File size: 77,689 Bytes
6fa4bc9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 |
{
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T01:13:27.399124Z"
},
"title": "",
"authors": [
{
"first": "Daniel",
"middle": [
"Benedict"
],
"last": "Dacanay",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Alberta",
"location": {
"addrLine": "4-32 Assiniboia Hall",
"postCode": "T6G 2E7",
"settlement": "Edmonton",
"region": "Alberta",
"country": "Canada"
}
},
"email": "dacanay@ualberta.ca"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "One problem in the task of automatic semantic classification is the problem of determining the level on which to group lexical items. This is often accomplished using already existing, hierarchical semantic ontologies. The following investigation explores the computational assignment of semantic classifications on the contents of a dictionary of n\u00eahiyaw\u00eawin / Plains Cree (ISO: crk, Algonquian, Western Canada and United States), using a semantic vector space model, and following two semantic ontologies, WordNet and SIL's Rapid Words, and compares how these computational results compare to manual classifications with the same two ontologies.",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "One problem in the task of automatic semantic classification is the problem of determining the level on which to group lexical items. This is often accomplished using already existing, hierarchical semantic ontologies. The following investigation explores the computational assignment of semantic classifications on the contents of a dictionary of n\u00eahiyaw\u00eawin / Plains Cree (ISO: crk, Algonquian, Western Canada and United States), using a semantic vector space model, and following two semantic ontologies, WordNet and SIL's Rapid Words, and compares how these computational results compare to manual classifications with the same two ontologies.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Despite the benefits and usages of semantically organised lexical resources such as dictionaries, ranging from uses as pedagogical tools (Lemnitzer and Kunze 2003) to aids for machine translation (Klyueva 2007) , fully elaborated semantic dictionaries remain less common than those assembled with more routine alphabetical ordering systems. Aside from the reason of convention, one prominent dissuasive factor towards creating semantic dictionaries is the sheer amount of effort necessary to create them if their lexical content is not already organised along some ontologically principled semantic lines; the manual semantic classification of even relatively small dictionaries of this nature frequently takes months. This may be a prohibitively costly procedure in situations where resources for linguistic analysis, be they temporal or economic, are limited. Thus, a dilemma faced by the prospective compiler of a semantic dictionary is that of selecting an ontology, that is, a principled system of semantic categories, typically (but not universally) arranged hierarchically, into which lexical items may be grouped. The following investigation aims to address potential remedies to both of these limitations, with vector semantics as a first-pass alternative to manual semantic classification, and with Princeton WordNet and SIL's Rapid Words as two practical contenders for pre-existing semantic ontologies. In practice, these methods are to be demonstrated on an existing bilingual dictionary of Plains Cree (n\u00eahiyaw\u00eawin), with results compared against human-made semantic classifications in both ontologies.",
"cite_spans": [
{
"start": 137,
"end": 163,
"text": "(Lemnitzer and Kunze 2003)",
"ref_id": "BIBREF14"
},
{
"start": 196,
"end": 210,
"text": "(Klyueva 2007)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The first, and perhaps most daunting, obstacle in the process of creating a semantic dictionary (or indeed any semantically organised lexical resource) is the issue of time; even with a welldefined ontology and ample resources, manual semantic classification is a lengthy and expensive process, with teams of linguists and native speakers often requiring years to produce fully annotated semantic dictionaries (Bosch and Griesl 2017) . Even with a more reduced ontology, semantically classifying an already existing full dictionary by hand takes months, and requires a thorough understanding of the chosen ontology (Dacanay et al. 2021) . Although the process of manually assigning semantic categories or correspondences to The More Detail, the Better? -Investigating the Effects of Semantic Ontology Specificity on Vector Semantic Classification with a Plains Cree / n\u00eahiyaw\u00eawin Dictionary dictionary entries is generally not an exceptionally difficult task for a human annotator (Basile et al. 2012) , the length of dictionaries, and the existence of highly polysemous lexical items, both complicate and lengthen the process of manual classification. As such, the mechanisation of the process of semantic classification assignment (or semantic annotation) appears to be one of the most direct routes to increasing overall efficiency with respect to time and resources, and to that end, the method of vector semantic classification is an alluring and well-attested alternative (Turney and Pantel 2010) .",
"cite_spans": [
{
"start": 410,
"end": 433,
"text": "(Bosch and Griesl 2017)",
"ref_id": null
},
{
"start": 615,
"end": 636,
"text": "(Dacanay et al. 2021)",
"ref_id": "BIBREF6"
},
{
"start": 981,
"end": 1001,
"text": "(Basile et al. 2012)",
"ref_id": "BIBREF2"
},
{
"start": 1478,
"end": 1502,
"text": "(Turney and Pantel 2010)",
"ref_id": "BIBREF23"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Vector Semantics",
"sec_num": "2"
},
{
"text": "In short, vector semantic classification is a method of computationally determining the semantic similarity between any two given lexical units based on commonalities in the usage contexts of those units in large corpora. This is accomplished by representing the meaning of a lexical unit (primarily a word) as a vector in multidimensional space, which is based on the co-occurrences of this lexical unit with other lexical units in its context, followed by a reduction of dimensionality using some heuristic to result in a compact, dense vector space (typically with several hundred dimensions).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Vector Semantics",
"sec_num": "2"
},
{
