File size: 88,243 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 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 |
{
"paper_id": "O12-1024",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T08:03:10.609961Z"
},
"title": "Translating Collocation using Monolingual and Parallel Corpus",
"authors": [
{
"first": "Ming-Zhuan",
"middle": [],
"last": "Jiang",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "National Tsing Hua Univ",
"location": {}
},
"email": ""
},
{
"first": "Tzu-Xi",
"middle": [],
"last": "Yen",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "National Tsing Hua Univ",
"location": {}
},
"email": ""
},
{
"first": "Chung-Chi",
"middle": [],
"last": "Huang",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "National Tsing Hua Univ",
"location": {}
},
"email": ""
},
{
"first": "Mei-Hua",
"middle": [],
"last": "Chen",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "National Tsing Hua Univ",
"location": {}
},
"email": "chen.meihua@gmail.com"
},
{
"first": "Jason",
"middle": [
"S"
],
"last": "Chang -\"",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "National Tsing Hua Univ",
"location": {}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "In this paper, we propose a method for translating a given verb-noun collocation based on a parallel corpus and an additional monolingual corpus. Our approach involves two models to generate collocation translations. The combination translation model generates combined translations of the collocate and the base word, and filters translations by a target language model from a monolingual corpus, and the bidirectional alignment translation model generates translations using bidirectional alignment information. At run time, each model generates a list of possible translation candidates, and translations in two candidate lists are re-ranked and returned as our system output. We describe the implementation of using method using Hong Kong Parallel Text. The experiment results show that our method improves the quality of top-ranked collocation translations, which could be used to assist ESL learners and bilingual dictionaries editors.",
"pdf_parse": {
"paper_id": "O12-1024",
"_pdf_hash": "",
"abstract": [
{
"text": "In this paper, we propose a method for translating a given verb-noun collocation based on a parallel corpus and an additional monolingual corpus. Our approach involves two models to generate collocation translations. The combination translation model generates combined translations of the collocate and the base word, and filters translations by a target language model from a monolingual corpus, and the bidirectional alignment translation model generates translations using bidirectional alignment information. At run time, each model generates a list of possible translation candidates, and translations in two candidate lists are re-ranked and returned as our system output. We describe the implementation of using method using Hong Kong Parallel Text. The experiment results show that our method improves the quality of top-ranked collocation translations, which could be used to assist ESL learners and bilingual dictionaries editors.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "A collocation is a recurrent combination of words that co-occur more frequently than expected by chance. Collocations can be classified into lexical and grammatical by the nature of their constituents. Another way of classifying collocations uses word positions to distinguish between rigid collocations and elastic collocations. Typically, a collocation consists of a base word and a collocate. Since collocations are used extensively, knowing the a right collocate for the base word plays an important role in second language learning as well as in machine translation. Translation of collocations is difficult for English as Second Language learners (ESL) because collocations are not always translated literally. For instance, the English collocation \"delegate authority\" can not be translated into \"7 :\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "INTRODUCTION",
"sec_num": "1"
},
{
"text": "Much previous work has been done on collocation translation by extracting bilingual collocations pairs from parallel corpora. Recently, researchers have also proposed methods for retrieve collocations and their translation based on parsers and bilingual dictionaries. However, previous works using parallel corpora are mostly heuristic and methods based on bilingual dictionaries may be limited by the availability of broad-coverage dictionaries.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "INTRODUCTION",
"sec_num": "1"
},
{
"text": "More recently, the mainstream Statistical Machine Translation (SMT) system like Moses has been widely used in many translation tasks such as translating texts and sentences. Unfortunately, the traditional SMT system does not take into consideration of the structure of collocations including variable word forms and non-contiguous phrases. Little work has been done on improving the SMT system for finding flexible collocations translation as a tool to assist ESL learners or to help the task of compiling bilingual collocation dictionaries.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "INTRODUCTION",
"sec_num": "1"
},
{
"text": "Consider the elastic collocation \"delegate ~ authority\" and its translations. The translations of \"authority\" can be \":\", \" \" and \".%\" which are found in parallel corpus. The traditional SMT system can find \"delegate some authority\" as \" \", but usually there is no continuous \"delegate authority\" phrase translation in the parallel corpus. The SMT system might translate the collocation word by word, resulting in a incorrect translation, such as \"7 :\" (Figure 2 ). Intuitively, a English collocation translation should be also a Chinese collocation, and using an appropriate Chinese collocation set might filter out the incorrect translations, and leads to better translations such as \" \". As shown in Figure 1 , Google Translate surely has a good translation in this example. In this paper, we propose a method that automatically translates the given collocation, by a combination word-based translation model and a bidirectional alignment translation model relying on aligned parallel corpora. A sample process of translating the collocation \"delegate authority\" is shown in Figure 2 . The output translation candidates are generated by these two models. At runtime, the given collocation is first decomposed into two parts as base words and collocates, in order to obtain a set of possible word translations. The combined translations of two words are then generated. The additional translations are also generated if available from the bidirectional alignment translation model. Finally, the top 3 Chinese translation candidates of these two models are combined, ranked and returned.",
"cite_spans": [],
"ref_spans": [
{
"start": 453,
"end": 462,
"text": "(Figure 2",
"ref_id": "FIGREF1"
},
{
"start": 703,
"end": 711,
"text": "Figure 1",
"ref_id": "FIGREF0"
},
{
"start": 1078,
"end": 1086,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "INTRODUCTION",
"sec_num": "1"
},
{
