File size: 76,151 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 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 | {
"paper_id": "U07-1018",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T03:08:49.875592Z"
},
"title": "Dictionary Alignment for Context-sensitive Word Glossing",
"authors": [
{
"first": "Willy",
"middle": [],
"last": "Yap",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Melbourne",
"location": {}
},
"email": ""
},
{
"first": "Timothy",
"middle": [],
"last": "Baldwin",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Melbourne",
"location": {}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "This paper proposes a method for automatically sense-to-sense aligning dictionaries in different languages (focusing on Japanese and English), based on structural data in the respective dictionaries. The basis of the proposed method is sentence similarity of the sense definition sentences, using a bilingual Japanese-to-English dictionary as a pivot during the alignment process. We experiment with various embellishments to the basic method, including term weighting, stemming/lemmatisation, and ontology expansion.",
"pdf_parse": {
"paper_id": "U07-1018",
"_pdf_hash": "",
"abstract": [
{
"text": "This paper proposes a method for automatically sense-to-sense aligning dictionaries in different languages (focusing on Japanese and English), based on structural data in the respective dictionaries. The basis of the proposed method is sentence similarity of the sense definition sentences, using a bilingual Japanese-to-English dictionary as a pivot during the alignment process. We experiment with various embellishments to the basic method, including term weighting, stemming/lemmatisation, and ontology expansion.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "In a multi-lingual environment such as the Internet, users often stumble across webpages authored in an unfamiliar language which potentially contain information of interest. While users can consult dictionaries to help them understand the content of the webpages, the process of looking up words in unfamiliar languages is at best time-consuming, and at worst impossible due to a range of reasons. First, the writing system of the language may be unfamiliar to the user, e.g. the Cyrillic alphabet for a monolingual English speaker. Second, the user may not be familiar with the non-segmenting nature of languages such as Chinese and Japanese, and hence be incapable of delimiting the words to look up in the dictionary in the first place. Third, the user may be unable to lemmatise the word to determine the form in which it is listed in a dictionary.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "There are several alternatives to help decipher webpages in unfamiliar languages. The first one is to use an online machine translation system such as Altavista's Babel Fish 1 or Google Translate. 2 Figure 1 : Multiple translations for the Japanese word",
"cite_spans": [],
"ref_spans": [
{
"start": 199,
"end": 207,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "[ageru] produced by rikai.com. The correct translation in this context is \"to raise\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "While web-based machine translation services occasionally produce good translations for linguisticallysimilar languages such as English and French, they do not perform very well in translating languages which are removed from one another (Koehn, 2005) .",
"cite_spans": [
{
"start": 238,
"end": 251,
"text": "(Koehn, 2005)",
"ref_id": "BIBREF7"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The second alternative is a pop-up glossing application. The application takes raw text or a URL, parses the words, and returns the pop-up translation of each word as the mouse hovers over it. Some example pop-up glossing applications for Japanese source text and English glosses are Rikai 3 and POPjisyo. 4 With the aid of these pop-up translations, the manual effort of segmenting words (if necessary) and looking up each can be avoided. This application is also useful as an educational aid for learners of that language.",
"cite_spans": [
{
"start": 306,
"end": 307,
"text": "4",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The drawback with these applications is they display all possible translations of a given word irrespective of context. Faced with the task of determining the correct translation themselves, users frequently misinterpret words. An illustration of this situation is given in Figure 1 .",
"cite_spans": [],
"ref_spans": [
{
"start": 274,
"end": 282,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We propose a context-sensitive dictionary glossing application to enhance the utility of on-line glossing applications by sensitising the presented glosses to the context of use. The proposed method works by combining a monolingual word sense disambiguation (WSD) system (Baldwin et al., to appear) with an automatically induced cross-lingual sense alignment table. Based on the prediction(s) of the WSD system, our application presents the corresponding set of context-sensitive glosses to the user dictionary glossing by analysing the output of the alignment process.",
"cite_spans": [
{
"start": 271,
"end": 298,
