sentence1
stringlengths 16
446
| sentence2
stringlengths 14
436
|
|---|---|
based metrics are robust to a variety of training conditions , such as the data volume and domain . | furthermore , the proposed metrics are robust to various training conditions , such as the data size and domain . |
for the document embedding , we use a doc2vec implementation that downsamples higher-frequency words for the composition . | with english gigaword corpus , we use the skip-gram model as implemented in word2vec 3 to induce embeddings . |
for english , there is no significant dependency treebank so we followed most previous work in using dependency trees automatically derived from constituent trees in the large penn treebank wsj corpus . | there exists no large-scale dependency treebank for english , and we thus had to construct a dependency-annotated corpus automatically from the penn treebank . |
we trained word embeddings for this dataset using word2vec on over around 10m documents of clinical records . | we trained the embedding vectors with the word2vec tool on the large unlabeled corpus of clinical texts provided by the task organizers . |
while wen et al . ’ s dataset is more than twice larger than ours , it is less diverse both in terms of input and in terms of text . | we show that while wen et al. ’ s dataset is more than twice larger than ours , it is less diverse both in terms of input and in terms of text . |
such that math-w-3-1-2-54 , define math-w-3-1-2-59 . | recall that a derivation takes the form math-w-8-1-1-7 . |
in this paper , we have provided a new perspective to predict the cqa answer quality . | in this paper , we address the problem for predicting cqa answer quality as a classification task . |
in this paper , we develop declarative rules which govern the translation of natural language description of these concepts . | in this paper , we introduce a framework for incorporating declarative knowledge in word problem solving . |
jansen et al describe answer reranking experiments on ya using a diverse range of lexical , syntactic and discourse features . | jansen et al report that answer reranking benefits from lexical semantic models , and describe experiments using skipavg embeddings pretrained using the english gigaword corpus . |
we then evaluate the effect of word alignment on machine translation quality using the phrase-based translation system moses . | we compare the model against the moses phrase-based translation system , applied to phoneme sequences . |
in section 6 we describe our machine learning approach and show results on pos tagging . | in section 6 we describe our machine learning approach and show results on pos tagging code-switched text . |
we used kappa statistics to evaluate the annotations made by the annotators in the second phase . | we evaluated annotation reliability by using the kappa statistic . |
therefore , we employ negative sampling and adam to optimize the overall objective function . | note that we employ negative sampling to transform the objective . |
statistical significance of difference from the baseline bleu score was measured by using paired bootstrap re-sampling . | we performed paired bootstrap sampling to test the significance in bleu score differences . |
we used the implementation of random forest in scikitlearn as the classifier . | for the feature-based system we used logistic regression classifier from the scikit-learn library . |
we pre-train the 200-dimensional word embeddings on each dataset in with skipgram . | all word vectors are trained on the skipgram architecture . |
we trained svm models with rbf kernel using scikit-learn . | we trained an svm with rbf kernel using scikit-learn . |
in the supervised phase , sentiment polarity labels of documents are used to guide bswe learning . | through the supervised learning phase , math-w-8-10-0-6 is optimized by maximizing sentiment polarity probability . |
weller et al use noun class information as tree labels in syntactic smt to model selectional preferences of prepositions . | weller et al propose using noun class information to model selectional preferences of prepositions in a string-to-tree translation system . |
birke and sarkar proposed the trope finder system to recognize verbs with non-literal meaning using word sense disambiguation and clustering . | birke and sarkar present a sentence clustering approach for non-literal language recognition implemented in the trofi system . |
the induction of selectional preferences from corpus data was pioneered by resnik . | one of the first approaches to the automatic induction of selectional preferences from corpora was the one by resnik . |
for language modeling , we computed 5-gram models using irstlm 7 and queried the model with kenlm . | after standard preprocessing of the data , we train a 3-gram language model using kenlm . |
we use the distance based logistic triplet loss which gave better results than a contrastive loss . | we use the distance based logistic triplet loss , which vo and hays report exhibits better performance in image similarity tasks . |
the penn discourse treebank is the largest available corpus of annotations for discourse relations , covering one million words of the wall street journal . | the penn discourse treebank is the largest manually annotated corpus of discourse relations on top of one million word tokens from the wall street journal . |