"text": "The rest of the paper is organized as follows. Chapter 2 reviews related works. Chapter 3 gives a formal statement of the problem that we attempt to resolve, and then present our method to extract translations from parallel corpus, involving generating translations by word alignment and filtering translation candidates using a dependency relation model. Chapter 4 describes the experimental settings and the data sets we utilize. In Chapter 5, we describe the evaluation results and present a further discussion. Finally, Chapter 6 gives the conclusion of this paper and points the future research direction.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "INTRODUCTION",
"sec_num": "1"
},
{
"text": "Machine Translation (MT) has been an area of active research since 1950's.. In the early years, rule-based approach is the state of the art for Machine Translation. Brown et al. (1993) propose a series of statistical models for improving MT performance and create a new approach called Statistical Machine Translation (SMT). Recently, much previous work have been done on phrase-based SMT (Marcu and Wong, 2002; Koehn et al. 2004) . While the traditional phrase-based SMT system which translates a paragraph of texts or a complete sentence, there are much previous work that consider translation of phrases, such as technical term translation (Dagan and Church, 1994) , noun phrase translation (Cao and Li, 2002; Koehn and Knight, 2003) , or bilingual collocation translation (Smadja et al. 1996) . These sub sentential translation tasks are helpful for assisting human translators or machine translation. In our work, we focus on retrieving bilingual collocations, similarly to what has been done by Smadja and McKeown 1996. Acquisition of bilingual collocation translation has been an active research topic recently. However, most previous work address translation of rigid collocation, such as technical terms and noun phrases (Kupiec, 1993; Ohmori and Higashida, 1999; Dagan and Church, 1994; Fung and Mckeown, 1997) . The traditional SMT and previous works also focus on translating continuous words in a sentence. Translating non-continuous words, such as elastic collocations, might result in an unseen phrase in training corpus and generate improper translations. In contrast, we focus on translating elastic collocations, which have intervening words between the base word and the collocate, such as verb-noun collocations.",
"cite_spans": [
{
"start": 178,
"end": 184,
"text": "(1993)",
"ref_id": null
},
{
"start": 389,
"end": 411,
"text": "(Marcu and Wong, 2002;",
"ref_id": "BIBREF11"
},
{
"start": 412,
"end": 430,
"text": "Koehn et al. 2004)",
"ref_id": "BIBREF4"
},
{
"start": 643,
"end": 667,
"text": "(Dagan and Church, 1994)",
"ref_id": "BIBREF2"
},
{
"start": 694,
"end": 712,
"text": "(Cao and Li, 2002;",
"ref_id": "BIBREF0"
},
{
"start": 713,
"end": 736,
"text": "Koehn and Knight, 2003)",
"ref_id": "BIBREF5"
},
{
"start": 776,
"end": 796,
"text": "(Smadja et al. 1996)",
"ref_id": "BIBREF17"
},
{
"start": 1001,
"end": 1025,
"text": "Smadja and McKeown 1996.",
"ref_id": "BIBREF17"
},
{
"start": 1230,
"end": 1244,
"text": "(Kupiec, 1993;",
"ref_id": "BIBREF8"
},
{
"start": 1245,
"end": 1272,
"text": "Ohmori and Higashida, 1999;",
"ref_id": "BIBREF12"
},
{
"start": 1273,
"end": 1296,
"text": "Dagan and Church, 1994;",
"ref_id": "BIBREF2"
},
{
"start": 1297,
"end": 1320,
"text": "Fung and Mckeown, 1997)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "RELATED WORK",
"sec_num": "2"
},
{
"text": "Many previous researchers have used bilingual dictionaries to generate collocation translations. L\u00fc and Zhou (2004) utilize bilingual dictionaries to generate collocation translation candidates and build a collocation translation triple model based on dependency parser using the EM algorithm. However, using bilingual dictionaries as the translation source might be limited by the coverage of dictionaries. In contrast, our method uses parallel corpora as source to generate collocation translations, in an attempt to avoid the problem of limited coverage of bilingual dictionaries.",
"cite_spans": [
{
"start": 97,
"end": 115,
"text": "L\u00fc and Zhou (2004)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "RELATED WORK",
"sec_num": "2"
},
{
"text": "Recently, retrieving collocation translation from sentence-aligned parallel corpora is a popular approach. Smadja et al. (1996) propose a statistical method based on DICE coefficient to measure the correlation of a collocation and its translations from sentence-aligned parallel corpus. However, using only statistical information, such as DICE, to translate collocations may generate translations which are not collocations in the target language. Intuitively, the translation of collocation is also a collocation in target language. For instance, the verb-noun collocation should have a translation which is also a verb-noun collocation in the target language. Zhou et al. (2001) found that about 70% of the Chinese translations have the same relation type with the source English collocations. Seretan and Wehrli (2007) introduce a similar method to identify verb-object collocation translation in sentence-aligned parallel corpus, using a parser to ensure that the both syntactical relations of the source collocation and the target translation are the same. Finally, an optional semantic filter using a bilingual dictionary can be used to validate the semantic head of collocations. Our approach, utilize a dependency parser, similar to Seretan and Wehrli' s (2007) method but with different experiment settings, to ensure that the target language translation has the same relation type as the source collocation using an additional monolingual corpus of the target language. The main difference between our work and previous works is that we extract word translations from a parallel corpus based on the word alignment information. More specifically, our method is based on statistical machine translation model, not statistical association measures such as DICE.",
"cite_spans": [
{
"start": 107,
"end": 127,
"text": "Smadja et al. (1996)",
"ref_id": "BIBREF17"
},
{
"start": 663,
"end": 681,
"text": "Zhou et al. (2001)",
"ref_id": "BIBREF18"
},
{
"start": 797,
"end": 822,
"text": "Seretan and Wehrli (2007)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "RELATED WORK",
"sec_num": "2"
},
{
"text": "In contrast to previous works, we present a model that generating collocation to assist ESL learners or bilingual dictionaries editors. The process of extracting word translation extraction is based on word alignment from parallel corpus. The translation candidates are filtered and ranked based on dependency relations, generated from a monolingual corpus using a target language dependency parser.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "RELATED WORK",
"sec_num": "2"
},
{
"text": "Submitting a collocation to the SMT system directly might not receive a correct or fluent translation. The traditional SMT system typically translates continuous phrases. Unfortunately, elastic collocations, such as verb-noun collocations, which contain intervening words, may be unseen phrases in the training corpus of an SMT system. The SMT system might translate unseen phrases word by word, and generates inappropriate translations. To generate a proper translation for elastic collocation, an effective approach is to consider the structure of collocations and various word forms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "We focus on finding translation equivalents of verb-noun collocation in a parallel corpus. These translations then are ranked and returned as output. The returned translations can be examined by a human user directly, or passed to an SMT system to improve translation quality. Therefore, our goal is to return a set of ranked collocation translations. We now formally state the problem we are addressing.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Statement",