"text": "(Baldwin et al., to appear)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "This paper focuses on the cross-lingual sense alignment aspect of the application. We take separate sense inventories for two distinct languages (Japanese and English in our case) and align the senses between the two. The basis of the alignment process is overlap in sense definitions. By adjusting a threshold for the required level of match, we are able to adjust the precision and recall of the alignment. In preliminary experimentation, we achieve promising results.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The remainder of this paper is structured as follows. We review previous research on dictionary alignment in Section 2, and outline the various resources we utilise during the alignment process in Section 3. We then describe the proposed basic sense-to-sense alignment method, along with various enhancements (Section 4), and present our experimental method and the results of our experiments (Sections 5 and 6, respectively). Finally we discuss our results and future research in Section 7.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "There has been a significant amount of research on bilingual dictionary alignment using a third language as a pivot. For example, built Japanese-French and Japanese-Korean dictionaries using English as the pivot language. In other research, used English and Chinese as pivots to generate a Korean-Japanese dictionary: English because of the accessibility of Korean-English and Japanese-English dictionaries, and Chinese because of the high overlap in orthography between Korean and Japanese, based on Chinese hanzi.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Previous Research",
"sec_num": "2"
},
{
"text": "There have been numerous attempts to manually develop multilingual resources that include crosslingual sense alignments (Vossen, 1998; Stamou et al., 2002) , and the import of cross-lingual semantic alignment has been ably demonstrated by the high impact of these resources. Due to the high overhead in manually constructing such resources, there have been various attempts at automatic crosslingual sense alignment. The methods are predominantly corpus-driven, based either on cross-lingual distributional similarity in a comparable corpus (e.g. Ngai et al. (2002) ) or word alignment over a parallel corpus (e.g. Gliozzo et al. (2005) ).",
"cite_spans": [
{
"start": 120,
"end": 134,
"text": "(Vossen, 1998;",
"ref_id": null
},
{
"start": 135,
"end": 155,
"text": "Stamou et al., 2002)",
"ref_id": "BIBREF15"
},
{
"start": 547,
"end": 565,
"text": "Ngai et al. (2002)",
"ref_id": "BIBREF10"
},
{
"start": 615,
"end": 636,
"text": "Gliozzo et al. (2005)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Previous Research",
"sec_num": "2"
},
{
"text": "There is a lesser amount of research on crosslingually aligning ontologies without using largescale corpus data, which we discuss in greater detail as it is more closely related to that proposed in this research. Asanoma (2001) aligned the Japanese Goi-Taikei ontology with WordNet by first translating a significant subset of the WordNet synonym sets (synsets) into Japanese, automatically matching these based on (monolingual Japanese) lexical overlap, and \"filling in the gaps\" for the remaining classes based on their hierarchical positioning relative to the aligned classes. Knight and Luk (1994) aligned Spanish and English senses based on: (1) overlap in sets of translations corresponding to each sense of a given Spanish word, with synsets in Word-Net; and (2) domain codes in the Spanish and English ontologies. They additionally aligned monolingual English dictionaries based on overlap in the definitions of each sense. The former cross-lingual case assumes a sense-discriminated bilingual dictionary, which we do not have access to. The latter case is similar to our research in that it compares definition sentences, but differs in that the definitions are in the same language. The most closely related work to our research is that of Nichols et al. (2005) , who aligned Lexeed senses with WordNet synsets as a by-product of the Lexeed ontology induction task (see Section 3.1), although they do not provide an explicit evaluation of the Lexeed-WordNet alignment for direct comparison.",
"cite_spans": [
{
"start": 213,
"end": 227,
"text": "Asanoma (2001)",
"ref_id": "BIBREF0"
},
{
"start": 580,
"end": 601,
"text": "Knight and Luk (1994)",
"ref_id": "BIBREF6"
},
{
"start": 1250,
"end": 1271,
"text": "Nichols et al. (2005)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Previous Research",
"sec_num": "2"
},
{
"text": "\u23a1 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a3 Word ryuu POS noun Sense 1 \u23a1 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a2 \u23a3 Lexical-type noun-lex Definition / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / / An imaginary animal.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Previous Research",
"sec_num": "2"
},
{
"text": "Dragons are like enormous snakes with 4 legs and horns. Dragons live in the sea, lakes and ponds, and are said to form clouds and cause rain when they fly up into the sky. Hypernym ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Previous Research",
"sec_num": "2"
},
{
"text": "ANIMAL \u23a4 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a6 Sense 2,3,4 ... Sense 5 \u23a1 \u23a2 \u23a2 \u23a3 Lexical-type noun-lex Definition / / / / / / / In shogi, a promoted rook. Domain SHOGI \u23a4 \u23a5 \u23a5 \u23a6 \u23a4 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a5 \u23a6",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Previous Research",
"sec_num": "2"
},
{
"text": "The Lexeed Semantic Database of Japanese is a machine-readable dictionary consisting of the most commonly-used words in Japanese (Kasahara et al., 2004) . In total, there are 28,000 words in Lexeed, and a total of 46,437 senses. Associated with each sense is a set of definition sentences, constructed entirely using the closed vocabulary of the 28,000 words found in Lexeed, such that 60% of the 28,000 words occur in the definition sentences (Tanaka et al., 2006) . In addition to the definition sentences, Lexeed also contains part of speech (POS), lexical relations between the senses (if any) and an example sentence, also based on the closed vocabulary of 28,000 words. All content words in the definition and example sentences are sense annotated.",
"cite_spans": [
{
"start": 129,
"end": 152,
"text": "(Kasahara et al., 2004)",
"ref_id": "BIBREF5"
},
{
"start": 444,
"end": 465,
"text": "(Tanaka et al., 2006)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The Lexeed semantic database of Japanese",
"sec_num": "3.1"
},
{
"text": "Automatic ontology acquisition methods have been applied to Lexeed to induce lexical relations between sense pairs, based on the sense-annotated definition sentences (Nichols et al., 2005) and comparison with both the Goi-Taikei thesaurus and WordNet 2.0.",