"text": "Since this vector space is based on common contextual features, one may compare the multidimensional vector of one word with that of another, calculating their cosine distance to determine similarity; the closer this value is to 1, the more similar the average contexts of those two words are, and thus the more similar those words are semantically. In this way, the model functions largely on the assumptions of the Distributional Hypothesis as put forth by Firth and Harris in the 1950s (Jurafsky and Martin 2019; Firth 1957; Harris 1954) , that semantic similarity begets distributional similarity, and vice versa. Vector generation is not monolithic, and various tools using various methods exist in common use, including frequency-weighted techniques such as tf-idf and Latent Semantic Analysis. In the context of this investigation, word2vec, a tool which makes use of predictionbased models rather than concurrence matrices to generate clusterable vector sets, has been used to generate all vectors; this decision was motivated chiefly by word2vec being readily available, easily applicable without lengthy training, and being able to leverage extensive, pre-existing pretraining on large English corpora, all advantages which largely offset the primary disadvantage of word2vec, being that it is a purely word-level vector generation tool, lacking the ability to model polysemy and contextual variances, a shortcoming which may possibly be addressed by using a sentence-level model such as BERT (see Section 5 and 7).",
"cite_spans": [
{
"start": 489,
"end": 515,
"text": "(Jurafsky and Martin 2019;",
"ref_id": "BIBREF11"
},
{
"start": 516,
"end": 527,
"text": "Firth 1957;",
"ref_id": "BIBREF8"
},
{
"start": 528,
"end": 540,
"text": "Harris 1954)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Vector Semantics",
"sec_num": "2"
},
{
"text": "The vector method is not a novelty, and its utility as a practical method of semantic classification assignment has been demonstrated on numerous occasions (Brixey et al. 2020; Vecchi et al. 2017) . However, useful as the method may be, in order to use vector semantics to classify entries in a dictionary, one requires a principled structure of semantic relationships into which to classify them. To this end, pre-existing semantic ontologies are a widespread and convenient solution.",
"cite_spans": [
{
"start": 156,
"end": 176,
"text": "(Brixey et al. 2020;",
"ref_id": "BIBREF5"
},
{
"start": 177,
"end": 196,
"text": "Vecchi et al. 2017)",
"ref_id": "BIBREF24"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Vector Semantics",
"sec_num": "2"
},
{
"text": "Although it is possible to computationally generate sets of semantic hierarchies, the results of such attempts generally indicate that humanmade, preset ontologies are preferable (Koper et. al 2015) . Many such premade ontologies exist, serving a wide variety of different classificational purposes; however, we will compare here only two, being a slightly modified version of the Princeton WordNet and SIL's Rapid Word Collection Method, both popular, general-purpose ontologies intended to cover the breadth of most semantic reference in a largely language-neutral fashion. A visual representation of the structures of both is detailed in Figure 1 (see next page).",
"cite_spans": [
{
"start": 179,
"end": 198,
"text": "(Koper et. al 2015)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [
{
"start": 641,
"end": 649,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Semantic Ontologies",
"sec_num": "3"
},
{
"text": "The Princeton WordNet is one of the oldest and most widely-used semantic classification systems, originating in the 1990s at Princeton University as a hierarchically organised structure wherein contextually synonymous word-senses (or individual word-senses) are grouped into 'synsets', each of which has a hypernymic Figure 1 , a visual demonstration of the differences in structure and specificity between WordNet (left) and Rapid Words (right).",
"cite_spans": [],
"ref_spans": [
{
"start": 317,
"end": 325,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Princeton WordNet",
"sec_num": "3.1"
},
{
"text": "synset above it in the hierarchy and possibly one or several hyponymic synsets below it (for example, the words (n) cod#2 and (n) codfish#1 form a synset with the definition \"lean white flesh of important North Atlantic food fish; usually baked or poached\"; this synset is a hyponym of the synset (n) saltwater fish#1, and is hypernymic to the synset (n) salt cod#1.). In this way, WordNet is essentially a hierarchy of hypernyms and hyponyms, with each level of hypernym and hyponym being populated by various contextually synonymous words. Although other semantic relations such as antonymy are also modelled in a 'full' WordNet, the three relations of hypernymy, hyponymy, and synonymy form the \"central organizing principle\" of WordNet as a whole (Miller 1993) , and a structurally complete, albeit semantically basic, WordNet can be constructed using only these three relationships; in Dacanay et al. 2021we referred to this core-level WordNet as a 'skeletal WordNet'.",
"cite_spans": [
{
"start": 751,
"end": 764,
"text": "(Miller 1993)",
"ref_id": "BIBREF19"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Princeton WordNet",
"sec_num": "3.1"
},
{
"text": "An alternative semantic classification scheme is the Rapid Word Collection Method of SIL, created as a framework for collecting native speaker vocabulary elicitations for dictionary creation, rather than the organisation of finished dictionaries (Moe 2003) . Despite this, the structure of Rapid Words is broadly similar to that of WordNet, consisting of various numbered, hierarchically organised, roughly hyper/hyponymic semantic domains, each of which is populated by highly semantically related (although in Rapid Words, not necessarily contextually synonymous) sets of words, which may be spread across various parts of speech. Broadly speaking, these domains are less specific than WordNet synsets. There are five 'tiers' of domains in RW, with the highest being the most general (e.g. 5 Daily Life, 7 Physical Actions, etc) and the lowest being the most specific (e.g. 5.2.3.3.3 Spice, 7.2.1.1.1 Run); for our purposes, only domains on the fourth tier (or level) were used for the vector classifications (see Section 5). These semantic domains are sub-organised into specific elicitation questions, each of which has a set of potential vocabulary items in English; for example, the domain 2.1.1.5 Tooth contains the elicitation question 'What are the parts of a tooth?', which would have with it the list of potential English answers as prompts 'enamel, root, crown, ivory'. Although not explicitly designed for it, Rapid Words has been used successfully for after-the-fact dictionary classification in the past (Reule 2018) .",
