File size: 87,271 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 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 |
{
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:12:56.910609Z"
},
"title": "The Highs and Lows of Simple Lexical Domain Adaptation Approaches for Neural Machine Translation",
"authors": [
{
"first": "Nikolay",
"middle": [],
"last": "Bogoychev",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Edinburgh",
"location": {}
},
"email": "n.bogoych@ed.ac.uk"
},
{
"first": "Pinzhen",
"middle": [],
"last": "Chen",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Edinburgh",
"location": {}
},
"email": "pinzhen.chen@ed.ac.uk"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Machine translation systems are vulnerable to domain mismatch, especially in a lowresource scenario. Out-of-domain translations are often of poor quality and prone to hallucinations, due to exposure bias and the decoder acting as a language model. We adopt two approaches to alleviate this problem: lexical shortlisting restricted by IBM statistical alignments, and hypothesis re-ranking based on similarity. The methods are computationally cheap, widely known, but not extensively experimented on domain adaptation. We demonstrate success on low-resource out-of-domain test sets, however, the methods are ineffective when there is sufficient data or too great domain mismatch. This is due to both the IBM model losing its advantage over the implicitly learned neural alignment, and issues with subword segmentation of out-of-domain words.",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "Machine translation systems are vulnerable to domain mismatch, especially in a lowresource scenario. Out-of-domain translations are often of poor quality and prone to hallucinations, due to exposure bias and the decoder acting as a language model. We adopt two approaches to alleviate this problem: lexical shortlisting restricted by IBM statistical alignments, and hypothesis re-ranking based on similarity. The methods are computationally cheap, widely known, but not extensively experimented on domain adaptation. We demonstrate success on low-resource out-of-domain test sets, however, the methods are ineffective when there is sufficient data or too great domain mismatch. This is due to both the IBM model losing its advantage over the implicitly learned neural alignment, and issues with subword segmentation of out-of-domain words.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Neural Machine translation (NMT) has achieved state-of-the-art performance in a variety of language pairs and settings (Bahdanau et al., 2015; Vaswani et al., 2017) , but it is vulnerable to domain mismatch, where the test set differs significantly from the training data in terms of vocabulary, genre, length, etc. This issue is exacerbated in a lowresource condition (Koehn and Knowles, 2017) .",
"cite_spans": [
{
"start": 119,
"end": 142,
"text": "(Bahdanau et al., 2015;",
"ref_id": "BIBREF1"
},
{
"start": 143,
"end": 164,
"text": "Vaswani et al., 2017)",
"ref_id": "BIBREF22"
},
{
"start": 369,
"end": 394,
"text": "(Koehn and Knowles, 2017)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Teacher forcing is used during traditional maximum likelihood neural network training, leading to a strong exposure bias, and model confusion when presented with unexpected sequences. This typically results in hallucinations in the output (M\u00fcller et al., 2020) , because the overly zealous language model component prefers a fluent translation, as opposed to an adequate one. A number of methods have been proposed in order to tackle the issue: exposing the model to its predictions during training (Ranzato et al., 2015; Shen et al., 2016; Zhang et al., * Equal contribution. 2019; Wang and Sennrich, 2020) ; tuning directly towards BLEU (Wiseman and Rush, 2016) or using minimum Bayes risk decoding (Kumar and Byrne, 2004; Stahlberg et al., 2017) . A common weakness of such methods is that they are computationally expensive.",
"cite_spans": [
{
"start": 239,
"end": 260,
"text": "(M\u00fcller et al., 2020)",
"ref_id": "BIBREF16"
},
{
"start": 499,
"end": 521,
"text": "(Ranzato et al., 2015;",
"ref_id": "BIBREF17"
},
{