"cite_spans": [
{
"start": 166,
"end": 188,
"text": "(Nichols et al., 2005)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The Lexeed semantic database of Japanese",
"sec_num": "3.1"
},
{
"text": "An example Lexeed entry for the word ryuu is given in Figure 2 .",
"cite_spans": [],
"ref_spans": [
{
"start": 54,
"end": 62,
"text": "Figure 2",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "The Lexeed semantic database of Japanese",
"sec_num": "3.1"
},
{
"text": "EDICT is a free machine-readable Japanese-to-English dictionary (Breen, 1995) . The project is highly active and has been extended to other target languages such as German, French and Russian.",
"cite_spans": [
{
"start": 64,
"end": 77,
"text": "(Breen, 1995)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "EDICT",
"sec_num": "3.2"
},
{
"text": "EDICT contains more than 170,000 Japanese entries, each of which is associated with one or more English glosses. It also optionally contains information such as the pronunciation of the entry, POS, and domain of application.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "EDICT",
"sec_num": "3.2"
},
{
"text": "WordNet is an electronic semantic lexical database of English (Fellbaum, 1998) . It is made up of more than 100,000 synsets, with each synset representing a group of synonyms. Its entries are categorised into four POS categories: nouns, verbs, adjectives and adverbs. Each POS is described in a discrete lexical network.",
"cite_spans": [
{
"start": 62,
"end": 78,
"text": "(Fellbaum, 1998)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "WordNet",
"sec_num": "3.3"
},
{
"text": "Every synset in WordNet has a definition sentence, and sample sentence(s) are provided for most of the synsets; in combination, these are termed the WordNet gloss. Semantic relations connect one synset to another, and include relation types such as hypernym, hyponymy, antonymy and meronymy. The majority of these relations do not cross POS boundaries.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "WordNet",
"sec_num": "3.3"
},
{
"text": "Since we only experiment with hypernyms (and, symmetrically, hyponyms), we provide a simple review of this relation. A synset A is a hypernym of a synset B iff B is a kind of A. For example, vehicle is a hypernym of car, while perceive is a hypernym of hear, sight, touch, smell, taste. 5 Figure 3 : Example of normalisation of the translation string; we stop at \"rook\" as WordNet has a matching entry for it When building the baseline for our evaluation, we used the SemCor corpus-a subset of the Brown corpus annotated with WordNet senses-to derive the frequency counts of each WordNet synset (Landes et al., 1998) . Section 5 discusses this process in more detail.",
"cite_spans": [
{
"start": 595,
"end": 616,
"text": "(Landes et al., 1998)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [
{
"start": 289,
"end": 297,
"text": "Figure 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "WordNet",
"sec_num": "3.3"
},
{
"text": "Our basic alignment method, along with various extensions, is outlined below.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Proposed Methods",
"sec_num": "4"
},
{
"text": "In this paper, we align a semantic database of Japanese (Lexeed) with a semantic network of English (WordNet) at the sense level. First, we use Lexeed to find all possible senses of a given word, and retrieve the definition sentences for each. Since all the definition sentences are in Japanese, we use EDICT as a pivot to convert Lexeed definition sentences into English. In this process, all possible translations of all Japanese words found in the definition sentences are returned, along with their POS classes. For every translation returned, we find entries in WordNet that match the translation and POS category. If there is no match for the given POS, we relax this constraint and search for entries in WordNet that match the translation but not the POS.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Basic alignment method using cosine similarity",
"sec_num": "4.1"
},
{
"text": "Problems arise when WordNet does not have a matching entry for the translation. This situation doesn't distinguish between hyponyms and troponyms, however, we treat the two identically.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Basic alignment method using cosine similarity",
"sec_num": "4.1"
},
{
"text": "usually happens when the translation returned by EDICT is comprised of more than one English word. For a Japanese verb, e.g., the English translation in EDICT almost always begins with the auxiliary to (e.g. nomu is translated as to drink). WordNet does not contain a verbal entry for to drink, but does contain an entry for drink. To handle this case of partial match, we locate the longest right word substring of the EDICT translation which is indexed in WordNet.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Basic alignment method using cosine similarity",
"sec_num": "4.1"
},
{
"text": "A related problem is when the translation contains domain or collocational information in parentheses. For example, ryuu is translated as both dragon and promoted rook (shogi). The first translation has a matching entry in WordNet but the second translation does not. In this second case, there is no right word substring which matches in WordNet, as we end up with rook (shogi) and then (shogi), neither of which is contained in WordNet. In order to deal with this situation, we first normalise the translation strings by removing all the brackets and query WordNet with the normalised string. Should there be a matching entry, we stop here. If not, we then remove all strings between brackets, and apply the longest right word substring heuristic as above. An illustration of this process is given in Figure 3 .",
"cite_spans": [],
"ref_spans": [
{
"start": 803,
"end": 811,
"text": "Figure 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Basic alignment method using cosine similarity",
"sec_num": "4.1"
},
{
