case frame": 15585,
      "patterns large": 85753,
      "disambiguation approach": 31956,
      "problem subproblems": 91253,
      "dependency learning": 29154,
      "acquisition problem": 3050,
      "disambiguation systems": 31987,
      "systems construct": 116809,
      "poses considerable": 88262,
      "corpusbased approach": 24075,
      "presents use": 89912,
      "lexicon used": 63911,
      "bottleneck paper": 14496,
      "shown fail": 108472,
      "corpus following": 23807,
      "genre topic": 47833,
      "disambiguation results": 31984,
      "better future": 13593,
      "sense ambiguous": 105648,
      "sensetagged corpora": 105726,
      "disambiguation explore": 31965,
      "using conceptual": 129469,
      "use generalized": 128065,
      "idea combination": 51319,
      "lead accurate": 61841,
      "sources word": 110931,
      "robust systems": 103077,
      "sources hand": 110893,
      "serve foundation": 107295,
      "base wordnet": 11492,
      "combining knowledge": 18730,
      "resolution lexical": 100766,
      "ambiguity arises": 5279,
      "commonly referred": 18962,
      "task wsd": 118858,
      "using electronic": 129618,
      "ambiguity words": 5288,
      "study experimental": 114375,
      "work evaluation": 134497,
      "methods combinations": 69375,
      "improved approach": 52594,
      "word correct": 132977,
      "word identified": 133312,
      "ambiguous sentences": 5296,
      "information accuracy": 54354,
      "nlp progress": 79673,
      "performance levels": 86495,
      "semeval 2007": 105493,
      "word single": 133576,
      "single sense": 109792,
      "correct meaning": 24106,
      "verb adjective": 131514,
      "adjective adverb": 3855,
      "extracted neural": 42168,
      "algorithms simple": 4882,
      "word text": 133602,
      "concepts methods": 20626,
      "employed tackle": 36308,
      "recognition performed": 96964,
      "words structural": 134243,
      "used pattern": 128683,
      "problem approaches": 90938,
      "feature words": 43331,
      "texts focus": 121521,
      "words special": 134233,
      "features employed": 43477,
      "algorithm particular": 4781,
      "particular cases": 85388,
      "useful improve": 128893,
      "learned cnn": 62207,
      "corpora small": 23588,
      "create concept": 24606,
      "representations compare": 99551,
      "fast easy": 43134,
      "key advantage": 57538,
      "highquality corpus": 50374,
      "definitions different": 28536,
      "combine semantic": 18641,
      "filtering relevant": 44087,
      "containing 200": 21781,
      "handlabeled examples": 49369,
      "concept mapping": 20578,
      "test instance": 120462,
      "recently efforts": 96682,
      "exception trend": 39928,
      "advances representation": 4017,
      "used meaning": 128625,
      "limitation using": 64167,
      "bagofwords approaches": 11370,
      "based experimentation": 11701,
      "english crosslingual": 37100,
      "comparisons paper": 19593,
      "task monolingual": 118406,
      "assist language": 9569,
      "learners applied": 62297,
      "lexical chain": 63735,
      "rate 78": 95783,
      "proximity measure": 94154,
      "f1 unsupervised": 42699,
      "predictions word": 89197,
      "knowledgebased systems": 58257,
      "tool features": 122388,
      "makes sense": 66804,
      "widelyused approach": 132603,
      "algorithm stateoftheart": 4802,
      "random choices": 95474,
      "information nlp": 54804,
      "disambiguation input": 31967,
      "fine granularity": 44323,
      "sense representation": 105695,
      "new rich": 79112,
      "classifier modified": 17559,
      "relationship training": 98311,
      "accuracy classifier": 2115,
      "obtained data": 81356,
      "ability quickly": 1565,
      "constantly evolving": 21518,
      "various senses": 131190,
      "vector semantic": 131366,
      "space consider": 110984,
      "extracted training": 42189,
      "release community": 98442,
      "wordnet proved": 133776,
      "data context": 25791,
      "context ignoring": 22125,
      "integrate context": 55748,
      "relationship context": 98288,
      "methods knowledgebased": 69569,
      "words frequent": 133965,
      "translation bitext": 124677,
      "introduce adaptive": 56376,
      "restaurant processes": 101321,
      "sense selection": 105697,
      "average sense": 11215,
      "task sense": 118687,
      "sense detection": 105659,
      "used new": 128659,
      "isolation results": 56980,
      "wordnet reduce": 133777,
      "disambiguation contextual": 31959,
      "meaning conflation": 67625,
      "focus better": 44738,
      "longstanding problem": 65193,
      "combining bert": 18707,
      "polysemic words": 88023,
      "effective downstream": 34668,
      "extracting contextual": 42206,
      "text analyzing": 120643,
      "words list": 134038,
      "known advance": 58285,
      "embeddings space": 35948,
      "task inherent": 118300,
      "building supervised": 14892,
      "appropriate sense": 8436,
      "approach taking": 7959,
      "highly overlapping": 50333,
      "sense learning": 105691,
      "senses label": 105716,
      "tasks dont": 119075,
      "lack coverage": 58697,
      "dataset stateoftheart": 27219,
      "provided model": 93975,
      "gap machine": 46462,
      "independently embeds": 53788,
      "finding nearest": 44279,
      "nearest sense": 76726,
      "parser produce": 84989,
      "resulting logical": 101451,
      "models taken": 74155,
      "semantic nuances": 105128,
      "respect lexical": 101090,
      "sense analysis": 105650,
      "finetuning feature": 44458,
      "proves robust": 93745,
      "studies illustrate": 114237,
      "languagespecific model": 60995,
      "tagging universal": 117459,
      "used predicting": 128689,
      "missing text": 70334,
      "drive future": 34146,
      "using mbert": 129870,
      "dependencies crosslingual": 29073,
      "context pretrained": 22225,
      "words incorporated": 133996,
      "proposed reduce": 93534,
      "high dimensionality": 50065,
      "humans natural": 51094,
      "discusses impact": 32311,
      "ambiguity rate": 5285,
      "multisense embeddings": 75776,
      "disambiguation process": 31983,
      "text resulting": 121263,
      "domains provides": 33841,
      "set containing": 107402,
      "patterns large corpus": 85754,
      "knowledge acquisition problem": 57744,
      "method word sense": 69228,
      "sense ambiguous word": 105649,
      "idea combination different": 51320,
      "different sources knowledge": 31441,
      "sources word sense": 110932,
      "task commonly referred": 117977,
      "sources different kinds": 110884,
      "processing nlp progress": 91750,
      "words related words": 134165,
      "noun verb adjective": 80435,
      "verb adjective adverb": 131515,
      "concepts methods complex": 20627,
      "methods complex networks": 69386,
      "proposed learning algorithm": 93322,
      "feature maps learned": 43297,
      "performance previous approaches": 86619,
      "entities knowledge bases": 37805,
      "absolute improvement best": 1735,
      "recent advances representation": 96423,
      "advances representation learning": 4018,
      "using supervised unsupervised": 130243,
      "pair word senses": 83466,
      "approach uses semantic": 8004,
      "naive bayesian classifier": 76022,
      "results obtained data": 102006,
      "vector semantic space": 131367,
      "neural network jointly": 78312,
      "stateoftheart systems english": 112989,
      "chinese restaurant processes": 16815,
      "sense selection mechanism": 105698,
      "mainly focuses improving": 66489,
      "order achieve results": 82268,
      "nearest neighbors knn": 76724,
      "propose use bert": 93139,
      "model word sense": 72324,
      "novel method encode": 80632,
      "method encode word": 68803,
      "similarity word vector": 109339,
      "disambiguation word senses": 31998,
      "languages models available": 60733,
      "learn appropriate sense": 61990,
      "approach taking advantage": 7960,
      "gap machine human": 46463,
      "machine human performance": 65759,
      "finding nearest sense": 44280,
      "target word embedding": 117746,
      "task semantic parsing": 118673,
      "number examples available": 80884,
      "finetuning feature extraction": 44459,
      "limited available training": 64214,
      "training sentences word": 123838,
      "using computational methods": 129464,
      "partofspeech tagging universal": 85563,
      "tagging universal dependency": 117460,
      "new dataset assessing": 78859,
      "extracting contextual information": 42207,
      "humans natural language": 51095,
      "paper discusses impact": 83879,
      "dataset automatically extracted": 26752,
      "language processing nlp progress": 59914,
      "noun verb adjective adverb": 80436,
      "concepts methods complex networks": 20628,
      "recent advances representation learning": 96424,
      "using supervised unsupervised approaches": 130244,
      "performance downstream nlp applications": 86320,
      "outperforms stateoftheart systems english": 83017,
      "finetune pretrained bert model": 44413,
      "novel method encode word": 80633,
      "similarity word vector representations": 109340,
      "gap machine human performance": 46464,
      "outperforms previous stateoftheart models": 82950,
      "limited available training data": 64215,
      "partofspeech tagging universal dependency": 85564,
      "natural language processing nlp progress": 76459,
      "problem natural language processing recently": 91142,
      "results model outperforms stateoftheart systems": 101948,
      "multiplesequence": 75758,
      "wieting": 132621,
      "lexicallyconstrained": 63859,
      "0709": 58,
      "blocking": 14363,
      "db": 27847,
      "warp": 132033,
      "digging": 31713,
      "multiplesequence alignment": 75759,
      "texttotext generation": 121669,
      "accurate paraphrases": 2349,
      "paraphrase model": 84828,
      "database ppdb": 26644,
      "used heuristic": 128572,
      "simultaneously improving": 109680,
      "query introduce": 94968,
      "identification work": 51460,
      "discontinuous linguistic": 32031,
      "identification neural": 51409,
      "generation conventional": 47345,
      "handwritten rules": 49431,
      "models bleu": 72859,
      "focus recent": 44809,
      "embeddings consider": 35608,
      "wieting et": 132622,
      "include training": 53240,
      "finding clear": 44269,
      "research lack": 100541,
      "method collect": 68698,
      "30000 new": 912,
      "utility downstream": 130478,
      "automatically proposed": 10818,
      "sentence quantitative": 106011,
      "demonstrates efficacy": 28944,
      "explored natural": 41627,
      "analysis machine": 5687,
      "noncompositional phrase": 80173,
      "easily generalized": 34454,
      "generate compositional": 46915,
      "data generator": 25978,
      "built sequencetosequence": 14938,
      "inverse reinforcement": 56702,
      "depending type": 29278,
      "improve downstream": 52375,
      "outperform supervised": 82765,
      "systems semeval": 117133,
      "makes contributions": 66766,
      "contextaware model": 22338,
      "previous techniques": 90510,
      "adversarial example": 4115,
      "example generation": 39783,
      "propose syntactically": 93100,
      "paraphrase sentence": 84834,
      "syntax possible": 116552,
      "paraphrase quality": 84831,
      "furthermore capable": 46150,
      "examples fool": 39837,
      "addressed literature": 3784,
      "hard time": 49482,
      "representing paraphrases": 100062,
      "modeling compare": 72396,
      "models claimed": 72904,
      "surprisingly improves": 116141,
      "performance twitter": 86817,
      "far larger": 43090,
      "networks effectively": 77578,
      "sentences specifically": 106503,
      "networks outperforms": 77685,
      "recognition generation": 96886,
      "englishfrench test": 37366,
      "different translations": 31516,
      "sentences paraphrase": 106435,
      "entropy diversity": 38163,
      "combining source": 18751,
      "makes data": 66767,
      "perspective computer": 87160,
      "sets large": 107682,
      "compiled automatically": 19716,
      "systems stateoftheart": 117159,
      "suffer performance": 115233,
      "based slu": 12051,
      "rnn sequencetosequence": 102931,
      "networks suitable": 77786,
      "test distributions": 120446,
      "comprising european": 20335,
      "cleaner data": 17713,
      "wikipedia overall": 132666,
      "similarity fluency": 109234,
      "lexicallyconstrained decoding": 63860,
      "rewriting tasks": 102700,
      "data fail": 25942,
      "fluency paraphrase": 44698,
      "capture nonlocal": 15349,
      "set values": 107632,
      "objectives promote": 81140,
      "baselines learn": 12422,
      "negative sentences": 76953,
      "fiction social": 43924,
      "texts dataset": 121491,
      "corpus diverse": 23762,
      "generation interesting": 47443,
      "identification existing": 51386,
      "adversarial data": 4108,
      "english plus": 37243,
      "robustness issues": 103103,
      "identification models": 51399,
      "significant robustness": 108862,
      "bert masked": 13161,
      "target models": 117675,
      "robustness issue": 103102,
      "issue unsupervised": 57030,
      "unsupervised paraphrase": 127689,
      "function involving": 46029,
      "editing method": 34545,
      "unsupervised does": 127625,
      "similarity parallel": 109280,
      "applications techniques": 7024,
      "graphbased technique": 48915,
      "paraphrasing model": 84848,
      "model semantically": 71968,
      "corpus improves": 23834,
      "introduce mechanism": 56454,
      "discriminator network": 32225,
      "model wide": 72315,
      "wide deep": 132493,
      "greater robustness": 49038,
      "noisy short": 80119,
      "description based": 29455,
      "current trend": 25381,
      "models reinforcement": 73902,
      "range text": 95610,
      "singlelabel classification": 109831,
      "building resources": 14882,
      "writing aid": 135086,
      "paraphrase ranking": 84832,
      "context used": 22312,
      "methodology build": 69250,
      "professional translation": 92080,
      "new versions": 79244,
      "improving bleu": 53074,
      "diverse automatic": 32796,
      "bleu reliance": 14318,
      "space valid": 111079,
      "outputs evaluated": 83166,
      "approach beats": 7387,
      "used sentencelevel": 128748,
      "use proposed": 128218,
      "constraints task": 21614,
      "limited syntactic": 64287,
      "accurately classifies": 2377,
      "understand models": 126760,
      "pair identical": 83435,
      "problem collect": 90964,
      "rank evaluate": 95624,
      "perform text": 86092,
      "data alignment": 25590,
      "study highlights": 114398,
      "highlights importance": 50283,
      "especially early": 38453,
      "diversity recent": 32892,
      "attempting generate": 9759,
      "present input": 89518,
      "better preserve": 13677,
      "additionally modify": 3616,
      "backtranslation approach": 11331,
      "context automatic": 22014,
      "identification requiring": 51431,
      "requires specialized": 100323,
      "syntactic level": 116428,
      "using bitext": 129417,
      "costly create": 24385,
      "tasks quantitatively": 119425,
      "representations suffer": 99914,
      "generating outputs": 47245,
      "decoding outperforms": 28118,
      "significantly narrowing": 108978,
      "narrowing gap": 76197,
      "architectures previous": 8838,
      "training pipeline": 123781,
      "datasets robust": 27690,
      "dataset augmentation": 26748,
      "result accurate": 101357,
      "problems despite": 91314,
      "scratch new": 104242,
      "technique named": 119801,
      "masked templates": 67306,
      "prevent model": 90364,
      "methods shows": 69747,
      "solving nlp": 110645,
      "generation adversarial": 47299,
      "embeddings concatenated": 35607,
      "dataset longer": 27003,
      "scientific field": 104006,
      "requires annotated": 100246,
      "diversity paper": 32889,
      "models finnish": 73255,
      "introduce fully": 56425,
      "model paraphrase database": 71708,
      "paraphrase database ppdb": 84809,
      "new manually annotated": 79001,
      "probabilistic contextfree grammars": 90803,
      "deep learning embeddings": 28270,
      "use statistical machine": 128297,
      "work explore deep": 134509,
      "models different datasets": 73068,
      "wieting et al": 132623,
      "prior work based": 90741,
      "method collect largescale": 68699,
      "address problem generating": 3735,
      "method based combination": 68664,
      "paraphrase generation given": 84815,
      "widely explored natural": 132545,
      "explored natural language": 41628,
      "sentiment analysis machine": 106605,
      "analysis machine translation": 5688,
      "reinforcement learning automatic": 97799,
      "generation specifically propose": 47629,
      "sequencetosequence learning model": 107163,
      "inverse reinforcement learning": 56703,
      "demonstrate proposed models": 28854,
      "paper makes contributions": 84042,
      "sources information used": 110900,
      "intrinsic evaluation tasks": 56356,
      "nlp applications recent": 79572,
      "adversarial example generation": 4116,
      "train neural encoderdecoder": 122978,
      "furthermore capable generating": 46151,
      "adversarial examples fool": 4119,
      "modeling compare performance": 72397,
      "conduct experiments benchmark": 20847,
      "model trained task": 72213,
      "achieve bleu score": 2437,
      "dialog systems stateoftheart": 30604,
      "suffer performance degradation": 115234,
      "model order improve": 71620,
      "experiments existing benchmark": 40944,
      "able produce multiple": 1675,
      "sentence rewriting tasks": 106055,
      "fiction social media": 43925,
      "neural models significantly": 78198,
      "paraphrase identification existing": 84818,
      "bert masked language": 13162,
      "unsupervised paraphrase generation": 127690,
      "optimization problem propose": 82204,
      "human evaluations approach": 50845,
      "language models leveraging": 59584,
      "models propose simple": 73821,
      "pretrain model largescale": 89977,
      "present data augmentation": 89427,
      "models reinforcement learning": 73903,
      "observe significant performance": 81210,
      "classification tasks including": 17445,
      "main contribution work": 66411,
      "semantic parser trained": 105135,
      "languages multiple domains": 60748,
      "using modern neural": 129893,
      "mt outputs evaluated": 74980,
      "neural paraphrasing model": 78625,
      "model given input": 71250,
      "using encoderdecoder model": 129626,
      "model perform poorly": 71720,
      "study highlights importance": 114399,
      "better preserve meaning": 13678,
      "source sentences contain": 110821,
      "datasets used train": 27779,
      "specialized domain knowledge": 111388,
      "human evaluation unsupervised": 50841,
      "models generate highquality": 73285,
      "original sentence pairs": 82543,
      "solving nlp tasks": 110646,
      "robustness nlp models": 103117,
      "model paraphrase database ppdb": 71709,
      "achieve stateoftheart results standard": 2568,
      "use statistical machine learning": 128298,
      "performance improvement stateoftheart methods": 86445,
      "widely explored natural language": 132546,
      "sentiment analysis machine translation": 106606,
      "important challenging task natural": 52118,
      "multiple sources information used": 75699,
      "create training data task": 24649,
      "train neural encoderdecoder model": 122979,
      "conduct experiments benchmark datasets": 20848,
      "stateoftheart neural network based": 112785,
      "unsupervised domain adaptation method": 127630,
      "generation experimental results proposed": 47401,
      "stateoftheart neural models significantly": 112783,
      "issues natural language processing": 57063,
      "bert masked language model": 13163,
      "language models propose simple": 59637,
      "text classification tasks including": 120777,
      "experimental results models achieve": 40655,
      "proposed approach outperforms competitive": 93205,
      "language model paper present": 59364,
      "important challenging task natural language": 52119,
      "generation experimental results proposed model": 47402,
      "extraordinary": 42586,
      "semanticallyrelated": 105387,
      "milne": 70130,
      "drt": 34193,
      "earlystage": 34399,
      "9205": 1396,
      "retrospect": 102476,
      "trex": 125759,
      "funding": 46136,
      "effective suitable": 34754,
      "collaborative development": 18357,
      "challenging emerging": 16246,
      "pose number": 88252,
      "systems noisy": 117026,
      "texts main": 121545,
      "main sources": 66464,
      "investigated improve": 56837,
      "approach open": 7756,
      "topical coherence": 122591,
      "topic paper": 122558,
      "knowledge entity": 57903,
      "disambiguation errors": 31964,
      "summarization semantic": 115563,
      "translation ambiguity": 124636,
      "entities contribute": 37761,
      "use effective": 128023,
      "token spans": 122268,
      "expensive training": 40437,
      "propagation perform": 92396,
      "task resolving": 118651,
      "conll dataset": 21087,
      "similarity entity": 109228,
      "mentions refer": 68466,
      "mentions context": 68448,
      "web forms": 132234,
      "report models": 99013,
      "web taking": 132262,
      "related concepts": 97846,
      "handling tasks": 49422,
      "structures various": 114117,
      "extraction embedding": 42318,
      "identifying mentions": 51605,
      "especially difficult": 38450,
      "additional text": 3583,
      "exploiting social": 41480,
      "incorporated neural": 53498,
      "reference distributed": 97512,
      "allows connect": 5134,
      "datadriven models": 26670,
      "symbolic models": 116316,
      "task test": 118783,
      "outperformed memory": 82782,
      "wikipedia sentences": 132675,
      "rank candidates": 95622,
      "framework advantages": 45417,
      "words query": 134151,
      "query second": 94987,
      "prior probability": 90725,
      "750 f1": 1253,
      "local neural": 64930,
      "representations key": 99708,
      "people entities": 85873,
      "pronouns entities": 92351,
      "fragments web": 45379,
      "graph algorithms": 48755,
      "languages presenting": 60808,
      "suggest automatic": 115303,
      "makes joint": 66790,
      "viability approach": 131638,
      "combined convolution": 18649,
      "2015 datasets": 588,
      "portuguese russian": 88201,
      "applications providing": 6994,
      "idioms languages": 51655,
      "disambiguation decisions": 31962,
      "decisions jointly": 27960,
      "accuracy popular": 2236,
      "training reranker": 123814,
      "music domain": 75964,
      "mixing context": 70414,
      "linking question": 64669,
      "data entity": 25899,
      "link entities": 64633,
      "final score": 44125,
      "modeling latent": 72458,
      "mentions named": 68460,
      "model induce": 71338,
      "scores standard": 104208,
      "recommending appropriate": 97099,
      "step approach": 113249,
      "domains computer": 33749,
      "application entity": 6850,
      "coreference model": 23351,
      "evaluation english": 39202,
      "2015 tac": 596,
      "document main": 33035,
      "potential mentions": 88574,
      "mentions learn": 68456,
      "learn contextual": 62007,
      "features empirically": 43475,
      "outperforms popular": 82931,
      "owing limited": 83390,
      "language extensive": 59038,
      "limited resource": 64272,
      "limitations zeroshot": 64198,
      "important disambiguation": 52147,
      "reference future": 97516,
      "encoding graph": 36697,
      "improving baseline": 53072,
      "mention source": 68430,
      "zeroshot scenario": 135521,
      "phonological representations": 87312,
      "usually fail": 130420,
      "mentions instead": 68455,
      "instead entire": 55659,
      "entire text": 37720,
      "ability conduct": 1498,
      "models calculate": 72871,
      "group entities": 49137,
      "mentions usually": 68471,
      "longterm influence": 65205,
      "decisions conduct": 27955,
      "types datasets": 126271,
      "dialogue humans": 30689,
      "task better": 117935,
      "architectural decisions": 8601,
      "lack guidance": 58715,
      "decisions specifically": 27965,
      "evidence related": 39665,
      "finally semantic": 44233,
      "decisions experimental": 27957,
      "particular knowledge": 85427,
      "assuming correct": 9657,
      "objectives furthermore": 81136,
      "holistic evaluation": 50600,
      "importance standardized": 52081,
      "techniques learning": 119920,
      "isa hierarchy": 56959,
      "use resources": 128247,
      "datadriven manner": 26667,
      "relations resources": 98249,
      "zeroresource crosslingual": 135484,
      "affects overall": 4246,
      "zeroshot entity": 135499,
      "mention context": 68418,
      "overcomes difficulties": 83302,
      "good candidate": 48466,
      "candidate mention": 15112,
      "trained joint": 123163,
      "assigning entity": 9555,
      "nonstandard noisy": 80295,
      "noisy long": 80112,
      "representations challenge": 99537,
      "end approach": 36794,
      "candidates represented": 15144,
      "topic level": 122537,
      "kgs based": 57664,
      "relation entity": 97975,
      "generation lowresource": 47464,
      "successful realm": 115170,
      "problems faced": 91323,
      "recall compared": 96334,
      "incorporating documentlevel": 53532,
      "techniques current": 119857,
      "focusing key": 44925,
      "step identifying": 113271,
      "indicates current": 53858,
      "effective zeroshot": 34783,
      "ranking present": 95681,
      "facilitate manual": 42780,
      "survey presents": 116183,
      "common themes": 18938,
      "including zeroshot": 53407,
      "meaning work": 67709,
      "techniques finally": 119889,
      "news comments": 79319,
      "facilitates quick": 42807,
      "topics popular": 122648,
      "popular opinions": 88117,
      "limited context": 64222,
      "training make": 123699,
      "domains collected": 33741,
      "shift language": 108177,
      "unknown training": 127363,
      "trained conll": 123097,
      "types corrections": 126267,
      "tested existing": 120570,
      "news titles": 79380,
      "context long": 22169,
      "effective concise": 34650,
      "called multiturn": 15060,
      "query generated": 94964,
      "framework sufficiently": 45703,
      "selection stage": 104829,
      "used state": 128780,
      "understand scientific": 126772,
      "scientific corpora": 103996,
      "expensive annotations": 40415,
      "deployed realworld": 29310,
      "retrieve rerank": 102453,
      "model retrieved": 71929,
      "candidates ranked": 15143,
      "time process": 122077,
      "mention span": 68431,
      "leveraging deep neural": 63671,
      "topic paper presents": 122559,
      "method specifically designed": 69161,
      "graph model learns": 48832,
      "networks key challenge": 77634,
      "recent success various": 96544,
      "structured learning framework": 114011,
      "symbolic models paper": 116317,
      "traditional neural network": 122852,
      "outperformed memory networks": 82783,
      "memory networks model": 68358,
      "sentences present simple": 106451,
      "data set consisting": 26403,
      "text fragments web": 120966,
      "results suggest automatic": 102228,
      "chinese paper introduce": 16799,
      "number different languages": 80869,
      "datasets proposed trained": 27647,
      "information past future": 54832,
      "features extracted external": 43507,
      "mentions named entities": 68461,
      "unlike previous approaches": 127440,
      "approaches recently developed": 8311,
      "models trained directly": 74200,
      "potential mentions learn": 88575,
      "owing limited availability": 83391,
      "target language additional": 117630,
      "available lowresource setting": 11039,
      "study recent neural": 114499,
      "reference future research": 97517,
      "future research end": 46301,
      "address problem investigate": 3738,
      "available datasets verify": 10977,
      "problem propose reinforcement": 91181,
      "model makes decisions": 71494,
      "datasets results model": 27687,
      "task models outperform": 118404,
      "models outperform state": 73683,
      "decisions experimental results": 27958,
      "indepth analysis various": 53796,
      "stateoftheart models code": 112729,
      "type information existing": 126206,
      "approach achieves performance": 7313,
      "including different ways": 53286,
      "lot research recent": 65326,
      "represent meaning work": 99121,
      "problem entity linking": 91032,
      "domains collected annotated": 33742,
      "existing approaches relying": 40062,
      "text paper propose": 121167,
      "used state art": 128781,
      "need expensive annotations": 76803,
      "benchmark datasets proposed trained": 12799,
      "used machine learning algorithm": 128619,
      "standard beam search algorithm": 112202,
      "publicly available datasets verify": 94302,
      "problem propose reinforcement learning": 91182,
      "datasets results model outperforms": 27688,
      "models outperform state art": 73684,
      "external knowledge paper propose": 42016,
      "pretrained language models improve": 90078,
      "lot research recent years": 65327,
      "framework achieves stateoftheart performance": 45411,
      "authorships": 10392,
      "pride": 90634,
      "attributing": 10208,
      "unethical": 127054,
      "gibberish": 47960,
      "aptness": 8482,
      "solicit": 110538,
      "stylebased": 114614,
      "unfeasible": 127069,
      "aside": 9293,
      "measurable": 67774,
      "plato": 87675,
      "latch": 61559,
      "sonority": 110663,
      "politically": 88006,
      "texts taken": 121630,
      "scheme open": 103933,
      "scheme provide": 103936,
      "style accuracy": 114565,
      "diversity english": 32873,
      "similarities native": 109182,
      "languages comparing": 60456,
      "enable studies": 36366,
      "studies dataset": 114201,
      "extracted wikipedia": 42195,
      "distribution work": 32691,
      "explicit forms": 41323,
      "studies assessed": 114181,
      "books using": 14425,
      "patterns useful": 85792,
      "authors use": 10388,
      "identify parts": 51532,
      "application technique": 6889,
      "symmetric patterns": 116327,
      "increase total": 53618,
      "number published": 80948,
      "discriminate real": 32193,
      "analysis features": 5621,
      "techniques allowed": 119832,
      "texts case": 121468,
      "potential features": 88555,
      "structural organization": 113784,
      "used studies": 128787,
      "styles using": 114623,
      "style author": 114569,
      "ones really": 81706,
      "used differently": 128492,
      "texts order": 121565,
      "prediction domain": 89052,
      "workers amazon": 134887,
      "finegrained category": 44340,
      "examples natural": 39856,
      "applied study": 7121,
      "distinguish writing": 32583,
      "correctly classified": 24173,
      "chance baseline": 16351,
      "taken findings": 117509,
      "authorship verification": 10391,
      "pan clef": 83689,
      "identified using": 51464,
      "applications study": 7021,
      "texts authors": 121459,
      "introduced allows": 56571,
      "solely word": 110537,
      "counts related": 24453,
      "measurements models": 67847,
      "networks considering": 77544,
      "using concepts": 129468,
      "authors based": 10378,
      "style text": 114598,
      "stateoftheart survey": 112984,
      "closely followed": 17856,
      "language consider": 58910,
      "literary novels": 64736,
      "style present": 114588,
      "addition combining": 3406,
      "covering various": 24551,
      "practice paper": 88734,
      "achieves 74": 2729,
      "styles paper": 114620,
      "languages measured": 60723,
      "accuracy 87": 2077,
      "individual characteristics": 53900,
      "recent applications": 96427,
      "led increased": 63279,
      "text far": 120948,
      "relatively harder": 98392,
      "leveraging discourse": 63673,
      "explore techniques": 41595,
      "nontrivial performance": 80313,
      "usefulness applying": 128955,
      "models act": 72708,
      "layer uses": 61752,
      "achieved precision": 2668,
      "author using": 10370,
      "multidimensional vector": 75071,
      "writing samples": 135096,
      "available second": 11099,
      "genre andor": 47829,
      "length parallel": 63372,
      "reviews machine": 102614,
      "identify author": 51479,
      "research aim": 100400,
      "classifier support": 17583,
      "expected word": 40403,
      "identification document": 51383,
      "evaluation purposes": 39345,
      "distinctive use": 32562,
      "style representations": 114592,
      "fictional characters": 43928,
      "literature analysis": 64744,
      "argue important": 8922,
      "aspects unified": 9415,
      "language networks": 59711,
      "according style": 2005,
      "models texts": 74187,
      "information case": 54409,
      "distributional term": 32739,
      "media author": 68080,
      "analysis regarding": 5800,
      "representations supported": 99917,
      "problem compare": 90968,
      "classic approaches": 17082,
      "results providing": 102092,
      "seminal work": 105585,
      "focus work": 44843,
      "contrast lexical": 22696,
      "annotated stylistic": 6240,
      "multiple styles": 75711,
      "nonparallel text": 80267,
      "transfer dataset": 124043,
      "verification task": 131572,
      "grammatical inconsistencies": 48710,
      "traditionally relied": 122891,
      "varied social": 130944,
      "approaches built": 8092,
      "network shows": 77433,
      "traditional linguistic": 122825,
      "robust approaches": 103020,
      "set existing": 107437,
      "examine application": 39735,
      "characters evaluate": 16606,
      "use experiments": 128037,
      "perspective work": 87172,
      "framework represent": 45673,
      "computational problem": 20420,
      "ml model": 70434,
      "develop pipeline": 30225,
      "thousands reviews": 121928,
      "detection rate": 30032,
      "people read": 85887,
      "domain style": 33663,
      "decades research": 27909,
      "research remains": 100614,
      "researched years": 100672,
      "approaches trying": 8384,
      "trying solve": 125904,
      "based writing": 12172,
      "cases performs": 15663,
      "texts use": 121642,
      "use types": 128346,
      "general categories": 46639,
      "properties text": 92485,
      "terms variability": 120404,
      "explored recent": 41634,
      "years shown": 135301,
      "learned contextual": 62208,
      "learning structural": 63062,
      "quality ratings": 94769,
      "performance provide": 86636,
      "psycholinguistic categories": 94202,
      "experiments texts": 41182,
      "metrics time": 70015,
      "significant perform": 108831,
      "style level": 114582,
      "modern approaches": 74381,
      "extensive experimental studies": 41889,
      "set novel features": 107513,
      "increase total number": 53619,
      "workers amazon mechanical": 134888,
      "examples natural language": 39857,
      "words play important": 134124,
      "taken findings suggest": 117510,
      "using random forest": 130079,
      "statistical methods based": 113127,
      "based solely word": 12058,
      "gold standard human": 48442,
      "writing style present": 135099,
      "simple linear classifier": 109457,
      "neural models generate": 78177,
      "natural language techniques": 76543,
      "reviews machine learning": 102615,
      "experiments methods achieve": 41008,
      "classifier support vector": 17584,
      "tree random forest": 125609,
      "probabilistic language models": 90815,
      "aspects unified framework": 9416,
      "social media author": 110350,
      "media author profiling": 68081,
      "introduce novel framework": 56512,
      "task social media": 118723,
      "nonparallel text style": 80268,
      "style transfer dataset": 114603,
      "proposed method able": 93335,
      "test time paper": 120547,
      "use word embedding": 128361,
      "based writing style": 12173,
      "structure sentences document": 113954,
      "recent years shown": 96637,
      "improves generalization different": 52986,
      "downstream tasks domains": 34035,
      "analysis tasks results": 5888,
      "prediction model using": 89081,
      "yields best performance": 135398,
      "workers amazon mechanical turk": 134889,
      "examples natural language processing": 39858,
      "words play important role": 134125,
      "applying machine learning techniques": 7253,