"start": 522,
"end": 540,
"text": "Shen et al., 2016;",
"ref_id": "BIBREF20"
},
{
"start": 541,
"end": 576,
"text": "Zhang et al., * Equal contribution.",
"ref_id": null
},
{
"start": 583,
"end": 607,
"text": "Wang and Sennrich, 2020)",
"ref_id": "BIBREF23"
},
{
"start": 639,
"end": 663,
"text": "(Wiseman and Rush, 2016)",
"ref_id": "BIBREF24"
},
{
"start": 701,
"end": 724,
"text": "(Kumar and Byrne, 2004;",
"ref_id": "BIBREF12"
},
{
"start": 725,
"end": 748,
"text": "Stahlberg et al., 2017)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper, we adopt and experiment with two approaches inspired by previous research. We use lexical shortlisting to interpolate a statistical alignment model with NMT; on top of it, we perform n-best list re-ranking by hypothesis agreement. Our aim is to constrain the lexical choice of the decoder, to prevent hallucinations from being generated. The methods are computationally simpler, as they require no change to the model or training. We analyse the effectiveness of these methods in different scenarios of domain adaptation. We show BLEU gains on a variety of out-of-domain datasets in a low-resource English-German setting. However, the methods show no improvements once the datasets are large, or the domains are too distant.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We purse and analyse two separate strategies for improving neural machine translation's performance in low-resource domain mismatched settings: lexical shortlisting and n-best list re-ranking based on inter hypothesis agreement.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Methodology",
"sec_num": "2"
},
{
"text": "Neural machine translation systems have a vocabulary size of tens of thousands, but for every single translation, most of the vocabulary items are improbable choices. Many researchers attempt to improve on this to speed up the computation in the output layer (Schwenk et al., 2007; Le et al., 2012; Devlin et al., 2014) . This is done by preparing a list of likely word-level translations for each sentence (commonly known as lexical shortlist), by using IBM alignment models such as fast-align (Dyer et al., 2013) and limiting the output layer choices to it. This has been implemented in NMT frameworks for efficient systems (Junczys-Dowmunt et al., 2018b) and since widely used (Bogoychev et al., 2020, inter alia) .",
"cite_spans": [
{
"start": 259,
"end": 281,
"text": "(Schwenk et al., 2007;",
"ref_id": "BIBREF18"
},
{
"start": 282,
"end": 298,
"text": "Le et al., 2012;",
"ref_id": "BIBREF13"
},
{
"start": 299,
"end": 319,
"text": "Devlin et al., 2014)",
"ref_id": "BIBREF4"
},
{
"start": 495,
"end": 514,
"text": "(Dyer et al., 2013)",
"ref_id": "BIBREF6"
},
{
"start": 626,
"end": 657,
"text": "(Junczys-Dowmunt et al., 2018b)",
"ref_id": "BIBREF10"
},
{
"start": 680,
"end": 716,
"text": "(Bogoychev et al., 2020, inter alia)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Lexical shortlisting",
"sec_num": "2.1"
},
{
"text": "Recently, Li et al. (2019) showed that even in a high-resource scenario, the quality of the IBM alignments can outperform those learned by a neural model. Therefore, we find it sensible to incorporate word alignments into neural models, especially in a low-resource situation. Although shortlists have a negligible impact on the quality of strong neural systems, we can successfully limit the output layer to the likely tokens according to an IBM model trained on a particular domain, and improve out-of-domain BLEU scores in a low-resource setting. This suggests that the IBM model provides complementary information to the neural model in this scenario.",
"cite_spans": [
{
"start": 10,
"end": 26,
"text": "Li et al. (2019)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Lexical shortlisting",
"sec_num": "2.1"
},
{
"text": "Fomicheva et al. (2020) estimate the quality of machine translation by measuring the agreement of the generated hypotheses for a given source sentence. The reason is that (higher) similarity between hypotheses reflects (higher) model confidence. In our problem, we assume that a hallucinated hypothesis will have a low agreement with the rest in the beam. Thus, we propose to re-rank the n-best list based on inter-hypothesis similarity and select the top one as the final translation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hypothesis re-ranking",