"sec_num": "3.1"
},
{
"text": "We are given a verb-noun collocation (V c , N c ) and a word-aligned parallel corpus PC, and a phrase table PT from a SMT system (e.g., Moses). Our goal is to retrieve a set of combined translations of the base word and the collocate ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Statement:",
"sec_num": null
},
{
"text": "CT combine = {(V t_comb , N t_comb ) 1 , (V t_comb , N t_comb ) 2 , \u2026, (V",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Statement:",
"sec_num": null
},
{
"text": "= {(V t_align , N t_align ) 1 , (V t_align , N t_align ) 2 , \u2026, (V t_align , N t_align ) n }",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Statement:",
"sec_num": null
},
{
"text": "from PC. These translations are finally ranked and returned as the system output.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Statement:",
"sec_num": null
},
{
"text": "In the rest of the paper, we describe the method for solving this problem in detail. First, we show the steps of extracting collocation translation from PC and building translation models (Section 3.2). Finally, we present how to generate collocation translations by these two models and ranks translation candidates at run time (Section 3.3).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Statement:",
"sec_num": null
},
{
"text": "We attempt to find translations of verb-noun collocations from a parallel corpus, and filter translation candidates using a monolingual corpus. Our training process is showed in Figure 3 . Generate word alignment from parallel corpus",
"cite_spans": [],
"ref_spans": [
{
"start": 178,
"end": 186,
"text": "Figure 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Extracting Collocation Translation from Parallel Corpus",
"sec_num": "3.2"
},
{
"text": "In the first stage of the training process (Step (1) in Figure 3 .), we generate word alignment data for each sentence pair in a parallel corpus using a alignment tool.",
"cite_spans": [],
"ref_spans": [
{
"start": 56,
"end": 64,
"text": "Figure 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Extracting Collocation Translation from Parallel Corpus",
"sec_num": "3.2"
},
{
"text": "The input for this stage of training is a parallel corpus, as we will describe in Section 4.1. For each sentence pair in the parallel corpus, we use a word alignment tool to align a source word to the corresponding target words. The same procedure is performed in the inverse direction, from the target language to the source language. The output of this stage is the alignment information in both directions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Extracting Collocation Translation from Parallel Corpus",
"sec_num": "3.2"
},
{
"text": "The alignment information in both directions is used to generate the phrase table (Section 3.2.2) and bidirectional alignment translation model (Section 3.2.3).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Extracting Collocation Translation from Parallel Corpus",
"sec_num": "3.2"
},
{
"text": "In the second stage of the training process, we build a combination translation model based on a word translation model using a parallel corpus and a model based on a separate target language corpus. The word translation model is used to generate translations of the base word and the collocate of a given collocation. To build this model, we need a phrase table PT, which is generated using an SMT tools, as our training data. A typical phrase table in the SMT system is usually contains the corresponding translation equivalents with direct and inverse translation probabilities for almost all the phrases up to a certain length in the training corpus. Figure 4 shows a sample part of the phrase table: Figure 4 as an example, for each translation pair t i in PT, the bidirectional translation probability P i_bidirect is calculated:",
"cite_spans": [],
"ref_spans": [
{
"start": 655,
"end": 663,
"text": "Figure 4",
"ref_id": "FIGREF4"
},
{
"start": 705,
"end": 713,
"text": "Figure 4",
"ref_id": "FIGREF4"
}
],
"eq_spans": [],
"section": "Build the combination translation model",
"sec_num": "3.2.2"
},
{
"text": "where P i_inverse (e_given_c in Figure 4 .) is the inverse translation probability and P i_direct (c_given_e in Figure 4 .) is the direct translation probability. Then we build the word translation model, which consists of translation pairs and corresponding bidirectional translation probabilities. Table 1 . shows an example of the word translation model. The target language model is also required to filter out inappropriate collocation translations. We build this model based on a target language monolingual corpus.",
"cite_spans": [],
"ref_spans": [
{
"start": 32,
"end": 40,
"text": "Figure 4",
"ref_id": "FIGREF4"
},
{
"start": 112,
"end": 120,
"text": "Figure 4",
"ref_id": "FIGREF4"
},
{
"start": 300,
"end": 307,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "Build the combination translation model",
"sec_num": "3.2.2"
},
{
"text": "In the first step of the procedure, we parse a monolingual corpus of the target language using a dependency parser to generate RelationPairs. For each relation pair in RelationPairs, we only count the frequency of the verb-noun relation pairs <w1, w2, VN>, since we aim at",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Build the combination translation model",
"sec_num": "3.2.2"
},
{
"text": "# \"! # \"! # $ # $",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Build the combination translation model",
"sec_num": "3.2.2"
},
{
"text": "Proceedings of the Twenty-Fourth Conference on Computational Linguistics and Speech Processing (ROCLING 2012) translating verb-noun collocation. Next, we generate the target language model VNPairsFrequency, consisting of the frequency of each verb-noun relation pair. The combination translation model is then generated by combining the word translation model and the target language model. We will describe the run time process of combination translation in Section 3.3.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Build the combination translation model",
"sec_num": "3.2.2"
},
{
"text": "In the third and final stage of training, we address building a bidirectional alignment translation model from word alignment for translating collocations. The input to this stage is the alignment information of both directions Align_StoT and Align_TtoS, generated in the previous section (Section 3.2.1). The algorithm is shown in Figure 5 . In Step (2) of the algorithm, we extract translations for each source word in each sentence pair. We first find target words aligned to the source word by the source to target alignment information. For each aligned target word, the target to source alignment information is then used to determine whether the source word is also aligned to this target word. We choose the target word as the translation of the source word if the source word is also aligned to it. The translations of each source word SrcTrans are generated.",
"cite_spans": [],
"ref_spans": [
{
"start": 332,
"end": 340,
"text": "Figure 5",
"ref_id": "FIGREF5"
}
],
"eq_spans": [],
"section": "Build the bidirectional alignment translation model",
"sec_num": "3.2.3"
},
{