      "neural probabilistic language models": 78636,
      "social media author profiling": 110351,
      "nonparallel text style transfer": 80269,
      "slotindependent": 110082,
      "twentyfive": 126078,
      "copymechanism": 23300,
      "598": 1138,
      "responded": 101182,
      "conditionals": 20796,
      "308": 916,
      "stimulated": 113346,
      "transcriptions training": 124009,
      "unique structure": 127195,
      "standard dialogue": 112231,
      "trivial preprocessing": 125830,
      "preprocessing performance": 89329,
      "systems architecture": 116736,
      "evaluation used": 39432,
      "frames used": 45401,
      "information presentation": 54862,
      "recently resources": 96757,
      "tracking task": 122762,
      "multiple frames": 75569,
      "different goals": 31166,
      "list frames": 64690,
      "types slot": 126359,
      "associated utterance": 9614,
      "act slot": 3070,
      "type slot": 126226,
      "slot value": 110071,
      "represent rich": 99133,
      "relies automatic": 98643,
      "systems slu": 117148,
      "multiturn information": 75919,
      "information separate": 54969,
      "separate phases": 106859,
      "avoid error": 11233,
      "propagation effectively": 92391,
      "contexts prior": 22419,
      "conversational information": 23008,
      "access common": 1903,
      "contain single": 21760,
      "values slot": 130805,
      "observations motivate": 81179,
      "current representation": 25329,
      "interaction state": 55969,
      "rely deep": 98683,
      "requires labeled": 100282,
      "makes building": 66762,
      "multidomain deep": 75088,
      "framework state": 45697,
      "candidate sets": 15125,
      "architecture transfer": 8764,
      "approach handling": 7596,
      "handling unknown": 49423,
      "generally assume": 46848,
      "assume predefined": 9648,
      "slu module": 110120,
      "multi domain": 75026,
      "shared domains": 107975,
      "collected multidomain": 18426,
      "automatically provide": 10819,
      "generation end": 47385,
      "parameters slots": 84787,
      "approach overcomes": 7784,
      "approaches multitask": 8242,
      "joint language": 57283,
      "encoding user": 36733,
      "slot annotations": 110033,
      "unbounded set": 126610,
      "values seen": 130801,
      "deployment production": 29317,
      "al uses": 4657,
      "called slots": 15065,
      "reduces latency": 97388,
      "tasks successful": 119533,
      "conversational state": 23034,
      "multidomain scenario": 75096,
      "exist multiple": 40027,
      "capture salient": 15377,
      "salient contextual": 103547,
      "slu dialogue": 110112,
      "stateoftheart contextual": 112618,
      "aims estimate": 4525,
      "refer previous": 97498,
      "necessitating use": 76767,
      "determine relevance": 30132,
      "slot carryover": 110034,
      "relevant current": 98535,
      "networks incorporate": 77625,
      "networks uses": 77810,
      "model slot": 72059,
      "longer distance": 65156,
      "values focus": 130794,
      "usually need": 130439,
      "severe challenge": 107879,
      "effectively decompose": 34803,
      "identify users": 51574,
      "increases paper": 53650,
      "models scalable": 73973,
      "defined domain": 28499,
      "predict dialogue": 88881,
      "maintaining highlevel": 66519,
      "resource domains": 100838,
      "works dialogue": 134939,
      "semisupervised bootstrapping": 105594,
      "detailed annotations": 29761,
      "investigate semisupervised": 56813,
      "dialogue analysis": 30639,
      "slots user": 110092,
      "definition target": 28528,
      "including intent": 53304,
      "intent prediction": 55906,
      "provide promising": 93898,
      "paradigm zeroshot": 84556,
      "slot prediction": 110066,
      "high margin": 50088,
      "definition dataset": 28520,
      "unseen ones": 127534,
      "primary reason": 90654,
      "randomly generated": 95528,
      "values unseen": 130807,
      "called explicit": 15058,
      "states experimental": 113041,
      "based multiwoz": 11874,
      "tasks understanding": 119581,
      "problem allows": 90930,
      "robustness noisy": 103120,
      "main metrics": 66442,
      "learning multidomain": 62809,
      "augmentation synthesized": 10308,
      "improve zeroshot": 52588,
      "learning state": 63053,
      "task considerably": 118006,
      "use value": 128352,
      "argue strategy": 8929,
      "words unknown": 134285,
      "historical information": 50525,
      "context lack": 22156,
      "powerful mechanism": 88670,
      "different slots": 31433,
      "training experimental": 123626,
      "joint accuracy": 57257,
      "respectively achieves": 101119,
      "performance considerable": 86251,
      "propose zeroshot": 93162,
      "task key": 118327,
      "explicitly expressed": 41370,
      "utterance generate": 130598,
      "state dependencies": 112489,
      "probabilistic graphical": 90809,
      "historical user": 50539,
      "demonstrated superiority": 28933,
      "policy improve": 87956,
      "compact representation": 19111,
      "estimate dialogue": 38630,
      "approaches define": 8118,
      "language metrics": 59281,
      "values experiments": 130792,
      "drawbacks paper": 34123,
      "tagging slot": 117447,
      "value position": 130785,
      "baselines greatly": 12406,
      "consistent dialogue": 21374,
      "used techniques": 128806,
      "reflect underlying": 97613,
      "severe data": 107881,
      "sparsity natural": 111256,
      "nlp corpora": 79594,
      "learning augmenting": 62375,
      "augmentation results": 10299,
      "use entire": 128030,
      "result loss": 101382,
      "124 points": 271,
      "stimulated research": 113347,
      "improved versions": 52655,
      "significantly outperform traditional": 108995,
      "rule based machine": 103349,
      "features paper presents": 43646,
      "baseline model task": 12260,
      "avoid error propagation": 11234,
      "error propagation effectively": 38316,
      "contexts prior work": 22420,
      "endtoend manner experiments": 36920,
      "dataset outperforms stateoftheart": 27081,
      "key component taskoriented": 57555,
      "requires labeled data": 100283,
      "multidomain deep learning": 75089,
      "understanding slu module": 126964,
      "dialogue generation task": 30683,
      "approach multitask learning": 7722,
      "performance reducing number": 86664,
      "slot values seen": 110073,
      "values seen training": 130802,
      "using recurrent networks": 130088,
      "model reduces latency": 71876,
      "understanding slu dialogue": 126959,
      "benchmark dataset used": 12763,
      "higher f1 scores": 50181,
      "demonstrates proposed methods": 28957,
      "based pointer networks": 11931,
      "based transformer networks": 12132,
      "proposed models able": 93490,
      "encoder representations transformer": 36567,
      "suggest promising direction": 115329,
      "zeroshot learning language": 135510,
      "state tracking task": 112531,
      "paper gives overview": 83969,
      "dialogue technology challenge": 30804,
      "occur training set": 81505,
      "states experimental results": 113042,
      "stateoftheart methods especially": 112708,
      "transfer learning multidomain": 124108,
      "data augmentation synthesized": 25650,
      "augmentation synthesized data": 10309,
      "improve zeroshot learning": 52589,
      "learning state art": 63054,
      "existing methods usually": 40199,
      "training experimental results": 123627,
      "stateoftheart performance considerable": 112815,
      "performance considerable improvements": 86252,
      "models effectively capture": 73118,
      "data annotation expensive": 25604,
      "sequence generation problem": 106951,
      "framework make use": 45608,
      "sequence generation model": 106949,
      "propose novel scalable": 92960,
      "conversational semantic parsing": 23029,
      "sparsity natural language": 111257,
      "nlp tasks benefit": 79709,
      "shows model performs": 108602,
      "validation set test": 130744,
      "machine learning based approach": 65781,
      "task dialog state tracking": 118090,
      "key component taskoriented dialogue": 57556,
      "component taskoriented dialogue systems": 20007,
      "language understanding slu module": 60273,
      "slot values seen training": 110074,
      "language understanding slu dialogue": 60269,
      "bidirectional encoder representations transformer": 13915,
      "achieve stateoftheart results datasets": 2567,
      "data augmentation synthesized data": 25651,
      "achieving state art performance": 2982,
      "stateoftheart performance considerable improvements": 112816,
      "paper propose novel scalable": 84307,
      "model performs better stateoftheart": 71737,
      "sparsity natural language processing": 111258,
      "transfer learning multitask learning": 124112,
      "analysis shows model performs": 5852,
      "validation set test set": 130745,
      "key component taskoriented dialogue systems": 57557,
      "spoken language understanding slu module": 112004,
      "spoken language understanding slu dialogue": 112001,
      "proposed model performs better stateoftheart": 93473,
      "rocktaschel": 103144,
      "subtreelevel": 115009,
      "languageand": 60356,
      "williams": 132705,
      "damaging": 25528,
      "fallible": 43030,
      "alice": 4896,
      "disjunction": 32367,
      "logloss": 65047,
      "00": 0,
      "bolster": 14405,
      "inflating": 54277,
      "available collection": 10961,
      "classifiers outperform": 17621,
      "benchmarks time": 12937,
      "models impose": 73363,
      "interpretable paper": 56248,
      "inference results": 54211,
      "analysis deriving": 5569,
      "carefully designing": 15522,
      "chain lstms": 15977,
      "improvement particularly": 52738,
      "architecture problem": 8730,
      "model mapped": 71502,
      "techniques increasingly": 119911,
      "discrete models": 32170,
      "particular recently": 85446,
      "rocktaschel et": 103145,
      "recursive composition": 97284,
      "version natural": 131613,
      "infer entailment": 54098,
      "accuracy understanding": 2310,
      "determine relationship": 30131,
      "difficulty using": 31710,
      "easy examples": 34477,
      "possible evaluate": 88400,
      "shared information": 107984,
      "evaluated neural": 38995,
      "williams et": 132706,
      "task indomain": 118292,
      "requires agent": 100244,
      "language premise": 59819,
      "nli multinli": 79548,
      "knowledge modeling": 58065,
      "needed perform": 76877,
      "knowledge build": 57820,
      "build nli": 14794,
      "version model": 131612,
      "encoding inference": 36698,
      "predictions finally": 89172,
      "solution experiments": 110554,
      "baseline number": 12277,
      "access context": 1904,
      "nli systems": 79551,
      "set shows": 107580,
      "sentences differ": 106276,
      "limited generalization": 64237,
      "nli proposed": 79549,
      "nli achieving": 79534,
      "decisions evaluation": 27956,
      "models nli": 73643,
      "taken corpus": 117507,
      "generalize task": 46819,
      "examples semantic": 39875,
      "fallible heuristics": 43031,
      "adopt heuristic": 3892,
      "set strong": 107591,
      "order machine": 82355,
      "garner widespread": 46492,
      "widespread public": 132618,
      "explanations decisions": 41292,
      "process output": 91543,
      "transferring outofdomain": 124243,
      "range research": 95598,
      "traditional bilstm": 122802,
      "different matching": 31247,
      "inference different": 54134,
      "interaction different": 55943,
      "memory component": 68301,
      "wrong reasons": 135163,
      "frequent example": 45860,
      "determining sentence": 30154,
      "contains examples": 21821,
      "motivate measure": 74834,
      "field experiments": 43951,
      "effective wide": 34780,
      "help dataset": 49707,
      "crowdsourced datasets": 25109,
      "despite efforts": 29687,
      "augmentation improved": 10280,
      "00 10": 1,
      "finally illustrate": 44194,
      "capture interaction": 15319,
      "way investigate": 132102,
      "stateoftheart nli": 112788,
      "transfer datasets": 124044,
      "methods known": 69570,
      "useful general": 128888,
      "negation words": 76919,
      "learning underlying": 63128,
      "underlying task": 126706,
      "examples experiment": 39832,
      "transfer textual": 124197,
      "datasets allows": 27306,
      "individual datasets": 53904,
      "representations abstract": 99485,
      "generalization improves": 46776,
      "examples significantly": 39876,
      "tasks notably": 119344,
      "remains understudied": 98828,
      "types use": 126382,
      "evidence pragmatic": 39658,
      "encourages models": 36772,
      "captured neural": 15434,
      "construction used": 21700,
      "network effectively": 77233,
      "learns perform": 63225,
      "incorporates explicit": 53505,
      "11 tasks": 228,
      "methods challenge": 69366,
      "structural evaluation": 113767,
      "evaluations models": 39476,
      "examples containing": 39817,
      "biases dataset": 13869,
      "outofdistribution test": 82644,
      "enables practitioners": 36398,
      "explore linguistic": 41557,
      "current pretrained": 25324,
      "performance contrast": 86257,
      "domain divergence": 33511,
      "measures common": 67857,
      "performing competitively": 86947,
      "minimal human": 70162,
      "datasets provided": 27650,
      "patterns results": 85777,
      "types current": 126268,
      "examples data": 39820,
      "learning research area": 62964,
      "average pooling used": 11205,
      "model outperformed existing": 71631,
      "inference results demonstrate": 54212,
      "results demonstrate efficacy": 101680,
      "perform extensive qualitative": 86005,
      "extensive qualitative analysis": 41950,
      "learning propose neural": 62922,
      "particular recently proposed": 85447,
      "rocktaschel et al": 103146,
      "development evaluation machine": 30387,
      "williams et al": 132707,
      "test set surprisingly": 120521,
      "variety linguistic phenomena": 130990,
      "representations sentence meaning": 99877,
      "inference task paper": 54232,
      "model shared task": 72002,
      "task requires agent": 118640,
      "natural language premise": 76384,
      "extracting semantic features": 42237,
      "inference nli data": 54176,
      "models external knowledge": 73219,
      "address natural language": 3718,
      "able significantly outperform": 1688,
      "world knowledge new": 135034,
      "limited generalization ability": 64238,
      "language understanding existing": 60222,
      "challenging linguistic phenomena": 16272,
      "current neural network": 25312,
      "using large pretrained": 129788,
      "garner widespread public": 46493,
      "research directions using": 100475,
      "attention mechanism proven": 9912,
      "network natural language": 77369,
      "right wrong reasons": 102838,
      "motivate measure progress": 74835,
      "measure progress area": 67814,
      "seen surge research": 104544,
      "effective wide range": 34781,
      "data augmentation improved": 25636,
      "datasets models perform": 27579,
      "learning underlying task": 63129,
      "stateoftheart models language": 112735,
      "order improve generalization": 82340,
      "considerably improves generalization": 21273,
      "requiring external knowledge": 100339,
      "evidence pragmatic reasoning": 39659,
      "encourages models learn": 36773,
      "crucial understanding natural": 25178,
      "neural models generalize": 78176,
      "setting natural language": 107760,
      "outofdistribution test sets": 82645,
      "model new domains": 71580,
      "domain adaptation scenarios": 33450,
      "semantic representations key": 105237,
      "key component natural": 57553,
      "provide finegrained analysis": 93829,
      "success deep neural": 115070,
      "train neural network based": 122986,
      "representation learning propose neural": 99311,
      "proposed model achieves better": 93426,
      "model achieves better accuracy": 70559,
      "language inference task paper": 59200,
      "language inference nli data": 59178,
      "present novel deep learning": 89605,
      "natural language understanding existing": 76559,
      "current neural network models": 25313,
      "motivate measure progress area": 74836,
      "largescale natural language inference": 61483,
      "crucial understanding natural language": 25179,
      "models trained large text": 74211,
      "key component natural language": 57554,
      "natural language understanding evaluation": 76558,
      "despite recent success deep": 29724,
      "recent success deep neural": 96532,
      "success deep neural networks": 115071,
      "natural language inference task paper": 76355,
      "natural language inference nli data": 76335,
      "models trained large text corpora": 74212,
      "models natural language understanding nlu": 73615,
      "despite recent success deep neural": 29725,
      "recent success deep neural networks": 96533,
      "success deep neural networks natural": 115072,
      "227": 783,
      "domaindependency": 33710,
      "appreciably": 7283,
      "multiclassification": 75058,
      "deaf": 27853,
      "0839": 85,
      "rgb": 102706,
      "roleplay": 103227,
      "codemix": 18165,
      "minimising": 70191,
      "henceforth": 49836,
      "theater": 121776,
      "improv": 52328,
      "cdac": 15869,
      "truly natural": 125863,
      "based 1000": 11496,
      "performance performance": 86606,
      "decoding discourse": 28085,
      "spontaneous dialogues": 112032,
      "dialogue language": 30700,
      "considered key": 21294,
      "egyptian dialect": 35212,
      "understanding spontaneous": 126975,
      "gained considerable": 46360,
      "resources systems": 101051,
      "need designing": 76792,
      "surveys different": 116212,
      "standard obtained": 112288,
      "way able": 132052,
      "work contribute": 134438,
      "act da": 3063,
      "approach users": 8000,
      "messages using": 68521,
      "scores 70": 104149,
      "describes hierarchical": 29407,
      "deep hierarchical": 28238,
      "test error": 120452,
      "obtained additional": 81341,
      "obtain data": 81273,
      "utterances task": 130668,
      "emission probabilities": 36077,
      "sequence inputs": 106963,
      "interactions speakers": 56002,
      "advantageous performance": 4066,
      "information reveal": 54954,
      "nature dataset": 76653,
      "understanding conversations": 126820,
      "applied fields": 7076,
      "identification vietnamese": 51458,
      "layer bilstmcrf": 61702,
      "facebook messages": 42738,
      "account previous": 2034,
      "training incorporate": 123659,
      "incorporate hierarchical": 53473,
      "modeling extend": 72431,
      "contextual utterances": 22512,
      "agreement score": 4406,
      "used participants": 128682,
      "believe dataset": 12697,
      "number differences": 80867,
      "wordlevel approaches": 133728,
      "tokenization levels": 122277,
      "iso standard": 56971,
      "testing outofdomain": 120599,
      "study collect": 114335,
      "theory called": 121814,
      "predict annotation": 88872,
      "task explored": 118187,
      "representations exploring": 99649,
      "considering different": 21322,
      "level addition": 63417,
      "functional level": 46062,
      "focus structure": 44825,
      "information considering": 54440,
      "results surpass": 102243,
      "surpass previous": 116093,
      "degradation domain": 28550,
      "shift paper": 108178,
      "acoustic textual": 3025,
      "adaptation da": 3212,
      "classification recognising": 17371,
      "classification adversarial": 17106,
      "examples correctly": 39818,
      "cast problem": 15678,
      "networks hierarchical": 77619,
      "baselines intent": 12416,
      "systems intent": 116957,
      "systems years": 117243,
      "users intents": 129136,
      "queries different": 94916,
      "combination vector": 18593,
      "latest investigations": 61656,
      "acousticprosodic cues": 3028,
      "importance prediction": 52071,
      "prediction spoken": 89127,
      "cues spoken": 25219,
      "hard hearing": 49461,
      "treated task": 125562,
      "instead predicting": 55673,
      "explore automated": 41522,
      "number endtoend": 80879,
      "poses problems": 88273,
      "problem posed": 91165,
      "prediction essential": 89057,
      "essential language": 38554,
      "ability paper": 1556,
      "annotations collected": 6415,
      "scheme leverage": 103930,
      "classifiers rich": 17632,
      "comparison model": 19556,
      "accurately predicting": 2388,
      "tag dependencies": 117334,
      "seq2seq approaches": 106885,
      "encoder novel": 36551,
      "complex costly": 19805,
      "propose context": 92597,
      "context encoding": 22077,
      "largescale general": 61439,
      "largescale multiturn": 61479,
      "overfitting improve": 83319,
      "role recognition": 103215,
      "attention hierarchical": 9854,
      "structure dialogue": 113834,
      "utterances based": 130626,
      "length context": 63357,
      "grounding process": 49123,
      "trained build": 123084,
      "dataset finetune": 26937,
      "exceptionally good": 39931,
      "consistent languages": 21392,
      "embeddings bring": 35581,
      "conclude possible": 20714,
      "filling intent": 44044,
      "models rapidly": 73854,
      "benefit tasks": 12991,
      "discuss current": 32244,
      "state research": 112513,
      "require attention": 100112,
      "predefined structure": 88836,
      "dataset japanese": 26989,
      "japanese dataset": 57186,
      "neural networks modeling": 78515,
      "models trained evaluated": 74203,
      "automatic speech recognizer": 10670,
      "language model improved": 59332,
      "information speech signal": 55002,
      "simple greedy decoding": 109442,
      "art performance dialogue": 9079,
      "dialogue language understanding": 30701,
      "gained considerable attention": 46361,
      "resources systems built": 101052,
      "systems built far": 116765,
      "need designing systems": 76793,
      "designing systems languages": 29651,
      "systems languages increasing": 116969,
      "paper surveys different": 84459,
      "classification approach based": 17119,
      "dialogue act da": 30630,
      "f1 scores 70": 42693,
      "text paper describes": 121163,
      "paper describes hierarchical": 83824,
      "inference present results": 54197,
      "dialog act classification": 30545,
      "learning approach using": 62355,
      "lstm crf layer": 65613,
      "crf layer bilstmcrf": 24751,
      "learning approaches study": 62364,
      "treated sequence labeling": 125560,
      "work build hierarchical": 134404,
      "field crf layer": 43943,
      "takes account previous": 117523,
      "endtoend training incorporate": 36988,
      "performance degradation domain": 86280,
      "domain shift paper": 33638,
      "domain adaptation da": 33431,
      "taskspecific attention mechanism": 119620,
      "auxiliary task model": 10889,
      "dynamic memory network": 34316,
      "adversarial examples correctly": 4118,
      "presents latest investigations": 89861,
      "performs competitively stateoftheart": 86996,
      "propose learning approach": 92746,
      "traditional methods tend": 122840,
      "dependencies work introduce": 29128,
      "transfer learning contextaware": 124088,
      "work shown incorporating": 134801,
      "different transfer learning": 31511,
      "approaches transfer learning": 8381,
      "largescale multiturn dialogue": 61480,
      "specifically propose use": 111586,
      "reduce overfitting improve": 97348,
      "improve performance results": 52482,
      "achieves promising performance": 2837,
      "transfer learning adapt": 124075,
      "adapt models trained": 3182,
      "language processing pipelines": 59931,
      "work new deep": 134654,
      "new deep neural": 78878,
      "slot filling intent": 110047,
      "filling intent classification": 44045,
      "various approaches proposed": 131041,
      "mutual benefit tasks": 75973,
      "current state research": 25346,
      "inference time experiments": 54240,
      "state art performance dialogue": 112466,
      "resources systems built far": 101053,
      "need designing systems languages": 76794,
      "designing systems languages increasing": 29652,
      "lstm crf layer bilstmcrf": 65614,
      "traditional machine learning approaches": 122830,
      "machine learning approaches study": 65776,
      "random field crf layer": 95479,
      "method achieves better performance": 68594,
      "spoken language understanding systems": 112008,
      "paper presents latest investigations": 84176,
      "address problems propose novel": 3753,
      "natural language processing pipelines": 76472,
      "new deep neural network": 78879,
      "slot filling intent classification": 110048,
      "language understanding dialogue systems": 60218,
      "paper explore different ways": 83916,
      "need designing systems languages increasing": 76795,
      "conditional random field crf layer": 20779,
      "voluntary": 131945,
      "humancrafted": 51009,
      "wnli": 132878,
      "aflite": 4265,
      "940": 1413,
      "dpr": 34077,
      "winogender": 132725,
      "cracking": 24586,
      "rightarrow": 102839,
      "1961": 448,
      "157": 368,
      "swag": 116251,
      "correspondingly": 24317,
      "interpolates": 56202,
      "creators": 24737,
      "perturbing": 87200,
      "incapability": 53208,
      "bird": 14226,
      "fish": 44573,
      "873": 1340,
      "mad": 66367,
      "text nontrivial": 121148,
      "people rarely": 85885,
      "influence way": 54318,
      "approach infer": 7637,
      "actions applied": 3088,
      "semantic plausibility": 105179,
      "events present": 39594,
      "models sophisticated": 74069,
      "remaining errors": 98783,
      "respect training": 101109,
      "art predicting": 9084,
      "inferential knowledge": 54258,
      "humancrafted features": 51010,
      "pronoun disambiguation": 92337,
      "sentence coreference": 105812,
      "benchmark commonsense": 12740,
      "everyday situations": 39631,
      "understanding commonsense": 126814,
      "strongly improves": 113746,
      "problem dataset": 90995,
      "leveraging language": 63685,
      "human explanations": 50857,
      "explanations including": 41294,
      "transfer outofdomain": 124170,
      "resolution problems": 100781,
      "spurious biases": 112062,
      "datasets lead": 27543,
      "dataset 44k": 26714,
      "systematic bias": 116661,
      "data allowed": 25593,
      "component models": 19987,
      "knowledge physical": 58100,
      "suggests neural": 115379,
      "reasoning rules": 96306,
      "kg reasoning": 57661,
      "effectively utilizes": 34862,
      "mechanism predicting": 68027,
      "brings improvement": 14648,
      "integrating contextual": 55789,
      "official leaderboard": 81601,
      "task score": 118667,
      "tasks increased": 119195,
      "information domainspecific": 54509,
      "type paper": 126218,
      "choice knowledge": 16888,
      "ability exploit": 1519,
      "types common": 126260,
      "understanding narratives": 126894,
      "narratives requires": 76185,
      "knowledge structures": 58194,
      "models vanilla": 74316,
      "various knowledge": 131110,
      "focused specifically": 44876,
      "injecting explicit": 55263,
      "bases improve": 12501,
      "underexplored paper": 126663,
      "fuses information": 46228,
      "amounts linguistic": 5348,
      "specialized architectures": 111384,
      "multitask setup": 75892,
      "problem datasets": 90996,
      "general recipe": 46704,
      "additional background": 3492,
      "interesting questions": 56087,
      "peak performance": 85833,
      "examples expensive": 39831,
      "aims achieve": 4496,
      "showing greater": 108406,
      "findings encourage": 44292,
      "knowledgeaware language": 58246,
      "task team": 118780,
      "challenge uses": 16116,
      "classifiers propose": 17628,
      "efficiently accurately": 35137,
      "commonsense understanding": 19011,
      "inference capability": 54120,
      "prohibit model": 92197,
      "accuracy subtask": 2292,
      "cues bert": 25209,
      "tasks importance": 119167,
      "cues model": 25214,
      "larger goal": 61367,
      "concepts reasoning": 20637,
      "capability generating": 15181,
      "explanations work": 41301,
      "multihop paths": 75139,
      "design bridge": 29523,
      "evaluations performance": 39479,
      "social events": 110338,
      "require different": 100128,
      "task validate": 118837,
      "symbolic neural": 116318,
      "graphs recent": 48970,
      "years brought": 135243,
      "brought renewed": 14715,
      "representation reasoning": 99394,
      "models tackling": 74154,
      "work posit": 134670,
      "reasoning unified": 96319,
      "underlying information": 126680,
      "generation use": 47689,
      "form dynamic": 45084,
      "couple language": 24455,
      "strategies enhancing": 113459,
      "relations largescale": 98199,
      "phrases identifying": 87430,
      "studies lms": 114250,
      "question train": 95229,
      "applying novel": 7265,
      "experiments 200": 40754,
      "experiments encompassing": 40929,
      "findings transfer": 44315,
      "inference requires": 54209,
      "humans reason": 51108,
      "contradictory statements": 22672,
      "implications paper": 51994,
      "providing novel": 94126,
      "law school": 61686,
      "significantly parameters": 109042,
      "simple models based": 109473,
      "models based distributional": 72808,
      "error analysis shows": 38286,
      "state art predicting": 112468,
      "incorporate hierarchical structure": 53474,
      "humancrafted features knowledge": 51011,
      "features knowledge bases": 43580,
      "benchmark commonsense reasoning": 12741,
      "models based pretrained": 72812,
      "compared human performance": 19377,
      "human performance 20": 50928,
      "language understanding commonsense": 60212,
      "language models automatically": 59506,
      "empirical results indicate": 36185,
      "training data allowed": 123425,
      "reasoning paper proposes": 96285,
      "language models semantic": 59656,
      "semantic similarity models": 105287,
      "commonsense reasoning recent": 19007,
      "language models demonstrated": 59536,
      "framework effectively utilizes": 45513,
      "incorporating commonsense knowledge": 53520,
      "understanding narratives requires": 126895,
      "models large pretrained": 73463,
      "knowledge bases improve": 57805,
      "amounts data training": 5334,
      "math word problem": 67452,
      "relevant knowledge propose": 98563,
      "consistently outperforms existing": 21432,
      "does make sense": 33367,
      "generative model language": 47738,
      "commonsense reasoning order": 19006,
      "results good performance": 101816,
      "pretrained models rely": 90152,
      "knowledge graphs recent": 57992,
      "recent years brought": 96602,
      "pretrained encoderdecoder model": 90032,
      "text generation challenging": 120981,
      "commonsense knowledge different": 18994,
      "language model named": 59352,
      "propose new multitask": 92872,
      "datasets pretrained language": 27627,
      "present comprehensive study": 89414,
      "previous state art predicting": 90467,
      "humancrafted features knowledge bases": 51012,
      "learning propose neural network": 62923,
      "improvement previous stateoftheart approaches": 52749,
      "natural language understanding commonsense": 76555,
      "language models semantic similarity": 59657,
      "introduce simple effective method": 56538,
      "stateoftheart deep learning architectures": 112632,
      "propose new evaluation framework": 92849,
      "results method significantly improves": 101927,
      "text generation challenging task": 120982,
      "pretrained language models demonstrated": 90071,
      "datasets pretrained language models": 27628,
      "outperforms previous state art predicting": 82946,
      "experimental results method significantly improves": 40643,
      "fullresource": 45931,
      "ganbased": 46431,
      "encoderbased": 36588,
      "hero": 49846,
      "verbbased": 131543,
      "malayalamenglish": 66877,
      "strands": 113441,
      "potentially beneficial": 88604,
      "space input": 111008,
      "input evaluate": 55333,
      "settings use": 107843,
      "representations multitask": 99762,
      "dualencoder model": 34246,
      "includes model": 53257,
      "simultaneously pretrained": 109693,
      "tool improving": 122390,
      "hand paper": 49322,
      "aforementioned features": 4267,
      "good transfer": 48504,
      "baselines considering": 12375,
      "features informative": 43568,
      "english benchmark": 37075,
      "improvements number": 52886,
      "given arbitrary": 47984,
      "proposed including": 93314,
      "word recovery": 133430,
      "learn mappings": 62086,
      "performed tasks": 86935,
      "averaged accuracy": 11221,
      "crosslingual bert": 24931,
      "2018 approach": 655,
      "agents recent": 4328,
      "complex nlp": 19848,
      "transfer provides": 124179,
      "language combination": 58887,
      "directly content": 31870,
      "learning scale": 62981,
      "xlm models": 135196,
      "required achieve": 100210,
      "general terms": 46720,
      "sentence retrieval": 106053,
      "estimation work": 38672,
      "recently multilingual": 96710,
      "learning share": 63021,
      "challenging setup": 16313,
      "approached using": 8036,
      "qa mlqa": 94506,
      "base versions": 11490,
      "instead investigate": 55665,
      "directly respect": 31903,
      "representations control": 99573,
      "ability multilingual": 1546,
      "transfer main": 124148,
      "bert xlmr": 13259,
      "art crosslingual": 9045,
      "simulate human": 109632,
      "evaluate range": 38908,
      "transfer finally": 124053,
      "adapt multilingual": 3183,
      "corpus bilingual": 23685,
      "representations enables": 99624,
      "setup larger": 107852,
      "involving languages": 56915,
      "typologically close": 126492,
      "transfer finetuning": 124054,
      "additional research": 3565,
      "using intermediate": 129764,
      "moderate improvements": 74374,
      "hindi multilingual": 50469,
      "subtask involves": 114947,
      "present article": 89384,
      "zeroshot monolingual": 135515,
      "prior model": 90719,
      "publish results": 94346,
      "codeswitching data": 18232,
      "languages mixing": 60728,
      "translation ranking": 125193,
      "model dramatically": 71029,
      "methods produces": 69690,
      "pairs aligned": 83487,
      "model mlm": 71519,
      "sentences distributed": 106283,
      "crosslingual data": 24936,
      "handle language": 49386,
      "single multiword": 109772,
      "stateoftheart lms": 112692,
      "allow transfer": 5097,
      "inherent limitation": 55189,
      "key goal": 57572,
      "encourage better": 36749,
      "masked word": 67309,
      "approach ability": 7287,
      "aligning contextual": 4934,
      "generation result": 47603,
      "strategies improve": 113468,
      "larger original": 61375,
      "strategies language": 113472,
      "previous zeroshot": 90584,
      "introduced multilingual": 56586,
      "set 10": 107339,
      "motivate wider": 74839,
      "wider investigation": 132608,
      "cues simple": 25218,
      "tasks leverage": 119252,
      "behaviour improves": 12680,
      "dealing underresourced": 27872,
      "latent knowledge": 61593,
      "fewshot experiments": 43900,
      "quite effective": 95401,
      "benefit knowledge": 12975,
      "unseen target": 127547,
      "lack proper": 58736,
      "proper understanding": 92427,
      "settings heterogeneous": 107806,
      "established benchmarks": 38599,
      "mbert trained": 67569,
      "trained 104": 123060,
      "handle codeswitched": 49375,
      "codeswitched languages": 18220,
      "strands research": 113442,
      "source label": 110770,
      "setting evaluate": 107746,
      "achieve surprisingly": 2583,
      "bert small": 13231,
      "results empirically": 101751,
      "detect given": 29805,
      "models xlmroberta": 74365,
      "multilingual distilled": 75244,
      "xlmr mbert": 135205,
      "performance document classification": 86310,
      "semantic space input": 105304,
      "ner pos tagging": 77068,
      "used improve accuracy": 128581,
      "task automatically selecting": 117917,
      "method code data": 68697,
      "examine different approaches": 39742,
      "yield best results": 135331,
      "english benchmark datasets": 37076,
      "significant improvements number": 108799,
      "available pretrained bert": 11076,