"sec_num": "2.2"
},
{
"text": "For every source sentence, NMT generates b hypotheses where b is the beam size. For each hypothesis, we measure its similarity against others in the beam, then sort all hypotheses by their final aggregated similarity scores. The similarity score score i for a certain hypothesis hyp i is calculated as Equation 1, where the similarity between hyp i and every other hyp j produced by the model is measured by an automatic metric similarity:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hypothesis re-ranking",
"sec_num": "2.2"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "score i = b j=0,i =j similarity(hyp i , hyp j )",
"eq_num": "(1)"
}
],
"section": "Hypothesis re-ranking",
"sec_num": "2.2"
},
{
"text": "3 Experimental setup",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hypothesis re-ranking",
"sec_num": "2.2"
},
{
"text": "For our experiments, we use OPUS English-German data (Lison and Tiedemann, 2016) , with preprocessing performed as per the work of M\u00fcller et al. (2020) . We follow the same data split as Wang and Sennrich (2020)'s low-resource, domain restricted setting: a model is trained on 1M sentences pairs from the medical domain, and is evaluated on medical (in-domain), Koran, IT, subtitles and law (out-of-domain). We use a joint vocabulary byte-pair encoding (BPE, Sennrich et al., 2016) trained on the medical domain with 32k merge operations. In this way, the vocabulary trained on the medical domain may be sub-optimal for the out-ofdomain test sets. In reality, monolingual data from unknown domains might be available, which can reduce the bias of the BPE training towards the indomain data. In order to simulate this scenario, we perform an analogous experiment where the BPE vocabulary is trained on all domains, except the subtitles corpus, as it is much larger than other domains and would dominate the vocabulary (reasons elaborated in Section 5). For training, we used the Transformer-base preset of the Marian toolkit (Junczys-Dowmunt et al., 2018a) with transformer preprocessing normalisation as opposed to postprocessing normalisation, and additional attention and feed-forward layer dropout.",
"cite_spans": [
{
"start": 53,
"end": 80,
"text": "(Lison and Tiedemann, 2016)",
"ref_id": "BIBREF15"
},
{
"start": 131,
"end": 151,
"text": "M\u00fcller et al. (2020)",
"ref_id": "BIBREF16"
},
{
"start": 453,
"end": 481,
"text": "(BPE, Sennrich et al., 2016)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Hypothesis re-ranking",
"sec_num": "2.2"
},
{
"text": "We performed ample hyperparameter search, to ensure that our neural network configuration achieves the best possible performance on this lowresource task. Decoding is always done with beam size 6 and length normalisation 0.6. On top of the baseline, we tried three different combinations of methods introduced in Section 2:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hypothesis re-ranking",
"sec_num": "2.2"
},
{
"text": "\u2022 A shortlisting configuration where we use a lexical shortlist generated by the fast-align model trained on the medical dataset. We try various configurations, and we settle on an optimal value of limiting the output layer to the 10 most probable unigram translations according to the IBM model.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hypothesis re-ranking",
"sec_num": "2.2"
},
{
"text": "\u2022 A re-ranking setup where we re-rank the nbest translation based on inter-hypothesis similarity. Like Fomicheva et al., we tested out a few different metrics to measure similarity: sentBLEU, ChrF, TER and METEOR. In an initial experiment on the medical domain, we found sentBLEU to have the best performance and stick to it for all experiments. We did not pick any neural metrics, as they may have a domain preference and are less interpretable.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hypothesis re-ranking",
"sec_num": "2.2"
},
{
"text": "\u2022 Both: shortlisting followed by re-ranking as specified above ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hypothesis re-ranking",