"text": "In Step (3a), source skip bigrams are generated for each source sentence. A skip bigram is combined by the head word and the tail word of a phrase. In order to limit the amount of the data processed, we only consider phrases with the distance 1 to 4 words in generating skip bigram. Then, in Step (3b), we retrieve the corresponding translations for each skip bigram.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Build the bidirectional alignment translation model",
"sec_num": "3.2.3"
},
{
"text": "In Step (4), we count the frequency of each skip bigram translation pair. Since we focus on translating verb-noun collocation, we only deal with verb-noun bigram and translation pairs to reduce processing time. Table 1 shows examples related to the skip bigram \"play role\"",
"cite_spans": [],
"ref_spans": [
{
"start": 211,
"end": 218,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "Build the bidirectional alignment translation model",
"sec_num": "3.2.3"
},
{
"text": "Finally, in Step (5), the frequency of skip bigram and translation pairs, BidirectionalTransFreq, is returned. Table 2 shows an example output of this stage. ",
"cite_spans": [],
"ref_spans": [
{
"start": 111,
"end": 118,
"text": "Table 2",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Build the bidirectional alignment translation model",
"sec_num": "3.2.3"
},
{
"text": "Once all collocation translation models are obtained, these models are combined and used to translate collocations. For a given collocation, we generate and evaluate translations using the procedure shown in Figure 6 . In the following, we first present the translating process of the combination translation model, and then the bidirectional alignment model. Finally, we describe the ranking algorithm to output the collocation translations. ",
"cite_spans": [],
"ref_spans": [
{
"start": 208,
"end": 216,
"text": "Figure 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "The Run Time Process",
"sec_num": "3.3"
},
{
"text": "In Step (1a), we first decompose the given collocation into the base word and the collocate. Consider the collocation \"delegate authority\" for example, \"authority\" is the base word and \"delegate\" is the collocate. A set of the base word translations is generated as BaseTransList, and translation list for the collocate CollocateTransList is also generated. We then generate possible collocation translations CombTransList using Cartesian product of each bTrans in BaseTransList and each cTrans in CombTransList. Each CombTrans (bTrans, cTrans) in CombTransList gets a word translation model score using the following formula:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Combination Translation Model",
"sec_num": "1"
},
{
"text": "and a target language model score as follows :",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Combination Translation Model",
"sec_num": "1"
},
{
"text": "Proceedings of the Twenty-Fourth Conference on Computational Linguistics and Speech Processing (ROCLING 2012) where is the smoothing weight to cope with the data sparse problem,",
"cite_spans": [
{
"start": 95,
"end": 109,
"text": "(ROCLING 2012)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Combination Translation Model",
"sec_num": "1"
},
{
"text": ". We combine the and by a weighting formula: where is the model weight and .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Combination Translation Model",
"sec_num": "1"
},
{
"text": "We retrieve the translations and rank translations in descending order of Score CombTM . The N top-ranked translations of combination translation model are produced.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Combination Translation Model",
"sec_num": "1"
},
{
"text": "In step (4a), we generate another set of translations using the bidirectional alignment model for the given collocation C. Translations of C are retrieved from the bidirectional alignment model, and each translation is scored as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Bidirectional Alignment Model",
"sec_num": "3.3.2"
},
{
"text": "The generated translations are ranked in descending order of Score BiModel , and N top-ranked translations of bidirectional alignment model are retrieved.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Bidirectional Alignment Model",
"sec_num": "3.3.2"
},
{
"text": "Once all translations of two models are generated, we merge the N top-ranked translations of two models and re-rank them. The ranking algorithm we use aims at retrieving the translations that two models have in common. The score of the top N translation of each model is re-calculated as the formula:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Bidirectional Alignment Model",
"sec_num": "3.3.2"
},
{
"text": "where N means the output rank of a translation in a model. We then merge all translations, and if there is a translation that both in output of two models, we add two scores together. Finally, the merged translations are ranked with their merged score (Step 5), and the K top-ranked translations are returned as the final result produced by our method.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Bidirectional Alignment Model",
"sec_num": "3.3.2"
},
{
"text": "We have proposed a new method to retrieve translations for a given collocation from parallel corpus that are likely to help ESL learners or bilingual collocation dictionary editors. As such, our method is trained and evaluated on top of word alignment information of parallel corpus and an additional monolingual corpus. Furthermore, since the goal of our model was to retrieve a set of good translations to assist bilingual collocation dictionary editors, we evaluated our method on a group of English collocations, which are selected from an English collocation dictionary. Finally, since we do not have reference answers for such translation advising task, we will use human judges to evaluate the quality of our generated collocation translations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Setting and Results",
"sec_num": "4"
},
{
"text": "In this chapter, we first present the details of training our system for the evaluation (Section 4.1) . Then, Section 4.2 describes the alternative methods that we used in our comparison. Section 4.3 introduces the datasets used in our experiments and the evaluation metrics for evaluating the performance of our system, and Section 4.4 describe the tuning process of our system module. Section 4.5 reports the results of our experiment evaluations. Finally, in Section 4.6, we analyze the experimental results in detail.",
"cite_spans": [],
"ref_spans": [
{
"start": 88,
"end": 101,
"text": "(Section 4.1)",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Experimental Setting and Results",
"sec_num": "4"
},
{
"text": "In our bidirectional alignment translation model, we used the Hong Kong Parallel Text (HKPT; LDC2004T08) as the training data, which contains approximately 222,000 sentence pairs,. English sentences of HKPT were lower-cased and performed lemmatization using For word translation model of our combination translation model, the phrase table of HKPT was built by the state-of-the-art phrase-based SMT system, Moses (Koehn et al., 2007) . Common settings are used to run Moses: GIZA++ was used for word alignment, grow-diagonal-final (Koehn et al., 2005) heuristics were used to combine bi-direction word alignment, and extract bilingual phrase (Koehn et al., 2005) .",
"cite_spans": [
{