      "embeddings substantially outperform": 35959,
      "achieve highly competitive": 2483,
      "performance complex nlp": 86246,
      "complex nlp tasks": 19850,
      "leads significant performance": 61957,
      "performs particularly lowresource": 87016,
      "models present detailed": 73770,
      "multilingual bert multilingual": 75206,
      "quality estimation work": 94653,
      "estimation work presents": 38673,
      "presents interesting challenges": 89857,
      "bert achieved impressive": 13057,
      "model finally evaluated": 71183,
      "best knowledge largescale": 13351,
      "improve downstream task": 52376,
      "little data available": 64801,
      "typological features languages": 126486,
      "propose simple methods": 93064,
      "methods word alignment": 69850,
      "multilingual bert xlmr": 75210,
      "languages zeroshot fewshot": 60987,
      "state art crosslingual": 112437,
      "evaluate range stateoftheart": 38909,
      "stateoftheart models novel": 112746,
      "models novel dataset": 73651,
      "adapt multilingual models": 3184,
      "linguistic properties languages": 64532,
      "achieve best overall": 2419,
      "codeswitching data augmentation": 18233,
      "method does rely": 68784,
      "embeddings combining best": 35601,
      "language model mlm": 59348,
      "existing methods use": 40198,
      "text input target": 121057,
      "translated text target": 124569,
      "tackle issue propose": 117291,
      "contextualized representations despite": 22566,
      "propose method improve": 92770,
      "times larger original": 122182,
      "adversarial training language": 4175,
      "motivate wider investigation": 74840,
      "transformer networks pretrained": 124360,
      "linguistic knowledge paper": 64503,
      "leads performance gains": 61949,
      "multilingual models lowresource": 75290,
      "improves performance compared": 53011,
      "mbert trained 104": 67570,
      "handle codeswitched data": 49376,
      "bert models effectiveness": 13178,
      "lead performance increase": 61866,
      "dependency parsing use": 29207,
      "understanding tasks models": 126985,
      "multilingual bert small": 75209,
      "belong different languages": 12715,
      "training data provided": 123520,
      "multilingual distilled bert": 75245,
      "models used feature": 74289,
      "recently achieved remarkable performance": 96652,
      "approaches machine translation mt": 8225,
      "publicly available pretrained bert": 94319,
      "leads significant performance gains": 61958,
      "multilingual bert multilingual bert": 75207,
      "bert multilingual bert mbert": 13186,
      "quality estimation work presents": 94654,
      "multilingual models multilingual bert": 75293,
      "evaluate range stateoftheart models": 38910,
      "range stateoftheart models novel": 95604,
      "stateoftheart models novel dataset": 112747,
      "masked language model mlm": 67290,
      "model makes predictions based": 71496,
      "translated text target language": 124570,
      "transformer networks pretrained language": 124361,
      "networks pretrained language modeling": 77705,
      "multilingual bert multilingual bert mbert": 75208,
      "evaluate range stateoftheart models novel": 38911,
      "range stateoftheart models novel dataset": 95605,
      "transformer networks pretrained language modeling": 124362,
      "withindocument": 132747,
      "combing": 18705,
      "dissect": 32389,
      "splitantecedent": 111946,
      "singleantecedent": 109818,
      "kuhn": 58358,
      "appended": 6813,
      "formal model": 45171,
      "evaluation reference": 39355,
      "processing rules": 91788,
      "texts thousand": 121636,
      "evaluated manually": 38986,
      "mental representations": 68412,
      "approaches discussed": 8131,
      "tested complex": 120564,
      "rulebased uses": 103376,
      "information constituent": 54441,
      "opensource publicly": 82026,
      "resolution traditional": 100791,
      "coreference resolvers": 23377,
      "coreference models": 23352,
      "sets demonstrate": 107661,
      "evidence coreference": 39641,
      "achieved systems": 2713,
      "useful prediction": 128918,
      "challenge coreference": 16012,
      "highdimensional vector": 50155,
      "representations pairs": 99794,
      "coreference clusters": 23344,
      "algorithm teaches": 4807,
      "require careful": 100116,
      "instead apply": 55653,
      "resolution used": 100792,
      "features implicitly": 43551,
      "implicitly model": 52034,
      "evaluation clearly": 39146,
      "resolution model": 100770,
      "mentions experiments": 68453,
      "adding manually": 3388,
      "resolution using": 100793,
      "understanding used": 127004,
      "recent coreference": 96441,
      "domains datasets": 33758,
      "specific embeddings": 111436,
      "coarsetofine approach": 18060,
      "mention paper": 68425,
      "labels model": 58620,
      "detection summarization": 30070,
      "article aim": 9108,
      "issues require": 57075,
      "information zero": 55114,
      "candidate antecedents": 15100,
      "make local": 66695,
      "single candidate": 109709,
      "candidate antecedent": 15099,
      "specified text": 111623,
      "systems subtask": 117167,
      "metrics performs": 69992,
      "types incorporating": 126299,
      "context external": 22100,
      "model pronoun": 71805,
      "validity effectiveness": 130753,
      "computational treatment": 20442,
      "task case": 117955,
      "resolution present": 100780,
      "modern knowledge": 74393,
      "baselines crossdomain": 12378,
      "knowledge coreference": 57843,
      "general coreference": 46645,
      "resolution significantly": 100783,
      "ner coreference": 77031,
      "standalone neural": 112181,
      "proposed able": 93164,
      "based mention": 11836,
      "second uses": 104464,
      "classifier approach": 17522,
      "detection f1": 29951,
      "improvements 17": 52793,
      "gap dataset": 46448,
      "resolution major": 100767,
      "major area": 66536,
      "labelling data": 58566,
      "especially conversational": 38445,
      "example people": 39790,
      "help resolve": 49780,
      "consistent world": 21406,
      "output makes": 83092,
      "lot easier": 65324,
      "forms supervision": 45261,
      "biased different": 13855,
      "art using": 9101,
      "problem span": 91240,
      "mention using": 68433,
      "provides flexibility": 94031,
      "score gap": 104073,
      "resolution anaphora": 100751,
      "key aspects": 57541,
      "task identification": 118262,
      "mechanism determine": 67969,
      "mentions cluster": 68447,
      "classifiers used": 17643,
      "gold mentions": 48436,
      "score equivalent": 104069,
      "dataset coreference": 26832,
      "works literature": 134959,
      "withindocument coreference": 132748,
      "challenges challenge": 16139,
      "different definitions": 31086,
      "corpus independent": 23837,
      "corpora evaluations": 23475,
      "algorithm selecting": 4797,
      "budget experiments": 14734,
      "domains code": 33739,
      "incremental neural": 53736,
      "text makes": 121099,
      "follow evaluation": 44937,
      "models withindocument": 74343,
      "challenge analyze": 15994,
      "analyze current": 5959,
      "used real": 128713,
      "correctly resolving": 24181,
      "splitantecedent anaphora": 111947,
      "resolution models": 100771,
      "splitantecedent anaphors": 111949,
      "21 percentage": 762,
      "lee et": 63293,
      "conll f1": 21088,
      "limitations task": 64193,
      "coreference reasoning": 23357,
      "long studied": 65134,
      "attempt evaluate": 9741,
      "does reflect": 33383,
      "phenomena furthermore": 87225,
      "existing coreference": 40092,
      "model appended": 70669,
      "complete documents": 19764,
      "document effective": 32997,
      "opensource publicly available": 82027,
      "coreference resolution traditional": 23374,
      "data sets demonstrate": 26421,
      "features instead propose": 43573,
      "challenge coreference resolution": 16013,
      "systems typically trained": 117213,
      "coreference evaluation metrics": 23347,
      "neural models detecting": 78172,
      "coreference resolution used": 23375,
      "using lexical features": 129809,
      "coreference resolution model": 23366,
      "network cnn model": 77195,
      "acoustic features extracted": 3005,
      "features extracted speech": 43509,
      "coreference resolution using": 23376,
      "role linguistic features": 103200,
      "domain specific embeddings": 33646,
      "additional features improve": 3516,
      "architecture attention mechanism": 8620,
      "shared task english": 108042,
      "semantic information zero": 105082,
      "make local decisions": 66696,
      "single candidate antecedent": 109710,
      "reinforcement learning agent": 97793,
      "context external knowledge": 22101,
      "external knowledge based": 42008,
      "validity effectiveness model": 130754,
      "coreference annotation task": 23340,
      "coreference resolution present": 23370,
      "general coreference resolution": 46646,
      "coreference resolution significantly": 23371,
      "ner coreference resolution": 77032,
      "neural models proposed": 78193,
      "detection approach based": 29880,
      "best model using": 13375,
      "specifically designed task": 111540,
      "different forms supervision": 31155,
      "obtain best performance": 81261,
      "formulate problem span": 45279,
      "problem span prediction": 91241,
      "span prediction task": 111129,
      "task like machine": 118354,
      "f1 score gap": 42681,
      "anaphora resolution anaphora": 6066,
      "resolution anaphora resolution": 100752,
      "multitask learning based": 75827,
      "neural model uses": 78156,
      "applied different corpora": 7060,
      "corpora evaluations different": 23476,
      "outperforming best reported": 82803,
      "graph neural networkbased": 48845,
      "future research task": 46309,
      "set baseline results": 107374,
      "models withindocument coreference": 74344,
      "withindocument coreference resolution": 132749,
      "detection important natural": 29971,
      "used real applications": 128714,
      "splitantecedent anaphora resolution": 111948,
      "coreference resolution models": 23367,
      "shows best model": 108556,
      "21 percentage points": 763,
      "lee et al": 63294,
      "models used paper": 74290,
      "used paper available": 128677,
      "existing coreference resolution": 40093,
      "model significantly outperforms previous": 72032,
      "significantly outperforms previous work": 109025,
      "neural network cnn model": 78285,
      "chinese zero pronoun resolution": 16845,
      "outperforms stateoftheart models large": 83010,
      "matches outperforms stateoftheart models": 67386,
      "natural language understanding systems": 76582,
      "formulate problem span prediction": 45280,
      "problem span prediction task": 91242,
      "anaphora resolution anaphora resolution": 6067,
      "models withindocument coreference resolution": 74345,
      "detection important natural language": 29972,
      "models used paper available": 74291,
      "convolutional neural network cnn model": 23186,
      "model outperforms stateoftheart models large": 71675,
      "outperforms stateoftheart models large margin": 83011,
      "formulate problem span prediction task": 45281,
      "invalidate": 56676,
      "manuallycollected": 67091,
      "ruling": 103455,
      "shootings": 108195,
      "agglutinating": 4334,
      "sd": 104277,
      "kfold": 57652,
      "crosswalks": 25091,
      "subdivision": 114664,
      "patent retrieval": 85666,
      "models regression": 73901,
      "usage multiple": 127866,
      "languages considered": 60466,
      "producing different": 92017,
      "selected sample": 104730,
      "reveals hidden": 102515,
      "tool facilitate": 122387,
      "corpus release": 23964,
      "event extractor": 39519,
      "task determine": 118079,
      "determine appropriate": 30120,
      "given case": 47990,
      "different expression": 31140,
      "investigate application": 56716,
      "textual form": 121707,
      "predicting case": 88978,
      "linear support": 64374,
      "data innovation": 26042,
      "overview natural": 83368,
      "quantitative data": 94865,
      "analysis provide": 5782,
      "obtain training": 81334,
      "ensemble combining": 37583,
      "multiple svm": 75715,
      "provide excellent": 93818,
      "treating problem": 125564,
      "attention problem": 9979,
      "following paper": 44976,
      "detect cases": 29796,
      "used authors": 128400,
      "cleaning data": 17715,
      "sentences varied": 106540,
      "treat problem": 125552,
      "prediction dataset": 89048,
      "inferred according": 54262,
      "scoring using": 104235,
      "model extraction": 71162,
      "use fast": 128050,
      "predict sentiments": 88935,
      "motivations study": 74881,
      "methodology achieves": 69241,
      "overall goal": 83233,
      "svm ensembles": 116239,
      "approaches classifying": 8096,
      "documents approaches": 33185,
      "predict accurately": 88869,
      "information ignored": 54663,
      "verification framework": 131570,
      "human rights": 50958,
      "evaluate broad": 38804,
      "broad variety": 14684,
      "tasks binary": 118968,
      "demographic information": 28650,
      "opinions related": 82130,
      "importance study": 52082,
      "opinion texts": 82106,
      "methodology evaluated": 69259,
      "safety incidents": 103537,
      "accident reports": 1936,
      "networks han": 77618,
      "shared organizations": 107998,
      "tools tasks": 122477,
      "adapted processing": 3291,
      "baselines propose": 12445,
      "differs substantially": 31604,
      "nli shared": 79550,
      "extraction identify": 42352,
      "proposed contrast": 93243,
      "studies addressed": 114178,
      "participants required": 85325,
      "competition best": 19622,
      "researchers better": 100679,
      "analyzed different": 6024,
      "particular classification": 85392,
      "addresses limitation": 3799,
      "complex case": 19802,
      "results ones": 102012,
      "example words": 39801,
      "scratch code": 104240,
      "analyses dataset": 5448,
      "available general": 11000,
      "technology artificial": 120034,
      "tasks rulebased": 119477,
      "especially machine": 38476,
      "data gap": 25968,
      "use associated": 127899,
      "expertannotated dataset": 41232,
      "ranking approach": 95664,
      "reranking better": 100368,
      "text challenges": 120701,
      "provide actionable": 93751,
      "actionable insights": 3085,
      "text requires": 121258,
      "requires controlling": 100257,
      "controlling confounding": 22851,
      "bias recently": 13839,
      "identify strengths": 51560,
      "parts query": 85580,
      "combination lexical": 18565,
      "domain explore": 33532,
      "available strategies": 11116,
      "strategies applying": 113451,
      "models intended": 73413,
      "documents hard": 33237,
      "needs small": 76898,
      "focus realworld": 44808,
      "develop sentiment": 30230,
      "domain low": 33578,
      "analyzing sentiments": 6052,
      "task developing": 118084,
      "lack domain": 58704,
      "structural text": 113797,
      "documents properly": 33275,
      "rely representations": 98735,
      "document providing": 33058,
      "representation downstream": 99215,
      "detection performing": 30022,
      "finetuning taskspecific": 44522,
      "hierarchical topic": 50016,
      "annotations use": 6477,
      "training results": 123821,
      "classification according": 17098,
      "investigated text": 56843,
      "resulted best": 101425,
      "results 12": 101483,
      "dynamic topic": 34331,
      "study analyzes": 114318,
      "technical words": 119757,
      "terms text": 120392,
      "study determined": 114354,
      "analyzed results": 6028,
      "data diversity": 25862,
      "makes finding": 66783,
      "dealing diverse": 27867,
      "intuition paper": 56665,
      "cases including": 15647,
      "effect proposed": 34607,
      "extraction need": 42407,
      "propose attentionbased neural": 92559,
      "linear support vector": 64375,
      "machine svm classifier": 65871,
      "overview natural language": 83369,
      "ensemble combining output": 37584,
      "combining output multiple": 18740,
      "output multiple svm": 83097,
      "multiple svm classifiers": 75716,
      "results effectiveness method": 101744,
      "framework based expectation": 45436,
      "experiments demonstrate benefits": 40879,
      "present method automatic": 89546,
      "proposed methodology achieves": 93409,
      "dataset provided organizers": 27129,
      "using combination words": 129451,
      "attention networks han": 9950,
      "propose new shared": 92883,
      "new shared task": 79135,
      "help researchers better": 49778,
      "researchers better understand": 100680,
      "applications text classification": 7026,
      "length input text": 63362,
      "present paper shows": 89638,
      "dataset available general": 26754,
      "technology artificial intelligence": 120035,
      "future directions research": 46272,
      "media data provide": 68101,
      "based model called": 11848,
      "bert models downstream": 13177,
      "texts case study": 121469,
      "labelled data domain": 58559,
      "human labeled data": 50890,
      "new dataset comprising": 78862,
      "task challenging large": 117958,
      "propose attentionbased neural network": 92560,
      "linear support vector machine": 64376,
      "vector machine svm classifier": 131317,
      "machine svm classifier trained": 65872,
      "ensemble combining output multiple": 37585,
      "combining output multiple svm": 18741,
      "output multiple svm classifiers": 83098,
      "experimental results effectiveness method": 40610,
      "framework based expectation maximization": 45437,
      "experiments demonstrate benefits proposed": 40880,
      "annotated dataset provided organizers": 6176,
      "hierarchical attention networks han": 49935,
      "propose new shared task": 92884,
      "help researchers better understand": 49779,
      "social media data provide": 110369,
      "linear support vector machine svm": 64377,
      "support vector machine svm classifier": 116007,
      "vector machine svm classifier trained": 131318,
      "ensemble combining output multiple svm": 37586,
      "combining output multiple svm classifiers": 18742,
      "geolocated": 47858,
      "specificities": 111612,
      "chile": 16733,
      "peru": 87201,
      "unibuckernel": 127086,
      "reloaded": 98671,
      "vardial": 130826,
      "postcompetition": 88462,
      "neo": 77008,
      "unreasonable": 127498,
      "nordic": 80321,
      "faroese": 43107,
      "farasa": 43103,
      "wanlp2021": 132011,
      "cxgs": 25478,
      "classification algorithm": 17110,
      "algorithm application": 4694,
      "main innovation": 66432,
      "make clear": 66631,
      "avoiding problems": 11247,
      "second application": 104383,
      "automatic dialect": 10513,
      "arabic egyptian": 8512,
      "errors context": 38360,
      "speakers provide": 111325,
      "years english": 135258,
      "handling diversity": 49414,
      "data comes": 25747,
      "december 2015": 27918,
      "words crawled": 133893,
      "acoustic phonetic": 3023,
      "data observed": 26185,
      "architecture hyperparameters": 8667,
      "investigate progress": 56801,
      "investigated human": 56836,
      "document written": 33113,
      "preparation training": 89296,
      "task complicated": 117995,
      "start analyzing": 112397,
      "parent node": 84856,
      "best classifier": 13318,
      "variety identification": 130982,
      "employed machine": 36303,
      "shortcomings traditional": 108284,
      "feature vectorization": 43328,
      "approach solution": 7921,
      "identification text": 51452,
      "trained character": 123085,
      "time native": 122053,
      "transcripts use": 124015,
      "results development": 101723,
      "similar machine": 109108,
      "participated tracks": 85351,
      "close accuracy": 17808,
      "rank overall": 95627,
      "research training": 100650,
      "techniques generally": 119892,
      "area recently": 8880,
      "brief history": 14611,
      "literature describing": 64751,
      "methods introduce": 69560,
      "training end": 123607,
      "vardial evaluation": 130828,
      "speech phonetic": 111723,
      "employ kernel": 36267,
      "better postcompetition": 13670,
      "postcompetition results": 88463,
      "features ranked": 43684,
      "features combine": 43406,
      "classifiers maximize": 17618,
      "age language": 4285,
      "issues encountered": 57045,
      "resources arabic": 100947,
      "arabic author": 8491,
      "profiling paper": 92104,
      "annotation pipeline": 6352,
      "sources covering": 110881,
      "24 million": 802,
      "process finally": 91487,
      "learning mechanisms": 62715,
      "leverage use": 63634,
      "vardial 2017": 130827,
      "considerably higher": 21269,
      "words iii": 133987,
      "measure robustness": 67821,
      "methods discriminate": 69452,
      "trained various": 123332,
      "18 different": 416,
      "filters identify": 44093,
      "identification tweets": 51454,
      "ml models": 70435,
      "secondary goal": 104469,
      "models accurately": 72655,
      "decisions taken": 27966,
      "uralic language": 127836,
      "machines naive": 66337,
      "dialect features": 30519,
      "feature detection": 43263,
      "unavailable making": 126599,
      "indian english": 53817,
      "demonstrate downstream": 28702,
      "nordic languages": 80322,
      "speaker traits": 111299,
      "dialect showing": 30525,
      "lstm convolutional": 65610,
      "arabic tweets": 8564,
      "farasa segmentation": 43104,
      "eacl wanlp2021": 34369,
      "wanlp2021 shared": 132012,
      "developing identifies": 30342,
      "arabic tweet": 8563,
      "involves preprocessing": 56900,
      "final approach": 44097,
      "findings results": 44310,
      "semantic categorization": 104999,
      "hosted github": 50662,
      "english measuring": 37207,
      "computational dialectology": 20375,
      "global scale": 48266,
      "previously limited": 90606,
      "construction grammars": 21689,
      "use public": 128221,
      "resources study": 101047,
      "areas social": 8903,
      "corpus future": 23814,
      "future finally": 46277,
      "language identification task": 59145,
      "important task paper": 52269,
      "neural network task": 78404,
      "dialect identification task": 30524,
      "multilingual language identification": 75267,
      "monolingual language identification": 74592,
      "language identification using": 59148,
      "language variety identification": 60326,
      "employed machine learning": 36304,
      "language identification text": 59146,
      "svm classifiers trained": 116234,
      "classifiers trained character": 17640,
      "time native language": 122054,
      "obtains better results": 81458,
      "based string kernels": 12078,
      "shared task participated": 108084,
      "processing pipelines paper": 91772,
      "natural language document": 76289,
      "vardial evaluation campaign": 130829,
      "better postcompetition results": 13671,
      "age language variety": 4286,
      "arabic author profiling": 8492,
      "author profiling paper": 10368,
      "effort create large": 35165,
      "models created using": 72999,
      "data language model": 26070,
      "evaluated using datasets": 39021,
      "participated shared tasks": 85347,
      "identification paper presents": 51418,
      "report results obtained": 99036,
      "languages paper presents": 60781,
      "using test sets": 130285,
      "automatically collected dataset": 10739,
      "learning ml models": 62752,
      "using ensemble learning": 129634,
      "shared task collected": 108029,
      "vector machines naive": 131323,
      "machines naive bayes": 66338,
      "learning approach automatic": 62340,
      "showing promising results": 108426,
      "classification tasks multitask": 17447,
      "memory lstm convolutional": 68328,
      "lstm convolutional neural": 65611,
      "pretrained models used": 90156,
      "eacl wanlp2021 shared": 34370,
      "wanlp2021 shared task": 132013,
      "various parts text": 131163,
      "final approach achieved": 44098,
      "shared task includes": 108053,
      "areas social media": 8904,
      "svm classifiers trained character": 116235,
      "language processing pipelines paper": 59932,
      "language identification paper presents": 59138,
      "machine learning ml models": 65805,
      "support vector machines naive": 116012,
      "vector machines naive bayes": 131324,
      "machines naive bayes classifier": 66339,
      "classification tasks multitask learning": 17448,
      "tasks multitask learning mtl": 119320,
      "memory lstm convolutional neural": 68329,
      "eacl wanlp2021 shared task": 34371,
      "natural language processing pipelines paper": 76473,
      "support vector machines naive bayes": 116013,
      "vector machines naive bayes classifier": 131325,
      "ldc2015e86": 61837,
      "testsuite": 120617,
      "274": 858,
      "tbru": 119674,
      "reentrant": 97484,
      "sequencetoaction": 107149,
      "graphaware": 48884,
      "pillars": 87504,
      "embedder": 35366,
      "currently important": 25406,
      "sentences abstract": 106192,
      "domains absolute": 33723,
      "task impact": 118274,
      "presented extension": 89782,
      "classifier produce": 17575,
      "results gain": 101804,
      "baseline parser": 12282,
      "parser errors": 84948,
      "carefully optimized": 15527,
      "manner based": 66940,
      "specific subtasks": 111493,
      "minimal recursion": 70172,
      "recursion semantics": 97279,
      "semantics mrs": 105436,
      "amr benchmark": 5386,
      "gpu batch": 48575,
      "highprecision grammarbased": 50365,
      "parsing making": 85154,
      "generation sequencetosequence": 47615,
      "careful preprocessing": 15512,
      "amr generation": 5387,
      "techniques leads": 119918,
      "input amr": 55296,
      "applying various": 7280,
      "entropy models": 38171,
      "probability individual": 90863,
      "best sequence": 13450,
      "data adding": 25573,
      "framework parsing": 45649,
      "unit tbru": 127215,
      "graph linearization": 48830,
      "result indicates": 101378,
      "indicates future": 53860,
      "semantic translation": 105337,
      "text code": 120792,
      "richardson kuhn": 102802,
      "languages mixed": 60727,
      "facilitate modeling": 42781,
      "type develop": 126194,
      "graphbased decoding": 48889,
      "challenging partly": 16291,
      "partly lack": 85515,
      "use variational": 128353,
      "performance overnight": 86592,
      "outperforms rulebased": 82976,
      "tags input": 117477,
      "score empirical": 104067,
      "proposed tackle": 93563,
      "techniques applying": 119838,
      "nodes directly": 80034,
      "data french": 25961,
      "builds previous": 14906,
      "work robust": 134789,
      "accuracy prediction": 2243,
      "represents sentences": 100073,
      "parser new": 84973,
      "results alternative": 101504,
      "perform case": 85957,
      "scores gold": 104176,
      "predicted parse": 88965,
      "parser generate": 84959,
      "silver data": 109064,
      "parsers rely": 85042,
      "evaluation evaluating": 39204,
      "graph labeled": 48827,
      "amr structure": 5399,
      "identification named": 51402,
      "parser compositional": 84928,
      "evaluation meaning": 39262,
      "propose normalization": 92891,
      "methods helping": 69531,
      "normalization algorithms": 80333,
      "grammars semantic": 48682,
      "present recurrent": 89662,
      "learns map": 63220,
      "errors model": 38385,
      "lemmatization pos": 63340,
      "approaches meaning": 8231,
      "design considerations": 29528,
      "design analysis": 29514,
      "task concerned": 117998,
      "problem structural": 91250,
      "measures extent": 67862,
      "pilot studies": 87509,
      "grounded image": 49106,
      "nlp domains": 79607,
      "deep graph": 28237,
      "community improving": 19078,
      "research abstract": 100395,
      "focus sequencetosequence": 44816,
      "selflearning techniques": 104924,
      "results amr": 101505,
      "robust natural": 103050,
      "using complementary": 129457,
      "yield surprisingly": 135364,
      "fully specified": 45970,
      "leverage advances": 63573,
      "advances crosslingual": 3979,
      "develop novel technique": 30224,
      "sentences abstract meaning": 106193,
      "representation amr graphs": 99166,
      "supervised manner based": 115797,
      "recognition semantic role": 96992,
      "propose neural encoderdecoder": 92817,
      "minimal recursion semantics": 70173,
      "recursion semantics mrs": 97280,
      "shown strong performance": 108530,
      "large corpus sentences": 61066,
      "meaning representations work": 67684,
      "input amr graph": 55297,
      "amr graph structure": 5389,
      "maximum entropy models": 67535,
      "models estimate probability": 73169,
      "transition based recurrent": 124483,
      "based recurrent unit": 11983,
      "recurrent unit tbru": 97269,
      "models trained multiple": 74217,
      "natural languages particular": 76609,
      "semantic parsing approach": 105143,
      "models semantic parsing": 73990,
      "stateoftheart performance overnight": 112838,
      "performance overnight dataset": 86593,
      "pos tags input": 88244,
      "builds previous work": 14907,
      "future work robust": 46326,
      "assign high scores": 9542,
      "produce natural language": 91912,
      "paper propose normalization": 84286,
      "present recurrent neural": 89663,
      "lemmatization pos tagging": 63341,
      "structural information input": 113772,
      "model encoder decoder": 71071,
      "text abstract meaning": 120625,
      "processing nlp domains": 91733,
      "research community improving": 100446,
      "paper focus sequencetosequence": 83951,
      "processing applications machine": 91618,
      "stateoftheart results amr": 112892,
      "generation generating text": 47417,
      "pretraining multilingual models": 90292,
      "sentences abstract meaning representation": 106194,
      "meaning representation amr graphs": 67670,
      "recognition semantic role labeling": 96993,
      "minimal recursion semantics mrs": 70174,
      "transition based recurrent unit": 124484,
      "based recurrent unit tbru": 11984,
      "achieves stateoftheart performance overnight": 2889,
      "stateoftheart performance overnight dataset": 112839,
      "achieve highly competitive performance": 2484,
      "present recurrent neural network": 89664,
      "map natural language utterances": 67112,
      "language processing nlp domains": 59898,
      "language processing applications machine": 59838,
      "processing applications machine translation": 91619,
      "generation generating text structured": 47418,
      "sentences abstract meaning representation amr": 106195,
      "abstract meaning representation amr graphs": 1771,
      "transition based recurrent unit tbru": 124485,
      "achieves stateoftheart performance overnight dataset": 2890,
      "natural language processing nlp domains": 76445,
      "natural language processing applications machine": 76395,
      "language processing applications machine translation": 59839,
      "indistinguishable": 53894,
      "perceptible": 85926,
      "populationbased": 88170,
      "untargeted": 127763,
      "jia": 57233,
      "attacker": 9707,
      "damages": 25527,
      "metropolishastings": 70022,
      "fools": 45003,
      "defence": 28470,
      "humanandmodelintheloop": 50993,
      "oftheart": 81632,
      "corporations": 23627,
      "text fully": 120967,
      "new abstract": 78762,
      "researchers different": 100687,
      "models disagree": 73080,
      "optimization algorithm": 82187,
      "models success": 74120,
      "fail yield": 42974,
      "diversity adversarial": 32866,
      "case general": 15587,
      "gan models": 46430,
      "attacks deep": 9712,
      "survey development": 116169,
      "applied texts": 7128,
      "data discrete": 25854,
      "select summarize": 104717,
      "models adversarial": 72730,
      "way assessing": 132059,
      "weaknesses model": 132209,
      "expected output": 40398,
      "related literature": 97879,
      "account using": 2042,
      "meaning general": 67635,
      "terms adversarial": 120268,
      "particular weaknesses": 85464,
      "set introduce": 107468,
      "method studying": 69167,
      "jia liang": 57234,
      "liang 2017": 63957,
      "difficult results": 31658,
      "likely token": 64144,
      "token text": 122269,
      "games multiple": 46423,
      "language game": 59058,
      "detecting target": 29854,
      "research adversarial": 100399,
      "large natural": 61163,
      "led rapid": 63281,
      "value model": 130783,
      "investigate difference": 56741,
      "flexible text": 44664,
      "words machine": 134043,
      "transferable domains": 124220,
      "popular classification": 88086,
      "emphasizes need": 36143,
      "susceptible adversarial": 116217,
      "examples recent": 39871,
      "properties adversarial": 92435,
      "detection adversarial": 29871,
      "accurately detecting": 2379,
      "methods adversarial": 69301,
      "robustness propose": 103126,
      "adversarial scenarios": 4162,
      "attack methods": 9701,
      "models exemplified": 73188,
      "virtual adversarial": 131740,
      "space discrete": 110993,
      "introduced nlp": 56593,
      "use tokenlevel": 128329,
      "improves score": 53041,
      "labeling text": 58550,
      "quality nlp": 94735,
      "capabilities ai": 15165,
      "paradigm using": 84555,
      "impact created": 51863,
      "better distribution": 13566,
      "process dynamic": 91464,
      "errors conduct": 38359,
      "design linguistic": 29545,
      "task reveal": 118659,
      "annotation natural": 6340,
      "local distributional": 64917,
      "perturbations added": 87195,
      "second set": 104453,
      "robust multitask": 103049,
      "data annotate": 25601,
      "methods code": 69370,
      "little studied": 64829,
      "real challenge": 96056,
      "neural learners": 77962,
      "detection email": 29933,
      "attack method": 9700,
      "targeted text": 117776,
      "generating examples": 47218,
      "fewer edits": 43887,
      "shows adversarial": 108548,
      "hard humans": 49462,
      "discrete words": 32184,
      "tasks specific": 119511,
      "reference models": 97526,
      "nlp adversarial": 79560,
      "research date": 100459,
      "determines validity": 30144,
      "suffer robustness": 115241,
      "text generates": 120975,
      "controllable attributes": 22825,
      "task labels": 118331,
      "reviews use": 102635,
      "product categories": 92030,
      "reviews experiments": 102607,
      "diverse fluent": 32813,
      "attacks adversarial": 9711,
      "benchmark works": 12873,
      "comparing performances": 19512,
      "models loss": 73527,
      "classification labels": 17245,
      "challenging state": 16314,
      "state oftheart": 112509,
      "text adversarial": 120633,
      "tuning pretrained": 125939,
      "rule experiments": 103352,
      "experiments classification": 40832,
      "text expansion": 120931,
      "attack strategy": 9703,
      "finds candidate": 44317,
      "craft adversarial": 24588,
      "hard label": 49465,
      "setting study": 107780,
      "overall semantic": 83256,
      "similarity original": 109277,
      "exponentially large": 41683,
      "discrete text": 32179,
      "despite deep": 29684,
      "input ii": 55348,
      "easily build": 34439,
      "different attacks": 31014,
      "model studies": 72097,
      "accuracy important": 2179,
      "fluency semantic": 44701,
      "researchers develop": 100684,
      "steps model": 113324,
      "text fully annotated": 120968,
      "noise adversarial attacks": 80043,
      "vulnerable adversarial examples": 131976,
      "adversarial attacks deep": 4104,
      "attacks deep learning": 9713,
      "current models capture": 25301,
      "jia liang 2017": 57235,
      "various realworld applications": 131181,
      "little attention devoted": 64793,
      "large natural language": 61164,
      "models trained standard": 74230,
      "value model size": 130784,
      "task deep neural": 118051,
      "neural networks domain": 78470,
      "networks based models": 77520,
      "models deployed real": 73043,
      "tested benchmark datasets": 120563,
      "adversarial examples improved": 4120,
      "neural text processing": 78709,
      "compare approach recently": 19222,