"sec_num": "2.2"
},
{
"text": "We present in Table 1 our models' BLEU scores on in-and out-of-domain test sets, with both BPE segmentation schemes. When applying the more restrictive BPE trained on the medical domain, our shortlisting mechanism always yields a slight increase in BLEU on out-of-domain sets and a small drop on the in-domain set. The law and IT domains benefit the most, whereas subtitles and Koran are largely unaffected. Re-ranking is not helpful in this BPE setting. When using the alternative BPE segmentation trained on all datasets, we see that the baseline scores are generally lower for the out-of-domain datasets, despite a stronger in-domain BLEU. This is potentially because the rare words from out-ofdomain datasets get insufficient exposure during training (on medical only). However, when using a lexical shortlist in this setting, we see greater improvements in terms of BLEU on the law and IT domains compared to the medical-only BPE scenario. Re-ranking is also much more effective, performing similarly to shortlisting. The combination of re-ranking and shortlisting delivers the best BLEU scores in nearly all out-of-domain splits.",
"cite_spans": [],
"ref_spans": [
{
"start": 14,
"end": 21,
"text": "Table 1",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Results and Analysis",
"sec_num": "4"
},
{
"text": "We see that the shortlisting method is always superior to the baseline method on all out-of-domain datasets, although the results vary with the data preprocessing. This is true even more so of reranking, which is much better when the BPE vocabulary is learned on all domains. Combining both shortlisting and re-ranking always brings in a slight improvement over just re-ranking on the outof-domain datasets, but its effectiveness is again preprocessing-dependent.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Analysis",
"sec_num": "4"
},
{
"text": "This clearly shows that the IBM model implemented by fast-align can learn information complementary to NMT, and interpolating them is beneficial for achieving higher BLEU scores in out-ofdomain settings. The lower BLEU scores on indomain test sets are related to the aggressive shortlist we used: by increasing the output layer limit from 10 to 50 most probable tokens, we achieve identical BLEU scores as the baseline.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Analysis",
"sec_num": "4"
},
{
"text": "According to BLEU scores reported earlier, shortlisting and re-ranking are beneficial to NMT domain adaptation (in a relatively low-resource condition). We try to understand what contributes to the increase in BLEU scores by breaking down BLEU scores into n-gram precisions and length (brevity) penalty. In Table 2 we list the numbers for law and IT domains, under the \"BPE trained on all\" setting in ",
"cite_spans": [],
"ref_spans": [
{
"start": 307,
"end": 314,
"text": "Table 2",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "BLEU breakdown",
"sec_num": "4.1"
},
{
"text": "sein Pilot hat nicht die volle Kontrolle . its p@@ il@@ ot is@@ n't in control .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "German English",
"sec_num": null
},
{
"text": "und Z@@ eth@@ rid ? nur einen Strei@@ f@@ sch@@ uss . and , Z@@ eth@@ rid , just gr@@ aze it . of BLEU scores. Additionally, we include ME-TEOR which focuses on n-gram overlap and is not influenced by the output length. From the table, shortlisting always significantly boosts n-gram precisions, whereas re-ranking alone decreases them. On the other hand, re-ranking \"rectifies\" the output length, leading to a better brevity penalty (closer to 1) compared to baseline or shortlisting. When using both together, we see slight improvement on both n-gram accuracies and length penalty over baseline. This is expected: shortlisting provides extra word alignment information which aids lexical accuracy; re-ranking favours the hypothesis with an average length, since too long or too short hypotheses will receive a lower similarity score. This implies that shortlisting and re-ranking enhance BLEU from different aspects.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "German English",
"sec_num": null
},
{