"cite_spans": [
{
"start": 246,
"end": 256,
"text": "(Moe 2003)",
"ref_id": "BIBREF20"
},
{
"start": 1519,
"end": 1531,
"text": "(Reule 2018)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Rapid Words",
"sec_num": "3.2"
},
{
"text": "Plains Cree (n\u00eahiyaw\u00eawin) is an Indigenous language of the Algonquian family, spoken b\u1ef9 30 000 throughout Saskatchewan, Alberta, and Northern Montana. Although slightly less critically endangered in comparison with other Canadian Indigenous languages, the majority of speakers are elderly, and intergenerational transmission remains low. Various revitalisation efforts have been undertaken in Cree communities, including bilingual education and the creation of online lexical resources (Arppe et al. 2018) ; however, digital resources for Cree remain limited overall. Like most Algonquian languages, Plains Cree is highly polysynthetic, with extensive morphology, particularly on verbs, which make up the bulk of the lexicon (e.g., Wolfart 1973 ).",
"cite_spans": [
{
"start": 486,
"end": 505,
"text": "(Arppe et al. 2018)",
"ref_id": "BIBREF0"
},
{
"start": 732,
"end": 744,
"text": "Wolfart 1973",
"ref_id": "BIBREF25"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Plains Cree / n\u00eahiyaw\u00eawin",
"sec_num": "4"
},
{
"text": "The lexical resource used for this investigation was a fully digitised copy of the database underlying n\u00eahiyaw\u00eawin: itw\u00eawina/Cree: Words (CW), a continually-updated bilingual Cree-English dictionary compiled by Arok Wolvengrey across the late 20th and early 21st centuries (Wolvengrey 2001) . Consisting currently of 21,347 words with morphological notes and PoS-tagging, CW is the most extensive modern dictionary of Plains Cree, and its contents may be accessed through the University of Alberta's online Cree dictionary, itw\u00eawina.",
"cite_spans": [
{
"start": 273,
"end": 290,
"text": "(Wolvengrey 2001)",
"ref_id": "BIBREF26"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Plains Cree / n\u00eahiyaw\u00eawin",
"sec_num": "4"
},
{
"text": "Word vectors were obtained for every Cree entry in CW using word2vec, a popular off-the-shelf vector generation tool . We used the pretrained Google News Corpus, which contains 3 million word embeddings trained on 3 billion tokens. Cree word (or rather, dictionary entry) vectors were obtained as a simple, dimension-wise average of the individual English word vectors as extracted from the English definition phrases/sentences (glosses) of their respective entries, rather than the Cree words themselves, as existing Cree corpora (Arppe 2020) are too small for meaningful dimensional vectors to be obtained (Harrigan and Arppe 2019) . For example, the vector for the Cree noun mahkahk (glossed in CW as 'tub, barrel; box') would be generated by averaging the vectors for the English words 'tub', 'barrel', and 'box', treated as a bag of words. Similarly, for the Cree verb n\u00e2tw\u00e2nam (glossed as 's/he breaks s.t. apart; s/he breaks s.t. off by hand'), the vector would be derived from the average of the vectors for 's/he', 'breaks', 's.t.', 'apart', 'off', and 'hand'. CW noun glosses tend to be either single words or extremely curt noun phrases, and verb glosses are usually brief, utilitarian verb phrases, with no grammatical or derivational information included in the gloss itself; this fact is a further justification for using a word-level vector generation tool such as word2vec rather than a sentence-level tool like BERT, as the pieces of linguistic information on which the CW vectors are based are typically either non-sentential or highly simplistic and formulaic, seemingly making the contextsensitivity of tools such as BERT much less useful.",
"cite_spans": [
{
"start": 608,
"end": 633,
"text": "(Harrigan and Arppe 2019)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "5"
},
{
"text": "The Google News Corpus and word2vec were similarly used to generate the vectors for the WordNet synsets, using the head words and synset description (definitions and example sentences) as context to create the vectors, and the head word(s) of the synset as labels (Dacanay et al. 2021) . For example, the vector for the synset (n) barrel#2 (glossed as \"barrel, cask (a cylindrical container that holds liquids)\") would be the average of the vectors for the words 'barrel', 'cask', 'cylindrical', 'container', 'holds', and 'liquids'. The vectors for Rapid Words were created using the semantic domain levels as labels, with all example words and elicitation questions contained therein as context. For example, for the word 'barrel' in Rapid Words, which is contained in the semantic domain 6.7.7 Container, the vector would be the average of the vectors for all of the English words in each elicitation question (i.e. \"What words refer to a container\", \"What words refer to what is in a container\", etc.), as well as all of the words listed as possible examples (such as 'container', 'vessel', 'bowl', 'pot', 'contents' etc.).",
"cite_spans": [
{
"start": 264,
"end": 285,
"text": "(Dacanay et al. 2021)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "5"
},
{
"text": "These sets of vectors were then compared against the CW vectors using cosine distance, and for every CW entry, two lists were created. For each entry on the first list (the WordNet list), all WordNet synsets were listed, ordered by cosine similarity to that entry. On the second list (the four-level Rapid Words list), for each CW entry, all Rapid Words semantic domains at the fourth tier of the hierarchy were ordered by similarity. To provide an example for the second list, even if the manually-selected RW domain for the Cree word acihkos ('caribou calf, fawn') was 1.6.1.1.3 Hoofed Animal, because, on this list, the vector method would only have access to the fourth hierarchy level, the ideal, most 'human-like' vector classification would instead be 1.6.1.1 Mammal, as this domain is at the fourth level of the hierarchy and is identical to the manual classification up to the fourth level (1.6.1.1). The reasoning behind limiting the RW domains to the fourth level of the hierarchy in the vector method was threefold; firstly, tests in which the vector method was allowed to select domain classifications from any of the five levels returned notably poorer results than those which limited the choice to only one tier. (see Table 1 Any-Level (AL) columns), secondly, the fourth level of the hierarchy had the largest number of domains (at 983 compared to the fifth level with 311 and the second level with 68), and thirdly, RW did not always provide fifth level domains throughout the hierarchy. Additionally, the fourth level of the hierarchy provided a useful middleground in terms of specificity compared with the other RW levels; fourth level domains are moderately, rather than highly, specific, and thus allow for a more informative comparison with WordNet's highly specific and complex synset structure. Still, investigating whether using the most specific Rapid Words domains as labels would provide more or less accurate results than the moderately specific four-level domains would be a worthwhile avenue of future study, as would be using the individual elicitation questions as labels instead of domains.",