      "pretraining large neural": 90273,
      "adversarial training methods": 4177,
      "performance sequence labeling": 86705,
      "sequence labeling text": 106988,
      "labeling text classification": 58551,
      "achieved human level": 2643,
      "snli dataset generalize": 110313,
      "dataset generalize better": 26951,
      "annotation natural language": 6341,
      "improvements existing data": 52854,
      "methods code publicly": 69371,
      "models provide comprehensive": 73835,
      "text generation requires": 121006,
      "model produces fluent": 71800,
      "attracted lot research": 10158,
      "models sensitive small": 73996,
      "sensitive small perturbations": 105747,
      "propose novel unified": 92973,
      "model using sequencetosequence": 72285,
      "models adversarial training": 72731,
      "adversarial attacks adversarial": 4103,
      "models despite success": 73054,
      "craft adversarial examples": 24589,
      "able achieve higher": 1595,
      "text classification experiments": 120721,
      "finetuned bert models": 44421,
      "adversarial training natural": 4180,
      "evaluating natural language processing": 39082,
      "adversarial attacks deep learning": 4105,
      "significantly outperforms baseline methods": 109004,
      "analysis natural language inference": 5716,
      "neural networks based models": 78446,
      "models deployed real world": 73044,
      "compare approach recently proposed": 19223,
      "sequence labeling text classification": 106989,
      "labeling text classification tasks": 58552,
      "snli dataset generalize better": 110314,
      "methods code publicly available": 69372,
      "pretrained masked language model": 90124,
      "models sensitive small perturbations": 73997,
      "sensitive small perturbations input": 105748,
      "adversarial training natural language": 4181,
      "sequence labeling text classification tasks": 106990,
      "models sensitive small perturbations input": 73998,
      "announcements": 6502,
      "selfreported": 104930,
      "moods": 74659,
      "hyperbole": 51205,
      "ambiguously": 5302,
      "trainingdevelopment": 123958,
      "disabling": 31929,
      "puzzle": 94471,
      "oversampled": 83353,
      "userlevel": 129080,
      "liwc": 64856,
      "150000": 361,
      "sensibility": 105730,
      "task3": 118866,
      "701": 1216,
      "hp": 50690,
      "prioritizing": 90775,
      "extraversion": 42589,
      "contacts": 21730,
      "train effective": 122928,
      "post detection": 88453,
      "mt technique": 75000,
      "collection phrases": 18483,
      "possibility utilizing": 88373,
      "media identify": 68119,
      "identification technique": 51451,
      "individuals social": 53947,
      "corresponding explanations": 24287,
      "resources support": 101049,
      "studying users": 114559,
      "users method": 129143,
      "dataset semisupervised": 27185,
      "health issues": 49621,
      "twitter terms": 126126,
      "data public": 26304,
      "challenging reasons": 16303,
      "typically short": 126460,
      "requires relatively": 100316,
      "twitter paper": 126112,
      "information received": 54899,
      "search tuning": 104360,
      "ensemble best": 37580,
      "teams microaveraged": 119726,
      "microaveraged fscore": 70044,
      "conversation threads": 22980,
      "classifying tweets": 17683,
      "sentiment domainspecific": 106733,
      "reduce class": 97316,
      "substantial impact": 114857,
      "impact overall": 51882,
      "twitter used": 126129,
      "based aggregated": 11500,
      "online language": 81778,
      "linguistic psychological": 64536,
      "conditions language": 20818,
      "temporal annotation": 120091,
      "approaches suggest": 8363,
      "extracting temporal": 42242,
      "benchmark setup": 12859,
      "sentiment specific": 106796,
      "classifier develop": 17539,
      "techniques did": 119870,
      "tweets articles": 126020,
      "develop classifier": 30184,
      "tweets train": 126062,
      "developed classifier": 30259,
      "works demonstrate": 134935,
      "demonstrate lstms": 28771,
      "tasks deploy": 119049,
      "attention unit": 10027,
      "second shared": 104454,
      "current future": 25284,
      "including recurrent": 53358,
      "regression support": 97712,
      "regression obtained": 97707,
      "testing task": 120604,
      "dataset combination": 26799,
      "combination features": 18561,
      "work identified": 134557,
      "media observing": 68132,
      "underlying meaning": 126687,
      "sentiments social": 106834,
      "case social": 15608,
      "corpus detecting": 23751,
      "corpus prerequisite": 23934,
      "challenges face": 16158,
      "corpus social": 24002,
      "adaptation sentiment": 3263,
      "data project": 26286,
      "analysis identification": 5645,
      "subset language": 114832,
      "strong binary": 113658,
      "linguistics methods": 64615,
      "collect analyse": 18373,
      "evidence computational": 39640,
      "classification social": 17412,
      "averaged f1": 11222,
      "learned classifiers": 62206,
      "informal unstructured": 54345,
      "unstructured nature": 127579,
      "users finding": 129119,
      "twolayered attention": 126147,
      "world experimental": 135031,
      "associated negative": 9603,
      "existing computational": 40089,
      "posts different": 88516,
      "complexity diversity": 19909,
      "data characteristics": 25727,
      "tweets contain": 126025,
      "parser outputs": 84979,
      "tweets used": 126068,
      "used results": 128732,
      "explore limitations": 41556,
      "phrases relevant": 87447,
      "studies inferred": 114240,
      "remains elusive": 98796,
      "result reproducible": 101401,
      "roberta achieves": 102992,
      "improvements average": 52807,
      "average scores": 11214,
      "use figurative": 128054,
      "extract contextual": 42073,
      "information diverse": 54502,
      "comment posts": 18801,
      "possible measure": 88412,
      "introducing opensource": 56650,
      "standardized schema": 112348,
      "detecting users": 29858,
      "tweet classification": 126009,
      "routinely used": 103320,
      "language analyses": 58827,
      "feasibility automatic": 43236,
      "users concerns": 129103,
      "expected provide": 40399,
      "develop evaluate": 30195,
      "annotated annotators": 6138,
      "f1score 0914": 42707,
      "0914 precision": 104,
      "classifiers predictions": 17627,
      "evaluated directly": 38970,
      "research using": 100661,
      "research raised": 100606,
      "produced data": 91952,
      "underrepresented groups": 126723,
      "labels concept": 58585,
      "frequent ones": 45861,
      "post detection model": 88454,
      "translation mt technique": 124988,
      "experimental results combination": 40569,
      "social media identify": 110385,
      "individuals social media": 53948,
      "platforms like twitter": 87670,
      "training data small": 123540,
      "twitter paper describes": 126113,
      "sentiment domainspecific features": 106734,
      "impact overall performance": 51883,
      "model performance addition": 71723,
      "paper submitted shared": 84450,
      "previous works demonstrate": 90567,
      "works demonstrate lstms": 134936,
      "demonstrate lstms achieved": 28772,
      "lstms achieved remarkable": 65699,
      "remarkable performance natural": 98836,
      "processing tasks deploy": 91812,
      "model attention unit": 70704,
      "second shared task": 104455,
      "recurrent convolutional networks": 97167,
      "regression support vector": 97713,
      "linear regression obtained": 64366,
      "regression obtained best": 97708,
      "social media observing": 110395,
      "case social media": 15609,
      "corpus social media": 24003,
      "case study demonstrate": 15617,
      "sentiment analysis widely": 106678,
      "analysis widely used": 5937,
      "domain adaptation sentiment": 33451,
      "adaptation sentiment analysis": 3264,
      "sentiment analysis tools": 106674,
      "purely datadriven approach": 94414,
      "classification social media": 17413,
      "macro averaged f1": 66355,
      "averaged f1 score": 11223,
      "providing new stateoftheart": 94125,
      "informal unstructured nature": 54346,
      "world experimental results": 135032,
      "word embeddings attention": 133052,
      "large dataset consisting": 61073,
      "attention weights using": 10040,
      "numerous studies inferred": 81030,
      "based social media": 12056,
      "use figurative language": 128055,
      "extract contextual information": 42074,
      "results language models": 101874,
      "expression propose novel": 41744,
      "improve models robustness": 52423,
      "f1score 0914 precision": 42708,
      "using classifiers predictions": 129442,
      "machine translation mt technique": 66073,
      "media platforms like twitter": 68147,
      "using social media text": 130195,
      "model outperforms stateoftheart techniques": 71678,
      "paper submitted shared task": 84451,
      "previous works demonstrate lstms": 90568,
      "works demonstrate lstms achieved": 134937,
      "demonstrate lstms achieved remarkable": 28773,
      "lstms achieved remarkable performance": 65700,
      "achieved remarkable performance natural": 2683,
      "remarkable performance natural language": 98837,
      "language processing tasks deploy": 59953,
      "models pretrained language model": 73781,
      "lstm model attention unit": 65641,
      "regression support vector regression": 97714,
      "linear regression obtained best": 64367,
      "supervised machine learning methods": 115792,
      "sentiment analysis widely used": 106679,
      "domain adaptation sentiment analysis": 33452,
      "classification social media posts": 17414,
      "macro averaged f1 score": 66356,
      "social media platforms like twitter": 110407,
      "previous works demonstrate lstms achieved": 90569,
      "works demonstrate lstms achieved remarkable": 134938,
      "demonstrate lstms achieved remarkable performance": 28774,
      "lstms achieved remarkable performance natural": 65701,
      "achieved remarkable performance natural language": 2684,
      "remarkable performance natural language processing": 98838,
      "natural language processing tasks deploy": 76492,
      "offense": 81534,
      "subcontinent": 114654,
      "0706": 57,
      "2454": 806,
      "speakerlistener": 111305,
      "transference": 124224,
      "tamilenglish": 117573,
      "8603": 1330,
      "7529": 1256,
      "disseminate": 32391,
      "events like": 39584,
      "using web": 130369,
      "medium communication": 68244,
      "hierarchical tagset": 50011,
      "identification best": 51367,
      "hindi data": 50462,
      "ranked 6th": 95640,
      "offense detection": 81535,
      "societies like": 110470,
      "indian subcontinent": 53826,
      "detection best": 29891,
      "tweets proposed": 126051,
      "subject type": 114685,
      "classifier specifically": 17582,
      "target identification": 117624,
      "techniques investigated": 119912,
      "subtasks subtask": 114983,
      "subtasks used": 114987,
      "used linear": 128612,
      "classifier paper": 17568,
      "special characters": 111356,
      "target present": 117692,
      "approach description": 7481,
      "subtask based": 114942,
      "content enhanced": 21879,
      "arabic social": 8554,
      "hindi indian": 50465,
      "comments youtube": 18831,
      "comments comments": 18811,
      "finetuning performance": 44489,
      "ranks 1st": 95700,
      "experiments albert": 40763,
      "participated subtask": 85348,
      "english nonenglish": 37226,
      "order enrich": 82317,
      "effect applying": 34585,
      "highlight great": 50260,
      "duluth semeval2020": 34261,
      "identification english": 51385,
      "english logistic": 37197,
      "participated semeval2020": 85344,
      "results class": 101576,
      "public communication": 94239,
      "nearly impossible": 76735,
      "word choices": 132949,
      "corpusbased study": 24081,
      "subtask identifying": 114944,
      "identifying categories": 51584,
      "submissions achieve": 114750,
      "cases used": 15670,
      "introduce various": 56566,
      "participation semeval2020": 85363,
      "topperforming systems": 122690,
      "extremely successful": 42612,
      "ability better": 1489,
      "score points": 104104,
      "points performing": 87868,
      "performing team": 86975,
      "subdivided multiple": 114663,
      "preparing proposed": 89302,
      "like lstms": 64057,
      "achieved proposed": 2673,
      "connected neural": 21127,
      "using markov": 129860,
      "codemixed tamilenglish": 18187,
      "tamilenglish language": 117574,
      "additionally model": 3614,
      "codemixed malayalamenglish": 18178,
      "labels generated": 58607,
      "generated supervised": 47133,
      "ranked terms": 95652,
      "popularity social": 88155,
      "platforms increased": 87665,
      "hindi paper": 50470,
      "respectively coarse": 101129,
      "adoption social": 3907,
      "multilingual bidirectional": 75212,
      "platforms increasing": 87666,
      "led need": 63280,
      "need proper": 76848,
      "languages quite": 60830,
      "divide conquer": 32906,
      "showing challenging": 108398,
      "final ensemble": 44102,
      "classify different": 17649,
      "content types": 21965,
      "work leverages": 134611,
      "different transformer": 31512,
      "detection multilingual": 29995,
      "multilingual codemixed": 75216,
      "codemixed text": 18188,
      "using transformers": 130317,
      "years systems": 135307,
      "number researches": 80956,
      "identify positive": 51535,
      "identify social": 51555,
      "regression ensemble": 97701,
      "f1score 093": 42709,
      "detection paper submitted": 30019,
      "task ranked 6th": 118600,
      "multilingual societies like": 75372,
      "language document describes": 58977,
      "used linear svm": 128613,
      "models lstm transformer": 73536,
      "task goal task": 118247,
      "goal task identify": 48390,
      "present approach description": 89377,
      "sub task semeval": 114645,
      "task semeval 2019": 118678,
      "detailed analysis results": 29756,
      "analysis results obtained": 5812,
      "obtained using trained": 81429,
      "arabic social media": 8555,
      "data best models": 25701,
      "models shared task": 74021,
      "shared task test": 108099,
      "bert models available": 13176,
      "learning approach ensemble": 62346,
      "duluth semeval2020 task": 34262,
      "english logistic regression": 37198,
      "manually labeled training": 67068,
      "language detection important": 58952,
      "contextualized language model": 22554,
      "improves classification performance": 52961,
      "transformer bert model": 124309,
      "dataset propose new": 27113,
      "detection paper describes": 30013,
      "describes participation semeval2020": 29423,
      "participation semeval2020 task": 85364,
      "sharing parameters languages": 108138,
      "corpus level information": 23862,
      "achieve score points": 2531,
      "score points performing": 104105,
      "points performing team": 87869,
      "language identification automatic": 59127,
      "preparing proposed use": 89303,
      "proposed use deep": 93583,
      "learning networks like": 62835,
      "networks like lstms": 77644,
      "performance achieved proposed": 86115,
      "achieved proposed model": 2674,
      "team ranked 7th": 119719,
      "speech detection paper": 111673,
      "popularity social media": 88156,
      "multilabel multiclass classification": 75156,
      "adoption social media": 3908,
      "platforms like facebook": 87668,
      "classification problem propose": 17350,
      "problem propose effective": 91175,
      "multilingual bidirectional encoder": 75213,
      "preprocessing techniques pretrained": 89338,
      "lot work recently": 65334,
      "multiclass classification tasks": 75053,
      "classification tasks respectively": 17455,
      "recent years systems": 96640,
      "logistic regression ensemble": 65036,
      "paper present approach description": 84083,
      "sub task semeval 2019": 114646,
      "task semeval 2019 task": 118679,
      "provide detailed analysis results": 93803,
      "detailed analysis results obtained": 29757,
      "analysis results obtained using": 5813,
      "results obtained using trained": 102010,
      "manually labeled training data": 67069,
      "representation transformer bert model": 99450,
      "paper describes participation semeval2020": 83839,
      "describes participation semeval2020 task": 29424,
      "achieve score points performing": 2532,
      "score points performing team": 104106,
      "preparing proposed use deep": 89304,
      "proposed use deep learning": 93584,
      "use deep learning networks": 127983,
      "deep learning networks like": 28302,
      "learning networks like lstms": 62836,
      "performance achieved proposed model": 86116,
      "hate speech detection paper": 49535,
      "media platforms like facebook": 68145,
      "platforms like facebook twitter": 87669,
      "multilingual bidirectional encoder representations": 75214,
      "multilabel multiclass classification tasks": 75157,
      "sub task semeval 2019 task": 114647,
      "provide detailed analysis results obtained": 93804,
      "detailed analysis results obtained using": 29758,
      "analysis results obtained using trained": 5814,
      "encoder representation transformer bert model": 36562,
      "paper describes participation semeval2020 task": 83840,
      "achieve score points performing team": 2533,
      "preparing proposed use deep learning": 89305,
      "proposed use deep learning networks": 93585,
      "use deep learning networks like": 127984,
      "deep learning networks like lstms": 28303,
      "using pretrained language models paper": 130049,
      "social media platforms like facebook": 110406,
      "media platforms like facebook twitter": 68146,
      "manuallybuilt": 67089,
      "mismatches": 70311,
      "concode": 20733,
      "sketchbased": 109965,
      "8447": 1318,
      "wikisql": 132694,
      "cky": 17003,
      "closure": 17887,
      "machineinterpretable": 66318,
      "lossy": 65312,
      "instruction following": 55705,
      "model interpreting": 71365,
      "actions conditioned": 3090,
      "highquality lexicons": 50391,
      "lexicons manuallybuilt": 63928,
      "manuallybuilt templates": 67090,
      "domains meaning": 33813,
      "datapoints sampled": 26696,
      "paraphrasing existing": 84847,
      "domains adapt": 33725,
      "problems small": 91373,
      "domains crossdomain": 33753,
      "domain representation": 33626,
      "augmented attention": 10321,
      "space learning": 111014,
      "algorithm searches": 4795,
      "abstract patterns": 1778,
      "produce response": 91923,
      "available experiments": 10995,
      "confidence modeling": 20992,
      "scores indicate": 104180,
      "lowlevel information": 65477,
      "lstm encoder": 65617,
      "process mapping": 91523,
      "human automated": 50758,
      "parsing generative": 85118,
      "forms paper": 45256,
      "general solutions": 46715,
      "search procedure": 104338,
      "model exploration": 71143,
      "recently received": 96746,
      "directly pairs": 31895,
      "present logical": 89537,
      "finitestate automata": 44544,
      "parse user": 84891,
      "slot label": 110064,
      "sequencetosequence approaches": 107152,
      "approaches dataset": 8114,
      "methods query": 69705,
      "obtaining broad": 81437,
      "humans create": 51075,
      "triples utterances": 125821,
      "utterances meaning": 130655,
      "learns compose": 63199,
      "executable semantic": 39968,
      "languages amenable": 60400,
      "31 improvement": 921,
      "class environment": 17036,
      "metalearning strategy": 68558,
      "focus specifically": 44823,
      "representations utterance": 99969,
      "parsing technique": 85261,
      "inhouse dataset": 55204,
      "model grammatical": 71264,
      "grammarbased neural": 48669,
      "missing components": 70319,
      "systems completely": 116799,
      "classification determines": 17181,
      "seq2seq method": 106898,
      "programs execution": 92137,
      "respect certain": 101075,
      "task explicit": 118181,
      "proposed process": 93529,
      "unlike works": 127455,
      "parsing recently": 85214,
      "novel domain": 80549,
      "attention encourage": 9827,
      "model guided": 71270,
      "domain far": 33535,
      "new structures": 79188,
      "comparable seq2seq": 19173,
      "ii parsing": 51713,
      "strong seq2seq": 113710,
      "parsing taskoriented": 85259,
      "understanding users": 127007,
      "parsers lowresource": 85030,
      "techniques representation": 119972,
      "release public": 98469,
      "data captured": 25717,
      "substantial work": 114880,
      "importance building": 52046,
      "long acknowledged": 65060,
      "domains learning": 33803,
      "approach handle": 7595,
      "shown struggle": 108531,
      "grammarbased approach": 48668,
      "parsing survey": 85250,
      "parsing current": 85092,
      "parsing stateoftheart": 85242,
      "finetune bart": 44402,
      "conditioning extracted": 20810,
      "neural modules": 78208,
      "framework does": 45502,
      "major bottleneck": 66540,
      "contemporary neural": 21850,
      "texttosql parsing": 121666,
      "results synthesized": 102248,
      "preprocessing ontology": 89328,
      "parser 15": 84905,
      "specialized model": 111392,
      "pretraining seq2seq": 90321,
      "success paper": 115111,
      "texttosql datasets": 121665,
      "model interpreting natural": 71366,
      "neural attention semantic": 77846,
      "attention semantic parsing": 10003,
      "highquality lexicons manuallybuilt": 50392,
      "lexicons manuallybuilt templates": 63929,
      "domains meaning representations": 33814,
      "logical form paper": 65003,
      "adapt target domain": 3195,
      "augmented attention mechanism": 10322,
      "learn word meanings": 62193,
      "learn semantic parser": 62139,
      "based sentence similarity": 12017,
      "semantic parsing process": 105158,
      "target sentence using": 117699,
      "process mapping natural": 91524,
      "recently received lot": 96747,
      "trained directly pairs": 123119,
      "parse user utterances": 84892,
      "semantic parsing systems": 105170,
      "obtaining broad coverage": 81438,
      "training data domain": 123453,
      "utterances meaning representations": 130656,
      "training data domains": 123454,
      "systems completely fail": 116800,
      "aims map natural": 4551,
      "paper addresses issue": 83714,
      "tasks domain adaptation": 119073,
      "ability generalize new": 1523,
      "comparable seq2seq models": 19174,
      "strong seq2seq baselines": 113711,
      "semantic parsing taskoriented": 105173,
      "domains propose novel": 33835,
      "propose new decoding": 92844,
      "semantic parsing survey": 105167,
      "task translating natural": 118808,
      "open problems challenges": 81923,
      "challenges future research": 16163,
      "parsing stateoftheart semantic": 85243,
      "stateoftheart semantic parsers": 112944,
      "framework does require": 45503,
      "framework consistently improve": 45465,
      "encourages model consider": 36770,
      "correspondence natural language": 24270,
      "model interpreting natural language": 71367,
      "neural attention semantic parsing": 77847,
      "highquality lexicons manuallybuilt templates": 50393,
      "new stateoftheart performance standard": 79164,
      "semantic parsing semantic parsing": 105164,
      "process mapping natural language": 91525,
      "recently received lot attention": 96748,
      "parse user utterances semantic": 84893,
      "aims map natural language": 4552,
      "domains propose novel method": 33836,
      "model outperforms existing approaches": 71646,
      "task translating natural language": 118809,
      "parsing stateoftheart semantic parsers": 85244,
      "fnn": 44731,
      "entropy training": 38175,
      "class word": 17053,
      "wellknown types": 132424,
      "task argue": 117892,
      "adequate representation": 3833,
      "reranking machine": 100372,
      "allows discover": 5144,
      "dropout demonstrate": 34175,
      "partition function": 85509,
      "lstm units": 65673,
      "finds optimal": 44318,
      "characterlevel inputs": 16568,
      "training rnns": 123823,
      "results google": 101817,
      "framework constructing": 45470,
      "basic module": 12530,
      "significantly accurate": 108879,
      "seq2seq attention": 106887,
      "networks particular": 77694,
      "context integration": 22149,
      "gated architectures": 46502,
      "layer weights": 61754,
      "set weights": 107635,
      "model character": 70818,
      "dutch outperform": 34282,
      "methods advantage": 69300,
      "advantages proposed": 4087,
      "provides real": 94063,
      "popular architectures": 88077,
      "treebank language": 125639,
      "challenges make": 16174,
      "future words": 46319,
      "slight degradation": 110016,
      "simple structure": 109524,
      "applications introduce": 6947,
      "terms perplexity": 120358,
      "demonstrate tradeoff": 28890,
      "size large": 109930,
      "applying state": 7271,
      "codemixed corpus": 18167,
      "space deep": 110988,
      "novel multilayer": 80655,
      "layers evaluate": 61775,
      "trained obtain": 123224,
      "quickly small": 95395,
      "does add": 33325,
      "basic modules": 12531,
      "learning hierarchical": 62631,
      "problem encoding": 91026,
      "network computation": 77203,
      "time standard": 122111,
      "modeling introduced": 72453,
      "layer fuses": 61722,
      "models bring": 72861,
      "internal behavior": 56160,
      "popular choices": 88085,
      "network train": 77457,
      "modeling datasets": 72411,
      "capacity paper": 15236,
      "sources make": 110907,
      "estimation inferring": 38650,
      "revisiting simple": 102655,
      "concatenates word": 20536,
      "long training times": 65145,
      "learning techniques neural": 63095,
      "prediction propose novel": 89110,
      "extensive experiments text": 41933,
      "reranking machine translation": 100373,
      "modeling work explore": 72590,
      "dropout demonstrate effectiveness": 34176,
      "memory lstm units": 68347,
      "language models english": 59545,
      "network language modeling": 77298,
      "lstm based language": 65598,
      "information long time": 54738,
      "models tasks language": 74165,
      "larger number parameters": 61373,
      "based word prediction": 12167,
      "provides real time": 94064,
      "models establish new": 73167,
      "improving language modeling": 53108,
      "bidirectional recurrent network": 13969,
      "proposing novel neural": 93629,
      "language models experiments": 59551,
      "vocabulary size large": 131903,
      "input output embedding": 55381,
      "applying state art": 7272,
      "tasks inspired propose": 119211,
      "inspired propose novel": 55579,
      "proposed model benchmark": 93436,
      "ensemble neural language": 37615,
      "datasets indicate proposed": 27525,
      "finetuning language models": 44470,
      "different architectures datasets": 31004,
      "paper examine effect": 83901,
      "character ngram embeddings": 16453,
      "ordinary word embeddings": 82452,
      "method achieves best": 68592,
      "language modeling datasets": 59429,
      "nlp tasks long": 79737,
      "memory capacity paper": 68298,
      "modeling datasets demonstrate": 72412,
      "language modeling work explore": 59485,
      "shortterm memory lstm units": 108342,
      "models tasks language modeling": 74166,
      "tasks inspired propose novel": 119212,
      "neural network language modeling": 78316,
      "language modeling datasets demonstrate": 59430,
      "long shortterm memory lstm units": 65123,
      "resilient": 100744,
      "constants": 21520,
      "fillings": 44063,
      "reduce ambiguity": 97312,
      "phrases represent": 87448,
      "represent information": 99113,
      "combined single": 18669,
      "approach rapidly": 7835,
      "based user": 12150,
      "network encoderdecoder": 77240,
      "translation going": 124842,
      "rnn learn": 102923,
      "grammar structure": 48657,
      "new encoderdecoder": 78898,
      "network better": 77168,
      "improve result": 52524,
      "sentence formal": 105876,
      "views problem": 131725,
      "architecture set": 8747,
      "parsing study": 85248,
      "empowers learn": 36342,
      "addition observe": 3456,
      "logarithmic relationship": 64985,
      "parser training": 85001,
      "paradigm model": 84540,
      "propose knowledgeaware": 92738,
      "based expert": 11703,
      "typing model": 126477,
      "nlp aims": 79561,
      "require high": 100144,
      "bias accuracy": 13792,
      "accuracy existing": 2156,
      "helpful designing": 49795,
      "designing future": 29639,
      "wrong annotations": 135161,
      "explicit logic": 41332,
      "form use": 45141,
      "use bertbased": 127923,
      "structure graph": 113873,
      "outofdomain words": 82669,
      "performs tokenlevel": 87036,
      "tokenlevel decoding": 122289,
      "decoding shown": 28132,
      "logical formalisms": 65004,
      "evergrowing number": 39621,
      "filling gap": 44042,
      "98 accuracy": 1447,
      "question translate": 95230,
      "time zeroshot": 122140,
      "set database": 107412,
      "messagepassing graph": 68502,
      "decoder incorporate": 28023,
      "amounts supervised": 5356,
      "knowledge limited": 58054,
      "number simple": 80967,
      "available access": 10933,
      "guarantee accuracy": 49214,
      "accuracy stability": 2283,
      "parsing highly": 85124,
      "multiple tables": 75719,
      "task area": 117891,
      "questions arise": 95276,
      "addition feature": 3430,
      "problem converting": 90985,
      "manipulation methods": 66930,
      "presented improve": 89786,
      "feedback given": 43830,
      "feedback correct": 43824,
      "dataset utterances": 27264,
      "flexibility natural": 44651,
      "leaves large": 63266,
      "improvement future": 52713,
      "argue previous": 8926,
      "stage propose": 112153,
      "place leaderboard": 87604,
      "networkbased semantic": 77492,
      "method helps": 68871,
      "increasingly sophisticated": 53712,
      "human input": 50865,
      "tool allow": 122369,
      "dataset resulting": 27166,
      "main content": 66408,
      "fully understanding": 45975,
      "crossdomain datasets": 24888,
      "information user": 55078,
      "importance different": 52054,
      "history paper": 50565,
      "capable effectively": 15197,
      "methods understanding": 69821,
      "language computers": 58907,
      "detailed performance": 29781,
      "groundtruth annotation": 49128,
      "usefulness annotated": 128954,
      "works require": 134974,
      "questions major": 95328,
      "baselines hierarchical": 12408,
      "setting users": 107784,
      "users correct": 129106,
      "neural network encoderdecoder": 78297,
      "machine translation going": 65994,
      "new semantic features": 79123,
      "neural network better": 78265,
      "input neural network": 55374,
      "new learning paradigm": 78991,
      "learning paradigm model": 62873,
      "based expert knowledge": 11704,
      "processing nlp aims": 91726,
      "require high quality": 100145,
      "outperforming previous state": 82819,
      "machine comprehension model": 65751,
      "training time zeroshot": 123931,
      "messagepassing graph neural": 68503,
      "structures different domains": 114070,
      "research questions arise": 100605,
      "accuracy achieve stateoftheart": 2092,
      "flexibility natural language": 44652,
      "neural networkbased semantic": 78431,
      "drawn attention recent": 34132,
      "corpora train large": 23604,
      "train large neural": 122946,
      "pilot study observe": 87511,
      "generation models generate": 47484,
      "previous works require": 90577,
      "strong baselines hierarchical": 113649,
      "discuss directions improvement": 32248,
      "language processing nlp aims": 59891,
      "outperforming previous state art": 82820,
      "using large pretrained models": 129789,
      "drawn attention recent years": 34133,
      "proposed method achieves better": 93340,
      "demonstrate proposed framework outperforms": 28839,
      "natural language processing nlp aims": 76440,
      "proposed method achieves better performance": 93341,
      "80s": 1296,
      "irstlm": 56957,
      "idiomaticity": 51653,
      "manmachine": 66936,
      "assertive": 9467,
      "mission": 70339,
      "mumbais": 75957,
      "cdacm": 15870,
      "collocated": 18514,
      "bing": 14145,
      "hybridization": 51201,
      "56k": 1124,
      "timetaking": 122206,
      "number machine": 80909,
      "results hindi": 101829,
      "sense use": 105703,
      "contextual feature": 22464,
      "activity machine": 3133,
      "article shall": 9157,
      "shall discuss": 107908,
      "evaluation important": 39238,
      "systems like": 116979,
      "like google": 64043,
      "translation various": 125422,
      "data basis": 25692,
      "good number": 48487,
      "number foreign": 80892,
      "languages needs": 60761,
      "harder english": 49486,
      "introduced translation": 56602,
      "correct mt": 24108,
      "features ranking": 43685,
      "difficulty handling": 31701,
      "sentence change": 105783,
      "researches shown": 100723,
      "preprocessing used": 89340,
      "machinereadable dictionaries": 66327,
      "rules ensure": 103396,
      "utilizing machine": 130572,
      "information material": 54752,
      "languages currently": 60478,
      "translation available": 124662,
      "language make": 59269,
      "assertive sentences": 9468,
      "simplification model": 109587,
      "describes algorithm": 29384,
      "advanced nonnative": 3955,
      "nonnative language": 80245,
      "language reflecting": 60030,
      "centre development": 15911,
      "advanced computing": 3937,
      "computing mumbais": 20519,
      "mumbais cdacm": 75958,
      "cdacm submission": 15871,
      "submission nlp": 114730,
      "collocated icon": 18515,
      "model reordering": 71902,
      "rules better": 103385,
      "better reordering": 13694,
      "process better": 91432,
      "used bleu": 128413,
      "develop statistical": 30235,
      "presents centre": 89825,
      "2015 collocated": 584,
      "icon 2015": 51309,
      "translation comparison": 124701,
      "hybrid machine": 51180,
      "presented shows": 89797,
      "work case": 134408,
      "evaluation performance": 39319,
      "verb phrases": 131524,
      "mt present": 74984,
      "performance detailed": 86290,
      "language perspective": 59810,
      "systems outperforms": 117039,
      "counterpart machine": 24435,
      "resolution english": 100760,
      "smt approaches": 110279,
      "translating morphologically": 124593,
      "models correctly": 72993,
      "automatic subjective": 10676,
      "resources help": 100986,
      "vocabulary provide": 131894,
      "sentences monolingual": 106401,
      "based moses": 11859,
      "englishhindi englishbengali": 37406,
      "complexity order": 19931,
      "consequently propose": 21184,
      "various transformer": 131228,
      "study languages": 114421,
      "translation involving": 124879,
      "costly timetaking": 24393,
      "vocabulary problem": 131893,
      "vocabulary oov": 131889,
      "address oov": 3721,
      "translation specific": 125280,
      "number machine translation": 80910,
      "activity machine translation": 3134,
      "human automatic evaluation": 50760,
      "machine translation development": 65951,
      "quality automatic evaluation": 94600,
      "subjective objective evaluation": 114699,
      "machine translation research": 66201,
      "translation systems like": 125312,
      "systems like google": 116980,
      "machine translation various": 66297,
      "model using features": 72277,
      "labels test data": 58648,
      "languages morphologically rich": 60739,
      "correct mt output": 24109,
      "language independent models": 59164,
      "using statistical methods": 130224,
      "improve accuracy translation": 52332,
      "nlp machine translation": 79633,
      "societies like india": 110471,
      "takes input text": 117535,
      "text simplification model": 121303,
      "translation systems used": 125320,
      "present computational analysis": 89417,
      "centre development advanced": 15912,
      "development advanced computing": 30368,
      "advanced computing mumbais": 3938,
      "computing mumbais cdacm": 20520,
      "mumbais cdacm submission": 75959,
      "cdacm submission nlp": 15872,
      "submission nlp tools": 114731,