"text": "On the contrary, according to METEOR, reranking is outperformed even by the baseline, leading to a negative METEOR change. This suggests that the bump in BLEU scores by re-ranking is only due to an improved brevity penalty. While it encourages a more desirable length, re-ranking does not produce better lexical choices. Adequacy wise, shortlisting is proven to be a more constructive method.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "German English",
"sec_num": null
},
{
"text": "Although our methods show promising gains on the low-resource domains, we found that they have limited application when the domain mismatch is too great or there is sufficient resource available.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Limitations",
"sec_num": "5"
},
{
"text": "In order to better interpret why our methods are much more helpful on some domains than on others (e.g. law versus Koran), we gather statistics of our test sets in different domains in Table 3 , to reflect the distance between domains. We compute vocabulary overlap with the in-domain medical data, for each cleaned corpus prior to BPE encoding. We count only words that are seen at least 20 times, and we sample 5% of the subtitles corpus, since it is orders of magnitude larger than the rest. After down-sampling, it has a similar size to the medical corpora. We also compute the average sentence length before and after BPE to determine the extent to which BPE transforms the original sentences. Both vocabulary overlap and sentence lengths implicitly reflect the degree of domain mismatch.",
"cite_spans": [],
"ref_spans": [
{
"start": 185,
"end": 192,
"text": "Table 3",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Large domain mismatch",
"sec_num": "5.1"
},
{
"text": "For our most difficult datasets, subtitles and Koran, we see the lowest vocabulary overlap with the medical domain of just 22-25%. When BPE is applied to these corpora, the sentence length increases by 20-30%. In practice, extensive and uneven BPE segmentation on named entities makes it difficult for the IBM model to produce interpretable and meaningful alignments to aid a translation model, as exemplified in Table 4 .",
"cite_spans": [],
"ref_spans": [
{
"start": 413,
"end": 420,
"text": "Table 4",
"ref_id": "TABREF4"
}
],
"eq_spans": [],
"section": "Large domain mismatch",
"sec_num": "5.1"
},
{
"text": "This means that lexical shortlisting is useful when we have a model trained on a relatively modest amount of data, and the out-of-domain dataset which we adapt to should share a reasonable amount of vocabulary. When the vocabulary overlap is too small, shortlisting drastically loses its effectiveness as a domain adaptation tool.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Large domain mismatch",
"sec_num": "5.1"
},
{
"text": "We confirm this by experimenting with another domain-distant, and extremely low-resource Burmese-English scenario. The training data consists 18k sentence pairs from news articles (Ding et al., 2019) , and the out-of-domain test set is a Bible corpus (Christodouloupoulos and Steedman, 2015) . The English side of the training set has 51k unique words, while the Bible set has 29k. The vocabulary overlap between the training and test sets is just 6k words, equivalent to 21% of the test vocabulary. After BPE, The average test sentence length increases from 25 to 38 tokens, seeing a 52% rise. We adopt the model trained by Aji et al. (2020) and add shortlisting during decoding. The results displayed in Table 5 show that out-of-domain performance does not improve with shortlisting, likely due to the high degree of vocabulary mismatch. baseline shortlist news (in-domain) 18.00 15.7 Bible 0.2 0.2 Table 5 : Very low-resource Burmese-English results.",
"cite_spans": [
{
"start": 180,
"end": 199,
"text": "(Ding et al., 2019)",
"ref_id": "BIBREF5"
},
{
"start": 251,
"end": 291,
"text": "(Christodouloupoulos and Steedman, 2015)",
"ref_id": "BIBREF3"
},
{
"start": 625,
"end": 642,
"text": "Aji et al. (2020)",
"ref_id": "BIBREF0"
},
{
"start": 864,
"end": 875,
"text": "(in-domain)",
"ref_id": null
}
],
"ref_spans": [
{
"start": 706,
"end": 713,
"text": "Table 5",
"ref_id": null
},
{
"start": 901,
"end": 908,