"cite_spans": [],
"ref_spans": [
{
"start": 1234,
"end": 1241,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "Method",
"sec_num": "5"
},
{
"text": "In total, applying the vector semantic method to this end requires access to a fully digitised copy of the target dictionary (with entries and their glosses clearly delineated), access to WordNet, Rapid Words, and word2vec (all of which are freely available online), and a computer capable of both generating vectors for the dictionary entries and comparing those vectors with the pre-existing ontology vectors. To this end, a 2core laptop with 8gb RAM is able to complete the cosine comparisons for the~16k CW entries with the~117k WordNet synsets in 4-5 days, and the same entries with the Rapid Words domains in no more than one and half days. On a highly parallelised computing cluster, such as ComputeCanada's Cedar (using 64 cores, each having 4-8gb RAM), performing all of the cosine comparisons takes less than 90 minutes. The computational cost of the actual vector cosine comparisons is fairly negligible, and the lengthy runtime of this operation on more basic machines is likely due to the inefficiency of retrieving each vector from large matrices.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "5"
},
{
"text": "To assess their quality, these vector classifications were compared against a gold standard of manual classifications for each entry in CW. These manual classifications were done following both WordNet and Rapid Words, with one or several synsets or RW elicitation questions assigned to each CW entry based on the meaning of the Cree word. For the WordNet classifications, the part of speech of the English WN synset was ignored; for example, the manual classification of the Cree verb mihkw\u00e2w (\"it is red\") was given in WordNet as the adjectival synset (adj) red#1. For Rapid Words classifications, given that RW elicitation questions do not have hard-coded parts of speech, whichever domain-internal elicitation question(s) were most semantically related to the target Cree word were used, regardless of their domain level in the hierarchy. For example, for mihkw\u00e2w, the question 8. 3.3.3.4.3 ",
"cite_spans": [
{
"start": 885,
"end": 894,
"text": "3.3.3.4.3",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "5"
},
{
"text": "Statistics: Overall, although the results of both ontologies are comparable, semantic classifications using Rapid Words appear noticeably more human-like than those with WordNet, with 'human-like' here referring to how high the rank of the manual classification(s) is among the total vector classifications for each entry on average. For the vector classifications of Cree nouns, the median position of the top manual classification was 2 for four-level RW domains (with 983 possible classes) and 36.5 when the vector method was allowed to choose from any level of domain (with 1789 possible classes). For Cree verbs, the median position of the top manual classification was 15 for the fourlevel domains and 222 for any-level domains. In cases where there was more than one manual RW classification, the median position of the median of the multiple classes for CW nouns was 3 for four-levels, and for CW verbs, the median of the medians of multiple classes was 36 for four-levels.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison of WordNet and Rapid Words Results",
"sec_num": "6"
},
{
"text": "For the WordNet vector classifications, the median computationally selected position for the top manual classification was 15th for Cree nouns and 333rd for verbs, and the median position of the manual classifications when there were several was 30.5 for the nouns and 1045 for the verbs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison of WordNet and Rapid Words Results",
"sec_num": "6"
},
{
"text": "From this, it is clear that vector classifications with Rapid Words domains are, on average, much more human-like than their WordNet counterparts, being up to 22 times more accurate in the case of Cree verbs, and that limiting the vector methods' potential selections to a single, moderately specific RW hierarchy level provides much more human-like results than allowing it to select from all domains at all levels. However, it is prudent to keep in mind that even with all of its domains, Rapid Words still has far fewer potential correspondences than WordNet (1789 total RW domains (with 983 four-levels) compared to 117,659 WN synsets), and in relative terms, relevant manual classifications occur on average in a higher position proportionate to the total number of possible choices in WN vector classifications than in those with RW; with four-level RW vector classifications, the median position of the top manual classification is in the top 0.203% for the nouns (2nd out of 983) and in the top 1.53% for the verbs (15th out of 983), compared with the top 0.0127% (15th out of 117659) and 0.283% (333rd out of 117659) respectively for WN.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison of WordNet and Rapid Words Results",
"sec_num": "6"
},
{
"text": "In general, the reduced specificity of Rapid Words, by virtue of both its inherently less detailed structure and its restriction here to a single hierarchical level of specificity, seemed to lend itself well to resolving a particular ill in the vector method, being its propensity to preferentially assign overly specific classifications to the high ranks of 'umbrellaterms', rather than the more appropriate general vocabulary. In this sense, Rapid Words semantic domains often represent concepts several steps higher in the hypernymic hierarchy than their WordNet equivalents. For example, with the WordNet classifications, the top classification for m\u00f4hkom\u00e2n (glossed as 'knife') was (n) knife blade#1, and the top 15 classifications consisted almost entirely of either specific types of knives or parts of knives, with the more appropriate generic term (n) knife#1 not appearing until 18th place. By contrast, in Rapid Words, in which such specific classifications are by nature impossible at the domain-level, the top ranking classifications are more appropriately general, with the any-level list, for example, having the appropriate 6.7.1 Cutting Tool as the top classification, and the similarly relevant 4.8.3.7 Weapon, Shoot in second place.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison of WordNet and Rapid Words Results",