"start": 413,
"end": 433,
"text": "(Koehn et al., 2007)",
"ref_id": "BIBREF7"
},
{
"start": 531,
"end": 551,
"text": "(Koehn et al., 2005)",
"ref_id": null
},
{
"start": 642,
"end": 662,
"text": "(Koehn et al., 2005)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Settings",
"sec_num": "4.1"
},
{
"text": "For the target language model of our combination translation model, we used Central News Agency (CNA) as the monolingual corpus, by using the CKIP Chinese Parser to produce dependency relations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Settings",
"sec_num": "4.1"
},
{
"text": "Our system uses some parameters during training. The parameters were tested with different values and finally the values were set as shown in Table 3 . We did not test these values exhaustively and further tuning may improve the performance of our system. ",
"cite_spans": [],
"ref_spans": [
{
"start": 142,
"end": 149,
"text": "Table 3",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Experimental Settings",
"sec_num": "4.1"
},
{
"text": "Recall that our method starts with an English verb-noun collocation given by a user, and find the Chinese translations of the collocation. The output of our system is a list of ranked translation candidates, which can either be shown to the user directly, or incorporated into the existing SMT systems.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methods Compared",
"sec_num": "4.2"
},
{
"text": "In this paper, we have introduced a hybrid method for generating collocation translations from a parallel corpus and an additional target language monolingual corpus for a given collocation. Therefore, we compare the results of different translation retrieval methods from a parallel corpus.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methods Compared",
"sec_num": "4.2"
},
{
"text": "We compared different methods for retrieving collocation translations from a parallel corpus, which are listed as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methods Compared",
"sec_num": "4.2"
},
{
"text": "-MOSES: The state-of-the-art SMT framework that are widely used recently. We build the Moses translating system using the same HKPT parallel corpus with default setting as our baseline system. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methods Compared",
"sec_num": "4.2"
},
{
"text": "The evaluation of the traditional SMT systems usually base on the quality of translated texts. Bilingual Evaluation Understudy (BLEU; Papineni et al, 2002) is a mainstream automatically scheme to evaluate quality of the MT translations. The translation of the input texts is compared the similarity with human-translated reference answers. However, since our system aims at assisting user to find appropriate translations for bilingual collocation dictionaries editors, the lack of reference translations results in a difficult situation of translation equivalents.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Data Sets and Metrics",
"sec_num": "4.3"
},
{
"text": "To evaluate our system, we randomly selected 55 English verb-noun collocations from the Oxford Collocations Dictionary (OCD; Oxford University Press, 2009), which collects about 25,000 common collocations. All nouns of collocations were chose from Academic Word List (AWL; Coxhead 2003). The testing data consisted of 80 collocations, which were selected in the same way.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Data Sets and Metrics",
"sec_num": "4.3"
},
{
"text": "We used two human judges to examine the generated translations for each collocation in the data sets for evaluation. The human judges were asked to examine retrieved collocation translation one at a time, and judge each translation candidate as \"correct\", \"partial acceptable\", or \"unacceptable\" for the collocations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Data Sets and Metrics",
"sec_num": "4.3"
},
{
"text": "By using the judgments from two human judges, we evaluate the translations using the Top-N accuracy, and Mean Reciprocal Rank (MRR) metrics that describes in the next. find the best MRR values by using development data. As the result. We make = 0.4 as our model weight.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Data Sets and Metrics",
"sec_num": "4.3"
},
{
"text": "In this section, we evaluate the performance of various systems in Section 4.2 using the testing data set and different metrics we described in Section 4.3.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Results",
"sec_num": "4.5"
},
{
"text": "For each compared system, we generated top 3 ranked translations for each collocation in the testing data. Samples of the system output for collocations in the testing data are listed in Appendix A. We first calculate the Kappa value to acquire the agreement between two judges. In order to calculate the Kappa value, we mixed all top 3 translations from various systems and generated a translation pool, which contains all generated translations from different systems for each collocation in test data. The human judges then evaluated on all 1451 translations in the translation pool, and we got the Kappa value , which indicates that the human judges have substantial agreement while judging translation results ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Results",
"sec_num": "4.5"
},
{
"text": ".78",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "CTM .76 Hybrid (CTM+BTM)",
"sec_num": null
},
{
"text": "We report the top-N accuracies from top-1 to top-5 in Table 4 . The results indicate that, except the top-1 accuracy, our Hybrid method has significantly better accuracy improvement than other three methods from top-2 to top-5. Compared with the baseline, our system improves 7% ~ 10% more accuracies. Hybrid, combined CTM and BTM, improves about more 6% accuracy than only CTM. This result indicates that although top-N accuracies of BTM is the lowest since it suffers from low translation coverage, BTM still improves Table 4 reports the MRR value for all compared methods. The reported MRR is an average value of the judgment by two judges. Hybrid has the best MRR 0.78 of all methods , which means that a correct answer can be found at the first 2 translations in ranked translation list by a human user. Also our HYBRID method, compared to the traditional SMT system MOSES, improves 0.06 MRR score.",
"cite_spans": [],
"ref_spans": [
{
"start": 54,
"end": 61,
"text": "Table 4",
"ref_id": null
},
{
"start": 520,
"end": 527,
"text": "Table 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "CTM .76 Hybrid (CTM+BTM)",
"sec_num": null
},
{
"text": "In this paper, we have introduced a new method for translating verb-noun collocations by using a parallel corpus and an additional monolingual corpus. The generated collocation translations can be used to assist ESL learners and bilingual collocation dictionaries editors with the choice of proper translations. Our method is based on a parallel corpus to extract collocation translations, and a monolingual corpus of the target language to filter out inappropriate translations. Evaluations of our experiments have shown that our method produce better translations for a given collocation than the traditional SMT system. Many avenues exist for future research and improvement of our system. For example, we could extend the parallel corpus by using more general corpora to increase the quality of collocation translations. The ranking algorithm to combine and rank outputs of two models could be used a better existing algorithm. Also, dealing with different types of collocation, such as Adjective-Noun and Phrasal Verb-Noun, could be considered to translate more collocations in our system. Additionally, an interesting direction to explore is to use more semantic information to improve translations. If example sentences of a collocation are available, we could use the word sense disambiguation technique to help us choose a precise translation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion and Future Work",