"text": "Table 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "Large domain mismatch",
"sec_num": "5.1"
},
{
"text": "Our previous German-English experiments are carried out under an artificially resource-constrained condition. We therefore verify the potential of our methods in a high-resource setting too, by applying shortlisting on a WMT19 German-English submission from Microsoft (Junczys-Dowmunt, 2019), and evaluating on the same out-of-domain datasets. Results in Table 6 show that shortlisting has no impact on BLEU comparing to the baseline. We conclude that given a high-resource setting and the apparent large domain mismatch, the IBM model's alignments do not contribute additional information to the model. This finding is corroborated by Li et al. (2019) : IBM model alignments are mostly better at capturing function words, not content words, compared to a neural model. Furthermore, unnatural and aggressive BPE segmentation on out-of-domain text (e.g. Table 4 ) could result in a lexical shortlist not capturing any meaningful alignment.",
"cite_spans": [
{
"start": 636,
"end": 652,
"text": "Li et al. (2019)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [
{
"start": 355,
"end": 362,
"text": "Table 6",
"ref_id": "TABREF6"
},
{
"start": 853,
"end": 860,
"text": "Table 4",
"ref_id": "TABREF4"
}
],
"eq_spans": [],
"section": "Availability of resources",
"sec_num": "5.2"
},
{
"text": "We add our low-resource baseline trained on medical with BPE on all data (Section 4 Table 1) to comparison, and find that it surpasses the huge WMT model on each domain. Two reasons account for this: a WMT model is heavily biased to the news domain only; and the BPE scheme learned from news data is inferior to one learned from the out-of-domain datasets, when being evaluated on these domains.",
"cite_spans": [],
"ref_spans": [
{
"start": 82,
"end": 92,
"text": "4 Table 1)",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Availability of resources",
"sec_num": "5.2"
},
{
"text": "We explore computationally cheap methods to improve neural machine translation performance in out-of-domain settings. We suggest that adding a lexical shortlist trained on the same data is always beneficial. While re-ranking also improves BLEU, it targets the BLEU brevity penalty, and does not produce better word choices. Although our results are promising in a low-resource condition, they do not transfer well to a scenario with very distant domains or sufficient resources. Our analysis shows that this is due to little vocabulary overlap, and the limited contribution from the IBM model under out-of-domain BPE segmentation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
}
],
"back_matter": [
{
"text": "We would like to thank the anonymous reviewers, as well as the members of the AGORA research group for their valuable comments.This work was conducted within the scope of the Connecting Europe Facility User-Focused Marian, which has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No INEA/CEF/ICT/A2019/1927024. Any communication or publication related to the action, made by the beneficiaries jointly or individually in any form and using any means, shall indicate that it reflects only the author's view and that the Agency is not responsible for any use that may be made of the information it contains.This research is based upon work supported in part by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via contract #FA8650-17-C-9117. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of ODNI, IARPA, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright annotation therein.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "In neural machine translation, what does transfer learning transfer?",
"authors": [
{
"first": "Alham",
"middle": [],
"last": "Fikri Aji",
"suffix": ""
},
{
"first": "Nikolay",
"middle": [],
"last": "Bogoychev",
"suffix": ""
},
{
"first": "Kenneth",
"middle": [],
"last": "Heafield",
"suffix": ""
},
{
"first": "Rico",
"middle": [],
"last": "Sennrich",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "7701--7710",
"other_ids": {
"DOI": [
"10.18653/v1/2020.acl-main.688"
]
},
"num": null,
"urls": [],