"text": "In the worst case of WordNet not having a matching entry for any right word substring, we discard the translation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Basic alignment method using cosine similarity",
"sec_num": "4.1"
},
{
"text": "At this point, we have aligned a given Japanese word with (hopefully) one or more English words, but are still no closer to inducing sense alignment pairs. In order to produce the sense alignments, we generate all pairings of Lexeed senses with WordNet synsets for each WordNet-matched word translation. For each such pair, we compile out the Lexeed definition sentence(s) word-translated into English, and the WordNet glosses, and convert each into a simple vector of term frequencies. We then measure the similarity of each vector pair using cosine similarity. An overview of this alignment process is presented in Figure 4 .",
"cite_spans": [],
"ref_spans": [
{
"start": 617,
"end": 625,
"text": "Figure 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Basic alignment method using cosine similarity",
"sec_num": "4.1"
},
{
"text": "The basic alignment method does not use any form of term weighting, and thus overemphasises common function words such as the, which and and, and downplays the impact of rare words. As we expect to have a large amount of noise in the word- Figure 4 : Overview of the Lexeed-WordNet sense alignment method translated Lexeed definition sentences, including spurious translations for Japanese function words such as ka, ga and no that have no literal translation in English, we predict that an appropriate form of term weighting should improve the performance of our method.",
"cite_spans": [],
"ref_spans": [
{
"start": 240,
"end": 248,
"text": "Figure 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Weighting terms using TF-IDF mechanism",
"sec_num": "4.2"
},
{
"text": "As a first attempt at term weighting, we experimented with the classic SMART formulation of TF-IDF (Salton, 1971) , treating the vector associated with each definition sentence as a single document.",
"cite_spans": [
{
"start": 99,
"end": 113,
"text": "(Salton, 1971)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Weighting terms using TF-IDF mechanism",
"sec_num": "4.2"
},
{
"text": "As mentioned in the previous section, commonlyoccurring semantically-bleached words are a source of noise in the naive cosine similarity scoring method. One conventional way of countering their impact is to filter them out of the vectors, based on a stop word list. For our experiments, we use the stop word list provided by the Snowball project. 6",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Word stopping",
"sec_num": "4.3"
},
{
"text": "Another source of possible noise is the translations of Japanese function words. As all the Lexeed definition sentences are POS tagged, it is a relatively simple process to filter out all Japanese function words, focusing on prefixes, suffixes and particles.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "POS filtering",
"sec_num": "4.4"
},
{
"text": "In its basic form, our vector space model treats distinct word as a unique term, including ignoring the 6 http://snowball.tartarus.org/ obvious similarity between inflectional variants of the same word, such as dragon and dragons. To remove such inflectional variation, we experiment with lemmatising all words found in both the Lexeed and WordNet vectors, using morph (Minnen et al., 2001 ). For similar reasons, we also experiment with the Porter stemmer, noting that stemming will further reduce the set of terms but potential introduce spurious matches. As part of this process (with both lemmatisation and stemming), we remove all punctuation from the definition sentences.",
"cite_spans": [
{
"start": 369,
"end": 389,
"text": "(Minnen et al., 2001",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Lemmatisation, stemming and normalisation",
"sec_num": "4.5"
},
{
"text": "Both the Lexeed and WordNet sense inventories are described in the form of hierarchies, making it possible to complement the sense definitions with those from neighbouring senses. The intuition behind this is that the sense granularity in the two sense inventories can vary greatly, such that a single sense in Lexeed is split across multiple WordNet synsets, which we can readily uncover by considering each sense as not a single point in WordNet but a semantic neighbourhood. For example, the second sense of the word kinou in Figure 5 , which literally means \"near past\", should be aligned with the second sense of yesterday, which is defined as \"the recent past\". This alignment is more self-evident, however, when we observe that the hypernym of each of the two senses is defined as \"past\".",
"cite_spans": [],
"ref_spans": [
{
"start": 529,
"end": 537,
"text": "Figure 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "Lexical relations",
"sec_num": "4.6"
},
{
"text": "In our current experiments, we only look at the utility of hypernymy. For a given sense Lexeed- Figure 5 : The output of word-translating Japanese definition sentences to English WordNet sense pairing, we extract the hypernyms of the respective senses and expand the definition sentences with the definition sentences from the hypernyms. The term vectors are then based on this expanded term set, similar to query expansion in information retrieval.",
"cite_spans": [],
"ref_spans": [
{
"start": 96,
"end": 104,
"text": "Figure 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "Lexical relations",
"sec_num": "4.6"
},
{
"text": "To evaluate the performance of our system, we randomly selected 100 words from Lexeed, extracted out the Lexeed-WordNet sense pairings as described above, and manually selected the goldstandard alignments from amongst them. The 100 words were associated with a total of 268 Lexeed senses and 772 WordNet senses, creating a total of 206,896 possible alignment pairs. Of these, 259 alignments were selected as our gold-standard.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Gold-standard data",
"sec_num": "5.1"
},
{