      "observed significant improvements": 81234,
      "paper presents centre": 84153,
      "presents centre development": 89826,
      "2015 collocated icon": 585,
      "collocated icon 2015": 18516,
      "hybrid machine translation": 51181,
      "counterpart machine translation": 24436,
      "machine translation tackle": 66258,
      "translating morphologically rich": 124594,
      "observed significant improvement": 81233,
      "english evaluate performance": 37130,
      "terms automatic evaluation": 120271,
      "level nmt models": 63482,
      "effective neural machine": 34722,
      "attentionbased encoderdecoder neural": 10073,
      "translation effective approach": 124776,
      "using multihead selfattention": 129908,
      "used bleu score": 128414,
      "address oov problem": 3722,
      "generating new words": 47242,
      "machine translation specific": 66234,
      "quality machine translation outputs": 94712,
      "machine translation systems like": 66250,
      "translation systems like google": 125313,
      "machine translation systems used": 66256,
      "centre development advanced computing": 15913,
      "development advanced computing mumbais": 30369,
      "advanced computing mumbais cdacm": 3939,
      "computing mumbais cdacm submission": 20521,
      "mumbais cdacm submission nlp": 75960,
      "cdacm submission nlp tools": 15873,
      "submission nlp tools contest": 114732,
      "paper presents centre development": 84154,
      "presents centre development advanced": 89827,
      "2015 collocated icon 2015": 586,
      "translating morphologically rich languages": 124595,
      "effective neural machine translation": 34723,
      "machine translation effective approach": 65964,
      "machine translation systems like google": 66251,
      "centre development advanced computing mumbais": 15914,
      "development advanced computing mumbais cdacm": 30370,
      "advanced computing mumbais cdacm submission": 3940,
      "computing mumbais cdacm submission nlp": 20522,
      "mumbais cdacm submission nlp tools": 75961,
      "cdacm submission nlp tools contest": 15874,
      "paper presents centre development advanced": 84155,
      "presents centre development advanced computing": 89828,
      "based statistical machine translation smt": 12074,
      "lagged": 58781,
      "positivenegative": 88352,
      "sir": 109870,
      "gyafc": 49291,
      "onthe": 81823,
      "inflexibility": 54300,
      "deanonymize": 27881,
      "mir": 70282,
      "textediting": 121435,
      "forwardtranslated": 45333,
      "data pretrain": 26269,
      "transfer nonparallel": 124168,
      "metrics tasks": 70012,
      "score comparing": 104063,
      "benchmarks metrics": 12920,
      "formality style": 45190,
      "metrics sentiment": 70006,
      "way expressed": 132083,
      "metrics response": 70003,
      "loss achieve": 65251,
      "style adaptation": 114566,
      "variety writing": 131032,
      "power model": 88637,
      "observation text": 81171,
      "inputs best": 55479,
      "rephrasing text": 98926,
      "aimed capturing": 4480,
      "stylistic similarity": 114632,
      "using wider": 130375,
      "trained explicitly": 123143,
      "semantics evaluate": 105414,
      "corpora compare": 23436,
      "compare automatic": 19225,
      "terms meaning": 120345,
      "appropriate style": 8441,
      "sentence style": 106094,
      "disentangled latent": 32347,
      "rephrases text": 98923,
      "systems control": 116815,
      "classifier introduced": 17552,
      "preservation style": 89918,
      "better improving": 13604,
      "transfer furthermore": 124055,
      "transferring text": 124246,
      "metrics generated": 69964,
      "quality evaluating": 94656,
      "metrics facilitate": 69961,
      "content desired": 21869,
      "margin especially": 67191,
      "data simultaneously": 26456,
      "different attributes": 31017,
      "aspects time": 9412,
      "use encoderdecoder": 128026,
      "techniques fail": 119888,
      "produce appropriate": 91874,
      "baselines semantic": 12458,
      "authors furthermore": 10380,
      "ubiquitous daily": 126509,
      "algorithm explicitly": 4732,
      "sentences scratch": 106484,
      "building new": 14872,
      "bert propose": 13211,
      "evaluations experimental": 39461,
      "sentences style": 106507,
      "quality validate": 94825,
      "realworld textual": 96186,
      "unsupervised generation": 127641,
      "generation significant": 47622,
      "transferred sentences": 124234,
      "content similarity": 21946,
      "bring researchers": 14633,
      "zeroshot finegrained": 135501,
      "transfer unseen": 124204,
      "model rewrite": 71933,
      "involves converting": 56889,
      "accuracy additionally": 2096,
      "grammaticality meaning": 48729,
      "obtain useful": 81335,
      "content tokens": 21963,
      "approach adapting": 7330,
      "computer generated": 20481,
      "competitively fully": 19702,
      "unsupervised style": 127725,
      "inputs meaning": 55485,
      "diverse styles": 32850,
      "styles use": 114622,
      "availability dataset": 10904,
      "loss especially": 65262,
      "encoderdecoder setup": 36623,
      "models discriminators": 73088,
      "applications field": 6930,
      "create taxonomy": 24645,
      "trends provide": 125754,
      "transfer direct": 124047,
      "makes impossible": 66787,
      "data dealing": 25818,
      "transfer code": 124040,
      "reliable method": 98617,
      "differences style": 30976,
      "style transfer nonparallel": 114604,
      "representations using adversarial": 99966,
      "formality style transfer": 45191,
      "style transfer task": 114606,
      "evaluation metrics response": 39294,
      "sentiment style transfer": 106799,
      "motivated observation text": 74857,
      "create benchmark dataset": 24603,
      "writing style using": 135100,
      "using sequence models": 130158,
      "input text explicitly": 55452,
      "natural language requires": 76525,
      "validate effectiveness model": 130717,
      "representation learning method": 99296,
      "stateoftheart models terms": 112751,
      "preservation style transfer": 89919,
      "sentencealigned parallel data": 106138,
      "unsupervised text style": 127737,
      "paper address challenge": 83702,
      "existing methods typically": 40197,
      "reinforcement learning use": 97830,
      "stateoftheart systems large": 112991,
      "large margin especially": 61144,
      "datasets human evaluations": 27507,
      "transfer nonparallel text": 124169,
      "positive negative sentences": 88334,
      "evaluations experimental results": 39462,
      "domain adaptation text": 33458,
      "meaning challenging problem": 67620,
      "real use cases": 96084,
      "existing methods human": 40185,
      "grammaticality meaning preservation": 48730,
      "provide insights future": 93860,
      "work introduce method": 134576,
      "real applications previous": 96055,
      "performs competitively fully": 86995,
      "unsupervised style transfer": 127726,
      "style transfer systems": 114605,
      "language models discriminators": 59540,
      "unsupervised text style transfer": 127738,
      "stateoftheart systems large margin": 112992,
      "demonstrate model achieves competitive": 28790,
      "important problem natural language processing": 52214,
      "demonstrate model achieves competitive performance": 28791,
      "rap": 95706,
      "artist": 9275,
      "displaced": 32376,
      "lfg": 63946,
      "treasure": 125548,
      "divine": 32918,
      "sonic": 110662,
      "explicitness": 41402,
      "synthesising": 116603,
      "opera": 82046,
      "tsinghua": 125909,
      "styled": 114615,
      "texts additional": 121452,
      "paper applies": 83736,
      "machine generate": 65757,
      "expressions used": 41767,
      "interpretation introduce": 56263,
      "simplification goal": 109582,
      "easily understandable": 34466,
      "creative writing": 24733,
      "control outputs": 22816,
      "compared classical": 19342,
      "approach unable": 7990,
      "complete semantic": 19767,
      "keywords text": 57649,
      "work produce": 134706,
      "text dialogue": 120872,
      "level syntactic": 63509,
      "convey specific": 23118,
      "function speech": 46045,
      "intonational contours": 56333,
      "approach given": 7590,
      "potential text": 88589,
      "different phases": 31333,
      "pipeline corpus": 87535,
      "generation sentiment": 47613,
      "humanwritten poems": 51129,
      "evaluation reveals": 39372,
      "reading learning": 96022,
      "memory processing": 68368,
      "reasonable good": 96209,
      "development technology": 30430,
      "research applied": 100406,
      "approach novel": 7744,
      "parameter controls": 84712,
      "approach alongside": 7351,
      "technology automatically": 120038,
      "method increase": 68900,
      "entities twitter": 37882,
      "twitter content": 126098,
      "characterize certain": 16548,
      "syllablebased neural": 116305,
      "available huge": 11012,
      "expressions deep": 41751,
      "input unlike": 55463,
      "200 thousand": 535,
      "video paper": 131659,
      "improves diversity": 52972,
      "generation generation": 47419,
      "weighting method": 132363,
      "forms longer": 45249,
      "generation developed": 47366,
      "tsinghua university": 125910,
      "defining characteristic": 28512,
      "training transformerbased": 123937,
      "preserve essential": 89921,
      "essential meaning": 38556,
      "literature focus": 64756,
      "current stateofthe": 25348,
      "domain concepts": 33482,
      "personality text": 87131,
      "features single": 43732,
      "used experimentation": 128532,
      "previous benchmark": 90390,
      "performance understand": 86821,
      "conditioned given": 20800,
      "vae text": 130701,
      "intended serve": 55870,
      "model dedicated": 70956,
      "problematic cases": 91291,
      "possibility automatic": 88367,
      "collection work": 18499,
      "direct interaction": 31788,
      "building annotating": 14822,
      "different bertbased": 31030,
      "beneficial task": 12953,
      "challenges introduce": 16168,
      "features usually": 43779,
      "autoencoder framework": 10403,
      "features generate": 43523,
      "able offer": 1666,
      "attentionbased sequencetosequence model": 10101,
      "goal paper develop": 48378,
      "text simplification goal": 121302,
      "information word structure": 55099,
      "semantic level addition": 105097,
      "model different settings": 70996,
      "stateoftheart models generating": 112734,
      "effectiveness proposed strategy": 34938,
      "defining characteristic human": 28513,
      "current stateofthe art": 25349,
      "autoencoder vae text": 10413,
      "opens new perspectives": 82010,
      "data augmentation sentiment": 25644,
      "beneficial task relations": 12954,
      "address challenges introduce": 3662,
      "generation outperforms stateoftheart": 47525,
      "propose hierarchical attention": 92710,
      "variational autoencoder framework": 130912,
      "variational autoencoder vae text": 130917,
      "likeliness": 64131,
      "backtransliteration": 11346,
      "seame": 104286,
      "mein": 68264,
      "sentimentlabeled": 106828,
      "backup": 11347,
      "1647": 387,
      "mandarinenglish": 66915,
      "iden": 51353,
      "community goal": 19076,
      "ones machine": 81694,
      "developing method": 30346,
      "word borrowed": 132934,
      "borrowing social": 14476,
      "switching languages": 116295,
      "codeswitching languages": 18237,
      "technologies multilingual": 120025,
      "likeliness word": 64132,
      "compare rankings": 19289,
      "rank correlation": 95623,
      "nearly 062": 76731,
      "speakers switch": 111326,
      "languages communication": 60452,
      "codeswitched speech": 18221,
      "labeled automatically": 58426,
      "exploiting monolingual": 41474,
      "annotated universal": 6255,
      "dependencies language": 29092,
      "modeling codeswitched": 72393,
      "communities limited": 19061,
      "intersentential intrasentential": 56314,
      "codeswitching text": 18241,
      "hindienglish codeswitched": 50486,
      "recognition codeswitching": 96840,
      "phenomenon mixing": 87252,
      "social constraints": 110328,
      "data differ": 25840,
      "degrades performance": 28564,
      "identification normalization": 51411,
      "dependencies scheme": 29110,
      "neural stacking": 78687,
      "treebanks present": 125665,
      "issue codeswitching": 56992,
      "introduce multitask": 56464,
      "perplexity seame": 87073,
      "resulting codemixed": 101434,
      "conversations specifically": 23071,
      "hindienglish bengalienglish": 50479,
      "implementations publicly": 51956,
      "identification tool": 51453,
      "used codemixed": 128427,
      "networks combining": 77540,
      "furthermore outperforms": 46197,
      "language tagging": 60142,
      "tagging codemixed": 117377,
      "models article": 72771,
      "tagger codemixed": 117356,
      "mechanism language": 68004,
      "capture real": 15366,
      "sentences improve": 106348,
      "perplexity score": 87071,
      "languages speech": 60889,
      "performance codeswitching": 86214,
      "code switching": 18156,
      "data overview": 26204,
      "discuss common": 32241,
      "improved sentiment": 52640,
      "codeswitched text": 18222,
      "text manually": 121100,
      "spans selected": 111210,
      "augmenting scarce": 10344,
      "improvements sentiment": 52908,
      "sentiment labeling": 106752,
      "text codeswitching": 120793,
      "adequately address": 3837,
      "constituent languages": 21540,
      "diverse dataset": 32804,
      "dataset written": 27281,
      "explore questions": 41580,
      "mainly addressed": 66478,
      "codeswitching cs": 18230,
      "understudied phenomenon": 127029,
      "codeswitched language": 18219,
      "mixing different": 70415,
      "multiple monolingual": 75619,
      "englishspanish codeswitching": 37428,
      "data complexity": 25758,
      "structure linguistic": 113900,
      "cope issue": 23274,
      "parallel monolingual": 84671,
      "limited source": 64282,
      "switch language": 116277,
      "modeling bilingual": 72383,
      "code switched": 18155,
      "phenomenon speaker": 87259,
      "control measures": 22811,
      "statistics various": 113211,
      "facilitate management": 42779,
      "egyptian dialectal": 35213,
      "systems performed": 117063,
      "task codemixed": 117969,
      "monolingual translation": 74627,
      "code mixing": 18115,
      "scale public": 103752,
      "languages attempt": 60414,
      "identifying english": 51596,
      "training cs": 123414,
      "synthesis machine": 116596,
      "dataset codemixed": 26791,
      "hinglish social": 50501,
      "media codemixed": 68088,
      "codemixing phenomenon": 18193,
      "phenomenon using": 87260,
      "pattern communication": 85712,
      "adds challenge": 3819,
      "text pos": 121182,
      "cs languages": 25184,
      "languages critical": 60474,
      "examples cs": 39819,
      "closer real": 17875,
      "cs text": 25185,
      "analysis benchmark": 5513,
      "noncanonical data": 80168,
      "languages utterance": 60959,
      "cs frequently": 25183,
      "normalization systems": 80347,
      "normalization pos": 80343,
      "tagging results": 117442,
      "mixed data": 70394,
      "intrasentential code": 56344,
      "world people": 135042,
      "people know": 85878,
      "particularly significant": 85495,
      "data considering": 25775,
      "mixed languages": 70399,
      "ones machine translation": 81695,
      "borrowing social media": 14477,
      "language speech technologies": 60117,
      "speech technologies multilingual": 111821,
      "technologies multilingual communities": 120026,
      "computational methods identify": 20406,
      "universal partofspeech tags": 127318,
      "present data set": 89428,
      "annotated universal dependencies": 6256,
      "language modeling codeswitched": 59428,
      "monolingual language model": 74593,
      "methods mitigate effects": 69621,
      "universal dependencies scheme": 127290,
      "introduce multitask learning": 56465,
      "model able identify": 70517,
      "outperforms standard lstm": 82990,
      "people speak language": 85892,
      "implementations publicly available": 51957,
      "language identification tool": 59147,
      "complex nlp systems": 19849,
      "neural networks combining": 78457,
      "using copy mechanism": 129504,
      "mechanism language modeling": 68005,
      "language models challenging": 59527,
      "generation generative adversarial": 47421,
      "improve performance codeswitching": 52460,
      "shared task shared": 108094,
      "significant improvements sentiment": 108803,
      "codeswitched text codeswitching": 18223,
      "continuous latent space": 22621,
      "large diverse dataset": 61082,
      "media platforms paper": 68148,
      "word embeddings cases": 133061,
      "speech recognition codeswitching": 111745,
      "language modeling bilingual": 59426,
      "data annotation important": 25605,
      "quality control measures": 94618,
      "egyptian dialectal arabic": 35214,
      "codemixed data social": 18169,
      "processing models work": 91714,
      "monolingual data lack": 74573,
      "speech synthesis machine": 111811,
      "synthesis machine translation": 116597,
      "hinglish social media": 50502,
      "social media codemixed": 110357,
      "codemixing phenomenon using": 18194,
      "text pos tagging": 121183,
      "multilingual models finetuned": 75288,
      "neural models generating": 78178,
      "data standard language": 26500,
      "using multiple languages": 129920,
      "multiple languages utterance": 75599,
      "languages utterance called": 60960,
      "codeswitching cs frequently": 18231,
      "normalization pos tagging": 80344,
      "code mixed data": 18111,
      "language identification best": 59128,
      "speech technologies multilingual communities": 111822,
      "generation generative adversarial networks": 47422,
      "shared task shared task": 108095,
      "achieve significant improvements sentiment": 2540,
      "different language pairs englishhindi": 31194,
      "social media platforms paper": 110408,
      "codemixed data social media": 18170,
      "language processing models work": 59889,
      "recognition speech synthesis machine": 97009,
      "speech synthesis machine translation": 111812,
      "using multiple languages utterance": 129921,
      "multiple languages utterance called": 75600,
      "speech recognition speech synthesis machine": 111778,
      "recognition speech synthesis machine translation": 97010,
      "tagging named entity recognition sentiment": 117419,
      "ideograms": 51642,
      "touched": 122704,
      "imaginary": 51818,
      "evidently": 39685,
      "sentimentbearing": 106827,
      "stemmed": 113233,
      "builders": 14818,
      "singlish": 109858,
      "envisioned": 38204,
      "velocity": 131507,
      "stopwords": 113376,
      "27000": 855,
      "word algorithm": 132905,
      "algorithm performance": 4783,
      "verified experiment": 131574,
      "users produce": 129158,
      "documents expressing": 33230,
      "expressing opinions": 41733,
      "semantic values": 105348,
      "nouns appropriate": 80440,
      "sentiment strength": 106797,
      "words tagged": 134254,
      "prior polarity": 90722,
      "polarity word": 87932,
      "broadcoverage resources": 14694,
      "problem compute": 90976,
      "literature assess": 64745,
      "great sentiment": 49022,
      "subjective language": 114695,
      "polarity lexicon": 87914,
      "adjectives analysis": 3862,
      "highly subjective": 50351,
      "opinions sentiment": 82131,
      "analysis discuss": 5580,
      "details various": 29792,
      "literature various": 64781,
      "movie product": 74899,
      "news opinion": 79352,
      "expressed target": 41724,
      "expressed explicitly": 41713,
      "extraction domainspecific": 42313,
      "work utilizes": 134876,
      "bootstrapping mechanism": 14458,
      "propagation algorithm": 92390,
      "combining sentiment": 18748,
      "political views": 88005,
      "events social": 39600,
      "social communication": 110327,
      "understanding potential": 126920,
      "studies obtained": 114258,
      "terms dictionary": 120306,
      "knowledge prior": 58121,
      "tradeoff precision": 122781,
      "lexicon manually": 63897,
      "manually built": 67041,
      "behavior given": 12658,
      "gain obtained": 46347,
      "process developing": 91459,
      "domain used": 33689,
      "150 years": 358,
      "stemmed text": 113234,
      "domain research": 33628,
      "results sa": 102143,
      "thorough understanding": 121898,
      "competitive robust": 19687,
      "annotated ones": 6217,
      "article addresses": 9107,
      "analysis processes": 5773,
      "negation handling": 76914,
      "classifier ensemble": 17541,
      "problems sentiment": 91371,
      "domains contextual": 33751,
      "built manually": 14926,
      "ranked higher": 95645,
      "bad ones": 11355,
      "created training": 24687,
      "models documentlevel": 73102,
      "including literature": 53312,
      "score 16": 104047,
      "original author": 82504,
      "resources sentiment": 101043,
      "situations require": 109901,
      "combine popular": 18638,
      "close best": 17810,
      "analysis stance": 5870,
      "assigning finegrained": 9556,
      "domains general": 33785,
      "set annotators": 107361,
      "veracity claim": 131511,
      "constantly growing": 21519,
      "need generating": 76814,
      "datasets particular": 27614,
      "annotation different": 6300,
      "including documentlevel": 53288,
      "sentences users": 106536,
      "time sentiment": 122099,
      "exists plethora": 40349,
      "sentiment identification": 106743,
      "news spread": 79371,
      "established method": 38602,
      "workshop 2016": 134997,
      "presents text": 89909,
      "classification achieves": 17102,
      "techniques prove": 119961,
      "study feature": 114382,
      "performance feature": 86380,
      "value corresponding": 130776,
      "known class": 58294,
      "models finegrained": 73251,
      "corpus professionally": 23941,
      "products movies": 92072,
      "examples present": 39866,
      "analysis interannotator": 5659,
      "analysis german": 5633,
      "thesis explores": 121850,
      "german tweets": 47922,
      "methods dictionary": 69446,
      "task examine": 118163,
      "examine common": 39738,
      "sources targets": 110923,
      "online consumer": 81752,
      "natural concepts": 76244,
      "ambiguous sentiment": 5297,
      "conceptnet semantic": 20600,
      "polarity annotations": 87905,
      "baselines validate": 12488,
      "validate usefulness": 130731,
      "great sentiment analysis": 49023,
      "sentiment analysis discuss": 106585,
      "topics sentiment analysis": 122656,
      "literature various approaches": 64782,
      "polarity sentiment analysis": 87928,
      "considered sentiment analysis": 21300,
      "experimental results domains": 40608,
      "combining sentiment analysis": 18749,
      "opinion mining applications": 82090,
      "time paper propose": 122065,
      "learning framework called": 62595,
      "word usage frequency": 133621,
      "largescale empirical studies": 61432,
      "problems sentiment analysis": 91372,
      "sentiment analysis new": 106617,
      "sentiment analysis far": 106596,
      "comments social media": 18827,
      "best results considered": 13440,
      "sentiment analysis stance": 106663,
      "analysis stance detection": 5871,
      "using supervised semisupervised": 130242,
      "semisupervised unsupervised methods": 105632,
      "annotated sentiment corpus": 6232,
      "detection twitter data": 30088,
      "fake news spread": 43019,
      "learning techniques prove": 63097,
      "movie review datasets": 74901,
      "does rely human": 33388,
      "models finegrained sentiment": 73252,
      "analysis interannotator agreement": 5660,
      "methods outperform stateoftheart": 69657,
      "using data explore": 129531,
      "classification major issue": 17263,
      "embedding model based": 35448,
      "conceptnet semantic network": 20601,
      "sentiment polarity annotations": 106771,
      "data set annotated": 26399,
      "experiments datasets different domains": 40870,
      "sentiment analysis stance detection": 106664,
      "analysis opinion mining important": 5731,
      "sentiment analysis opinion mining important": 106621,
      "milestones": 70087,
      "incongruity": 53432,
      "turkers": 125956,
      "reverses": 102529,
      "humour": 51146,
      "words employed": 133934,
      "sentiwordnet sentiment": 106841,
      "conducted machine": 20934,
      "compilation past": 19709,
      "observe milestones": 81202,
      "milestones research": 70088,
      "sentiment use": 106814,
      "discuss representative": 32284,
      "performance values": 86832,
      "values shared": 130803,
      "pointers future": 87835,
      "given prior": 48090,
      "useful illustrations": 128892,
      "table summarizes": 117262,
      "techniques data": 119859,
      "subtle forms": 114999,
      "comparison embeddings": 19540,
      "approach past": 7791,
      "naive baseline": 76012,
      "book snippets": 14416,
      "baselines comparison": 12373,
      "expressed subtle": 41723,
      "ways requires": 132173,
      "topics work": 122667,
      "linguistic stylistic": 64562,
      "algorithms evaluation": 4849,
      "present sign": 89697,
      "automatic measures": 10589,
      "scored higher": 104143,
      "discussion future": 32322,
      "detecting incongruity": 29841,
      "case performance": 15605,
      "news highly": 79345,
      "performing classifier": 86946,
      "develop theory": 30244,
      "use number": 128179,
      "number common": 80854,
      "portions text": 88188,
      "dialog understand": 30616,
      "twitter represent": 126121,
      "different discourse": 31097,
      "083 f1": 83,
      "linguistic pattern": 64517,
      "identifying classifying": 51589,
      "establishing comprehensive": 38615,
      "fit type": 44583,
      "detection tweets": 30086,
      "tweets particular": 126047,
      "pretraining schemes": 90317,
      "improve obtained": 52445,
      "tweets finally": 126032,
      "detection mainly": 29986,
      "person person": 87104,
      "utilizes user": 130563,
      "corpus englishhindi": 23775,
      "texts social": 121612,
      "text gaining": 120971,
      "gaining importance": 46375,
      "traditionally designed": 122888,
      "maximum performance": 67546,
      "approaches existing": 8151,
      "detection identify": 29967,
      "sentences identify": 106345,
      "context consider": 22034,
      "discuss differences": 32245,
      "based elmo": 11669,
      "offering competitive": 81578,
      "humorous responses": 51145,
      "carefully curated": 15519,
      "learn patterns": 62109,
      "chatbot human": 16665,
      "context detecting": 22053,
      "using author": 129362,
      "research develop": 100464,
      "context metadata": 22177,
      "network implemented": 77279,
      "metaphorical meaning": 68565,
      "meaning content": 67627,
      "advanced deep": 3941,
      "proposed pretrained": 93522,
      "linguistic observation": 64515,
      "potentially helpful": 88614,
      "similar nlp": 109114,
      "conversation thread": 22979,
      "datasets social": 27722,
      "baselines best": 12364,
      "highest performing": 50229,
      "reddit twitter": 97303,
      "people world": 85905,
      "tweets training": 126063,
      "approaches experimented": 8154,
      "multilingual perspective": 75333,
      "perspective english": 87161,
      "demonstrate modeling": 28800,
      "aims developing": 4519,
      "pre processing": 88767,
      "experimenting multiple": 40750,
      "conducted machine learning": 20935,
      "compilation past work": 19710,
      "observe milestones research": 81203,
      "milestones research far": 70089,
      "discuss representative performance": 32285,
      "representative performance values": 100017,
      "performance values shared": 86833,
      "values shared tasks": 130804,
      "pointers future work": 87836,
      "future work given": 46323,
      "work given prior": 134546,
      "given prior works": 48091,
      "key task natural": 57604,
      "finally apply model": 44147,
      "detection paper propose": 30017,
      "challenging research problem": 16306,
      "algorithms evaluation measures": 4850,
      "best performing classifier": 13401,
      "input features including": 55338,
      "english tweets paper": 37322,
      "external features provide": 42003,
      "improve obtained results": 52446,
      "data nlp tasks": 26172,
      "texts social media": 121613,
      "text poses challenge": 121185,
      "developed using dataset": 30318,
      "outofdomain training data": 82667,
      "test model different": 120473,
      "multitask learning sentiment": 75866,
      "potential future research": 88558,
      "investigate impact using": 56765,
      "advanced deep learning": 3942,
      "tackle problem identifying": 117304,
      "recently proposed pretrained": 96741,
      "datasets social media": 27723,
      "improvements baselines best": 52813,
      "limited availability data": 64210,
      "multilingual perspective english": 75334,
      "perspective english spanish": 87162,
      "stateoftheart results widely": 112936,
      "widely used benchmark": 132564,
      "used benchmark datasets": 128409,
      "tweets paper presents": 126046,
      "conducted machine learning algorithms": 20936,
      "observe milestones research far": 81204,
      "discuss representative performance values": 32286,
      "representative performance values shared": 100018,
      "performance values shared tasks": 86834,
      "pointers future work given": 87837,
      "future work given prior": 46324,
      "work given prior works": 134547,
      "key task natural language": 57605,
      "paper present novel task": 84127,
      "multilingual perspective english spanish": 75335,
      "stateoftheart results widely used": 112937,
      "results widely used benchmark": 102333,
      "widely used benchmark datasets": 132565,
      "discuss representative performance values shared": 32287,
      "representative performance values shared tasks": 100019,
      "pointers future work given prior": 87838,
      "future work given prior works": 46325,
      "key task natural language processing": 57606,
      "task natural language processing works": 118442,
      "results widely used benchmark datasets": 102334,
      "codalab": 18068,
      "domainadaptation": 33701,
      "fares": 43106,
      "copyaugmented": 23294,
      "singlecorpus": 109822,
      "bea": 12595,
      "clumsy": 17922,
      "personalize": 87139,
      "uncorrected": 126646,
      "proofreaders": 92376,
      "showing gains": 108405,
      "problems arise": 91299,
      "metric does": 69876,
      "comes ability": 18775,
      "text transformations": 121376,
      "discrete word": 32183,
      "achieve statistically": 2570,
      "correction grammatical": 24153,
      "detecting correcting": 29829,
      "benchmark test": 12866,
      "rely goldstandard": 98700,
      "written nonnative": 135141,
      "nonnative writers": 80248,
      "idea cast": 51318,
      "svms convolutional": 116248,
      "corpus benchmark": 23680,
      "developing evaluating": 30340,
      "taskspecific evaluation": 119625,
      "metric avoiding": 69870,
      "bias issue": 13810,
      "issue mle": 57006,
      "usage local": 127865,
      "sentence correction": 105814,
      "ngram information": 79434,
      "superiority convolutional": 115700,
      "model edit": 71039,
      "terms grammaticality": 120332,
      "correction lowresource": 24157,
      "mt successfully": 74994,
      "trustable results": 125872,
      "domainadaptation techniques": 33702,
      "jfleg test": 57228,
      "systems outperformed": 117038,
      "nmt hybrid": 79873,
      "benchmarks gec": 12905,
      "analysis metric": 5696,
      "support claims": 115960,
      "use rnns": 128254,
      "grammatical mistakes": 48713,
      "writing support": 135103,
      "language esl": 59011,
      "esl learners": 38426,
      "scholars world": 103962,
      "evaluate usefulness": 38939,
      "domainspecific scientific": 33914,
      "propose copyaugmented": 92606,
      "traditional symbolic": 122877,
      "systems combining": 116791,
      "rescoring neural": 100393,
      "systems annotated": 116730,
      "viewed monolingual": 131714,
      "sequencetosequence transformation": 107212,
      "backtranslation combination": 11335,
      "errors grammatical": 38372,
      "correction using": 24167,
      "introduce unsupervised": 56563,
      "f05 score": 42644,
      "target subwords": 117717,
      "obtain higher": 81290,
      "methods control": 69404,
      "correction english": 24147,
      "research error": 100488,
      "potential drawbacks": 88548,
      "conll2014 benchmarks": 21105,
      "stronger baselines": 113734,
      "use generic": 128069,
      "proficiency level": 92095,
      "domains text": 33873,
      "thousand annotated": 121913,
      "levels different": 63531,
      "relative strong": 98378,
      "correction existing": 24149,
      "correction propose": 24164,
      "cost inference": 24361,
      "low error": 65362,
      "speakers varying": 111334,
      "levels proficiency": 63558,
      "classifying syntactic": 17679,
      "learner english": 62292,
      "generate pseudo": 46990,
      "metalearning fewshot": 68549,
      "using pseudo": 130063,
      "domain 200": 33424,
      "task showing gains": 118710,
      "using ngram overlap": 129970,
      "achieve statistically significant": 2571,
      "correction grammatical error": 24154,
      "sentence level instead": 105931,
      "grammatical errors text": 48702,
      "error detection correction": 38303,
      "unlike conventional maximum": 127429,
      "evaluation metric avoiding": 39277,
      "models sentence correction": 74001,
      "superiority convolutional neural": 115701,
      "networks long shortterm": 77647,
      "trustable results neural": 125873,
      "network based feature": 77159,
      "classification model trained": 17274,
      "feature embeddings learned": 43269,
      "artificial training data": 9268,
      "second language esl": 104416,
      "language esl learners": 59012,
      "systems stateoftheart approaches": 117160,
      "data achieve high": 25559,
      "published stateoftheart results": 94374,
      "hybrid systems combining": 51197,
      "outperforms best published": 82859,
      "error correction using": 38300,
      "methods experiments proposed": 69490,
      "humanlevel performance datasets": 51051,
      "using thousand annotated": 130291,
      "time cost inference": 122006,
      "essays written nonnative": 38536,
      "speakers varying levels": 111335,
      "varying levels proficiency": 131264,
      "generate pseudo data": 46991,
      "achieve statistically significant improvement": 2572,
      "correction grammatical error correction": 24155,
      "unlike conventional maximum likelihood": 127430,
      "conventional maximum likelihood estimation": 22887,
      "recurrent neural networks long": 97234,
      "neural networks long shortterm": 78508,
      "networks long shortterm memory": 77648,
      "neural network based feature": 78257,
      "classification model trained using": 17275,
      "english second language esl": 37270,
      "second language esl learners": 104417,
      "data achieve high accuracy": 25560,
      "methods experiments proposed method": 69491,
      "speakers varying levels proficiency": 111336,
      "correction grammatical error correction gec": 24156,
      "unlike conventional maximum likelihood estimation": 127431,
      "recurrent neural networks long shortterm": 97235,
      "neural networks long shortterm memory": 78509,
      "english second language esl learners": 37271,
      "9179": 1392,
      "9355": 1409,
      "configure": 21008,
      "filed": 44021,
      "intentslot": 55928,
      "words classification": 133858,
      "filling critical": 44035,
      "critical tasks": 24839,
      "relatively long": 98403,
      "domain intent": 33549,
      "intent slots": 55910,
      "bound number": 14510,
      "number quality": 80949,
      "learning effectiveness": 62530,
      "available evaluate": 10993,
      "effective exploiting": 34671,
      "semantic hashing": 105069,
      "attempt overcome": 9750,
      "arise use": 9001,
      "vocabulary use": 131916,
      "models intent": 73414,
      "classifiers label": 17614,
      "truth dataset": 125881,
      "detect intent": 29811,
      "filling essential": 44039,
      "simple finetuning": 109433,
      "filling model": 44047,
      "bert experimental": 13120,
      "hierarchical decoding": 49949,