named entity recognition ( ner ) is the task of identifying and typing phrases that contain the names of persons , organizations , locations , and so on . | named entity recognition ( ner ) is the task of identifying and classifying phrases that denote certain types of named entities ( nes ) , such as persons , organizations and locations in news articles , and genes , proteins and chemicals in biomedical literature . |
we use word2vec as the vector representation of the words in tweets . | in this run , we use a sentence vector derived from word embeddings obtained from word2vec . |
we used the moses decoder , with default settings , to obtain the translations . | we obtained a phrase table out of this data using the moses toolkit . |
one is the bilexical dependency model and the other is the generative model . | one is a bilexical model , which is a kind of discriminative model , and the other is a generative model . |
hierarchical machine translation extends the phrase-based model by allowing the use of non-contiguous phrase pairs . | a hierarchical phrase-based translation model reorganizes phrases into hierarchical ones by reducing sub-phrases to variables . |
recent work has focused on a much larger set of fine grained labels . | recent work has focused on a much larger set of fine-grained types . |
coreference resolution is the task of grouping mentions to entities . | coreference resolution is the process of determining whether two expressions in natural language refer to the same entity in the world . |
the encoder and decoder are two-layer lstms with a 500-dimension hidden size and 500-dimension word embeddings . | the decoder and encoder word embeddings are of size 500 , the encoder uses a bidirectional lstm layer with 1k units to encode the source side . |
for dependency grammar induction , smith and eisner favored short attachments using a fixed-weight feature whose weight was optionally annealed during learning . | smith and eisner propose structural annealing , in which a strong bias for local dependency attachments is enforced early in learning , and then gradually relaxed . |
we represent each citation as a feature set in a support vector machine framework and use n-grams of length 1 to 3 as well as dependency triplets as features . | we represent each citation as a feature set in a support vector machine framework which has been shown to produce good results for sentiment classification . |
phrase-based models excel at capturing local reordering phenomena and memorizing multi-word translations . | the phrase-based translation systems rely on language model and lexicalized reordering model to capture lexical dependencies that span phrase boundaries . |
disfluency detection is the task of recognizing non-fluent word sequences in spoken language transcripts ( cite-p-25-3-15 , cite-p-25-3-10 , cite-p-25-3-12 ) . | disfluency detection is the task of detecting these infelicities in spoken language transcripts . |
gram language models are trained over the target-side of the training data , using srilm with modified kneser-ney discounting . | the targetside 4-gram language model was estimated using the srilm toolkit and modified kneser-ney discounting with interpolation . |
a typical approach for sentiment classification is to use supervised machine learning algorithms with bag-of-words as features . | a typical approach for sentiment classification is to use supervised machine learning algorithms with bag-of-words as features , which is widely used in topic-based text classification . |
faruqui et al introduce a graph-based retrofitting method where they post-process learned vectors with respect to semantic relationships extracted from additional lexical resources . | faruqui et al demonstrated that embeddings learned without supervision can be retro-fitted to better conform to some semantic lexicon . |
finkel and manning modeled the task of named entity recognition together with parsing . | finkel and manning show how to model parsing and named entity recognition together . |
we conceptualized the induction problem as one of detecting alternate linkings and finding their canonical syntactic form . | we formulate the role induction problem as one of detecting alternations and finding a canonical syntactic form for them . |
a simile is a figure of speech comparing two essentially unlike things . | a simile is a figure of speech comparing two fundamentally different things . |
case-insensitive bleu4 was used as the evaluation metric . | case-insensitive nist bleu was used to measure translation performance . |
we present a novel interactive visualisation that we have developed for displaying collaborations . | we also present a novel visualisation interface for browsing collaborations . |
we preprocessed the corpus with tokenization and true-casing tools from the moses toolkit . | we used the moses toolkit for performing statistical machine translation . |
we have presented a technique for creating a ? estimates for inference . | we present a novel method for creating a ? estimates for structured search problems . |
we considered one layer and used the adam optimizer for parameter optimization . | for optimization , we used adam with default parameters . |
luong and manning , 2016 ) proposes a hybrid architecture for nmt that translates mostly at the word level and consults the character components for rare words when necessary . | luong and manning proposed a hybrid scheme that consults character-level information whenever the model encounters an oov word . |
for our parsing experiments , we use the berkeley parser . | we adopt berkeley parser 1 to train our sub-models . |