"text": "We encountered a number of partial matches that were caused by the Japanese word being more specific than its English counterparts (as identified by our WordNet matching method). For example, kakkazan is translated as \"active volcano\". Since WordNet does not have any entry for active volcano, the longest right word substring that matches in WordNet is simply volcano. The definition sentences returned by Lexeed describe kakkazan as \"a volcano which still can erupt\" and \"a volcano that will soon erupt\", while volcano is described as \"a fissure in the earth's crust (or in the surface of some other planet) through which molten lava and gases erupt\" and \"a mountain formed by volcanic material\". Although there is some similarity between these definitions (namely key words such as erupt and volcano), we do not include this pairing in our gold-standard alignment data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Gold-standard data",
"sec_num": "5.1"
},
{
"text": "As a baseline, we take the most-frequent sense of each of the 100 random words from Lexeed, and match it with the synset with the highest SemCor frequency count out of all the candidate synsets.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Baseline",
"sec_num": "5.2"
},
{
"text": "All our calculations are based on cosine similarity, which returns a similarity between 0 and 1, with 1 being an exact match. In its simplest form, we would identify the unique WordNet sense with highest similarity to each Lexeed sense, irrespective of the magnitude of the similarity. This has the dual disadvantage of allowing only one WordNet sense for each Lexeed sense, and potentially forcing alignments to be made on low similarity values. A more reasonable approach is to apply a threshold x, and treat all WordNet senses with similarity greater than x as being aligned with the Lexeed sense. Thresholding also gives us more flexibility in terms of tuning the performance of our method: at higher threshold values, we can hope to increase precision at the expense of recall, and at lower threshold values, we can hope to increase recall at the expense of precision.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Thresholding",
"sec_num": "5.3"
},
{
"text": "To evaluate the performance of our system, we use precision, recall and F-score. In an alignment context, precision is defined as the proportion of correct alignments to all alignments returned by the system, and recall is defined as the proportion of the correct alignments returned by our system to all the align- Table 1 : Best system F-score of combination of all features using the basic model vs. the basic model with TF-IDF weighting ments in our gold-standard. F-score is the harmonic mean of precision and recall, and provides a single figure-of-merit rating of the balance between these two factors. We evaluate our system using unbiased (\u03b2 = 1) F-score.",
"cite_spans": [],
"ref_spans": [
{
"start": 316,
"end": 323,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Evaluation metrics",
"sec_num": "5.4"
},
{
"text": "Throughout our experimentation, we evaluate relative to the 100 manually sense-aligned Japanese words.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "6"
},
{
"text": "Our baseline method predicts 100 alignments (as it is guaranteed to produce a unique alignment per source-language word), of which 60 are correct. Hence, the precision is 60 100 = 0.600, the recall is 60 259 = 0.231, and the F-score is 0.334.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "6"
},
{
"text": "With the basic alignment model, the highest Fscore achieved with thresholding is 0.228 at a threshold value of 0.19, well below the baseline F-score. The recall and precision value at this threshold are 0.263 and 0.202, respectively.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "6"
},
{
"text": "The basic model with TF-IDF weighting performed considerably better, scoring the highest Fscore of 0.292 (recall = 0.382 and precision = 0.236) at a threshold value of 0.04, but is still well below the baseline F-score. To confirm that TF-IDF term weighting is always beneficial to overall alignment performance, we took the unweighted model combined with each of the proposed extensions, and compared it with the same extension but with the inclusion of TF-IDF (without lexical relations at this point). The result of these experiments can be found in Table 1 . As we can see, TF-IDF weighting constantly improves alignment performance. Also note that, with the exception of simple (punctuation) normalisation, all extensions improve over the basic model both with and without TF-IDF weighting.",
"cite_spans": [],
"ref_spans": [
{
"start": 553,
"end": 560,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "6"
},
{
"text": "We extended our experiments by considering all possible combinations of 2 or more proposed extensions (excluding lexical relations for the time being) with TF-IDF weighting. Table 2 . The best result is achieved by combining all the proposed extensions, at an F-score of 0.364, which is significantly above baseline. It is also interesting to see that not all methods are fully complementary. By excluding stemming, e.g., the system actually performs better, producing a higher F-score of 0.372.",
"cite_spans": [],
"ref_spans": [
{
"start": 174,
"end": 181,
"text": "Table 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "6"
},
{
"text": "We then experimented with the addition of lexical relations to the different combinations of extensions explored above. The 5 top-performing combinations are presented in Table 3 . The best F-score of 0.408 is achieved with the combination of all the extensions proposed. When lexical relations are used exclusively or combined with less than three of the proposed extensions, the performance tends to decline.",
"cite_spans": [],
"ref_spans": [
{
"start": 171,
"end": 178,
"text": "Table 3",
"ref_id": "TABREF4"
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "6"
},
{
"text": "In our best performing combination, we outperformed the baseline F-score by 22%. 349 alignments were returned for this F-score, of which 124 matched the gold-standard. The precision and recall scores are 0.355 and 0.478, respectively.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "6"