      "decoding model": 28108,
      "easier obtained": 34421,
      "triples propose": 125817,
      "incremental natural": 53734,
      "flow management": 44683,
      "argue currently": 8913,
      "techniques applicable": 119835,
      "building spoken": 14889,
      "capture intent": 15318,
      "intent semantic": 55908,
      "bert fixed": 13131,
      "finding corresponding": 44271,
      "tapping potential": 117580,
      "means online": 67768,
      "events previous": 39595,
      "ways improving": 132161,
      "source weak": 110862,
      "addition limited": 3448,
      "smart speaker": 110256,
      "documents automatically": 33188,
      "indicate better": 53833,
      "potential usage": 88592,
      "technique example": 119785,
      "domains frequently": 33784,
      "augmentation yields": 10318,
      "modern taskoriented": 74420,
      "website models": 132295,
      "using 100": 129305,
      "examples datasets": 39822,
      "datasets composed": 27366,
      "component spoken": 20000,
      "deep contextualised": 28213,
      "stance expressed": 112173,
      "evaluate nlu": 38877,
      "captures mutual": 15451,
      "make selection": 66722,
      "finally practical": 44215,
      "commercial solutions": 18840,
      "requiring fraction": 100340,
      "deep open": 28390,
      "scores benchmark": 104156,
      "novel lightweight": 80621,
      "traditionally tasks": 122892,
      "research different": 100469,
      "number classes": 80853,
      "simpler ones": 109562,
      "input classifies": 55307,
      "information suffer": 55016,
      "obtained significant": 81410,
      "identification crucial": 51375,
      "transcription using": 124003,
      "present public": 89653,
      "intent context": 55901,
      "information slot": 54986,
      "language understanding investigate": 60231,
      "stateoftheart results f1score": 112902,
      "slot filling critical": 110039,
      "provide additional information": 93754,
      "models previously used": 73797,
      "bound number quality": 14511,
      "problems arise use": 91300,
      "models intent classification": 73415,
      "difficult task requires": 31663,
      "used ground truth": 128567,
      "ground truth dataset": 49098,
      "slot filling essential": 110043,
      "wide variety natural": 132525,
      "slot filling model": 110049,
      "bert experimental results": 13121,
      "intent classification accuracy": 55894,
      "hierarchical decoding model": 49950,
      "triples propose novel": 125818,
      "incremental natural language": 53735,
      "seen training paper": 104547,
      "results standard datasets": 102203,
      "events previous work": 39596,
      "source weak supervision": 110863,
      "approach shows improvement": 7895,
      "datasets different evaluation": 27429,
      "different tasks proposed": 31481,
      "stateoftheart generation models": 112664,
      "datasets achieves competitive": 27295,
      "modern taskoriented dialog": 74421,
      "new domains new": 78893,
      "models able predict": 72646,
      "labeled data achieve": 58430,
      "propose new loss": 92859,
      "method yields significant": 69233,
      "fixed word embeddings": 44611,
      "language models intent": 59568,
      "lowresource language research": 65505,
      "paper present public": 84130,
      "information slot filling": 54987,
      "publicly release dataset": 94331,
      "achieves stateoftheart results f1score": 2903,
      "attentionbased neural network model": 10090,
      "seen training paper propose": 104548,
      "experimental results standard datasets": 40705,
      "results standard datasets model": 102204,
      "large indomain training data": 61116,
      "modern taskoriented dialog systems": 74422,
      "propose new loss function": 92860,
      "achieves stateoftheart results benchmark": 2900,
      "models intent classification slot": 73416,
      "language models intent classification": 59569,
      "propose attentionbased neural network model": 92561,
      "experimental results standard datasets model": 40706,
      "achieves stateoftheart results benchmark datasets": 2901,
      "models intent classification slot filling": 73417,
      "ness": 77101,
      "automatically large": 10796,
      "predicate logic": 88851,
      "temporal expressions": 120104,
      "method translating": 69200,
      "expressions english": 41752,
      "considered parts": 21296,
      "represent relative": 99132,
      "extended analysis": 41803,
      "discuss use": 32291,
      "minimal models": 70169,
      "automatically determining": 10758,
      "90 cases": 1366,
      "automatic temporal": 10689,
      "signals paper": 108703,
      "resources limited": 101005,
      "format resources": 45209,
      "difficulties arising": 31678,
      "introduce resources": 56531,
      "combines use": 18701,
      "ones does": 81683,
      "event annotation": 39495,
      "lexicon turkish": 63909,
      "processing crucial": 91651,
      "resources techniques": 101055,
      "discussed including": 32297,
      "structure compositional": 113823,
      "important distinguishing": 52149,
      "net model": 77113,
      "status identification": 113216,
      "identifying event": 51597,
      "target event": 117621,
      "scheme event": 103923,
      "annotated subset": 6241,
      "predicting relative": 89006,
      "current leading": 25292,
      "prediction temporal": 89140,
      "model minimal": 71515,
      "news reports": 79360,
      "relations event": 98155,
      "understanding learning": 126877,
      "annotated event": 6191,
      "event relations": 39535,
      "ordering models": 82438,
      "dataset frequent": 26945,
      "text annotated": 120645,
      "tools annotation": 122431,
      "trained stateoftheart": 123293,
      "received limited": 96364,
      "limited attention": 64207,
      "hope new": 50631,
      "code resources": 18151,
      "crisis management": 24785,
      "turkish language": 125962,
      "mining analysis": 70223,
      "categories result": 15751,
      "structures classify": 114064,
      "segments correspond": 104672,
      "implementation translates": 51952,
      "improve automatic classification": 52339,
      "use conditional random": 127956,
      "like information retrieval": 64053,
      "importantly propose novel": 52300,
      "perform extensive experimentation": 86003,
      "word embeddings enhanced": 133099,
      "event temporal relation": 39546,
      "neural model event": 78148,
      "serve strong baseline": 107303,
      "key challenge task": 57545,
      "code resources available": 18152,
      "newly collected dataset": 79266,
      "relations event coreference": 98156,
      "semantic representation format": 105227,
      "using dynamic programming": 129611,
      "extraction natural language texts": 42406,
      "event temporal relation extraction": 39547,
      "strong baseline future research": 113630,
      "ascending": 9281,
      "447": 1030,
      "bookingcom": 14420,
      "mt benchmark": 74960,
      "domain making": 33579,
      "translation cat": 124688,
      "new notion": 79060,
      "notion domain": 80406,
      "network maximizes": 77316,
      "maximizes translation": 67516,
      "wmt evaluation": 132792,
      "domain nmt": 33598,
      "obtaining improvements": 81441,
      "adaptation nmt": 3242,
      "gain bleu": 46338,
      "approaches techniques": 8372,
      "style fact": 114578,
      "presented different": 89781,
      "thesis explore": 121849,
      "idea treat": 51340,
      "finetuning explore": 44457,
      "outofdomain model": 82655,
      "structured sparsity": 114043,
      "incremental adaptation": 53722,
      "adaptation multiple": 3237,
      "personalized machine": 87145,
      "information domain": 54508,
      "common encoder": 18870,
      "introduce curriculum": 56403,
      "approach adapt": 7328,
      "adapt generic": 3180,
      "pairs domain": 83520,
      "words lack": 134024,
      "domainspecific words": 33925,
      "englishgerman tasks": 37384,
      "report novel": 99016,
      "adaptive nmt": 3335,
      "best submitted": 13458,
      "known highly": 58303,
      "regardless domain": 97666,
      "approach adapts": 7331,
      "accordingly experimental": 2012,
      "knowledge response": 58156,
      "transformation propose": 124271,
      "reveal domain": 102496,
      "studies domain": 114215,
      "indomain nmt": 53973,
      "ascending order": 9282,
      "particular translation": 85460,
      "changes translation": 16395,
      "decisions using": 27969,
      "setting domain": 107743,
      "challenging build": 16229,
      "simple datadriven": 109397,
      "translation diverse": 124761,
      "terminology paper": 120260,
      "performance backtranslation": 86162,
      "nmt adaptation": 79828,
      "approach make": 7700,
      "specialized data": 111386,
      "used millions": 128634,
      "complex sophisticated": 19880,
      "architectures adapted": 8778,
      "applications far": 6929,
      "domains cases": 33737,
      "domains movie": 33821,
      "translation respectively": 125220,
      "challenging traditional": 16342,
      "present documentlevel": 89457,
      "domainspecific translation": 33920,
      "corpora scarce": 23581,
      "systems vulnerable": 117231,
      "approaches alleviate": 8055,
      "results years": 102350,
      "years production": 135290,
      "translate articles": 124524,
      "development maintenance": 30398,
      "specific examples": 111440,
      "train pruned": 123000,
      "architecture parameter": 8723,
      "nmt research": 79955,
      "propose new technique": 92887,
      "covering multiple domains": 24546,
      "translation domain adaptation": 124769,
      "introduce new notion": 56491,
      "nmt train large": 79991,
      "neural network maximizes": 78323,
      "network maximizes translation": 77317,
      "maximizes translation performance": 67517,
      "translation performance given": 125110,
      "performance given parallel": 86420,
      "parallel corpus nmt": 84622,
      "report significant gains": 99044,
      "simple domain adaptation": 109400,
      "score neural machine": 104096,
      "finetuning indomain data": 44466,
      "domain adaptation approaches": 33430,
      "data sets language": 26428,
      "sets language pairs": 107680,
      "personalized machine translation": 87146,
      "adversarial training model": 4179,
      "introduce curriculum learning": 56404,
      "learning approach adapt": 62338,
      "models specific domain": 74077,
      "machine translation lexicon": 66031,
      "model multiple domains": 71543,
      "supervised neural machine": 115815,
      "auxiliary language modeling": 10876,
      "approach allows model": 7349,
      "accordingly experimental results": 2013,
      "experimental results domain": 40607,
      "nmt model trained": 79900,
      "empirical results chineseenglish": 36182,
      "different users different": 31547,
      "translation task approach": 125329,
      "machine translation diverse": 65954,
      "tasks evaluate proposed": 119103,
      "25 bleu points": 812,
      "results stateoftheart models": 102206,
      "domain domain adaptation": 33513,
      "translation nmt propose": 125050,
      "additionally provide analysis": 3624,
      "parallel corpora scarce": 84608,
      "domains general domain": 33786,
      "multiple domains languages": 75545,
      "test set evaluation": 120505,
      "approaches alleviate problem": 8056,
      "years production systems": 135291,
      "training domain specific": 123593,
      "evaluation results models": 39371,
      "nmt models extremely": 79913,
      "data training time": 26577,
      "training time systems": 123929,
      "domain adaptation neural machine": 33442,
      "neural machine translation domain": 77993,
      "machine translation domain adaptation": 65960,
      "nmt train large neural": 79992,
      "train large neural network": 122947,
      "large neural network maximizes": 61171,
      "neural network maximizes translation": 78324,
      "network maximizes translation performance": 77318,
      "maximizes translation performance given": 67518,
      "translation performance given parallel": 125111,
      "performance given parallel corpus": 86421,
      "given parallel corpus nmt": 48078,
      "bleu score neural machine": 14329,
      "score neural machine translation": 104097,
      "data sets language pairs": 26429,
      "neural machine translation translation": 78122,
      "supervised neural machine translation": 115816,
      "neural machine translation demonstrate": 77988,
      "target domain domain adaptation": 117609,
      "machine translation nmt propose": 66113,
      "translation nmt models extremely": 125042,
      "domain adaptation neural machine translation": 33443,
      "neural machine translation domain adaptation": 77994,
      "nmt train large neural network": 79993,
      "train large neural network maximizes": 122948,
      "large neural network maximizes translation": 61172,
      "neural network maximizes translation performance": 78325,
      "network maximizes translation performance given": 77319,
      "maximizes translation performance given parallel": 67519,
      "translation performance given parallel corpus": 125112,
      "performance given parallel corpus nmt": 86422,
      "bleu score neural machine translation": 14330,
      "standard neural machine translation nmt": 112280,
      "neural machine translation nmt propose": 78056,
      "machine translation nmt models extremely": 66106,
      "disinformation": 32364,
      "textually": 121756,
      "clinton": 17799,
      "clubs": 17911,
      "openlyavailable": 82001,
      "wmd": 132763,
      "sentiment tweets": 106812,
      "submissions 19": 114748,
      "expressed tweet": 41727,
      "models scenarios": 73976,
      "determining veracity": 30156,
      "tweets different": 126029,
      "f1score identifying": 42716,
      "political debates": 87992,
      "hillary clinton": 50446,
      "useful stance": 128934,
      "completely wrong": 19786,
      "collection tweets": 18496,
      "decades recently": 27908,
      "tweets results": 126055,
      "boost classifier": 14431,
      "extraction semisupervised": 42479,
      "identify particular": 51531,
      "detection news": 30005,
      "provide efficient": 93811,
      "weighted ngram": 132356,
      "uses various": 129295,
      "task broken": 117946,
      "components make": 20032,
      "final outcome": 44114,
      "allows joint": 5166,
      "main auxiliary": 66398,
      "automated fake": 10441,
      "dataset identify": 26968,
      "imbalanced dataset": 51829,
      "stance prediction": 112174,
      "extraction stance": 42493,
      "claim validation": 17014,
      "topic analysis": 122497,
      "concepts political": 20631,
      "prevent spreading": 90367,
      "purposes paper": 94448,
      "news important": 79346,
      "involves estimating": 56890,
      "reflect characteristics": 97602,
      "algorithms created": 4838,
      "generator based": 47767,
      "distance approach": 32405,
      "news past": 79355,
      "assess news": 9484,
      "claim existing": 17010,
      "solutions combat": 110573,
      "detection claim": 29894,
      "differences dataset": 30952,
      "simple adversarial": 109355,
      "extract largescale": 42095,
      "years work": 135316,
      "shows benefits": 108553,
      "largest available": 61526,
      "embeddings provides": 35895,
      "veracity prediction": 131512,
      "variation previous": 130898,
      "implicitly captures": 52026,
      "submission task": 114737,
      "settings incorporating": 107811,
      "twitter popular": 126114,
      "media networks": 68129,
      "develop annotated": 30172,
      "submissions 19 teams": 114749,
      "describes novel approach": 29421,
      "language processing text": 59977,
      "data set evaluation": 26407,
      "opinions social media": 82134,
      "uses various machine": 129296,
      "automated fake news": 10442,
      "highly imbalanced dataset": 50325,
      "paper presents baseline": 84152,
      "concepts political science": 20632,
      "dataset annotated named": 26736,
      "annotated named entity": 6212,
      "annotated datasets used": 6181,
      "approach based word": 7385,
      "deep ensemble model": 28232,
      "detection claim validation": 29895,
      "train single model": 123022,
      "results multilingual bert": 101966,
      "paper addresses issues": 83715,
      "topic modeling approach": 122543,
      "dataset method outperforms": 27023,
      "examples paper present": 39863,
      "task pretrained models": 118556,
      "twitter popular social": 126115,
      "popular social media": 88129,
      "social media networks": 110392,
      "paper describes novel approach": 83837,
      "results demonstrate proposed approach": 101699,
      "natural language processing text": 76506,
      "uses various machine learning": 129297,
      "automated fake news detection": 10443,
      "dataset annotated named entity": 26737,
      "annotated named entity recognition": 6213,
      "twitter popular social media": 126116,
      "experimental results demonstrate proposed approach": 40600,
      "dataset annotated named entity recognition": 26738,
      "hien": 49921,
      "sail": 103539,
      "teluguenglish": 120074,
      "bnen": 14389,
      "331": 945,
      "krippendorffs": 58351,
      "nbsvm": 76697,
      "using codemixed": 129448,
      "hindienglish hien": 50488,
      "quite popular": 95403,
      "dataset having": 26960,
      "english bengali": 37077,
      "trained codemixed": 123092,
      "tool contest": 122377,
      "participated contest": 85336,
      "bengalienglish hindienglish": 13043,
      "tag sentiment": 117339,
      "raw twitter": 95866,
      "sentiment properties": 106784,
      "codemixed indian": 18176,
      "enhance sentiment": 37492,
      "used recommendation": 128717,
      "analysis classification": 5527,
      "impact audience": 51858,
      "dataset ensemble": 26892,
      "data ensemble": 25897,
      "reallife user": 96130,
      "data reveals": 26362,
      "respectively submitted": 101169,
      "submitted models": 114765,
      "tweet level": 126013,
      "media based": 68083,
      "benchmark sentiment": 12856,
      "create models": 24628,
      "used semisupervised": 128746,
      "krippendorffs alpha": 58352,
      "codemixed texts": 18189,
      "various platforms": 131169,
      "sentiment classes": 106691,
      "written roman": 135145,
      "using grid": 129721,
      "grid search": 49089,
      "phonetic typing": 87300,
      "phenomenon poses": 87257,
      "attempted develop": 9757,
      "given codemixed": 47993,
      "communication different": 19029,
      "used providing": 128708,
      "python used": 94489,
      "combine languages": 18629,
      "making text": 66868,
      "intelligible familiar": 55863,
      "approaches pretrained": 8287,
      "language increasing": 59161,
      "dictionary outperforms": 30886,
      "data phenomenon": 26236,
      "encode character": 36421,
      "texts extracted": 121518,
      "contribution semeval2020": 22778,
      "incorporates pretrained": 53511,
      "bert multitask": 13189,
      "discusses results": 32313,
      "using count": 129514,
      "voting classifiers": 131952,
      "classifiers data": 17604,
      "tweets model": 126038,
      "methods social media": 69763,
      "paper report results": 84403,
      "report results various": 99040,
      "languages english bengali": 60529,
      "codemixed indian languages": 18177,
      "different social media": 31435,
      "dataset ensemble model": 26893,
      "annotated corpus support": 6154,
      "regression random forests": 97711,
      "investigate various training": 56828,
      "various training strategies": 131227,
      "social media based": 110353,
      "benchmark sentiment analysis": 12857,
      "different levels text": 31235,
      "unstructured text data": 127583,
      "written roman script": 135146,
      "media text work": 68173,
      "using grid search": 129722,
      "media text paper": 68170,
      "describes contribution semeval2020": 29390,
      "contribution semeval2020 task": 22779,
      "bert multitask learning": 13190,
      "paper discusses results": 83881,
      "machine learning neural network": 65819,
      "logistic regression random forests": 65042,
      "recent deep learning approaches": 96448,
      "investigate various training strategies": 56829,
      "social media text work": 110426,
      "social media text paper": 110423,
      "media text paper describes": 68171,
      "paper describes contribution semeval2020": 83813,
      "describes contribution semeval2020 task": 29391,
      "using machine learning neural network": 129849,
      "codemixed social media text using": 18186,
      "social media text paper describes": 110424,
      "paper describes contribution semeval2020 task": 83814,
      "duplicated": 34271,
      "straightthrough": 113436,
      "annotation graph": 6318,
      "set trained": 107613,
      "stateoftheart information": 112676,
      "able accurately": 1592,
      "document help": 33020,
      "phrases best": 87411,
      "embeddings extracting": 35718,
      "comprising short": 20336,
      "uses minimum": 129244,
      "representation candidate": 99183,
      "labeled samples": 58465,
      "titles articles": 122224,
      "articles experimental": 9187,
      "stateoftheart generative": 112667,
      "document title": 33096,
      "overall document": 83228,
      "information massive": 54750,
      "methods does": 69457,
      "study generating": 114392,
      "generating multiple": 47232,
      "concatenating multiple": 20538,
      "different orders": 31312,
      "study examined": 114373,
      "model running": 71946,
      "summarize large": 115606,
      "documents traditional": 33303,
      "network measures": 77322,
      "finetuning contextualized": 44451,
      "bert scibert": 13222,
      "overview content": 83362,
      "extraction traditional": 42520,
      "utilize context": 130498,
      "capable overcoming": 15210,
      "specific components": 111419,
      "produce short": 91932,
      "attract readers": 10140,
      "sequential decoding": 107223,
      "incorporate available": 53451,
      "datasets illustrate": 27509,
      "rely copying": 98682,
      "learns rank": 63227,
      "techniques demonstrated": 119865,
      "inputs neural": 55487,
      "networks consequently": 77543,
      "selfattentive architecture": 104908,
      "set terms": 107604,
      "various centrality": 131056,
      "work follow": 134540,
      "position text": 88295,
      "using syntactical": 130261,
      "language taken": 60144,
      "taken twitter": 117519,
      "extract sequence": 42117,
      "studies persian": 114261,
      "lack human": 58717,
      "techniques tfidf": 119999,
      "study sequencetosequence": 114514,
      "analyze extent": 5972,
      "outputs generated": 83168,
      "training alleviate": 123354,
      "large diversity": 61083,
      "topics given": 122629,
      "sentences address": 106203,
      "straightthrough estimator": 113437,
      "gap paper": 46470,
      "representation phrase": 99374,
      "pertaining domain": 87187,
      "standard publicly": 112292,
      "available digital": 10984,
      "words multiword": 134069,
      "stateoftheart information extraction": 112677,
      "using sentence embeddings": 130148,
      "articles experimental results": 9188,
      "conventional approaches task": 22877,
      "benchmark datasets terms": 12808,
      "gathers relevant information": 46532,
      "framework jointly learns": 45591,
      "model generative model": 71246,
      "generative model produce": 47739,
      "model learns meaningful": 71449,
      "stateoftheart generative models": 112668,
      "study generating multiple": 114393,
      "text documents traditional": 120888,
      "proposed method effectively": 93365,
      "words input text": 134006,
      "context information effectively": 22142,
      "models capture local": 72881,
      "neural networks consequently": 78458,
      "experimental results seven": 40697,
      "words phrases text": 134122,
      "comprehensive comparison different": 20253,
      "analyze extent information": 5973,
      "domains demonstrate proposed": 33761,
      "scientific documents using": 104003,
      "standard publicly available": 112293,
      "publicly available digital": 94303,
      "words multiword expressions": 134070,
      "models achieved remarkable performance": 72688,
      "results benchmark datasets proposed": 101544,
      "proposed model able generate": 93424,
      "different domains demonstrate proposed": 31110,
      "results model significantly outperforms stateoftheart": 101951,
      "datasets different domains demonstrate proposed": 27428,
      "wat2016": 132043,
      "sentencepiece": 106188,
      "305": 915,
      "stochastically": 113361,
      "making nmt": 66856,
      "compositional translation": 20137,
      "factored neural": 42863,
      "architecture addresses": 8612,
      "words means": 134051,
      "training produce": 123797,
      "rely statistics": 98746,
      "words leads": 134029,
      "used preprocessing": 128691,
      "measuring accuracy": 67906,
      "alleviated using": 5054,
      "nmt case": 79845,
      "suffers performance": 115263,
      "different nmt": 31302,
      "nmt emerged": 79855,
      "alternative conventional": 5227,
      "exploit existing": 41413,
      "subword candidates": 115015,
      "ambiguous multiple": 5293,
      "addressed paper": 3786,
      "robustness nmt": 103118,
      "subword segmentations": 115033,
      "nmt bidirectional": 79843,
      "level granularity": 63454,
      "feature information": 43289,
      "paper enable": 83893,
      "shape semantic": 107939,
      "apache license": 6768,
      "task participate": 118517,
      "finnish estonian": 44556,
      "names introduce": 76144,
      "systems utilizing": 117229,
      "east asian": 34469,
      "unfortunately traditional": 127079,
      "sequences highly": 107125,
      "segmentations neural": 104659,
      "translations improves": 125470,
      "multiple tokens": 75730,
      "regularization latent": 97749,
      "translation directly": 124757,
      "set approximately": 107364,
      "decoder learn": 28029,
      "subwordlevel models": 115050,
      "need finegrained": 76811,
      "lot common": 65320,
      "common terms": 18936,
      "motivated propose": 74859,
      "solution improve": 110557,
      "morphologically motivated": 74762,
      "language complexity": 58899,
      "translation effectively": 124777,
      "characters different": 16605,
      "contain unseen": 21766,
      "generally offer": 46870,
      "factored neural machine": 42864,
      "translation present new": 125134,
      "network architecture addresses": 77140,
      "conventional methods aim": 22890,
      "morphologically rich agglutinative": 74767,
      "23 bleu points": 788,
      "machine translation proved": 66177,
      "problem alleviated using": 90928,
      "terms bleu meteor": 120287,
      "previous work english": 90528,
      "translation nmt emerged": 125032,
      "improve robustness nmt": 52531,
      "evaluation metrics including": 39289,
      "14 bleu points": 311,
      "impressive results language": 52323,
      "semantic information characters": 105074,
      "recurrent convolutional models": 97166,
      "text processing paper": 121210,
      "text processing including": 121209,
      "translation article describes": 124650,
      "proper names introduce": 92418,
      "translation quality particularly": 125183,
      "diverse target languages": 32853,
      "subword segmentations neural": 115034,
      "segmentations neural machine": 104660,
      "datasets language pairs": 27538,
      "models despite recent": 73053,
      "nmt models capture": 79911,
      "comparison baseline models": 19530,
      "nmt achieved impressive": 79822,
      "motivated word segmentation": 74865,
      "models recent works": 73879,
      "factored neural machine translation": 42865,
      "machine translation present new": 66165,
      "neural network architecture addresses": 78245,
      "machine translation nmt emerged": 66096,
      "proposed method improves translation": 93375,
      "subword segmentations neural machine": 115035,
      "segmentations neural machine translation": 104661,
      "translation nmt achieved impressive": 125022,
      "neural machine translation nmt emerged": 78047,
      "proposed method improves translation quality": 93376,
      "subword segmentations neural machine translation": 115036,
      "machine translation nmt achieved impressive": 66088,
      "axiom": 11269,
      "dialectical": 30532,
      "gametheoretic": 46424,
      "algebras": 4681,
      "epsilon": 38218,
      "lan": 58798,
      "guage": 49212,
      "thrown": 121951,
      "printing": 90688,
      "neodavidsonian": 77009,
      "pertains": 87189,
      "computation based": 20344,
      "meanings constituents": 67742,
      "semantics gives": 105423,
      "evaluation order": 39316,
      "grammar present": 48651,
      "logic paper": 64994,
      "single discrete": 109726,
      "grammars based": 48671,
      "minimalist grammars": 70183,
      "processing complexity": 91641,
      "grammars mg": 48677,
      "steps giving": 113320,
      "rich type": 102798,
      "terms work": 120412,
      "theoretical practical": 121799,
      "natural lan": 76261,
      "lan guage": 58799,
      "compose meaning": 20059,
      "granularity type": 48752,
      "languages pronouns": 60818,
      "pragmatic phenomena": 88754,
      "formal compositional": 45154,
      "meanings assigned": 67740,
      "similar type": 109163,
      "various modern": 131136,
      "study grammar": 114395,
      "typed events": 126237,
      "necessary background": 76751,
      "truth value": 125886,
      "dynamic model": 34317,
      "grounded commonsense": 49105,
      "knowledge extracting": 57929,
      "type grammar": 126202,
      "types experiments": 126285,
      "frobenius algebras": 45897,
      "logic paper explores": 64995,
      "algorithm selecting best": 4798,
      "minimalist grammars mg": 70184,
      "terms work propose": 120413,
      "natural lan guage": 76262,
      "meanings words paper": 67753,
      "fundamentally different types": 46134,
      "semanticallymotivated": 105386,
      "141": 326,
      "sememebased": 105490,
      "model paraphrasing": 71710,
      "reading level": 96023,
      "studied impact": 114164,
      "sentences like": 106381,
      "generally better": 46850,
      "length frequency": 63359,
      "2017 sentence": 638,
      "order collect": 82295,
      "examples usage": 39894,
      "interactive process": 56027,
      "models updated": 74273,
      "task mere": 118385,
      "pair simple": 83455,
      "structure comparing": 113820,
      "vast space": 131281,
      "particular neural": 85434,
      "extensive automatic": 41859,
      "layer utilizes": 61753,
      "complexity given": 19912,
      "parameters sentence": 84782,
      "referenceless quality": 97555,
      "metrics require": 70002,
      "sentence resulting": 106052,
      "issue use": 57031,
      "sentence complexity": 105794,
      "splitting rephrasing": 111955,
      "rephrasing complex": 98925,
      "mt information": 74969,
      "linked rhetorical": 64648,
      "proposed syntactic": 93560,
      "automatic readability": 10639,
      "simplification german": 109581,
      "set minimal": 107495,
      "minimal semantic": 70176,
      "models realistic": 73863,
      "motivate need": 74837,
      "translation blackbox": 124678,
      "evaluation verify": 39440,
      "pair measured": 83441,
      "survey text": 116203,
      "utilizing deep": 130568,
      "sentence simpler": 106076,
      "need explained": 76804,
      "faster higher": 43178,
      "wikipedia simple": 132676,
      "endtoend perform": 36951,
      "quality judgments": 94694,
      "predict complexity": 88876,
      "prlms bert": 90792,
      "notion consistency": 80404,
      "process easier": 91466,
      "compare texts": 19303,
      "time generated": 122032,
      "correlations traditional": 24262,
      "translation systems use": 125319,
      "using complementary approaches": 129458,
      "datasets terms automatic": 27751,
      "used collect training": 128429,
      "text simplification task": 121305,
      "text simplification systems": 121304,
      "simple effective text": 109417,
      "extensive automatic human": 41860,
      "referenceless quality estimation": 97556,
      "metrics bleu metrics": 69934,
      "simplification task rewriting": 109594,
      "splitting rephrasing complex": 111956,
      "linked rhetorical relations": 64649,
      "analysis automatic evaluation": 5508,
      "standard automatic metrics": 112198,
      "automatic readability assessment": 10640,
      "text simplification german": 121301,
      "automatic text simplification": 10695,
      "previously proposed text": 90613,
      "complex linguistic structure": 19829,
      "complexity syntactic complexity": 19940,
      "machine translation blackbox": 65916,
      "translation performance compared": 125107,
      "human evaluation verify": 50843,
      "english wikipedia simple": 37341,
      "human quality judgments": 50938,
      "require human references": 100148,
      "remove spurious correlations": 98878,
      "machine translation systems use": 66255,
      "extensive automatic human evaluation": 41861,
      "automatic human evaluation shows": 10556,
      "automatic text simplification german": 10696,
      "extensive automatic human evaluation shows": 41862,
      "focussed": 44930,
      "visualising": 131822,
      "laid": 58787,
      "standardisation": 112341,
      "customisation": 25454,
      "survey wide": 116205,
      "existing annotation": 40045,
      "tools interact": 122451,
      "graphs graph": 48955,
      "signals text": 108707,
      "mapped annotation": 67119,
      "support analysis": 115955,
      "text engineering": 120910,
      "order aid": 82275,
      "data associated": 25624,
      "associated linguistic": 9601,
      "suite report": 115419,
      "documents german": 33235,
      "strong emphasis": 113668,
      "techniques deal": 119860,
      "exploit methods": 41429,
      "document discusses": 32993,
      "subsequent processing": 114816,
      "certain extent": 15942,
      "extent proposed": 41982,
      "grammars language": 48674,
      "nlp survey": 79695,
      "representing human": 100049,
      "markup languages": 67272,
      "chapter present": 16411,
      "set general": 107450,
      "order reach": 82394,
      "guidelines allow": 49259,
      "conceptual technical": 20673,
      "series queries": 107278,
      "annotation text": 6387,
      "digital resources": 31725,
      "provides systematic": 94086,
      "systematic solution": 116685,
      "present major": 89542,
      "characters tokens": 16626,
      "available present": 11074,
      "paper dictionaries": 83872,
      "problems involve": 91328,
      "survey wide variety": 116206,
      "wide variety existing": 132522,
      "mapped annotation graph": 67120,
      "annotation graph model": 6319,
      "input output data": 55380,
      "solve complex tasks": 110593,
      "used provide insights": 128707,
      "different user groups": 31545,
      "linguistic annotation framework": 64423,
      "annotation framework based": 6315,
      "annotation tool paper": 6395,
      "tool paper introduce": 122398,
      "evaluating machine learning": 39070,
      "article presents results": 9148,
      "mapped annotation graph model": 67121,
      "annotation tool paper introduce": 6396,
      "realword": 96147,
      "keystrokes": 57628,
      "counters": 24444,
      "userfacing": 129067,
      "kingdom": 57697,
      "carried various": 15546,
      "human spoken": 50968,
      "asr operating": 9441,
      "dataset detecting": 26867,
      "detecting word": 29860,
      "remarkable reduction": 98844,
      "distributed systems": 32628,
      "terms technical": 120391,
      "electronic version": 35262,
      "google web": 48527,
      "new contextsensitive": 78840,
      "ngram word": 79456,
      "correction patterns": 24163,
      "focus instead": 44775,
      "using stochastic": 130226,
      "stochastic language": 113359,
      "important phenomena": 52205,
      "profound impact": 92109,
      "classification ii": 17225,
      "little cost": 64799,
      "significantly robust": 109050,
      "correction word": 24168,
      "experiment human": 40468,
      "employs multilayer": 36330,
      "tuning models": 125937,
      "largely improve": 61341,
      "word imputation": 133315,
      "model case": 70815,
      "checking task": 16697,
      "writing speaking": 135097,
      "literature past": 64764,
      "web present": 132245,
      "target error": 117619,
      "pairs makes": 83582,
      "phonetic similarity": 87296,
      "email spam": 35341,
      "evaluation basic": 39132,
      "specific error": 111439,
      "detection context": 29907,
      "previously suggested": 90625,
      "dataset serves": 27192,
      "based scores": 12003,
      "matching language": 67408,
      "missing previous": 70331,