"sec_num": "5"
},
{
"text": "Proceedings of the Twenty-Fourth Conference on Computational Linguistics and Speech Processing(ROCLING 2012)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "Definition 4.1. The Top-N accuracy of a translation model for K collocations in test data, in our definition, is the percentage of all collocations with translation results, where Top-N translations contain a correct translation.Example 4.1. Consider top 3 translations returned by the system for 10 collocations in test data. If there are 3 collocations with correct translations at first place, 2 at second place, and 1 at third place, the Top-N accuracy of this system is (3+2+1)/10 = 60%.We also compute Mean Reciprocal Rank (MRR), a measure of how much effort needed for a user to find a compatible translation in the returned order of collocation translations (Voorhees and Tice, 1999). The MRR value is a real number between 0 and 1, where 1 denotes the compatible translations always occur at first place. We report the MRR results to examine the effectiveness of our system being used to assist bilingual dictionaries editors. We also calculate Kappa statistics (Cohen, 1960) to evaluate the agreement between two human judges. Cohen's Kappa coefficient is a statistical measure of the inter-judge agreement, which consider the agreement occurring by chance and the agreement of observed judgment result. If the judges are in complete agreement with each other for the classification totally, then = 1. If there is no agreement between the judges, then 0. where Pr(a) is relative observed agreement between judges, and Pr(e) is the hypothetical probability of agreement by chance, which is calculated by using the observed judgments by each human judge.",
"cite_spans": [
{
"start": 971,
"end": 984,
"text": "(Cohen, 1960)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "annex",
"sec_num": null
},
{
"text": "In this section, we describe the process of tuning the parameter (weight of word translation model in the combination translation model (CTM) ) by using the development data. Recall that the score of CTM is calculated as the following:The different weights of determine whether the word translation model (WTM) or the target language model (TLM) has more influence on the collocation translations score Score CombTM . A higher value of means that Score CombTM relies more on WTM than TLM. In contrast, TLM has more influence for a lower .To select a suitable weight , we choose a set values in the division between 0 and 1 to Proceedings of the Twenty-Fourth Conference on Computational Linguistics and Speech Processing (ROCLING 2012) ",
"cite_spans": [
{
"start": 721,
"end": 735,
"text": "(ROCLING 2012)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Tuning Parameters",
"sec_num": "4.4"
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Base noun phrase translation using web data and the EM algorithm",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Cao",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Li",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the 19th international conference on computational linguistics",
"volume": "1",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cao Y. and Li H. 2002. Base noun phrase translation using web data and the EM algorithm. In Proceedings of the 19th international conference on computational linguistics-volume 1, Association for Computational Linguistics. 1 p.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "A coefficient of agreement for nominal scales",
"authors": [
{
"first": "J",
"middle": [],
"last": "Cohen",
"suffix": ""
}
],
"year": 1960,
"venue": "Educational and Psychological Measurement",
"volume": "20",
"issue": "1",
"pages": "37--46",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cohen J. 1960. A coefficient of agreement for nominal scales. Educational and Psychological Measurement 20(1):37-46.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Termight: Identifying and translating technical terminology",
"authors": [
{
"first": "I",
"middle": [],
"last": "Dagan",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Church",
"suffix": ""
}
],
"year": 1994,
"venue": "Proceedings of the fourth conference on applied natural language processing",
"volume": "34",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dagan I. and Church K. 1994. Termight: Identifying and translating technical terminology. In Proceedings of the fourth conference on applied natural language processing, Association for Computational Linguistics. 34 p.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "A technical word-and term-translation aid using noisy parallel corpora across language groups",
"authors": [
{
"first": "P",
"middle": [],
"last": "Fung",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Mckeown",
"suffix": ""
}
],
"year": 1997,
"venue": "Machine Translation",
"volume": "12",
"issue": "1",
"pages": "53--87",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Fung P and McKeown K. 1997. A technical word-and term-translation aid using noisy parallel corpora across language groups. Machine Translation 12(1):53-87.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Pharaoh: A beam search decoder for phrase-based statistical machine translation models. Machine Translation: From Real Users to Research",
"authors": [
{
"first": "P",
"middle": [],
"last": "Koehn",
"suffix": ""
}
],
"year": 2004,
"venue": "",
"volume": "",
"issue": "",
"pages": "115--139",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Koehn P. 2004. Pharaoh: A beam search decoder for phrase-based statistical machine translation models. Machine Translation: From Real Users to Research :115-24.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Feature-rich statistical translation of noun phrases",
"authors": [
{
"first": "P",
"middle": [],
"last": "Koehn",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Knight",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of the 41st annual meeting on association for computational linguistics-volume 1Association for Computational Linguistics",
"volume": "311",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Koehn P. and Knight K. 2003. Feature-rich statistical translation of noun phrases. Proceedings of the 41st annual meeting on association for computational linguistics-volume 1Association for Computational Linguistics. 311 p.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Statistical phrase-based translation",
"authors": [
{
"first": "P",
"middle": [],
"last": "Koehn",
"suffix": ""
},
{
"first": "F",
"middle": [
"J"
],
"last": "Och",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Marcu",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of the 2003 conference of the north american chapter of the association for computational linguistics on human language technology-volume 1Association for Computational Linguistics",