},
{
"text": "We carried out more detailed analysis of the precision-recall trade-off. While we expect the pre- cision to go up to 1 as we increase our threshold, we found out that it is in fact not the case. The precision peaks at 0.625 at a threshold level of 0.265. At this level, there are 10 correct alignments out of 16 alignments returned. Upon investigating the six non-matching entries, we found that they all contain similar words but that the literal meaning of the senses are very different. Below, we present two of the six non-matching entries. The first one relates to a sense of the Japanese word shanpuu \"shampoo\". The definition sentences for this sense found in Lexeed are directly translated as \"shampoo medicine, drug, or dose; detergent or washing material that is used to wash hair or fur\". The corresponding match in WordNet is \"the act of washing your hair with shampoo\". We can see that there are similar terms in the two vectors, such as shampoo, washing and hair, but that the literal meaning of the two senses is quite different.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "6"
},
{
"text": "The second example is very similar to the kakkazan example presented in Section 5. One sense of sengetsu (\"last month\") is defined as \"the previous month\", and is aligned to the WordNet synset of month (WordNet does not have an entry for last month). It does not help that the hypernym of sengetsu is tsuki which translates to \"month\", boosting the similarity of this alignment.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": null
},
{
"text": "In terms of F-score, the best-performing combination of extensions performed better than the baseline. However, the recall seems to be the dominant factor in the F-score calculations for the proposed method. This is in sharp contrast to what we have in our baseline, where precision dominates the F-score calculation. There are several reasons for the baseline scores. First, there are 259 alignments in our gold-standard for 100 random words, corresponding to approximately 2.6 alignment per word. Given how we created our baseline, with one alignment per word, the maximum recall that the baseline can achieve is 100 259 = 0.386. On the other hand, the first-sense basis of the baseline method leads to high precision, largely due to the design process for ontologies and dictionaries. Namely, there is usually good coverage of frequent word senses in ontologies and dictionaries, and additionally, the translations for a given word are generally selected to be highly biased towards common senses (i.e. even if a polysemous word is chosen as a translation, its predominant sense is almost always that which corresponds to the source language word, for obvious accessibility/usability reasons). For this reason, there is a very high probability that these frequent senses for each of the two languages align with each other.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion",
"sec_num": "7"
},
{
"text": "In this paper, we proposed a cross-lingual senseto-sense alignment method, based on similarity of definition sentences as calculated via a bilingual dictionary. We explored various extensions to a simple lexical overlap method, and achieved promising results in preliminary experiments.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion",
"sec_num": "7"
},
{
"text": "In future work, we plan to exploit more lexical relations, such as synonymy and hyponymy. We also plan to experiment with weighting up alignments where both the sense pairing and the hypernym pairing match well. Nichols et al. (2005) linked Lexeed senses to WordNet in their evaluation on ontology induction. Comparison with their method would be very interesting and is an area for future research.",
"cite_spans": [
{
"start": 212,
"end": 233,
"text": "Nichols et al. (2005)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion",
"sec_num": "7"
},
{
"text": "http://babelfish.altavista.com/ 2 http://www.google.com/translate t",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "http://www.rikai.com/perl/Home.pl 4 http://www.popjisyo.com Proceedings of the Australasian Language Technology Workshop 2007, pages 125-133",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "ResourcesIn this section, we review the key resources used in this research.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Strictly speaking, hear, etc. are troponyms of perceive, i.e. they denote specific ways of perceiving. Because WordNet",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "We thank Francis Bond for his insights and suggestions for our experiment and Sanae Fujita for her help with the data. We are also grateful to anonymous reviewers for their helpful comments. This research was supported by NTT Communication Science Laboratories, Nippon Telegraph and Telephone Corporation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Alignment of ontologies: Word-Net and Goi-Taikei",
"authors": [
{
"first": "Naoki",
"middle": [],
"last": "Asanoma",
"suffix": ""
}
],
"year": 2001,
"venue": "Proc. of the NAACL 2001 Workshop on WordNet and Other Lexical Resources: Applications, Extensions and Customizations",
"volume": "",
"issue": "",
"pages": "89--94",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Naoki Asanoma. 2001. Alignment of ontologies: Word- Net and Goi-Taikei. In Proc. of the NAACL 2001 Workshop on WordNet and Other Lexical Resources: Applications, Extensions and Customizations, pages 89-94, Pittsburgh, USA.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "MRD-based word sense disambiguation: Further #2 extending #1 Lesk",
"authors": [
{
"first": "Timothy",
"middle": [],
"last": "Baldwin",
"suffix": ""
},
{
"first": "Su",
"middle": [
"Nam"
],
"last": "Kim",
"suffix": ""
},
{
"first": "Francis",
"middle": [],
"last": "Bond",
"suffix": ""
},
{
"first": "Sanae",
"middle": [],
"last": "Fujita",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Martinez",
"suffix": ""
},
{
"first": "Takaaki",
"middle": [],
"last": "Tanaka",
"suffix": ""
}
],
"year": null,