      "text spelling": 121326,
      "spelling normalization": 111910,
      "given users": 48168,
      "approach real": 7837,
      "perturbations using": 87197,
      "rate 34": 95777,
      "systems context": 116813,
      "data vulnerable": 26624,
      "errors corrected": 38361,
      "output based": 83053,
      "low quality paper": 65381,
      "error correction method": 38297,
      "experiments carried various": 40816,
      "text paper proposes": 121168,
      "text documents different": 120885,
      "high accuracy using": 50036,
      "shared task machine": 108063,
      "spelling correction word": 111902,
      "encoder decoder demonstrate": 36507,
      "performance compared state": 86235,
      "perform experiments languages": 85996,
      "potential directions future": 88547,
      "email spam detection": 35342,
      "information present methods": 54859,
      "approach real world": 7838,
      "work address task": 134357,
      "error rate 34": 38325,
      "text documents different domains": 120886,
      "performance compared state art": 86236,
      "outperforms previous stateoftheart result": 82952,
      "semeval16": 105519,
      "sst5": 112097,
      "probabilistic way": 90837,
      "formal description": 45156,
      "best sentiment": 13449,
      "semeval16 task": 105520,
      "words tweet": 134282,
      "trained sentencelevel": 123269,
      "treestructured long": 125723,
      "models required": 73936,
      "clear information": 17725,
      "architectures bidirectional": 8788,
      "language deep": 58938,
      "information wordlevel": 55100,
      "map using": 67114,
      "feedback using": 43840,
      "gru network": 49200,
      "second rank": 104449,
      "business reviews": 14981,
      "demonstrate approaches": 28673,
      "classification significant": 17405,
      "profile paper": 92100,
      "combining convolution": 18717,
      "architecture addition": 8609,
      "leveraging new": 63694,
      "onehot encoding": 81675,
      "wordnet build": 133768,
      "stateoftheart semeval": 112945,
      "task 17": 117818,
      "makes effective": 66777,
      "robust superiority": 103075,
      "classification analyze": 17113,
      "sentences ambiguous": 106206,
      "classification improved": 17229,
      "features final": 43516,
      "negation sentiment": 76916,
      "directly affected": 31860,
      "order correctly": 82304,
      "related performance": 97886,
      "encoders using": 36664,
      "examples help": 39842,
      "variety sentiment": 131014,
      "advancement artificial": 3962,
      "response sentiment": 101244,
      "ai intelligent": 4421,
      "setting given": 107753,
      "attentionbased architectures": 10059,
      "architectures employed": 8804,
      "squad race": 112079,
      "improving documentlevel": 53090,
      "sentences previous": 106453,
      "support important": 115975,
      "propose documentlevel": 92630,
      "differences sentences": 30971,
      "attention sentiment analysis": 10005,
      "sentiment analysis provide": 106639,
      "introduces novel approach": 56622,
      "treestructured long shortterm": 125724,
      "language deep neural": 58939,
      "embedding bidirectional long": 35384,
      "method based convolutional": 68665,
      "analysis paper present": 5738,
      "vector representations text": 131359,
      "sentiment analysis key": 106601,
      "classification analyze performance": 17114,
      "analyze performance different": 5991,
      "models neural sequence": 73636,
      "words annotated finegrained": 133817,
      "annotations proposed method": 6456,
      "text classification improved": 120726,
      "end propose use": 36828,
      "negation sentiment analysis": 76917,
      "network models achieved": 77343,
      "assessed human evaluators": 9501,
      "stateoftheart performance variety": 112854,
      "performance variety sentiment": 86839,
      "analysis tasks deep": 5887,
      "advancement artificial intelligence": 3963,
      "glue squad race": 48316,
      "improving documentlevel sentiment": 53091,
      "sentence classification model": 105785,
      "classification model based": 17272,
      "paper introduces novel approach": 84011,
      "treestructured long shortterm memory": 125725,
      "embedding bidirectional long shortterm": 35385,
      "method based convolutional neural": 68666,
      "sentiment analysis paper present": 106625,
      "achieve comparable performance stateoftheart": 2442,
      "neural network models achieved": 78344,
      "sentiment analysis tasks deep": 106668,
      "advancement artificial intelligence ai": 3964,
      "embedding bidirectional long shortterm memory": 35386,
      "bidirectional long shortterm memory network": 13943,
      "reap": 96194,
      "813": 1300,
      "predictiveness": 89209,
      "unpredictability": 127492,
      "heated": 49647,
      "007": 4,
      "used indicate": 128588,
      "campaign speeches": 15090,
      "presidential campaign": 89946,
      "al 2006": 4598,
      "predict reading": 88922,
      "corpora web": 23619,
      "addressing issue": 3811,
      "higher previously": 50196,
      "sentencelevel data": 106153,
      "features resulted": 43706,
      "method infer": 68906,
      "surface lexical": 116064,
      "challenges data": 16145,
      "ngrams highlevel": 79465,
      "workshop proceedings": 135015,
      "corpus database": 23743,
      "play significant": 87705,
      "levels shows": 63565,
      "differences english": 30957,
      "word grammar": 133304,
      "word acquisition": 132904,
      "language reading": 60022,
      "language powerful": 59817,
      "french chinese": 45801,
      "general used": 46729,
      "complex messages": 19832,
      "complexity classification": 19903,
      "texts suitable": 121624,
      "applications example": 6925,
      "esl texts": 38427,
      "texts overall": 121566,
      "corpus lexical": 23863,
      "hard understand": 49484,
      "helps lexical": 49824,
      "single lexical": 109754,
      "nature multiple": 76660,
      "according rules": 2000,
      "process comparing": 91440,
      "expressed using": 41729,
      "predicting lexical": 88991,
      "nature lexical": 76658,
      "new protocol": 79089,
      "level given": 63453,
      "scheme task": 103938,
      "mae score": 66369,
      "additionally used": 3636,
      "score subtask": 104129,
      "complex process": 19862,
      "advanced nlp": 3954,
      "embeddings graph": 35735,
      "targets results": 117792,
      "2016 presidential campaign": 613,
      "et al 2006": 38689,
      "second language learning": 104420,
      "language learning material": 59243,
      "range linguistic features": 95578,
      "using different sets": 129576,
      "surface lexical syntactic": 116065,
      "task approach based": 117887,
      "approach based combining": 7381,
      "character ngrams highlevel": 16456,
      "12 teams submitted": 259,
      "play significant role": 87706,
      "data set containing": 26405,
      "second language reading": 104423,
      "english french chinese": 37140,
      "combining deep learning": 18720,
      "complex process requires": 19863,
      "requires proper understanding": 100311,
      "word embeddings graph": 133121,
      "second language learning material": 104421,
      "12 teams submitted results": 260,
      "approaches natural language processing tasks": 8251,
      "wmt2019": 132873,
      "sdl": 104278,
      "332": 946,
      "postediting ape": 88471,
      "decoded target": 27989,
      "exploration neural": 41502,
      "specialized neural": 111393,
      "interactions human": 55985,
      "providing evidence": 94112,
      "interactive machine": 56021,
      "varying type": 131270,
      "post edit": 88455,
      "gold training": 48446,
      "contribute generation": 22752,
      "sentence properly": 106006,
      "used extra": 128542,
      "submission wmt2018": 114745,
      "improvements based": 52808,
      "based weak": 12156,
      "ranked subtask": 95650,
      "submission wmt2019": 114746,
      "financial texts": 44261,
      "translation postediting": 125126,
      "translation production": 125148,
      "integrating machine": 55797,
      "effort machine": 35176,
      "set computational": 107397,
      "process user": 91585,
      "translation mistakes": 124935,
      "finegrained understanding": 44393,
      "analyze new": 5988,
      "postediting time": 88476,
      "exploiting context": 41461,
      "architecture automatic": 8623,
      "ape model": 6772,
      "mt iii": 74968,
      "aims correct": 4513,
      "contrasting views": 22724,
      "understand potential": 126768,
      "neural decoding": 77885,
      "conference computational": 20974,
      "step important": 113272,
      "work required": 134782,
      "quality metric": 94717,
      "nmt translations": 80001,
      "participants read": 85324,
      "version historical": 131607,
      "blackbox nmt": 14256,
      "automatic postediting ape": 10624,
      "results training data": 102280,
      "interactive machine translation": 56022,
      "human effort needed": 50798,
      "machine translation production": 66174,
      "release analyze new": 98430,
      "analyze new dataset": 5989,
      "decisions paper present": 27964,
      "shared task work": 108108,
      "conference computational linguistics": 20975,
      "ape model trained": 6773,
      "generating new version": 47239,
      "new version historical": 79241,
      "version historical document": 131608,
      "historical document written": 50521,
      "document written modern": 33114,
      "neural machine translation production": 78082,
      "neural machine translation reinforcement": 78094,
      "release analyze new dataset": 98431,
      "shared task wmt 2019": 108106,
      "generating new version historical": 47240,
      "new version historical document": 79242,
      "version historical document written": 131609,
      "historical document written modern": 50522,
      "document written modern version": 33115,
      "generating new version historical document": 47241,
      "new version historical document written": 79243,
      "version historical document written modern": 131610,
      "historical document written modern version": 50523,
      "trans": 123980,
      "acknowledging": 2992,
      "machinetranslation": 66351,
      "exacerbate": 39709,
      "language produce": 59989,
      "trait author": 123970,
      "author gender": 10363,
      "texts translations": 121640,
      "authors gender": 10381,
      "tool helps": 122389,
      "influenced number": 54321,
      "perform machine": 86028,
      "syntactic supervision": 116489,
      "supervision need": 115901,
      "tune bert": 125927,
      "simplicity architecture": 109567,
      "architecture make": 8690,
      "speakers different": 111313,
      "aspects world": 9419,
      "order use": 82424,
      "contribution twofold": 22780,
      "2000 sentences": 538,
      "fewer sentences": 43891,
      "training achieving": 123345,
      "adaptation elastic": 3218,
      "allows performance": 5178,
      "general test": 46721,
      "model augment": 70706,
      "determined human": 30137,
      "presents largest": 89859,
      "19 systems": 437,
      "polish russian": 87984,
      "tackling problem": 117326,
      "differ sharing": 30931,
      "discuss practical": 32278,
      "pronouns translation": 92362,
      "texts meaning": 121552,
      "pronoun prediction": 92340,
      "translations existing": 125463,
      "new field": 78918,
      "related disciplines": 97853,
      "inference procedures": 54201,
      "likely output": 64142,
      "sentence applying": 105773,
      "decoding improve": 28097,
      "techniques allows": 119833,
      "translates english": 124581,
      "especially egregious": 38455,
      "influenced number factors": 54322,
      "perform machine translation": 86029,
      "techniques neural machine": 119942,
      "fine tune bert": 44325,
      "information nmt systems": 54806,
      "bias neural machine": 13826,
      "adaptation elastic weight": 3219,
      "methods machine translation": 69607,
      "using data train": 129533,
      "manually annotated large": 67032,
      "realworld scenarios results": 96178,
      "techniques neural machine translation": 119943,
      "bias neural machine translation": 13827,
      "adaptation elastic weight consolidation": 3220,
      "sc": 103676,
      "ci": 16976,
      "finest": 44400,
      "measure improvement": 67803,
      "useful languages": 128899,
      "different pos": 31339,
      "statistical rulebased": 113159,
      "framework recent": 45668,
      "taggers paper": 117367,
      "assign multiple": 9544,
      "multiple tags": 75720,
      "probabilities second": 90850,
      "transducers present": 124023,
      "disambiguation compared": 31957,
      "tagging assign": 117374,
      "time directly": 122016,
      "structure speech": 113961,
      "examined methods": 39763,
      "use cluster": 127941,
      "punctuation marks": 94395,
      "expected improve": 40391,
      "markov chain": 67252,
      "chain monte": 15978,
      "evidence achieve": 39633,
      "doing able": 33418,
      "tagging process": 117440,
      "trigger words": 125788,
      "translation application": 124640,
      "based incremental": 11767,
      "time tagging": 122119,
      "universal framework": 127308,
      "compare simple": 19294,
      "tasks uniform": 119583,
      "predicted models": 88964,
      "differ words": 30937,
      "processing problem": 91775,
      "negatively affected": 76963,
      "stateoftheart taggers": 112995,
      "speech present": 111727,
      "rules statistical": 103441,
      "achieves precision": 2835,
      "annotation partofspeech": 6348,
      "previously collected": 90593,
      "task pos": 118535,
      "tagging new": 117430,
      "monolingual pos": 74606,
      "strategy selects": 113539,
      "languages available resources": 60420,
      "different pos taggers": 31340,
      "level machine learning": 63472,
      "using little training": 129823,
      "experiments using methods": 41195,
      "french german italian": 45809,
      "markov chain monte": 67253,
      "chain monte carlo": 15979,
      "tagging paper propose": 117433,
      "open class words": 81885,
      "method performs best": 69047,
      "semantic models paper": 105116,
      "universal pos tag": 127322,
      "pos tag set": 88214,
      "major problems natural": 66580,
      "language processing problem": 59934,
      "pos tagging techniques": 88239,
      "task pos tagging": 118536,
      "pos tagging new": 88236,
      "paper demonstrate efficacy": 83801,
      "using little training data": 129824,
      "markov chain monte carlo": 67254,
      "distributional semantic models paper": 32722,
      "universal pos tag set": 127323,
      "major problems natural language": 66581,
      "natural language processing problem": 76475,
      "major problems natural language processing": 66582,
      "npr": 80801,
      "baselines existing": 12393,
      "important utterances": 52289,
      "selected generate": 104726,
      "assessments human": 9534,
      "baselines unsupervised": 12484,
      "domain requires": 33627,
      "propose abstractive": 92525,
      "capabilities neural": 15171,
      "propose topicaware": 93114,
      "experiments annotated": 40772,
      "incremental contributions": 53723,
      "corpus humanannotated": 23828,
      "professionally edited": 92086,
      "nlp attention": 79579,
      "discussed specific": 32299,
      "dialogue rewriting": 30736,
      "minutes using": 70281,
      "model summarize": 72113,
      "crucial users": 25180,
      "input abstractive": 55291,
      "topics addition": 122613,
      "issues specific": 57079,
      "information matters": 54757,
      "leverage textual": 63626,
      "summaries create": 115433,
      "domains conduct": 33750,
      "truth summaries": 125885,
      "content long": 21901,
      "using unsupervised supervised": 130347,
      "generated using supervised": 47150,
      "generating abstractive summaries": 47197,
      "outperforms stateoftheart abstractive": 82996,
      "modeling text summarization": 72565,
      "conversations paper propose": 23065,
      "automatic evaluations human": 10542,
      "version paper present": 131616,
      "pretrained sentence embedding": 90183,
      "quality training instances": 94815,
      "challenges existing models": 16157,
      "existing models paper": 40206,
      "ground truth summaries": 49101,
      "recent advances word embeddings": 96426,
      "future research dataset available": 46297,
      "scl": 104034,
      "dependence word": 29059,
      "domainspecific aspects": 33888,
      "using learnt": 129807,
      "convey useful": 23119,
      "generalization examples": 46775,
      "requires learning": 100291,
      "important achieving": 52096,
      "semeval 2013": 105495,
      "classifier learned": 17555,
      "various experimental": 131093,
      "studies explicitly": 114224,
      "feature alignment": 43249,
      "subtask use": 114959,
      "manner domain": 66943,
      "important limitation": 52180,
      "multiple new": 75628,
      "knowledge lead": 58043,
      "differ context": 30928,
      "unique challenge": 127182,
      "using 11": 129311,
      "include novel": 53233,
      "training graph": 123645,
      "domain event": 33523,
      "uses adversarial": 129191,
      "especially domain": 38452,
      "leverage useful": 63635,
      "domains help": 33789,
      "feature framework": 43284,
      "twostage training": 126165,
      "framework sequential": 45678,
      "domains leads": 33802,
      "domains robust": 33854,
      "crossdomain crosslanguage": 24885,
      "approaches directly": 8129,
      "utilize structure": 130529,
      "word representations proposed": 133469,
      "convey useful information": 23120,
      "using adversarial training": 129326,
      "various experimental settings": 131094,
      "demonstrate approach consistently": 28668,
      "2019 task suggestion": 714,
      "neural networks significantly": 78578,
      "text previous work": 121202,
      "model simple structure": 72044,
      "effectiveness method text": 34909,
      "open domain event": 81895,
      "leverage useful information": 63636,
      "information sharing tasks": 54979,
      "improve nlp models": 52442,
      "demonstrate approach consistently outperforms": 28669,
      "pretrained language model used": 90059,
      "semeval 2019 task suggestion": 105506,
      "2019 task suggestion mining": 715,
      "task suggestion mining online": 118765,
      "unsupervised domain adaptation framework": 127628,
      "semeval 2019 task suggestion mining": 105507,
      "2019 task suggestion mining online": 716,
      "task suggestion mining online reviews": 118766,
      "local translation": 64940,
      "inconsistencies paper": 53435,
      "21 bleu": 757,
      "translation extended": 124818,
      "iterative decoding": 57128,
      "ignoring crosssentence": 51683,
      "crosssentence links": 25073,
      "links dependencies": 64679,
      "sentences entire": 106293,
      "nmt improved": 79876,
      "chineseenglish datasets": 16848,
      "context improving": 22133,
      "specific phenomena": 111476,
      "translation including": 124865,
      "evaluation schemes": 39377,
      "contextaware nmt": 22345,
      "history extra": 50556,
      "representation hierarchical": 99250,
      "missing context": 70320,
      "particular consider": 85395,
      "nmt usually": 80008,
      "avoid models": 11237,
      "using constraints": 129478,
      "data run": 26365,
      "experiments parallel": 41052,
      "did make": 30916,
      "new documentlevel": 78887,
      "preference baseline": 89238,
      "model consistent": 70889,
      "predict source": 88938,
      "learn contextualized": 62008,
      "help case": 49702,
      "represent contextual": 99101,
      "noise generation": 80052,
      "fact different": 42819,
      "independent translation": 53780,
      "represented graph": 100029,
      "lexical consistency": 63745,
      "construct document": 21630,
      "multiencoder models": 75101,
      "makes efficient": 66778,
      "model search": 71954,
      "contrastive test": 22741,
      "substantial improvements translation": 114867,
      "augmented neural network": 10330,
      "inconsistencies paper propose": 53436,
      "information performance neural": 54836,
      "21 bleu points": 758,
      "machine translation extended": 65988,
      "ignoring crosssentence links": 51684,
      "crosssentence links dependencies": 25074,
      "nmt models paper": 79919,
      "context experiments evaluation": 22096,
      "model effective improving": 71043,
      "neural machine translations": 78136,
      "improve translation coherence": 52570,
      "experiments parallel corpora": 41053,
      "allows model trained": 5173,
      "contextualized sentence embeddings": 22568,
      "help case study": 49703,
      "represent contextual information": 99102,
      "translation performance strong": 125116,
      "representation contextual information": 99198,
      "substantial improvements translation quality": 114868,
      "information performance neural machine": 54837,
      "neural machine translation extended": 78004,
      "ignoring crosssentence links dependencies": 51685,
      "experimental results chineseenglish englishgerman": 40567,
      "demonstrate proposed model effective": 28850,
      "neural machine translation documentlevel": 77991,
      "improves translation performance strong": 53059,
      "information performance neural machine translation": 54838,
      "neural machine translation documentlevel machine": 77992,
      "documentlevel neural machine translation documentlevel": 33156,
      "959": 1427,
      "warranted": 132036,
      "ssl": 112092,
      "capable achieving": 15190,
      "classification combining": 17149,
      "scale document": 103715,
      "mark sentences": 67223,
      "model naturally": 71560,
      "explanations predictions": 41298,
      "existing recursive": 40264,
      "analyze sentences": 6005,
      "framework detecting": 45491,
      "framework obtains": 45633,
      "different rnn": 31391,
      "sequential encoding": 107225,
      "sa task": 103520,
      "dataset transferred": 27247,
      "task perspective": 118532,
      "design dynamic": 29533,
      "introduce class": 56391,
      "rnns convolution": 102966,
      "account language": 2028,
      "language compositionality": 58904,
      "count vectors": 24416,
      "learning ssl": 63049,
      "core concept": 23317,
      "concept discovery": 20565,
      "set domainspecific": 107424,
      "3d scene": 992,
      "grounding natural": 49121,
      "challenges context": 16143,
      "parameters furthermore": 84752,
      "study interpretability": 114409,
      "typical text": 126406,
      "sentences articles": 106218,
      "enrich semantic": 37557,
      "representations end": 99628,
      "adversarial losses": 4142,
      "obtain sentence representation": 81319,
      "classification tasks primarily": 17453,
      "text classification combining": 120714,
      "existing recursive neural": 40265,
      "dependency information sentence": 29148,
      "proposed models achieves": 93493,
      "stateoftheart performances datasets": 112861,
      "analysis sa task": 5826,
      "networks convolutional neural": 77556,
      "models sentence classification": 74000,
      "applying neural network": 7264,
      "semisupervised learning ssl": 105610,
      "new approach natural": 78793,
      "different neural modules": 31294,
      "novel text classification": 80751,
      "neural network layer": 78318,
      "paper present experimental": 84100,
      "propose sequential convolutional": 93045,
      "model comparison existing": 70860,
      "enrich semantic representation": 37558,
      "trained crossentropy loss": 123104,
      "parsing natural language inference": 85171,
      "sentiment analysis sa task": 106647,
      "new approach natural language": 78794,
      "results text classification task": 102271,
      "networks long shortterm memory networks": 77649,
      "hyperpartisan": 51238,
      "publishers": 94378,
      "eyecatching": 42634,
      "oh": 81633,
      "60000": 1152,
      "guardian": 49222,
      "sway": 116256,
      "soul": 110699,
      "hyperpartisan news": 51239,
      "media outlets": 68137,
      "focus automatic": 44737,
      "effective computational": 34649,
      "elusive challenge": 35332,
      "representations attending": 99508,
      "incorporating different": 53530,
      "detecting unreliable": 29857,
      "number instances": 80898,
      "systems detecting": 116844,
      "traditional classification": 122806,
      "identifying understanding": 51633,
      "unreliable information": 127508,
      "reliable unreliable": 98623,
      "articles finally": 9189,
      "regular news": 97728,
      "set news": 107509,
      "cost manual": 24363,
      "manual factchecking": 66997,
      "bert ensemble": 13118,
      "propaganda spreads": 92382,
      "sentences article": 106217,
      "making informed": 66847,
      "use differences": 128000,
      "classifiers classifier": 17602,
      "outlets news": 82625,
      "spread misinformation": 112049,
      "political news": 87996,
      "languages remain": 60841,
      "used conduct": 128446,
      "stock price": 113365,
      "2020 challenge": 722,
      "final submitted": 44129,
      "negative impacts": 76930,
      "disciplines including": 32018,
      "linguistics study": 64626,
      "facilitate detection": 42764,
      "news website": 79397,
      "short news": 108224,
      "fake ones": 43020,
      "research contribution": 100452,
      "news sites": 79365,
      "business models": 14979,
      "truth make": 125883,
      "paper focus automatic": 83947,
      "paper make use": 84038,
      "remains elusive challenge": 98797,
      "study feature selection": 114383,
      "structure news articles": 113923,
      "regular news articles": 97729,
      "test set news": 120509,
      "novel data set": 80527,
      "performance adversarial training": 86131,
      "2016 presidential election": 614,
      "generalizes different types": 46838,
      "neural networks paper presents": 78532,
      "different social media platforms": 31436,
      "algorithm checking": 4703,
      "worst case": 135072,
      "introducing large": 56640,
      "size significant": 109944,
      "models included": 73376,
      "permitting simple": 87051,
      "grammars use": 48684,
      "model optimality": 71615,
      "active candidates": 3105,
      "multiple context": 75527,
      "new results": 79111,
      "having certain": 49553,
      "random variable": 95514,
      "recently explored": 96689,
      "normal form": 80327,
      "study parsing": 114464,
      "parser input": 84964,
      "specified annotation": 111621,
      "aspects english": 9381,
      "express meaning": 41702,
      "process prior": 91552,
      "able parse sentences": 1669,
      "parse sentence syntactic": 84886,
      "nounphrase": 80437,
      "pythonbased": 94490,
      "cqs": 24585,
      "sumo": 115662,
      "web order": 132240,
      "concepts existing": 20617,
      "existing ontologies": 40247,
      "concepts word": 20651,
      "useful effective": 128882,
      "present complete": 89409,
      "content existing": 21880,
      "concepts problem": 20635,
      "used independently": 128587,
      "generic domainspecific": 47797,
      "chains using": 15985,
      "method linking": 68944,
      "means identifying": 67764,
      "include comparison": 53222,
      "ontologies paper": 81831,
      "documents preprocessing": 33268,
      "concepts experimental": 20619,
      "semantics languages": 105429,
      "event ontology": 39530,
      "fact review": 42832,
      "relations plain": 98235,
      "postevaluation runs": 88495,
      "detecting hypernymy": 29838,
      "using improved": 129748,
      "annotation nlp": 6343,
      "require domainspecific": 100130,
      "languages comparison": 60457,
      "data drawing": 25868,
      "isa relations": 56960,
      "identifying concept": 51591,
      "ones previous work": 81704,
      "concepts existing ontologies": 20618,
      "concepts word senses": 20652,
      "concepts experimental results": 20620,
      "relations plain text": 98236,
      "perform better random": 85955,
      "methods suffer error": 69780,
      "domains approach outperforms": 33733,
      "building knowledge base": 14850,
      "perform better random baseline": 85956,
      "methods suffer error propagation": 69781,
      "phobert": 87269,
      "efficiently use": 35158,
      "use past": 128195,
      "sparse coding": 111223,
      "evaluate sequence": 38924,
      "presents stateoftheart": 89901,
      "recognition event": 96876,
      "best discrete": 13326,
      "discriminative sequence": 32217,
      "models practical": 73763,
      "previous tags": 90508,
      "nlp annotation": 79563,
      "present featurebased": 89494,
      "useful practitioners": 128916,
      "main shortcomings": 66462,
      "covering languages": 24543,
      "experiments increasing": 40962,
      "presents neural": 89871,
      "sentences supported": 106511,
      "problem compared": 90969,
      "datasets empirically": 27444,
      "phobert vietnamese": 87270,
      "network word embedding": 77476,
      "multilingual sequence labeling": 75365,
      "improves performance supervised": 53021,
      "different network design": 31287,
      "network design choices": 77223,
      "entity recognition event": 38015,
      "recognition event detection": 96877,
      "tasks including word": 119194,
      "paper present featurebased": 84101,
      "proposed attentive neural": 93231,
      "neural network achieves": 78236,
      "paper presents neural": 84183,
      "stateoftheart performances tasks": 112865,
      "treated sequence tagging": 125561,
      "problem compared existing": 90970,
      "classification based methods": 17132,
      "new model architectures": 79027,
      "released open source": 98489,
      "largescale monolingual language": 61465,
      "nlp research applications": 79682,
      "different network design choices": 31288,
      "entity recognition event detection": 38016,
      "achieves stateoftheart performances tasks": 2897,
      "results demonstrate proposed approach significantly": 101700,
      "tagging named entity recognition dependency": 117417,
      "evidential": 39684,
      "worthiness": 135074,
      "increased focus": 53629,
      "conducted variety": 20958,
      "entailment given": 37665,
      "identifying appropriate": 51580,
      "asks predict": 9310,
      "participate lab": 85331,
      "true facts": 125847,
      "frames framenet": 45399,
      "articles best": 9174,
      "labels solely": 58643,
      "scores second": 104201,
      "clues identify": 17920,
      "quality core": 94620,
      "finally support": 44236,
      "future systems": 46316,
      "develop baseline": 30178,
      "features world": 43794,
      "data highly": 26002,
      "identifying beneficial": 51583,
      "pair separately": 83453,
      "input claim": 55305,
      "model scientific": 71949,
      "task asks predict": 117897,
      "detection news articles": 30006,
      "labels solely based": 58644,
      "test set work": 120525,
      "improvement future systems": 52714,
      "additional features including": 3517,
      "new publicly available dataset": 79093,
      "proposed model outperforms stateoftheart models": 93470,
      "friedman": 45887,
      "scored quality": 104144,
      "nlg training": 79530,
      "evaluation relies": 39357,
      "traditional automatic": 122798,
      "nlg use": 79531,
      "despite challenging": 29679,
      "automatic scores": 10644,
      "evaluating generation": 39059,
      "propose syntactic": 93099,
      "evaluate using": 38940,
      "metrics finally": 69963,
      "nlg research": 79526,
      "highlight promising": 50272,
      "promising approaches": 92266,
      "nonneural systems": 80256,
      "directly evaluate": 31875,
      "showing method": 108413,
      "dataset relative": 27151,
      "model reducing": 71877,
      "growing natural": 49174,
      "single parameter": 109779,
      "workers generate": 134891,
      "new human": 78948,
      "adequacy scores": 3827,
      "different nlg": 31299,
      "key contributions": 57562,
      "gap proposing": 46475,
      "surprisingly high": 116140,
      "better outputs": 13641,
      "generation conversational responses": 47347,
      "automatic scores human": 10645,
      "cost human evaluation": 24359,
      "systems commonly evaluated": 116794,
      "growing natural language": 49175,
      "new human evaluation": 78949,
      "bridge gap proposing": 14595,
      "generation evaluation metrics": 47393,
      "evaluation language generation": 39248,
      "machine learning approaches natural": 65774,
      "natural language generation evaluation": 76302,
      "language generation evaluation metrics": 59072,
      "machine learning approaches natural language": 65775,
      "language generation natural language generation": 59079,
      "formalizes": 45201,
      "inputs using": 55498,
      "discriminator used": 32226,
      "sound like": 110704,
      "training rnn": 123822,
      "generation experiment": 47396,
      "uninformative responses": 127169,
      "increases diversity": 53642,
      "response generated": 101197,
      "generation grammatical": 47425,
      "editing model": 34546,
      "studies benchmarks": 114189,
      "usually built": 130410,
      "approaches benefit": 8085,
      "generation distinct": 47372,
      "generation difficult": 47370,
      "proposed generative": 93304,
      "success generating": 115082,
      "compared various": 19487,
      "level shown": 63505,
      "process achieve": 91416,
      "turn decoded": 125971,
      "response sentence": 101243,
      "number problems": 80944,
      "previously observed": 90607,
      "generation posterior": 47547,
      "language taskoriented": 60150,
      "uses straightthrough": 129283,
      "straightthrough gumbelsoftmax": 113438,
      "gumbelsoftmax estimator": 49279,
      "training effectively": 123598,
      "iterative translation": 57138,
      "significantly diverse": 108906,
      "relevance context": 98505,
      "semi autoregressive": 105564,
      "learning multiturn": 62820,
      "responses address": 101253,
      "space generate": 111007,
      "task suffer": 118758,
      "address robustness": 3761,
      "dialogue generation paper": 30681,
      "conclude future work": 20713,
      "generation experiment results": 47397,
      "systems usually built": 117226,
      "generation distinct existing": 47373,
      "empirical study large": 36202,
      "benchmark dataset shows": 12760,
      "generation difficult train": 47371,
      "multiturn response generation": 75923,
      "attention mechanism incorporates": 9900,
      "evaluation experiments demonstrate": 39214,
      "previous research task": 90451,
      "outperform stateoftheart generation": 82753,
      "generation models terms": 47488,
      "models terms response": 74178,
      "relevant diverse responses": 98542,
      "demonstrate efficacy approach": 28728,
      "model uses straightthrough": 72269,
      "uses straightthrough gumbelsoftmax": 129284,
      "straightthrough gumbelsoftmax estimator": 113439,
      "human evaluation showed": 50833,
      "human evaluation experiments demonstrate": 50815,
      "significantly outperform stateoftheart generation": 108989,
      "outperform stateoftheart generation models": 82754,
      "stateoftheart generation models terms": 112665,
      "generation models terms response": 47489,
      "based generative adversarial nets": 11737,
      "welleck et al 2019": 132408,
      "model uses straightthrough gumbelsoftmax": 72270,
      "uses straightthrough gumbelsoftmax estimator": 129285,
      "model significantly outperform stateoftheart generation": 72027,
      "significantly outperform stateoftheart generation models": 108990,
      "outperform stateoftheart generation models terms": 82755,
      "stateoftheart generation models terms response": 112666,
      "model uses straightthrough gumbelsoftmax estimator": 72271,
      "rightbranching": 102840,
      "uncertaintyaware": 126628,
      "basic nmt": 12532,
      "major topics": 66591,
      "topics cover": 122619,
      "uncertainty neural": 126623,
      "metrics assess": 69923,
      "weights neural": 132381,
      "right right": 102835,
      "containing substantial": 21808,
      "require costly": 100124,
      "considering robustness": 21331,
      "defined types": 28506,
      "robustness machine": 103107,
      "difficulty exploits": 31696,
      "sentences surprisingly": 106512,
      "use augment": 127903,
      "tasks reveal": 119475,
      "choices models": 16917,
      "noise robust": 80065,
      "sequence fed": 106942,
      "unseen types": 127554,
      "spelling grammatical": 111906,
      "properties paper": 92472,
      "draw attention": 34110,
      "attention fact": 9836,
      "observe strong": 81212,
      "possibility developing": 88368,
      "conduct indepth analysis": 20881,
      "basic nmt models": 12533,
      "uncertainty neural machine": 126624,