"volume": "48",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Koehn P., Och F. J. and Marcu D. 2003. Statistical phrase-based translation. Proceedings of the 2003 conference of the north american chapter of the association for computational linguistics on human language technology-volume 1Association for Computational Linguistics. 48 p.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Moses: Open source toolkit for statistical machine translation",
"authors": [
{
"first": "P",
"middle": [],
"last": "Koehn",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Hoang",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Birch",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Callison-Burch",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Federico",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Bertoldi",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Cowan",
"suffix": ""
},
{
"first": "W",
"middle": [],
"last": "Shen",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Moran",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Zens",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the 45th annual meeting of the ACL on interactive poster and demonstration sessionsAssociation for Computational Linguistics",
"volume": "177",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Koehn P., Hoang H., Birch A., Callison-Burch C., Federico M., Bertoldi N., Cowan B., Shen W., Moran C. and Zens R. 2007. Moses: Open source toolkit for statistical machine translation. Proceedings of the 45th annual meeting of the ACL on interactive poster and demonstration sessionsAssociation for Computational Linguistics. 177 p.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "An algorithm for finding noun phrase correspondences in",
"authors": [
{
"first": "J",
"middle": [],
"last": "Kupiec",
"suffix": ""
}
],
"year": 1993,
"venue": "bilingual Proceedings of the Twenty-Fourth Conference on Computational Linguistics and Speech Processing (ROCLING 2012) corpora. Proceedings of the 31st annual meeting on association for computational linguisticsAssociation for Computational Linguistics",
"volume": "17",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kupiec J. 1993. An algorithm for finding noun phrase correspondences in bilingual Proceedings of the Twenty-Fourth Conference on Computational Linguistics and Speech Processing (ROCLING 2012) corpora. Proceedings of the 31st annual meeting on association for computational linguisticsAssociation for Computational Linguistics. 17 p.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "NLTK: The natural language toolkit",
"authors": [
{
"first": "E",
"middle": [],
"last": "Loper",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Bird",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the ACL-02 workshop on effective tools and methodologies for teaching natural language processing and computational linguistics-volume 1Association for Computational Linguistics",
"volume": "63",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Loper E. and Bird S. 2002. NLTK: The natural language toolkit. Proceedings of the ACL-02 workshop on effective tools and methodologies for teaching natural language processing and computational linguistics-volume 1Association for Computational Linguistics. 63 p.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Collocation translation acquisition using monolingual corpora",
"authors": [
{
"first": "Y",
"middle": [],
"last": "L\u00fc",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Zhou",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of the 42nd annual meeting on association for computational linguisticsAssociation for Computational Linguistics",
"volume": "167",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "L\u00fc Y. and Zhou M. 2004. Collocation translation acquisition using monolingual corpora. Proceedings of the 42nd annual meeting on association for computational linguisticsAssociation for Computational Linguistics. 167 p.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "A phrase-based, joint probability model for statistical machine translation",
"authors": [
{
"first": "D",
"middle": [],
"last": "Marcu",
"suffix": ""
},
{
"first": "W",
"middle": [],
"last": "Wong",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the ACL-02 conference on empirical methods in natural language processing",
"volume": "10",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marcu D. and Wong W. 2002. A phrase-based, joint probability model for statistical machine translation. In Proceedings of the ACL-02 conference on empirical methods in natural language processing-volume 10. Association for Computational Linguistics. 133 p.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Extracting bilingual collocations from non-aligned parallel corpora",
"authors": [
{
"first": "K",
"middle": [],
"last": "Ohmori",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Higashida",
"suffix": ""
}
],
"year": 1999,
"venue": "Proeeding. of the 8th international conference on theoretical and methodological issues in machine translation (TMI99). Citeseer. 88 p",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ohmori K. and Higashida M. 1999. Extracting bilingual collocations from non-aligned parallel corpora. In Proeeding. of the 8th international conference on theoretical and methodological issues in machine translation (TMI99). Citeseer. 88 p.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Oxford collocations dictionary 2nd",
"authors": [],
"year": 2009,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Oxford University Press. 2009. Oxford collocations dictionary 2nd . USA: Oxford University Press.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "BLEU: A method for automatic evaluation of machine translation",
"authors": [
{
"first": "K",
"middle": [],
"last": "Papineni",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Roukos",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Ward",
"suffix": ""
},
{
"first": "W",
"middle": [
"J"
],
"last": "Zhu",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the 40th annual meeting on association for computational linguistics",
"volume": "311",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Papineni K., Roukos S., Ward T. and Zhu W. J. 2002. BLEU: A method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting on association for computational linguistics, Association for Computational Linguistics. 311 p.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "The Measurement of Observer Agreement for Categorical Data",
"authors": [
{
"first": "J",
"middle": [],
"last": "",
"suffix": ""
},
{
"first": "Richard",
"middle": [],
"last": "Landis",
"suffix": ""
},
{
"first": "Gary",
"middle": [
"G"
],
"last": "Koch",
"suffix": ""
}
],
"year": 1977,
"venue": "Biometrics",
"volume": "33",
"issue": "1",
"pages": "159--174",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Richard Landis and Gary G. Koch. 1977. The Measurement of Observer Agreement for Categorical Data. Biometrics, 33(1):159-174. International Biometric Society",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Collocation translation based on sentence alignment and parsing. Actes de la 14e conf\u00e9rence sur le traitement automatique des langues naturelles",
"authors": [
{
"first": "V",
"middle": [],
"last": "Seretan",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Wehrli",
"suffix": ""
}
],
"year": 2007,
"venue": "Citeseer",