"sec_num": "6"
},
{
"text": "The 'regift' problem: The in-built simplicity of Rapid Words also seems to have partially remedied, if not entirely solved, the so-called 'regift problem' which was prevalent in WordNet classifications; we discuss this problem in more detail in Dacanay et al. (2021) , but simply put, a small number of extremely low frequency WordNet synsets occurred disproportionately frequently in the high-ranking classifications of target Cree words. The problem was so named due to such one low-frequency synset, (v) regift#1, being present in the top 1000 computational classifications of 65% of all Cree verbs, despite almost always being entirely unrelated semantically to the target Cree word. (v) regift#1 is not the only WordNet entry to exhibit this behaviour, and other words, such as (n) Rumpelstiltskin#1 occurred in as many as 72% of the top 1000 vector classifications of Cree verbs; other common regift words include (n) Dido#1, (n) gumption#1, and (n) dingbat#1. As a rule, these 'regift' words were both low frequency in corpora and highly specific, often being proper nouns, however, there did not appear to be any pattern in the formatting or content of these entries' glosses. The Rapid Words vector classifications also exhibited this problem to an extent; for example, subdomains of the domain 4.1.9 Kinship occurred in the top 1000 vector classifications of CW nouns and verbs an average of~12 times, and appeared in the top 10 classifications 33.9% and 35.7% of the time for CW nouns and verbs respectively. However, as a whole, the regift problem was markedly less notable with RW classifications of both types than with WN classifications, with both fewer different regift words (or domains) and fewer occurrences of these words/domains overall. This broadly supported our initial theory that the 'regift' problem was at least partially caused by the excessive degree of specificity in WordNet synsets overwhelming the vector method and providing it with a large number of potential classification choices with poorly defined vectors (due the low frequency of 'regift' words in the Google News Corpus) which muddy the optimal, human-like choices.",
"cite_spans": [
{
"start": 245,
"end": 266,
"text": "Dacanay et al. (2021)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison of WordNet and Rapid Words Results",
"sec_num": "6"
},
{
"text": "By contrast, since Rapid Words generally lacks highly specific vocabulary and is instead structured by more generic categories or 'domains', fewer of these low-frequency words are factored into the Rapid Words vectors, and these vectors are thus, in general, based on higher frequency, more contextually attested vocabulary, and are therefore (in theory) more accurate. In general, the lack of highly specific vocabulary in Rapid Words seems to contribute both to diminishing the number of semanticallyrelated, but overly specific correspondences in the computational classifications, as well as to reducing the prominence of semanticallyunrelated, overly specific 'regift' words (or in the case of Rapid Words, domains). One potential method to imitate this degree of simplicity in WordNet could involve using the hypernymic synsets of the current WordNet correspondences as labels, in essence, shifting all classifications one or more levels up in the WordNet hierarchy. This would appear to at least partially resolve the over-specificity issue (although it would do nothing to reduce the number of outright irrelevant classifications), despite incurring an obvious cost in terms of semantic richness.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison of WordNet and Rapid Words Results",
"sec_num": "6"
},
{
"text": "Vector Content: Broadly speaking, the improved results with Rapid Words seem to be due not only to its simpler hierarchical structure and reduced level of specificity, but also due to its domain internal structure, in which domains generally include fewer irrelevant content words than WordNet synsets do. WordNet synsets frequently include example sentences in their glosses; although useful for human clarification, these inclusions inevitably lead to large amounts of semantically unrelated vocabulary influencing the respective synset vectors. As an example, the gloss for the synset (v) drive#2 (defined as \"travel or be transported in a vehicle\") includes the example sentences \"We drove to the university every morning\" and \"They motored to London for the theater\". As such, the semantically irrelevant words \"university\", \"morning\", \"London\", and \"theater\" are all factored equally into the vector for (v) drive#2 as the semantically relevant terms \"drive\", \"motor\", \"vehicle\". While the inclusion of these less relevant words may more accurately simulate natural linguistic contexts, the otherwise terse nature of WordNet synset glosses means that they introduce a potentially significant amount of distracting information, possibly skewing synset vectors towards the contexts of their irrelevant example sentence vocabulary rather than their relevant gloss vocabulary. By contrast, with the exception of infrequent descriptions of lexicalisation patterns, Rapid Words domains and questions contain only semantically related vocabulary, lessening potential 'distractions' for their vectors.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison of WordNet and Rapid Words Results",
"sec_num": "6"
},
{