      "human reference translations": 50953,
      "nmt systems improve": 79975,
      "neural statistical machine": 78689,
      "attention weights neural": 10038,
      "machine translation usually": 66296,
      "information source sentences": 54993,
      "models trained limited": 74214,
      "recently attracted great": 96661,
      "robustness machine translation": 103108,
      "paper present adversarial": 84076,
      "nmt models based": 79908,
      "nature neural machine": 76663,
      "current nmt models": 25318,
      "machine translation modeling": 66050,
      "translation demonstrate proposed": 124742,
      "model robustness noisy": 71944,
      "paper draw attention": 83885,
      "uncertainty neural machine translation": 126625,
      "neural statistical machine translation": 78690,
      "neural machine translation usually": 78131,
      "performance experiment results method": 86355,
      "nature neural machine translation": 76664,
      "neural machine translation modeling": 78028,
      "experiments chineseenglish translation demonstrate": 40828,
      "chineseenglish translation demonstrate proposed": 16861,
      "experiments chineseenglish translation demonstrate proposed": 40829,
      "variegated": 130945,
      "corpusbased natural": 24079,
      "method getting": 68863,
      "method tested": 69185,
      "18 training": 421,
      "location extraction": 64963,
      "scores proposed": 104195,
      "using main": 129852,
      "operations propose": 82073,
      "combines bert": 18679,
      "autoregressive transformer": 10870,
      "new writing": 79256,
      "creative way": 24732,
      "methods inferior": 69553,
      "tokens order": 122322,
      "order output": 82377,
      "missing tokens": 70335,
      "employs novel": 36332,
      "editing process": 34551,
      "openvocabulary approach": 82042,
      "indicates new": 53863,
      "normalized text": 80359,
      "important technology": 52278,
      "mechanism input": 67997,
      "normalization model": 80339,
      "existing nonautoregressive": 40241,
      "encoding entire": 36691,
      "outputs experiments": 83167,
      "build interactive": 14775,
      "corpusbased natural language": 24080,
      "unsupervised pretraining large": 127697,
      "autoregressive transformer decoder": 10871,
      "data improve models": 26023,
      "establish baseline performance": 38578,
      "overcome challenges propose": 83277,
      "seq2seq learning task": 106897,
      "model employs novel": 71065,
      "target sentence experiments": 117698,
      "model smaller model": 72063,
      "fast inference speed": 43140,
      "approach yields better performance": 8025,
      "data improve models performance": 26024,
      "proposed approach yields better performance": 93223,
      "natural language understanding tasks paper": 76587,
      "kay": 57498,
      "sproat": 112056,
      "producer": 91982,
      "archived": 8862,
      "fundamentals": 46135,
      "intersect": 56303,
      "apparatus": 6784,
      "slopes": 110031,
      "gestural": 47947,
      "semiotic": 105586,
      "sigtyp": 109059,
      "words analyzed": 133815,
      "special mechanisms": 111365,
      "counts given": 24452,
      "producer consumer": 91983,
      "relation described": 97965,
      "information special": 54999,
      "underlying theoretical": 126707,
      "languages set": 60864,
      "main difficulties": 66416,
      "compared obtained": 19401,
      "words persian": 134114,
      "conveys information": 23128,
      "phonological phonetic": 87311,
      "properties linguistic": 92463,
      "identification cases": 51368,
      "timing patterns": 122209,
      "language selection": 60074,
      "focus fundamental": 44767,
      "structures present": 114097,
      "present human": 89511,
      "sound paper": 110705,
      "age old": 4287,
      "age group": 4284,
      "conditions including": 20817,
      "communication discussed": 19030,
      "semantics used": 105478,
      "suggest ways": 115338,
      "sigtyp 2020": 109060,
      "model scale": 71948,
      "ii conditional": 51700,
      "compared obtained using": 19402,
      "paper present probabilistic": 84129,
      "human language work": 50903,
      "night": 79484,
      "shops": 108198,
      "blogger": 14369,
      "users presence": 129155,
      "communication analyzed": 19026,
      "affect language": 4218,
      "stages feature": 112157,
      "annotated authors": 6140,
      "words modelling": 134058,
      "swedish language": 116264,
      "uses text": 129292,
      "broad field": 14671,
      "geographic entities": 47850,
      "semantics facilitate": 105416,
      "semantics different": 105408,
      "algorithms implementation": 4858,
      "datasets sufficient": 27737,
      "restaurants hotels": 101328,
      "reveal valuable": 102507,
      "studies examine": 114220,
      "disambiguation work": 31999,
      "rigorous evaluation": 102852,
      "dataset suitable": 27225,
      "blogger users": 14370,
      "different media": 31252,
      "dimensions state": 31773,
      "expressions words": 41769,
      "information twitter": 55061,
      "process mining": 91527,
      "related natural": 97884,
      "english tweet": 37320,
      "error approach": 38289,
      "investigate information present": 56770,
      "using latent variable": 129800,
      "model parameters training": 71706,
      "years lot research": 135270,
      "compared stateoftheart systems": 19463,
      "event detection paper": 39510,
      "work conduct empirical": 134427,
      "rigorous evaluation including": 102853,
      "propose formal representation": 92685,
      "information twitter social": 55062,
      "neural networks finally": 78483,
      "approach based bert": 7378,
      "work conduct empirical study": 134428,
      "characterwise": 16634,
      "texts joint": 121534,
      "based constraints": 11599,
      "exploited variety": 41455,
      "represent events": 99108,
      "derived various": 29366,
      "mentions types": 68470,
      "event mention": 39525,
      "types zeroshot": 126390,
      "model post": 71751,
      "data leads": 26081,
      "event nugget": 39528,
      "traditional event": 122817,
      "2015 event": 589,
      "proposal networks": 92520,
      "kbp 2017": 57507,
      "tackbp 2015": 117281,
      "learning trained": 63114,
      "arguments event": 8984,
      "propagated event": 92385,
      "trigger argument": 125782,
      "collected manually": 18425,
      "method event": 68819,
      "model financial": 71184,
      "exhibit following": 39999,
      "restricts applications": 101352,
      "analyses source": 5468,
      "integrating event": 55792,
      "templates approach": 120083,
      "entity coreference": 37916,
      "higher computational": 50169,
      "result learned": 101380,
      "effectively generalize": 34814,
      "target types": 117737,
      "type attack": 126188,
      "human informationseeking": 50864,
      "neural event": 77912,
      "features derived various": 43445,
      "training data leads": 123484,
      "event nugget detection": 39529,
      "capture structural semantic": 15401,
      "tac kbp 2017": 117278,
      "errors propagated event": 38402,
      "extraction using neural": 42534,
      "dataset collected manually": 26796,
      "argument extraction event": 8943,
      "supervised sequence labeling": 115835,
      "significantly higher computational": 108919,
      "fully supervised models": 45973,
      "recurrent neural networks traditional": 97251,
      "event argument extraction event": 39499,
      "translation texts": 125379,
      "model infers": 71341,
      "matching algorithms": 67395,
      "output experiment": 83074,
      "agreement rates": 4402,
      "biases work": 13890,
      "correlated translation": 24211,
      "edition shared": 34557,
      "czech polish": 25497,
      "outputs translation": 83187,
      "able increase": 1657,
      "simulated environment": 109637,
      "translations specific": 125494,
      "language speak": 60110,
      "significant human": 108762,
      "mathematical text": 67463,
      "implemented machine": 51966,
      "universal document": 127304,
      "translation wmt20": 125429,
      "metrics proposed literature": 69997,
      "interannotator agreement rates": 56046,
      "correlated translation quality": 24212,
      "large improvements baselines": 61108,
      "edition shared task": 34558,
      "test data performance": 120436,
      "reports paper describes": 99086,
      "general news domain": 46686,
      "implemented machine translation": 51967,
      "machine translation wmt20": 66301,
      "evaluation metrics proposed literature": 39293,
      "wmt19 news translation task": 132858,
      "9146": 1388,
      "combining various": 18757,
      "machines classification": 66332,
      "sets obtained": 107691,
      "manually collected": 67044,
      "semantic orientation": 105130,
      "reviews websites": 102639,
      "specific user": 111515,
      "section devoted": 104485,
      "entitylevel sentiment": 38142,
      "articles goal": 9191,
      "main input": 66433,
      "data readers": 26317,
      "algorithm detect": 4717,
      "reviews specifically": 102631,
      "accuracy 956": 2087,
      "task authors": 117906,
      "related subtasks": 97899,
      "develop hybrid": 30205,
      "classifiers terms": 17637,
      "applied datasets": 7055,
      "lead wrong": 61877,
      "particular class": 85391,
      "trained combination": 123094,
      "predicting rating": 89004,
      "vector machines classification": 131321,
      "lexical resources natural": 63803,
      "paper present set": 84132,
      "classifiers terms accuracy": 17638,
      "results showed best": 102172,
      "develop algorithm detect": 30171,
      "review sentiment analysis": 102576,
      "using text mining": 130287,
      "parts speech tagging": 85589,
      "support vector machines classification": 116010,
      "lexical resources natural language": 63804,
      "lexical resources natural language processing": 63805,
      "conceptualizing": 20678,
      "illequipped": 51730,
      "communication provide": 19041,
      "embeddings currently": 35628,
      "task suggesting": 118762,
      "communication automatic": 19027,
      "features tried": 43767,
      "chat second": 16660,
      "follows given": 44992,
      "vastly improve": 131284,
      "words argue": 133825,
      "study exploit": 114376,
      "task iii": 118272,
      "computational models able": 20411,
      "prediction task result": 89136,
      "sentences automatically annotated": 106223,
      "headsup": 49612,
      "trained discriminatively": 123120,
      "neural parsing": 78628,
      "approximates original": 8467,
      "combination improve": 18562,
      "tree learning": 125599,
      "work problem": 134704,
      "use parse": 128193,
      "grammars lexicons": 48676,
      "makes explicit": 66781,
      "propose potential": 92992,
      "accuracy figures": 2164,
      "sentences composing": 106251,
      "according induced": 1983,
      "based downstream": 11663,
      "baselines use": 12485,
      "success latent": 115087,
      "motivated development": 74847,
      "efficiently incorporating": 35148,
      "trees provided": 125707,
      "trees better": 125683,
      "does produce": 33380,
      "demonstrated useful": 28934,
      "attention enhance": 9830,
      "plms learn": 87755,
      "final tree": 44137,
      "decoding framework model": 28095,
      "latent tree learning": 61622,
      "tree learning possible": 125600,
      "use external data": 128043,
      "sentences composing words": 106252,
      "composing words according": 20087,
      "words according induced": 133802,
      "rise deep learning": 102865,
      "success latent tree": 115088,
      "able achieve lower": 1596,
      "better understand performance": 13757,
      "latent tree learning possible": 61623,
      "sentences composing words according": 106253,
      "composing words according induced": 20088,
      "shen et al 2018": 108169,
      "success latent tree learning": 115089,
      "sentences composing words according induced": 106254,
      "disambiguate different": 31940,
      "cognitive semantics": 18282,
      "access world": 1927,
      "high flexibility": 50073,
      "expressions tackle": 41764,
      "expressions use": 41766,
      "model events": 71117,
      "technical considerations": 119743,
      "model simulations": 72048,
      "compositions model": 20149,
      "jointly reason": 57386,
      "setting train": 107782,
      "employ reinforcement": 36276,
      "instructions complex": 55709,
      "automatically discovering": 10762,
      "error 50": 38276,
      "action generation": 3077,
      "symbolic world": 116323,
      "reasoning entities": 96246,
      "entities makes": 37814,
      "parameters dynamically": 84744,
      "humans approach": 51068,
      "learning naturallanguage": 62830,
      "finally generator": 44190,
      "natural language usually": 76600,
      "expressions tackle problem": 41765,
      "improvements supervised learning": 52926,
      "employ reinforcement learning": 36277,
      "language instructions complex": 59213,
      "addition propose new": 3465,
      "natural language communication": 76276,
      "uses language model": 129238,
      "natural language instructions complex": 76361,
      "goodman": 48513,
      "stricter": 113589,
      "improvements perplexity": 52896,
      "challenging critical": 16236,
      "model deliver": 70962,
      "model users": 72261,
      "user given": 128990,
      "context users": 22315,
      "use short": 128274,
      "containing various": 21811,
      "traits large": 123975,
      "schemes designed": 103951,
      "traits different": 123973,
      "generation personalized": 47542,
      "propose personalized": 92989,
      "model personality": 71747,
      "random guessing": 95498,
      "slots second": 110091,
      "multiturn user": 75928,
      "studies field": 114227,
      "characteristics human": 16525,
      "corpus confirm": 23714,
      "target distribution": 117602,
      "basic human": 12524,
      "corroborate previous": 24322,
      "way speaking": 132132,
      "influences language": 54324,
      "capture differences": 15288,
      "traits large number": 123976,
      "large number speakers": 61190,
      "introduce task automatic": 56550,
      "model generate natural": 71235,
      "study multiturn": 114445,
      "fact words": 42841,
      "generation sequential": 47616,
      "problem response": 91207,
      "domains require": 33849,
      "given product": 48093,
      "aspects address": 9371,
      "contexts response": 22425,
      "methods time": 69805,
      "time enjoys": 122023,
      "speed paper": 111873,
      "response utterances": 101250,
      "building knowledgegrounded": 14852,
      "representation best": 99179,
      "consists autoencoder": 21470,
      "research content": 100451,
      "personalized attention": 87141,
      "study multiturn response": 114446,
      "empirical studies automatic": 36196,
      "narrow domains require": 76191,
      "learn matching model": 62088,
      "time cost human": 122005,
      "study multiturn response generation": 114447,
      "memberships": 68275,
      "features observe": 43634,
      "choice training": 16907,
      "score previously": 104109,
      "method despite": 68768,
      "predicted mentions": 88962,
      "harms performance": 49511,
      "aims assign": 4503,
      "integrating symbolic": 55804,
      "continuous variables": 22640,
      "heuristic search": 49877,
      "experimentally approach": 40736,
      "flat predictions": 44637,
      "community encourage": 19073,
      "schema includes": 103899,
      "pipelined manner": 87565,
      "hyperbolic space": 51208,
      "hyperbolic embeddings": 51207,
      "tree prediction": 125606,
      "contextual clues": 22448,
      "symbolic data": 116313,
      "relations inventory": 98192,
      "attempts deal": 9767,
      "trained using different": 123327,
      "score previously proposed": 104110,
      "network model jointly": 77332,
      "methods attempt address": 69323,
      "features achieves stateoftheart": 43353,
      "work propose zeroshot": 134749,
      "datasets including standard": 27519,
      "neural network model jointly": 78333,
      "financial news": 44257,
      "proposes principled": 93615,
      "understanding financial": 126844,
      "time support": 122115,
      "way test": 132135,
      "sentence frame": 105880,
      "world proposed": 135044,
      "span surrounding": 111134,
      "human baseline": 50762,
      "reached performance": 95909,
      "little progress": 64826,
      "baseline previous": 12289,
      "tools research": 122471,
      "tackling task": 117327,
      "attention english": 9829,
      "examples sampled": 39874,
      "single model best": 109760,
      "finnishenglish": 44560,
      "fragile": 45369,
      "evaluation pretrained": 39332,
      "monolingual dutch": 74587,
      "network pipeline": 77389,
      "evaluations focused": 39463,
      "small treebank": 110222,
      "effectively propose": 34839,
      "including pos": 53351,
      "answering sequence": 6696,
      "languagespecific features": 60992,
      "results suite": 102237,
      "different strengths weaknesses": 31455,
      "models low resource": 73530,
      "language model modern": 59351,
      "question answering sequence": 95102,
      "answering sequence labeling": 6697,
      "multitask learning pretrained": 75861,
      "results models based": 101957,
      "question answering sequence labeling": 95103,
      "policymakers": 87978,
      "2017 paper": 634,
      "prediction techniques": 89139,
      "naturallanguage processing": 76634,
      "assess incremental": 9481,
      "category based": 15782,
      "headlines use": 49600,
      "complement current": 19727,
      "consisting manual": 21455,
      "based clusters": 11578,
      "certain term": 15958,
      "different events": 31135,
      "extractor extract": 42572,
      "news commodity": 79320,
      "international conference": 56183,
      "using different evaluation": 129571,
      "2017 paper describes": 635,
      "nlp tasks automatic": 79708,
      "current work propose": 25392,
      "dqn": 34079,
      "performance open": 86586,
      "hierarchical reinforcement": 49997,
      "ultimate goals": 126557,
      "task retrieving": 118658,
      "inference state": 54224,
      "distribution higher": 32657,
      "explore role": 41583,
      "contexts order": 22413,
      "translation multitask": 124995,
      "rarely considered": 95756,
      "improves dialogue": 52971,
      "multimodal features": 75428,
      "hierarchical reinforcement learning": 49998,
      "simple elegant approach": 109423,
      "significantly outperforms state": 109029,
      "inference state art": 54225,
      "multihop reasoning model": 75143,
      "objective subjective evaluations": 81119,
      "highly competitive benchmark": 50297,
      "competitive benchmark datasets": 19645,
      "translation multitask learning": 124996,
      "approach significantly outperforms state": 7911,
      "significantly outperforms state art": 109030,
      "highly competitive benchmark datasets": 50298,
      "competitive benchmark datasets related": 19646,
      "approach significantly outperforms state art": 7912,
      "highly competitive benchmark datasets related": 50299,
      "sensespecific": 105724,
      "learned predict": 62242,
      "tagging semantic": 117444,
      "countbased distributional": 24418,
      "representing semantics": 100063,
      "overcome inherent": 83285,
      "models seamlessly": 73981,
      "sentences embedding": 106290,
      "proposes address": 93594,
      "proposed modular": 93499,
      "words generate": 133969,
      "clusters embeddings": 17965,
      "represent single": 99142,
      "information sense": 54964,
      "learning sense": 62999,
      "similarity measure designed": 109258,
      "partofspeech tagging semantic": 85561,
      "proposed modular approach": 93500,
      "model generalizes previous": 71231,
      "method learn word": 68935,
      "multiple word senses": 75748,
      "language model predict": 59369,
      "word embedding paper": 133028,
      "embeddings present novel": 35871,
      "results model outperforms previous models": 101944,
      "gamified": 46426,
      "realizes": 96124,
      "speaker uses": 111301,
      "steps training": 113338,
      "needs successfully": 76901,
      "capabilities work": 15177,
      "introduce language": 56443,
      "learning biases": 62409,
      "humangenerated data": 51022,
      "indicate progress": 53848,
      "sentences evaluated": 106296,
      "meaning input": 67643,
      "tree encoders": 125588,
      "apply domain": 7173,
      "information associated": 54381,
      "features learn": 43589,
      "representations empirically": 99622,
      "points wmt14": 87885,
      "maintaining model": 66522,
      "standard decoders": 112228,
      "applies existing": 7152,
      "representing meaning input": 100057,
      "bleu points wmt14": 14316,
      "mellon": 68266,
      "mbr": 67576,
      "pros": 93636,
      "differentiable experiments": 31583,
      "mellon university": 68267,
      "bayesrisk decoding": 12590,
      "translation according": 124616,
      "new hypotheses": 78951,
      "contribution release": 22776,
      "decoding tasks": 28139,
      "hypotheses contain": 51252,
      "research research": 100617,
      "numerous research": 81028,
      "pros cons": 93637,
      "example decoder": 39779,
      "shorter translations": 108298,
      "right beam": 102828,
      "new technology": 79213,
      "translation challenges": 124690,
      "explanation methods": 41283,
      "impact online": 51881,
      "machine translation unlike": 66283,
      "minimum bayesrisk decoding": 70204,
      "machine translation allows": 65892,
      "large beam size": 61038,
      "right beam search": 102829,
      "machine translation challenges": 65920,
      "languages deep neural": 60486,
      "neural machine translation unlike": 78123,
      "languages deep neural networks": 60487,
      "data providing": 26303,
      "performance train": 86806,
      "approach challenge": 7414,
      "features architecture": 43366,
      "sufficient examples": 115278,
      "havent explored": 49550,
      "nlp approach": 79575,
      "includes different": 53253,
      "deal imbalanced": 27858,
      "characters text": 16625,
      "regular english": 97724,
      "propose unique": 93132,
      "considers similarity": 21342,
      "learning approach text": 62351,
      "intrinsic extrinsic measures": 56362,
      "various applications like": 131038,
      "benchmark datasets results demonstrate": 12802,
      "domain natural language processing": 33593,
      "answers observe": 6738,
      "pretrained millions": 90127,
      "report objective": 99018,
      "paper derive": 83806,
      "quantify level": 94852,
      "groundwork future": 49132,
      "data using neural": 26610,
      "text analysis research": 120641,
      "source information users": 110767,
      "effectiveness proposed approaches": 34928,
      "models pretrained millions": 73786,
      "problem previous studies": 91170,
      "previous research shows": 90450,
      "probabilitybased": 90874,
      "new translations": 79230,
      "problem translation": 91270,
      "quality smaller": 94793,
      "does learn": 33364,
      "introduce structured": 56545,
      "decoding machine": 28102,
      "word step": 133588,
      "complex lexical": 19827,
      "approach powerful": 7803,
      "decoding beam search": 28078,
      "decoding machine translation": 28103,
      "machine translation recent advances": 66188,
      "neural machine translation recent advances": 78087,
      "superordinate": 115711,
      "corpusindependent": 24084,
      "curie": 25250,
      "saliency scores": 103545,
      "distribution investigate": 32662,
      "generalized word": 46833,
      "approaches enable": 8142,
      "distributional contexts": 32697,
      "large realworld": 61236,
      "class task": 17050,
      "set finetuning": 107446,
      "classification methods based": 17268,
      "paper propose way": 84344,
      "tasks existing work": 119116,
      "nlp tasks existing work": 79718,
      "prediction online": 89095,
      "science literature": 103978,
      "accuracy 88": 2078,
      "content task": 21958,
      "public awareness": 94235,
      "model informational": 71343,
      "introduce classification": 56392,
      "researchers analyze": 100677,
      "predict news": 88907,
      "target news": 117679,
      "improvements applying": 52805,
      "inference propose novel": 54204,
      "patterns different levels": 85739,
      "generation recurrent": 47589,
      "networks generation": 77612,
      "baseline text": 12324,
      "bagofwords document": 11373,
      "embeddings corpora": 35619,
      "baselines code": 12368,
      "capture quality": 15365,
      "working tandem": 134906,
      "difficult directly": 31621,
      "discriminator guide": 32224,
      "feature distributions": 43265,
      "generation related": 47592,
      "language generation recurrent": 59094,
      "generation model called": 47479,
      "existing baselines code": 40078,
      "baselines code available": 12369,
      "results using language": 102302,
      "sentence generation method": 105889,
      "generation method based": 47472,
      "conduct extensive experimental": 20869,
      "existing text generation methods": 40316,
      "existing baselines code available": 40079,
      "results using language model": 102303,
      "sentence generation method based": 105890,
      "using unigram": 130336,
      "dictionary word": 30900,
      "specifically selected": 111590,
      "effect automatic": 34586,
      "just word": 57474,
      "proposed pipeline": 93520,
      "text using recurrent": 121398,
      "overall average f1": 83215,
      "finegrained pos tags": 44373,
      "text using recurrent neural": 121399,
      "overall average f1 score": 83216,
      "accuracy suggesting": 2293,
      "emerging text": 36075,
      "edited headlines": 34541,
      "classifiers predict": 17626,
      "attempt generate": 9745,
      "labels produced": 58631,
      "classification method based": 17266,
      "make use knowledge": 66745,
      "shown good performance": 108475,
      "meaning similar": 67691,
      "increasingly applied": 53691,
      "human expectations": 50853,
      "patterns demonstrating": 85737,
      "relatively standard": 98418,
      "forms linguistic": 45248,
      "experiments study": 41156,
      "high context": 50049,
      "sentences labels": 106367,
      "space lastly": 111012,
      "independent test": 53778,
      "time average": 121989,
      "model preferred": 71764,
      "corpora encoding": 23468,
      "performance yielding": 86866,
      "significantly improving previous": 108964,
      "data integrate": 26048,
      "information similar": 54983,
      "issue perform": 57019,
      "particular contexts": 85397,
      "use subsequence": 128304,
      "rural communities": 103484,
      "strategies building": 113453,
      "successful nonautoregressive": 115167,
      "ar models": 8484,
      "training language modeling": 123674,
      "autoregressive ar models": 10860,
      "largely absent": 61332,
      "articles model": 9199,
      "observations training": 81185,
      "better f1": 13585,
      "dataset notably": 27058,
      "news articles model": 79300,
      "generalization ability paper": 46764,
      "ability paper propose": 1557,
      "generalization ability paper propose": 46765,
      "rankaware": 95631,
      "built training": 14942,
      "review detection": 102547,
      "fake reviews": 43024,
      "product business": 92028,
      "model team": 72150,
      "greatly help": 49052,
      "study follow": 114388,
      "expert rules": 41229,
      "users experiences": 129114,
      "traditional way": 122883,
      "reviews mainly": 102616,
      "abstractive opinion": 1805,
      "ineffective paper": 54067,
      "challenges noisy": 16186,
      "representative sample": 100020,
      "systems outperform baselines": 117037,
      "built training data": 14943,
      "fake review detection": 43023,
      "product business reviews": 92029,
      "ineffective paper propose": 54068,
      "neural network model able": 78329,
      "automatons": 10849,
      "precision 82": 88788,
      "82 recall": 1304,
      "trained social": 123284,
      "outperforms monolingual": 82919,
      "task attracted": 117905,
      "participating runs": 85355,
      "ensemble technique": 37621,
      "majority submissions": 66608,
      "precision 82 recall": 88789,
      "trained social media": 123285,
      "construal": 21617,
      "zero pronouns": 135479,
      "improving pronoun": 53149,
      "reference implementation": 97522,
      "role application": 103157,
      "method rulebased": 69112,
      "precision identification": 88793,
      "sets news": 107690,
      "traditional systems": 122878,
      "approach detection": 7486,
      "available systems": 11120,
      "proposal new": 92521,
      "high precision identification": 50105,
      "needs translation": 76902,
      "method resolve": 69107,
      "properties compositionality": 92440,
      "increase overall": 53608,
      "translations translation": 125502,
      "special domain": 111358,
      "obtaining comparable": 81439,
      "corpus currently": 23739,
      "corpora texts": 23602,
      "difficulties analysis": 31676,
      "based approach outperforms": 11515,
      "processing recent years": 91781,
      "better performance various": 13665,
      "language processing recent years": 59938,
      "natural language processing recent years": 76479,
      "cataphora": 15707,
      "article explore": 9129,
      "translation learns": 124902,
      "evaluation contextaware": 39160,
      "pronouns paper": 92360,
      "generate translation": 47037,
      "compared aligned": 19321,
      "twofold provide": 126143,
      "curated datasets": 25245,
      "datasets surprisingly": 27742,
      "machine translation learns": 66028,
      "previous works using": 90581,
      "properly evaluate performance": 92432,
      "neural machine translation learns": 78017,
      "articulated": 9233,
      "concepts emotions": 20615,
      "challenge humans": 16040,
      "meaning computational": 67624,
      "models lstms": 73538,
      "conditional language modeling task": 20762,
      "models imitation": 73354,
      "propose imitation": 92716,
      "conditioned previously": 20804,
      "score relative": 104114,
      "inference computationally": 54126,
      "translation mechanism": 124929,
      "models imitation learning": 73355,
      "propose imitation learning": 92717,
      "construed": 21711,
      "discuss plans": 32275,
      "quality nlg": 94734,
      "contextaware natural": 22340,
      "performance transferring": 86812,
      "scores providing": 104196,
      "preceding context": 88776,
      "language responses": 60061,
      "multidomain datasets": 75087,
      "talk dynamically": 117557,
      "leveraging linguistic": 63687,
      "models researchers": 73943,
      "mrs paired": 74937,
      "slot descriptions": 110035,
      "contextaware natural language": 22341,
      "natural language responses": 76528,
      "generalize new unseen": 46815,
      "fail generate complex": 42964,
      "fix errors": 44592,
      "language teachers": 60152,
      "highlevel concepts": 50243,
      "make acceptable": 66617,
      "includes annotations": 53251,
      "label using": 58419,
      "corpora construction": 23441,
      "translation parsing": 125100,
      "provides direct": 94020,
      "loss quality": 65295,
      "dataset includes annotations": 26978,
      "machine translation parsing": 66150,
      "zhang et al 2017": 135539,
      "periodically": 87039,
      "gaelic": 46333,
      "summary workshop": 115653,
      "corpora identify": 23497,
      "language tree indoeuropean": 60192,
      "optimises": 82183,
      "assigning score": 9558,
      "automatic essay": 10523,
      "domain hand": 33543,
      "similarity scoring": 109302,
      "grade essays": 48584,
      "extracted multiple": 42166,
      "causes poor": 15839,
      "features extracted multiple": 43508,
      "results transfer learning": 102282,
      "treebank annotation": 125626,
      "arbitrary source": 8591,
      "treebank line": 125640,
      "treebanks based": 125658,
      "based treebank": 12137,
      "improvements single": 52909,
      "evaluation ranked": 39349,
      "inducing dependency": 54015,
      "method based syntactic": 68669,
      "based syntactic parse": 12102,
      "data target languages": 26543,
      "method based syntactic parse": 68670,
      "continuous semantic": 22631,
      "preferences account": 89243,
      "require additional": 100110,
      "autoencoders vaes": 10419,
      "topics associated": 122614,
      "apply adaptive": 7157,
      "experiment results realworld": 40506,
      "variational autoencoders vaes": 130919,
      "experiment results realworld datasets": 40507,
      "provo": 94148,
      "demonstrate time": 28888,
      "comprehension end": 20178,
      "provo corpus": 94149,
      "reading comprehension end": 95982,
      "tutorials": 126001,
      "nlp specialists": 79691,
      "study previous": 114479,
      "generalizable natural": 46754,
      "techniques end": 119880,
      "approach scoring": 7868,
      "generalizable natural language": 46755,
      "generalizable natural language processing": 46756,
      "datadependent": 26653,
      "generation advanced": 47298,
      "stateoftheart models new": 112745,
      "psychologists": 94216,
      "systems construction": 116811,
      "representation semantics": 99403,
      "representation develop": 99209,
      "meanings text": 67749,
      "interpretable way": 56256,
      "partial orders": 85295,
      "limiting number": 64309,
      "test settings": 120537,
      "represent sentence meaning": 99139,
      "presents solution": 89899,
      "adapts specific": 3346,
      "bring closer": 14628,
      "points rouge": 87875,
      "profit": 92105,
      "model multilabel": 71536,
      "jointly particular": 57376,
      "performed text": 86936,
      "label refinement": 58408,
      "labels high": 58609,
      "learned jointly particular": 62222,
      "model multilabel text": 71537,
      "text classification multilabel": 120744,
      "classification multilabel text": 17288,
      "task training data": 118801,
      "model multilabel text classification": 71538,
      "multilabel text classification multilabel": 75162,
      "text classification multilabel text": 120745,
      "classification multilabel text classification": 17289,
      "multilabel text classification multilabel text": 75163,
      "text classification multilabel text classification": 120746,
      "classification multilabel text classification mltc": 17290,
      "work article": 134380,
      "validity hypothesis": 130755,
      "use unsupervised methods": 128351,
      "played important role": 87712,
      "explicit syntax": 41354,
      "using linearized": 129815,
      "rnnbased seq2seq": 102949,
      "largely determined": 61337,
      "case study machine": 15620,
      "explainers": 41272,
      "datahungry models": 26688,
      "step evaluating": 113263,
      "cases exist": 15642,
      "highlevel text": 50252,
      "structures nlp": 114092,
      "model sensitivity": 71972,
      "locally interpretable": 64951,
      "evaluate different models": 38822,
      "diet": 30925,
      "classification subtasks": 17423,
      "optimize classification": 82224,
      "requires extensive": 100270,
      "report experiments using": 99001,
      "study effectiveness different": 114365,
      "used classifier": 128423,
      "task small": 118721,
      "knowledgegraph": 58272,
      "semantics recent": 105454,
      "rulebased information": 103363,
      "nlp growing": 79619,
      "tasks scientific": 119478,
      "contributions research": 22793,
      "methods datasets evaluation": 69420,
      "methods datasets evaluation metrics": 69421,
      "publish new": 94344,
      "correlation language": 24238,
      "distributional compositional": 32695,
      "way create": 132069,
      "open community": 81886,
      "disambiguation homographs": 31966,
      "provide enhanced": 93814,
      "ungrounded": 127083,
      "select knowledge": 104704,
      "model discrete": 71005,
      "important capability": 52108,
      "classification training": 17472,
      "utility sentence": 130490,
      "tasks currently": 119025,
      "jointly learning sentence": 57357,
      "learning semantically": 62997,
      "explain various": 41262,
      "level constituent": 63430,
      "resources aim": 100940,
      "sentences various topics": 106542,
      "introducing new dataset": 56646,
      "paper propose syntactic": 84329,
      "syntax unsupervised": 116568,
      "sentence possible": 105994,
      "specifically model achieves": 111571,
      "improve pretrained models": 52500,
      "combination powerful": 18579,
      "relevant new": 98565,
      "classifiers trained using": 17641,
      "type classification task": 126191,
      "method knowledge": 68926
    }
  }
}