"venue": "Proc. of the Third International Joint Conference on Natural Language Prcoessing (IJCNLP 2008)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Timothy Baldwin, Su Nam Kim, Francis Bond, Sanae Fujita, David Martinez, and Takaaki Tanaka. to appear. MRD-based word sense disambiguation: Further #2 extending #1 Lesk. In Proc. of the Third International Joint Conference on Natural Language Prcoessing (IJCNLP 2008), Hyderabad, India.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Building an electronic Japanese-English dictionary",
"authors": [
{
"first": "Jim",
"middle": [],
"last": "Breen",
"suffix": ""
}
],
"year": 1995,
"venue": "Japanese Studies Association of Australia Conference",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jim Breen. 1995. Building an electronic Japanese- English dictionary. In Japanese Studies Association of Australia Conference.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "WordNet: An Electronic Lexical Database",
"authors": [],
"year": 1998,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Christiane Fellbaum, editor. 1998. WordNet: An Elec- tronic Lexical Database. MIT Press, Cambridge, USA.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Crossing parallel corpora and multilingual lexical databases for WSD",
"authors": [
{
"first": "Alfio",
"middle": [],
"last": "Massimiliano Gliozzo",
"suffix": ""
},
{
"first": "Marcello",
"middle": [],
"last": "Ranieri",
"suffix": ""
},
{
"first": "Carlo",
"middle": [],
"last": "Strapparava",
"suffix": ""
}
],
"year": 2005,
"venue": "Proc. of the 6th International Conference on Intelligent Text Processing and Computational Linguistics (CICLing-2005)",
"volume": "",
"issue": "",
"pages": "242--247",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Alfio Massimiliano Gliozzo, Marcello Ranieri, and Carlo Strapparava. 2005. Crossing parallel corpora and multilingual lexical databases for WSD. In Proc. of the 6th International Conference on Intelligent Text Processing and Computational Linguistics (CICLing- 2005), pages 242-5, Mexico City, Mexico.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Construction of a Japanese semantic lexicon: Lexeed",
"authors": [
{
"first": "Kaname",
"middle": [],
"last": "Kasahara",
"suffix": ""
},
{
"first": "Hiroshi",
"middle": [],
"last": "Sato",
"suffix": ""
},
{
"first": "Francis",
"middle": [],
"last": "Bond",
"suffix": ""
},
{
"first": "Takaaki",
"middle": [],
"last": "Tanaka",
"suffix": ""
},
{
"first": "Sanae",
"middle": [],
"last": "Fujita",
"suffix": ""
},
{
"first": "Tomoko",
"middle": [],
"last": "Kasunagi",
"suffix": ""
},
{
"first": "Shigeaki",
"middle": [],
"last": "Amano",
"suffix": ""
}
],
"year": 2004,
"venue": "Proc. of SIG NLC-159",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kaname Kasahara, Hiroshi Sato, Francis Bond, Takaaki Tanaka, Sanae Fujita, Tomoko Kasunagi, and Shigeaki Amano. 2004. Construction of a Japanese seman- tic lexicon: Lexeed. In Proc. of SIG NLC-159, IPSJ, Tokyo, Japan.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Building a largescale knowledge base for machine translation",
"authors": [
{
"first": "Kevin",
"middle": [],
"last": "Knight",
"suffix": ""
},
{
"first": "Steve",
"middle": [
"K"
],
"last": "Luk",
"suffix": ""
}
],
"year": 1994,
"venue": "Proc. of the 12th Annual Conference on Artificial Intelligence (AAAI-94)",
"volume": "",
"issue": "",
"pages": "773--781",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kevin Knight and Steve K. Luk. 1994. Building a large- scale knowledge base for machine translation. In Proc. of the 12th Annual Conference on Artificial Intelli- gence (AAAI-94), pages 773-8, Seattle, USA.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Europarl: A parallel corpus for statistical machine translation",
"authors": [
{
"first": "Philipp",
"middle": [],
"last": "Koehn",
"suffix": ""
}
],
"year": 2005,
"venue": "Proc. of the Tenth Machine Translation Summit",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Philipp Koehn. 2005. Europarl: A parallel corpus for statistical machine translation. In Proc. of the Tenth Machine Translation Summit (MT Summit X), Phuket, Thailand.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Building semantic concordances",
"authors": [
{
"first": "Shari",
"middle": [],
"last": "Landes",
"suffix": ""
},
{
"first": "Claudia",
"middle": [],
"last": "Leacock",
"suffix": ""
},
{
"first": "Randes",
"middle": [],
"last": "Tengi",
"suffix": ""
}
],
"year": 1998,
"venue": "WordNet: An Electronic Lexical Database",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shari Landes, Claudia Leacock, and Randes Tengi. 1998. Building semantic concordances. In Chris- tiane Fellbaum, editor, WordNet: An Electronic Lex- ical Database. MIT Press, Cambridge, USA.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Applied morphological processing of English",
"authors": [
{
"first": "Guido",
"middle": [],
"last": "Minnen",
"suffix": ""
},
{
"first": "John",
"middle": [],
"last": "Carroll",
"suffix": ""
},
{
"first": "Darren",
"middle": [],
"last": "Pearce",
"suffix": ""
}
],
"year": 2001,
"venue": "Natural Language Engineering",
"volume": "7",
"issue": "3",
"pages": "207--230",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Guido Minnen, John Carroll, and Darren Pearce. 2001. Applied morphological processing of English. Natu- ral Language Engineering, 7(3):207-23.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Identifying concepts across languages: A first step towards a corpus-based approach to automatic ontology alignment",
"authors": [
{
"first": "Grace",
"middle": [],
"last": "Ngai",
"suffix": ""
},
{
"first": "Marine",
"middle": [],
"last": "Carpuat",
"suffix": ""
},
{
"first": "Pascale",
"middle": [],
"last": "Fung",
"suffix": ""
}
],
"year": 2002,
"venue": "Proc. of the 19th International Conference on Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Grace Ngai, Marine Carpuat, and Pascale Fung. 2002. Identifying concepts across languages: A first step to- wards a corpus-based approach to automatic ontology alignment. In Proc. of the 19th International Confer- ence on Computational Linguistics (COLING 2002), Taipei, Taiwan.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Robust ontology acquisition from machine-readable dictionaries",