"text": "Given the state of current results, it remains unfeasible to fully replace manual semantic annotators using the vector method; even with the best possible RW results, the vector method still only selects the most human-like classification as the top classification less than 50% of the time for Cree nouns, and less than 30% of the time for Cree verbs. Rather, the vector method in its present state seems most immediately usable as an accessory to manual classification, with the method being applied on dictionary resources as a preparatory step for manual annotators, who would then select the best classification for each entry based on the pre-generated vector classification lists. Using only the top 15 vector selected four-level RW classifications, the most human-like classification would be present on this list 50% of the time for Cree verbs, and over 70% of the time for nouns, preventing the annotator from needing to search through the entire ontology every time they wished to classify a word. In this way, present vector results are best suited as a time-saving addition to manual semantic annotation, rather than as a replacement for it.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Utility of Results",
"sec_num": "6.1"
},
{
"text": "The vector semantic method is a significantly faster and cheaper alternative to manual semantic annotation for tasks of semantic classification. However, the method is not yet capable of producing reliably human-like results across target-language parts of speech, and struggles to match natural levels of semantic specificity. To this end, using a consistent hierarchical level of a simpler, more generalistic semantic ontology, such as Rapid Words, seems to make vector semantic classifications appear more human-like, as restricting the breadth of choices available to the method as labels for correspondences seems to both reduce the number of potentially unrelated classifications and make the remaining classifications general enough that a less precise vector is necessary to generate a human-like correspondence.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "7"
},
{
"text": "Future avenues of research into dictionary vector semantics include the use of sentence-based vector generation tools such as BERT which can more accurately model polysemy, although it should be kept in mind that even a model like BERT cannot be expected to generate humanlike results for dictionary glosses if those glosses are non-sentential or otherwise overly brief. It may also prove productive to experiment with the further modification of existing semantic ontologies such as WordNet and Rapid Words (such as reducing the specificity of WN by using only synsets one or several levels higher in the hypernym hierarchy as correspondences), with one of the ultimate goals of this being the integration of the results of automatic vector classifications into online dictionaries in a form which is easily navigable and understandable to an untrained user.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "7"
}
],
"back_matter": [
{
"text": "We would like to thank the Roger S. Smith Undergraduate Researcher Award for the generous funding they provided for the manual classification process and the early stages of this study.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "A morphologically intelligent online dictionary for Plains Cree, Presentation conducted at the meeting of Stabilizing Indigenous Languages Symposium (SILS)",
"authors": [
{
"first": "Antti",
"middle": [],
"last": "Arppe",
"suffix": ""
},
{
"first": "Atticus",
"middle": [],
"last": "Harrigan",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Arppe, Antti, Atticus Harrigan, Katherine Schmirler & Arok Wolvengrey. 2018. A morphologically intelligent online dictionary for Plains Cree, Presentation conducted at the meeting of Stabilizing Indigenous Languages Symposium (SILS), University of Lethbridge, Lethbridge, Alberta.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Harrigan & Arok Wolvengrey. 2020. A Morphosyntactically Tagged Corpus for Plains Cree**",
"authors": [
{
"first": "Antti",
"middle": [],
"last": "Arppe",
"suffix": ""
},
{
"first": "Katherine",
"middle": [],
"last": "Schmirler",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Atticus",
"suffix": ""
}
],
"year": null,
"venue": "Papers of the 49th Algonquian Conference (PAC49)",
"volume": "49",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Arppe, Antti, Katherine Schmirler, Atticus G. Harrigan & Arok Wolvengrey. 2020. A Morphosyntactically Tagged Corpus for Plains Cree**. In M. Macaulay & M. Noodin (eds), Papers of the 49th Algonquian Conference (PAC49), 49: 1-",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Developing a large semantically annotated corpus",
"authors": [
{
"first": "",
"middle": [],
"last": "Basile",
"suffix": ""
},
{
"first": "Johan",
"middle": [],
"last": "Valerio",
"suffix": ""
},
{
"first": "Kilian",
"middle": [],
"last": "Bos",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Evang & Noortje",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Venhuizen",
"suffix": ""
}
],
"year": null,
"venue": "Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Basile, Valerio, Johan Bos, Kilian Evang & Noortje J. Venhuizen. 2012. \"Developing a large semantically annotated corpus.\" Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12), May 2012, 3196-200, doi:http://www.lrec- conf.org/proceedings/lrec2012/pdf/534_Paper.pdf.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Rapid Word Collection, dictionary production, and community well-being",
"authors": [
{
"first": "Brenda",
"middle": [
"H"
],
"last": "Boerger",
"suffix": ""
}
],
"year": 2017,
"venue": "5th International Conference on Language Documentation & Conservation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Boerger, Brenda H. 2017. \"Rapid Word Collection, dictionary production, and community well-being.\" 5th International Conference on Language Documentation & Conservation, Mar. 2017, doi:https://scholarspace.manoa.hawaii.edu/bitstream/1 0125/41988/41988-b.pdf.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Strategies for building wordnets for underresourced languages: The case of African languages",
"authors": [
{
"first": "Sonja",
"middle": [
"E"
],
"last": "Bosch",
"suffix": ""
},
{
"first": "Marissa",
"middle": [],
"last": "Griesel",
"suffix": ""
}
],
"year": 2017,
"venue": "Literator -Journal of Literary Criticism, Comparative Linguistics and Literary Studies",
"volume": "38",
"issue": "1",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bosch, Sonja E & Marissa Griesel. 2017. \"Strategies for building wordnets for underresourced languages: The case of African languages.\" Literator -Journal of Literary Criticism, Comparative Linguistics and Literary Studies, vol. 38, no. 1, 31, 8, doi:https://literator.org.za/index.php/literator/arti cle/view/1351/2294. Accessed 12 Sept. 2020.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Exploring a Choctaw Language Corpus with Word Vectors and Minimum Distance Length",
"authors": [
{
"first": "Jacqueline",
"middle": [],