in the parse tree , strong evidence about either aspect of the model should positively impact the other aspect . | because a named entity should correspond to a node in the parse tree , strong evidence about either aspect of the model should positively impact the other aspect . |
kendall ¡¯ s math-w-2-5-2-97 as a performance measure for evaluating the output of information-ordering components . | kendall¡¯s math-w-11-5-2-1 can be easily used to evaluate the output of automatic systems , irrespectively of the domain or application at hand . |
we use the skll and scikit-learn toolkits . | for all classifiers , we used the scikit-learn implementation . |
grammatical information for the sentential context is obtained using the dependency relation output of the stanford parser . | sentences are passed through the stanford dependency parser to identify the dependency relations . |
due to the imbalanced characteristic of the training data , we specifically adopted a two-step classifier to deal with subtask a . | since the training data is imbalanced , we specifically designed a two-step classifier to address subtask a . |
and the results demonstrate the good effectiveness of the proposed model . | moreover , the results show the robustness of the proposed model . |
while we extend the seq2seq framework to conduct template reranking and template-aware summary generation . | then , we extend the seq2seq framework to jointly conduct template reranking and template-aware summary generation ( rewriting ) . |
in our experiments , we used the srilm toolkit to build 5-gram language model using the ldc arabic gigaword corpus . | for both languages , we used the srilm toolkit to train a 5-gram language model using all monolingual data provided . |
the srilm toolkit was used to build this language model . | srilm toolkit is used to build these language models . |
active learning is a framework that makes it possible to efficiently train statistical models by selecting informative examples from a pool of unlabeled data . | active learning is a promising way for sentiment classification to reduce the annotation cost . |
a total of 42 systems were submitted from 21 distinct teams , and nine . | a total of 42 systems were submitted to the task . |
in this demo , we introduce need4tweet , a twitterbot for a combined system for nee and ned in tweets . | in this demo paper , we present need4tweet , a twitterbot for nee and ned in tweets . |
for this task , we use a deep learning method to obtain final predict answer . | to address this machine comprehension task , we utilized rule-based methods and a deep learning method . |
semantic role labeling ( srl ) consists of finding the arguments of a predicate and labeling them with semantic roles ( cite-p-9-1-5 , cite-p-9-3-0 ) . | semantic role labeling ( srl ) is the task of automatically annotating the predicate-argument structure in a sentence with semantic roles . |
where math-w-8-3-0-1 is a fresh nonterminal symbol , the characteristic string math-w-8-3-0-12 is the string obtained from math-w-8-3-0-22 . | for math-w-2-6-2-13 , we write math-w-2-6-2-21 to denote the interval math-w-2-6-2-30 , and use [ i ] as a shorthand for math-w-2-6-2-51 . |
on all datasets and models , we use 300-dimensional word vectors pre-trained on google news . | we use distributed word vectors trained on the wikipedia corpus using the word2vec algorithm . |
in particular , we use a rnn based on the long short term memory unit , designed to avoid vanishing gradients and to remember some long-distance dependences from the input sequence . | we consider both long short-term memory networks and gated recurrent unit networks , two variants of rnns that use gating to mitigate vanishing gradients . |
cattoni et al apply statistical language models to da classification . | cattoni et al also apply statistical language models to da classification . |
we first used a variant of the lesk algorithm , which is based on word exact match . | we used lesk as the similarity measure in our algorithm which is based on lesk . |
to extract the features of the rule selection model , we parse the english part of our training data using the berkeley parser . | for samt grammar extraction , we parsed the english training data using the berkeley parser with the provided treebank-trained grammar . |
we experiment with word2vec and glove for estimating similarity of words . | we use pre-trained glove vector for initialization of word embeddings . |
we use the logistic regression classifier as implemented in the skll package , which is based on scikitlearn , with f1 optimization . | we use the logistic regression classifier in the skll package , which is based on scikit-learn , optimizing for f 1 score . |
he et al proposes maximum entropy models which combine rich context information for selecting translation rules during decoding . | besides , he et al built a maximum entropy model which combines rich context information for selecting translation rules during decoding . |
according to the metrics of semeval 2018 , our system gets the final scores of 0 . 636 , 0 . 531 , 0 . 731 , 0 . 708 , and 0 . 408 in terms of pearson correlation . | according to semeval 2018¡¯s metrics , our model runs got final scores of 0.636 , 0.531 , 0.731 , 0.708 , and 0.408 in terms of pearson correlation on 5 subtasks , respectively . |
novel metaphors are marked by their unusualness in a given context . | metaphorical instances tend to have personal topics . |