"authors": [
{
"first": "Eric",
"middle": [],
"last": "Nichols",
"suffix": ""
},
{
"first": "Francis",
"middle": [],
"last": "Bond",
"suffix": ""
},
{
"first": "Daniel",
"middle": [],
"last": "Flickinger",
"suffix": ""
}
],
"year": 2005,
"venue": "Proc. of the 19th International Join Conference on Artificial Intelligence (IJCAI-2005)",
"volume": "",
"issue": "",
"pages": "1111--1117",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Eric Nichols, Francis Bond, and Daniel Flickinger. 2005. Robust ontology acquisition from machine-readable dictionaries. In Proc. of the 19th International Join Conference on Artificial Intelligence (IJCAI-2005), pages 1111-6, Edinburgh, UK.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Using multiple pivots to align Korean and Japanese lexical resources",
"authors": [
{
"first": "Kyonghee",
"middle": [],
"last": "Paik",
"suffix": ""
},
{
"first": "Francis",
"middle": [],
"last": "Bond",
"suffix": ""
},
{
"first": "Shirai",
"middle": [],
"last": "Satoshi",
"suffix": ""
}
],
"year": 2001,
"venue": "Proc. of the NLPRS-2001 Workshop on Language Resources in Asia",
"volume": "",
"issue": "",
"pages": "63--70",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kyonghee Paik, Francis Bond, and Shirai Satoshi. 2001. Using multiple pivots to align Korean and Japanese lexical resources. In Proc. of the NLPRS-2001 Work- shop on Language Resources in Asia, pages 63-70, Tokyo.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "The SMART Retrieval System: Experiments in Automatic Document Processing",
"authors": [
{
"first": "Gerald",
"middle": [],
"last": "Salton",
"suffix": ""
}
],
"year": 1971,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gerald Salton. 1971. The SMART Retrieval Sys- tem: Experiments in Automatic Document Processing. Prentice-Hall.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Overlapping constraints of two step selection to generate a transfer dictionary",
"authors": [
{
"first": "Satoshi",
"middle": [],
"last": "Shirai",
"suffix": ""
},
{
"first": "Kazuhide",
"middle": [],
"last": "Yamamoto",
"suffix": ""
},
{
"first": "Kyonghee",
"middle": [],
"last": "Paik",
"suffix": ""
}
],
"year": 2001,
"venue": "ICSP-2001",
"volume": "",
"issue": "",
"pages": "731--736",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Satoshi Shirai, Kazuhide Yamamoto, and Kyonghee Paik. 2001. Overlapping constraints of two step selection to generate a transfer dictionary. In ICSP-2001, pages 731-736.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "BALKANET: A multilingual semantic network for the Balkan languages",
"authors": [
{
"first": "Sofia",
"middle": [],
"last": "Stamou",
"suffix": ""
},
{
"first": "Kemal",
"middle": [],
"last": "Oflazer",
"suffix": ""
},
{
"first": "Karel",
"middle": [],
"last": "Pala",
"suffix": ""
},
{
"first": "Dimitris",
"middle": [],
"last": "Christoudoulakis",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Cristea",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Tufi\u015f",
"suffix": ""
},
{
"first": "Svetla",
"middle": [],
"last": "Koeva",
"suffix": ""
},
{
"first": "George",
"middle": [],
"last": "Totkov",
"suffix": ""
},
{
"first": "Dominique",
"middle": [],
"last": "Dutoit",
"suffix": ""
},
{
"first": "Maria",
"middle": [],
"last": "Grigoriadou",
"suffix": ""
}
],
"year": 2002,
"venue": "Proc. of the International Wordnet Conference",
"volume": "",
"issue": "",
"pages": "12--16",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sofia Stamou, Kemal Oflazer, Karel Pala, Dimitris Chris- toudoulakis, Dan Cristea, Dan Tufi\u015f, Svetla Koeva, George Totkov, Dominique Dutoit, and Maria Grigo- riadou. 2002. BALKANET: A multilingual semantic network for the Balkan languages. In Proc. of the In- ternational Wordnet Conference, pages 12-4, Mysore, India.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "The hinoki sensebank -a large-scale word sense tagged corpus of Japanese",
"authors": [
{
"first": "Takaaki",
"middle": [],
"last": "Tanaka",
"suffix": ""
},
{
"first": "Francis",
"middle": [],
"last": "Bond",
"suffix": ""
},
{
"first": "Sanae",
"middle": [],
"last": "Fujita",
"suffix": ""
}
],
"year": 2006,
"venue": "Proceedings of the Workshop on Frontiers in Linguistically Annotated Corpora",
"volume": "",
"issue": "",
"pages": "62--71",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Takaaki Tanaka, Francis Bond, and Sanae Fujita. 2006. The hinoki sensebank -a large-scale word sense tagged corpus of Japanese -. In Proceedings of the Workshop on Frontiers in Linguistically Annotated Corpora 2006, pages 62-9, Sydney, Australia.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "EuroWordNet: A Multilingual Database with Lexical Semantic Networks",
"authors": [],
"year": 1998,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Piek Vossen, editor. 1998. EuroWordNet: A Multilingual Database with Lexical Semantic Networks. Kluwer Academic Publishers, Dordrecht, Netherlands.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"text": "A partial view of the Lexeed entry for [ryuu] (with English glosses)",
"num": null,
"type_str": "figure",
"uris": null
},
"FIGREF1": {
"text": "2: Top-5 combinations of extensions, excluding lexical relations (WS = Word stopping, PF = POS filtering, L = Lemmatisation, S = Stemming, N = Normalisation) periment is to investigate whether the proposed extensions are complementary in improving alignment performance. The 5 top-performing combinations are presented in",
"num": null,
"type_str": "figure",
"uris": null
},
"TABREF2": {
"num": null,
"text": "",
"html": null,
"type_str": "table",
"content": "<table/>"
},
"TABREF4": {
"num": null,
"text": "",
"html": null,
"type_str": "table",
"content": "<table><tr><td>: Top-5 performing combinations of exten-</td></tr><tr><td>sions, including lexical relations (WS = stopping, PF</td></tr><tr><td>= POS filtering, L = Lemmatisation, S = Stemming,</td></tr><tr><td>N = Normalisation, H = Hypernym)</td></tr></table>"
}
}
}
} |