"last": "Brixey",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Sides",
"suffix": ""
},
{
"first": "Timothy",
"middle": [],
"last": "Vizthum",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020)",
"volume": "",
"issue": "",
"pages": "2746--53",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Brixey, Jacqueline, David Sides, Timothy Vizthum, David Traum & Khalil Iskarous. 2020. \"Exploring a Choctaw Language Corpus with Word Vectors and Minimum Distance Length.\" Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), May 2020, 2746-53.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Computational Analysis versus Human Intuition: A Critical Comparison of Vector Semantics with Manual Semantic Classification in the Context of Plains Cree",
"authors": [
{
"first": "Daniel",
"middle": [],
"last": "Dacanay",
"suffix": ""
},
{
"first": "Atticus",
"middle": [],
"last": "Harrigan & Antti Arppe",
"suffix": ""
}
],
"year": 2021,
"venue": "Proceedings of the 4th Workshop on the Use of Computational Methods in the Study of Endangered Languages",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dacanay, Daniel, Atticus Harrigan & Antti Arppe. 2021. \"Computational Analysis versus Human Intuition: A Critical Comparison of Vector Semantics with Manual Semantic Classification in the Context of Plains Cree.\" Proceedings of the 4th Workshop on the Use of Computational Methods in the Study of Endangered Languages, doi:https://computel- workshop.org/wpcontent/uploads/2021/02/2021.comp utel-1.5.pdf",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "WordNet: An Electronic Lexical Database",
"authors": [
{
"first": "Christiane",
"middle": [],
"last": "Fellbaum",
"suffix": ""
}
],
"year": 1998,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Fellbaum, Christiane. 1998, ed. WordNet: An Electronic Lexical Database. Cambridge, MA: MIT Press.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "A Synopsis of Linguistic Theory",
"authors": [
{
"first": "J",
"middle": [
"R"
],
"last": "Firth",
"suffix": ""
}
],
"year": 1930,
"venue": "Selected Papers of J. R. Firth",
"volume": "",
"issue": "",
"pages": "168--205",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Firth, J. R. A Synopsis of Linguistic Theory, 1930- 1955. 1957. In: Firth, J. R. 1968. Selected Papers of J. R. Firth 1952-1959. London: Logmans, 168-205.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Automatic Semantic Classification of Plains Cree Verbs",
"authors": [
{
"first": "Atticus & Antti",
"middle": [],
"last": "Harrigan",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Arppe",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "24--27",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Harrigan, Atticus & Antti Arppe. 2019. Automatic Semantic Classification of Plains Cree Verbs. Paper presented at the 51st Algonquian Conference in Montreal, Canada, 24-27 October.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Distributional Structure",
"authors": [
{
"first": "Zellig",
"middle": [
"S"
],
"last": "Harris",
"suffix": ""
}
],
"year": 1954,
"venue": "Word",
"volume": "10",
"issue": "2-3",
"pages": "146--62",
"other_ids": {
"DOI": [
"10.1080/00437956.1954.11659520"
]
},
"num": null,
"urls": [],
"raw_text": "Harris, Zellig S. 1954. \"Distributional Structure.\" Word, vol. 10, no. 2-3, 146-62, doi:https://www.tandfonline.com/doi/pdf/10.1080/004 37956.1954.11659520.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Speech and Language Processing: An Introduction to Natural Language Processing",
"authors": [
{
"first": "Dan",
"middle": [
"&"
],
"last": "Jurafsky",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "James",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Martin",
"suffix": ""
}
],
"year": 2019,
"venue": "Computational Linguistics, and Speech Recognition",
"volume": "",
"issue": "",
"pages": "94--119",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jurafsky, Dan & James H. Martin. 2019. Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition. 3rd ed., 94-119",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Semantics in Machine Translation",
"authors": [
{
"first": "Natalia",
"middle": [],
"last": "Klyueva",
"suffix": ""
}
],
"year": 2007,
"venue": "WDS'07 Proceedings of Contributed Papers, Part I",
"volume": "",
"issue": "",
"pages": "141--185",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Klyueva, Natalia. \"Semantics in Machine Translation.\" WDS'07 Proceedings of Contributed Papers, Part I, 2007, pp. 141-44, doi:https://www.mff.cuni.cz/veda/konference/wds/pro c/pdf07/WDS07_123_i3_Klyueva.pdf.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Multilingual Reliability and \"Semantic\" Structure of Continuous Word Spaces",
"authors": [
{
"first": "Maximilian",
"middle": [],
"last": "Koper",
"suffix": ""
},
{
"first": "Christian",
"middle": [],
"last": "Scheible & Sabine Schulte Im Walde",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of the 11th International Conference on Computational Semantics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Koper, Maximilian, Christian Scheible & Sabine Schulte im Walde. 2015. \"Multilingual Reliability and \"Semantic\" Structure of Continuous Word Spaces.\" Proceedings of the 11th International Conference on Computational Semantics, 15 Apr. 2015, doi:https://www.aclweb.org/anthology/W15-0105.pdf.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Using WordNets in Teaching Virtual Courses of Computational Linguistics",
"authors": [
{
"first": "Lothar & Claudia",
"middle": [],
"last": "Lemnitzer",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Kunze",
"suffix": ""
}
],
"year": 2003,
"venue": "Seminar f\u00fcr Sprachwissenschaft",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lemnitzer, Lothar & Claudia Kunze. 2003. \"Using WordNets in Teaching Virtual Courses of Computational Linguistics.\" Seminar f\u00fcr Sprachwissenschaft, Universit\u00e4t T\u00fcbingen, Jan. 2003",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Improving Word Vector with Prior Knowledge in Semantic Dictionary",
"authors": [
{
"first": "Wei",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Yunfang",
"middle": [],
"last": "Wu & Xueqiang",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Lv",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Li, Wei, Yunfang Wu & Xueqiang Lv. 2018. Improving Word Vector with Prior Knowledge in Semantic Dictionary. Beijing, Key Laboratory of Computational Linguistics, Peking University.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Efficient Estimation of Word Representations in Vector Space",
"authors": [
{
"first": "Tomas",