"volume": "401",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Seretan V. and Wehrli E. 2007. Collocation translation based on sentence alignment and parsing. Actes de la 14e conf\u00e9rence sur le traitement automatique des langues naturelles (TALN 2007). Citeseer. 401 p.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Translating collocations for bilingual lexicons: A statistical approach",
"authors": [
{
"first": "F",
"middle": [],
"last": "Smadja",
"suffix": ""
},
{
"first": "K",
"middle": [
"R"
],
"last": "Mckeown",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Hatzivassiloglou",
"suffix": ""
}
],
"year": 1996,
"venue": "Computational Linguistics",
"volume": "22",
"issue": "1",
"pages": "1--38",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Smadja F, McKeown KR, Hatzivassiloglou V. 1996. Translating collocations for bilingual lexicons: A statistical approach. Computational Linguistics 22(1):1-38.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Improving translation selection with a new translation model trained by independent monolingual corpora",
"authors": [
{
"first": "M",
"middle": [],
"last": "Zhou",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Ding",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Huang",
"suffix": ""
}
],
"year": 2001,
"venue": "Computational Linguistics and Chinese Language Processing",
"volume": "6",
"issue": "1",
"pages": "1--26",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Zhou M, Ding Y, Huang C. 2001. Improving translation selection with a new translation model trained by independent monolingual corpora. Computational Linguistics and Chinese Language Processing 6(1):1-26.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"num": null,
"text": "Submitting a English Collocation \"delegate authority\" to Google Translate",
"uris": null,
"type_str": "figure"
},
"FIGREF1": {
"num": null,
"text": "An example of translating \"delegate authority\"",
"uris": null,
"type_str": "figure"
},
"FIGREF2": {
"num": null,
"text": "t_comb , N t_comb ) m } from PT, and another set of aligned collocation translations CT align",
"uris": null,
"type_str": "figure"
},
"FIGREF3": {
"num": null,
"text": "Generate word alignment from parallel corpus PC. (Section 3.2.1) (2) Build the combination translation model. (Section 3.2.2) (3) Build the alignment translation model from word alignment. (Section 3.2.3) Outline of the training process 3.2.1",
"uris": null,
"type_str": "figure"
},
"FIGREF4": {
"num": null,
"text": "An example of phrase table from English to Chinese Take the phrase table in",
"uris": null,
"type_str": "figure"
},
"FIGREF5": {
"num": null,
"text": "procedure BuildBidirectionalModel(PC, Align_StoT, Align_TtoS) (1) LemmatizePC = Lemmatize( PC ) for each src_sentence i , tgt_sentence i in fulfill their functions for each src_word j in src_sentece i(2) SrcTrans [j] = FindIntersection( src_word j , Align_StoT, Align_TtoS ) for each src_word j in src_sentece i (3a) SkipBigramList = GenerateSkipBgram1toN (src_word j , src_word j+N ) The algorithm of building bidirectional alignment translation modelIn Step (1) of the algorithm, we first lemmatize all source sentences to generate the lemmatized parallel corpus LemmatizePC.",
"uris": null,
"type_str": "figure"
},
"FIGREF6": {
"num": null,
"text": "procedure TranslatingCollocation ( C, CombTM, BiTM ) (1a) Base, Collocate = DecomposeCollocation(C) (1b) BaseTransList = GenerateCombBaseTranslation(Base) (1c) CollocateTransList = GenerateCombCollocateTranslation(Collocate) (1d) CombTransList = for each bTrans in BaseTransList, for each cTrans in CollocateTransList (2a) Score = CalculateCombTM_Score(cTrans, bTrans) (2b) CombTrans = (bTrans, cTrans) (2c) CombTransList += (CombTrans, Score) (3) Sort CombTransList in decreasing order (4a) BiTransList = GenerateListOfBiTransWithScore( C ) (4b) Sort BiTransList in decreasing (5) RankedCandidates = Rank( CombTransList, BiTransList) (6) Return top N RankedCandidates Figure 6. Generation and Ranking Procedure at run time 3.3.",
"uris": null,
"type_str": "figure"
},
"FIGREF7": {
"num": null,
"text": "Nature Language Toolkit (NLTK), a suite of open source modules written in Python. Chinese sentences of HKPT were word-segmented by the CKIP Chinese word segmentation system (Ma and Chen, 2003). To obtain word alignment information of English and Chinese sentences, we used GIZA++ (Och and Ney, 2003) as the word alignment tool.",
"uris": null,
"type_str": "figure"
},
"FIGREF8": {
"num": null,
"text": "Combination Translation Model (CTM): The system based on translating the base word and the collocate separately and then combined them to generate candidates. The candidates are filtered by the target language model as output. -Bidirectional Alignment Translation Model (BTM): The system extracts translation based on the bidirectional alignment information of a word-aligned parallel corpus using GIZA++. -Hybrid Translation Model (HYBRID): Our system based on both CTM and BTM by combining the results of each model with the translation ranking scheme as described in Section 3.3.",
"uris": null,
"type_str": "figure"
},
"TABREF0": {
"type_str": "table",
"num": null,
"text": "An example of the word translation model",
"content": "<table><tr><td>Word</td><td>Translation</td><td>Bidirectional Probability</td></tr><tr><td>authority</td><td>.%</td><td>-1.454484</td></tr><tr><td/><td>*+</td><td>-3.747178</td></tr><tr><td/><td>.%</td><td>-5.029688</td></tr><tr><td/><td>*+</td><td>-5.439792</td></tr><tr><td/><td>.</td><td>-5.586563</td></tr><tr><td/><td/><td>-5.652645</td></tr><tr><td/><td>.% 9</td><td>-5.814680</td></tr><tr><td/><td/><td>-5.833058</td></tr></table>",
"html": null
},
"TABREF1": {
"type_str": "table",
"num": null,
"text": "An example of bidirectional alignment translation model",
"content": "<table><tr><td>Collocation</td><td>Translation</td><td>Frequency</td></tr><tr><td/><td>( &</td><td>1193</td></tr><tr><td/><td># 52</td><td>612</td></tr><tr><td>play role</td><td>' 52</td><td>475</td></tr><tr><td/><td>52</td><td>46</td></tr><tr><td/><td>( 1</td><td>37</td></tr></table>",
"html": null
},
"TABREF2": {
"type_str": "table",
"num": null,
"text": "Parameter used in training",
"content": "<table><tr><td>Parameter</td><td>Value</td><td>Description</td><td/><td/></tr><tr><td>minBidiretionProb</td><td>-15.0</td><td>Minimum</td><td>bidirectional</td><td>translation</td></tr><tr><td colspan=\"5\">probability of numWordTrans 100 Number of the base word and the</td></tr><tr><td/><td/><td colspan=\"3\">collocate translations used to generate</td></tr><tr><td/><td/><td colspan=\"2\">collocation translations.</td><td/></tr></table>",
"html": null
},
"TABREF3": {
"type_str": "table",
"num": null,
"text": "Top-N precision of different systems MRR value for all translations for collocations in test data by seeing \"correct",
"content": "<table><tr><td/><td>Top-1</td><td>Top-2</td><td>Top-3</td><td>Top-4</td><td>Top-5</td></tr><tr><td>Moses (baseline)</td><td>.55</td><td>.73</td><td>.77</td><td>.78</td><td>.82</td></tr><tr><td>BTM</td><td>.49</td><td>.56</td><td>.59</td><td>.59</td><td>.60</td></tr><tr><td>CTM</td><td>.67</td><td>.75</td><td>.80</td><td>.82</td><td>.83</td></tr><tr><td>Hybrid (CTM+BTM)</td><td>.65</td><td>.81</td><td>.85</td><td>.88</td><td>.89</td></tr><tr><td>System</td><td/><td/><td>MRR</td><td/><td/></tr><tr><td colspan=\"2\">Moses (baseline)</td><td/><td>.72</td><td/><td/></tr><tr><td>BTM</td><td/><td/><td>.55</td><td/><td/></tr></table>",
"html": null
}
}
}
} |