in this paper , we have proposed a deep belief network based approach to model the semantic relevance for the question answering pairs . | to solve the first problem , we present a deep belief network ( dbn ) to model the semantic relevance between questions and their answers . |
named entity recognition ( ner ) is a fundamental task in text mining and natural language understanding . | named entity recognition ( ner ) is a challenging learning problem . |
we used 300 dimensional skip-gram word embeddings pre-trained on pubmed . | to encode the original sentences we used word2vec embeddings pre-trained on google news . |
nuclearity in rhetorical structure theory is explained in terms of relative importance of text spans . | rhetorical structure theory posits a hierarchical structure of discourse relations between spans of text . |
semantic role labeling ( srl ) is a kind of shallow sentence-level semantic analysis and is becoming a hot task in natural language processing . | semantic role labeling ( srl ) is defined as the task to recognize arguments for a given predicate and assign semantic role labels to them . |
we use case-insensitive bleu as evaluation metric . | for evaluation metric , we used bleu at the character level . |
due to the underspecified representation we are using . | this is largely due to the underspecified representation we are using . |
we train an english language model on the whole training set using the srilm toolkit and train mt models mainly on a 10k sentence pair subset of the acl training set . | for both languages , we used the srilm toolkit to train a 5-gram language model using all monolingual data provided . |
we formalize the problem of reference page selection . | we propose an automatic method that can select reference pages . |
for sampling nodes , non-interactive active learning algorithms exclude expert annotators ’ human labels from the protocol . | note that , unlike active learning used in the nlp community , non-interactive active learning algorithms exclude expert annotators ’ human labels from the protocol . |
therefore , we used bleu and rouge as automatic evaluation measures . | additionally , we used bleu , a very popular machine translation evaluation metric , as a feature . |
part-of-speech ( pos ) tagging is a critical task for natural language processing ( nlp ) applications , providing lexical syntactic information . | part-of-speech ( pos ) tagging is a fundamental nlp task , used by a wide variety of applications . |
shen et al describe the result of filtering rules by insisting that target-side rules are well-formed dependency trees . | shen et al proposed a target dependency language model for smt to employ target-side structured information . |
abstract meaning representations are a graph-based representation of the semantics of sentences . | abstract meaning representation is a compact , readable , whole-sentence semantic annotation . |
cite-p-17-3-16 tackled this issue by allowing the number to be dynamically adjusted for each word . | cite-p-17-5-4 modified the skip-gram model in order to learn multiple embeddings for each word type . |
in tmhmm , tmhmms and tmhmmss , the number of ¡° topics ¡± in the latent states . | in tmhmm , tmhmms and tmhmmss , the number of ¡°topics¡± in the latent states and a dialogue is a hyperparameter . |
the parameter weights are optimized with minimum error rate training . | the model parameters are trained using minimum error-rate training . |
if the anaphor is a pronoun , the cache is searched for a plausible referent . | the anaphor is a definite noun phrase and the referent is in focus , that is . |
we use the english penn treebank to evaluate our model implementations and yamada and matsumoto head rules are used to extract dependency trees . | we extract dependency structures from the penn treebank using the head rules of yamada and matsumoto . |
predicate models such as framenet are core resources in most advanced nlp tasks , such as question answering , textual entailment or information extraction . | predicate models such as framenet , verbnet or propbank are core resources in most advanced nlp tasks , such as question answering , textual entailment or information extraction . |
wordnet is a large lexical database of english , where open class words are grouped into concepts represented by synonyms that are linked to each other by semantic relations such as hyponymy and meronymy . | wordnet is a large semantic lexicon database of english words , where nouns , verbs , adjectives and adverbs are grouped into sets of cognitive synonyms . |
the bleu score measures the precision of n-grams with respect to a reference translation with a penalty for too short sentences . | this score measures the precision of unigrams , bigrams , trigrams and fourgrams with respect to a reference translation with a penalty for too short sentences . |
we use randomization test to calculate statistical significance . | for assessing significance , we apply the approximate randomization test . |
we used the moses decoder , with default settings , to obtain the translations . | for all submissions , we used the phrase-based variant of the moses decoder . |
specifically , we generalise the model of cohn and lapata to our abstractive task . | our work builds on the model developed by cohn and lapata . |
we consider a phrase-based translation model and a hierarchical translation model . | the disadvantage of word-to-word translation is overcome by phrase-based translation and log-linear model combination . |