"middle": [],
"last": "Mikolov",
"suffix": ""
},
{
"first": "Kai",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Greg Corrado & Jeffrey",
"middle": [],
"last": "Dean",
"suffix": ""
}
],
"year": 2013,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mikolov, Tomas, Kai Chen, Greg Corrado & Jeffrey Dean. 2013. Efficient Estimation of Word Representations in Vector Space. arxiv.org/pdf/1301.3781.pdf.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Distributed representations of words and phrases and their compositionality",
"authors": [
{
"first": "Tomas",
"middle": [],
"last": "Mikolov",
"suffix": ""
},
{
"first": "Ilya",
"middle": [],
"last": "Sutskever",
"suffix": ""
},
{
"first": "Kai",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Greg Corrado & Jeffrey",
"middle": [],
"last": "Dean",
"suffix": ""
}
],
"year": 2013,
"venue": "Advances in Neural Information Processing Systems",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mikolov, Tomas, Ilya Sutskever, Kai Chen, Greg Corrado & Jeffrey Dean. 2013. \"Distributed representations of words and phrases and their compositionality.\" Advances in Neural Information Processing Systems, https://arxiv.org/pdf/1310.4546.pdf.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "WordNet: A Lexical Database for English",
"authors": [
{
"first": "George",
"middle": [
"A"
],
"last": "Miller",
"suffix": ""
}
],
"year": 1995,
"venue": "Communications of the ACM",
"volume": "38",
"issue": "11",
"pages": "39--41",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Miller, George A. 1995. \"WordNet: A Lexical Database for English\".Communications of the ACM, vol. 38, no. 11: 39-41.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Introduction to WordNet: An On-line Lexical Database",
"authors": [
{
"first": "George",
"middle": [],
"last": "Miller",
"suffix": ""
},
{
"first": "Richard",
"middle": [],
"last": "Beckwith",
"suffix": ""
},
{
"first": "Christiane",
"middle": [],
"last": "Fellbaum",
"suffix": ""
}
],
"year": 1993,
"venue": "",
"volume": "",
"issue": "",
"pages": "1--9",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Miller, George, Richard Beckwith, Christiane Fellbaum, Derek Gross & Katherine Miller. 1993. Introduction to WordNet: An On-line Lexical Database. Princeton University, 1-9",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Compiling dictionaries using semantic domains",
"authors": [
{
"first": "Ronald",
"middle": [],
"last": "Moe",
"suffix": ""
}
],
"year": 2003,
"venue": "Lexikos",
"volume": "13",
"issue": "",
"pages": "215--223",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Moe, Ronald. 2003. Compiling dictionaries using semantic domains. Lexikos 13, 215-223, doi:http://lexikos.journals.ac.za/pub/article/view/731",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Elicitation and Speech Acts in the Maskwac\u00ees Spoken Cree Dictionary Project",
"authors": [
{
"first": "Tanzi",
"middle": [],
"last": "Reule",
"suffix": ""
}
],
"year": 2018,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Reule, Tanzi. 2018. Elicitation and Speech Acts in the Maskwac\u00ees Spoken Cree Dictionary Project. Department of Linguistics, University of Alberta.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "A vector space model for semantic similarity calculation and OWL ontology alignment",
"authors": [
{
"first": "Ruben & Jaime",
"middle": [],
"last": "Tous",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Delgado",
"suffix": ""
}
],
"year": 2006,
"venue": "Proceedings of the 17th International Conference on Database and Expert Systems Applications",
"volume": "",
"issue": "",
"pages": "307--323",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tous, Ruben & Jaime Delgado. 2006. \"A vector space model for semantic similarity calculation and OWL ontology alignment.\" Proceedings of the 17th International Conference on Database and Expert Systems Applications, Sept. 2006, 307-16, doi:https://citeseerx.ist.psu.edu/viewdoc/download?do i=10.1.1.61.9727&rep=rep1&type=pdf.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "From Frequency to Meaning:Vector Space Models of Semantics",
"authors": [
{
"first": "Peter",
"middle": [
"D"
],
"last": "Turney",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Patrick Pantel",
"suffix": ""
}
],
"year": 2010,
"venue": "Journal of Artificial Intelligence Research",
"volume": "37",
"issue": "",
"pages": "141--1888",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Turney, Peter D. & Patrick Pantel. 2010. \"From Frequency to Meaning:Vector Space Models of Semantics.\" Journal of Artificial Intelligence Research, vol. 37, 141-1888, doi:https://www.jair.org/index.php/jair/article/view/10 640/25440.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Spicy Adjectives and Nominal Donkeys: Capturing Semantic Deviance Using Compositionality in Distributional Spaces",
"authors": [
{
"first": "E",
"middle": [
"M"
],
"last": "Vecchi",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Marelli",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Zamparelli",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Baroni",
"suffix": ""
}
],
"year": 2017,
"venue": "Cognitive Science",
"volume": "41",
"issue": "",
"pages": "102--136",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Vecchi EM, Marelli M, Zamparelli R & Baroni M. 2017. \"Spicy Adjectives and Nominal Donkeys: Capturing Semantic Deviance Using Compositionality in Distributional Spaces. \" Cognitive Science 41, 102-136",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Plains Cree: A Grammatical Study",
"authors": [
{
"first": "H",
"middle": [],
"last": "Wolfart",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Christoph",
"suffix": ""
}
],
"year": 1973,
"venue": "Transactions of the American Philosophical Society",
"volume": "63",
"issue": "5",
"pages": "1--90",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Wolfart, H. Christoph. 1973. \"Plains Cree: A Grammatical Study.\" Transactions of the American Philosophical Society, New Series, vol.63, no. 5, Nov. 1973, 1-90.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "n\u00eahiyaw\u00eawin: itw\u00eawina -Cree: Words",
"authors": [
{
"first": "Arok",
"middle": [],
"last": "Wolvengrey",
"suffix": ""
}
],
"year": 2001,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Wolvengrey, Arok. 2001. n\u00eahiyaw\u00eawin: itw\u00eawina - Cree: Words. 11th ed., University of Regina Press.",
"links": null
}
},
"ref_entries": {
"TABREF0": {
"text": "",
"type_str": "table",
"content": "<table><tr><td>What are the</td></tr></table>",
"html": null,
"num": null
}
}
}
} |