"raw_text": "Alham Fikri Aji, Nikolay Bogoychev, Kenneth Heafield, and Rico Sennrich. 2020. In neural ma- chine translation, what does transfer learning trans- fer? In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7701-7710, Online. Association for Compu- tational Linguistics.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Neural machine translation by jointly learning to align and translate",
"authors": [
{
"first": "Dzmitry",
"middle": [],
"last": "Bahdanau",
"suffix": ""
},
{
"first": "Kyunghyun",
"middle": [],
"last": "Cho",
"suffix": ""
},
{
"first": "Yoshua",
"middle": [],
"last": "Bengio",
"suffix": ""
}
],
"year": 2015,
"venue": "3rd International Conference on Learning Representations",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Ben- gio. 2015. Neural machine translation by jointly learning to align and translate. In 3rd Inter- national Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Emmanouil-Ioannis Farsarakis, and Mateusz Chudyk. 2020. Edinburgh's submissions to the 2020 machine translation efficiency task",
"authors": [
{
"first": "Nikolay",
"middle": [],
"last": "Bogoychev",
"suffix": ""
},
{
"first": "Roman",
"middle": [],
"last": "Grundkiewicz",
"suffix": ""
},
{
"first": "Alham",
"middle": [],
"last": "Fikri Aji",
"suffix": ""
},
{
"first": "Maximiliana",
"middle": [],
"last": "Behnke",
"suffix": ""
},
{
"first": "Kenneth",
"middle": [],
"last": "Heafield",
"suffix": ""
},
{
"first": "Sidharth",
"middle": [],
"last": "Kashyap",
"suffix": ""
}
],
"year": null,
"venue": "Proceedings of the Fourth Workshop on Neural Generation and Translation",
"volume": "",
"issue": "",
"pages": "218--224",
"other_ids": {
"DOI": [
"10.18653/v1/2020.ngt-1.26"
]
},
"num": null,
"urls": [],
"raw_text": "Nikolay Bogoychev, Roman Grundkiewicz, Al- ham Fikri Aji, Maximiliana Behnke, Kenneth Heafield, Sidharth Kashyap, Emmanouil-Ioannis Farsarakis, and Mateusz Chudyk. 2020. Ed- inburgh's submissions to the 2020 machine translation efficiency task. In Proceedings of the Fourth Workshop on Neural Generation and Translation, pages 218-224, Online. Association for Computational Linguistics.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "A massively parallel corpus: The bible in 100 languages",
"authors": [
{
"first": "Christos",
"middle": [],
"last": "Christodouloupoulos",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Steedman",
"suffix": ""
}
],
"year": 2015,
"venue": "Lang. Resour. Eval",
"volume": "49",
"issue": "2",
"pages": "375--395",
"other_ids": {
"DOI": [
"10.1007/s10579-014-9287-y"
]
},
"num": null,
"urls": [],
"raw_text": "Christos Christodouloupoulos and Mark Steedman. 2015. A massively parallel corpus: The bible in 100 languages. Lang. Resour. Eval., 49(2):375-395.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Fast and robust neural network joint models for statistical machine translation",
"authors": [
{
"first": "Jacob",
"middle": [],
"last": "Devlin",
"suffix": ""
},
{
"first": "Rabih",
"middle": [],
"last": "Zbib",
"suffix": ""
},
{
"first": "Zhongqiang",
"middle": [],
"last": "Huang",
"suffix": ""
},
{
"first": "Thomas",
"middle": [],
"last": "Lamar",
"suffix": ""
},
{
"first": "Richard",
"middle": [],
"last": "Schwartz",
"suffix": ""
},
{
"first": "John",
"middle": [],
"last": "Makhoul",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "1370--1380",
"other_ids": {
"DOI": [
"10.3115/v1/P14-1129"
]
},
"num": null,
"urls": [],
"raw_text": "Jacob Devlin, Rabih Zbib, Zhongqiang Huang, Thomas Lamar, Richard Schwartz, and John Makhoul. 2014. Fast and robust neural network joint models for sta- tistical machine translation. In Proceedings of the 52nd Annual Meeting of the Association for Compu- tational Linguistics (Volume 1: Long Papers), pages 1370-1380, Baltimore, Maryland. Association for Computational Linguistics.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Towards Burmese (Myanmar) morphological analysis: Syllable-based tokenization and part-of-speech tagging",
"authors": [
{
"first": "Chenchen",
"middle": [],
"last": "Ding",
"suffix": ""
},
{
"first": "Hnin",
"middle": [],
"last": "Thu Zar Aye",
"suffix": ""
},
{
"first": "Win",
"middle": [
"Pa"
],
"last": "Pa",
"suffix": ""
},
{
"first": "Khin",
"middle": [
"Thandar"
],
"last": "Nwet",
"suffix": ""
},
{
"first": "Khin",
"middle": [],
"last": "Mar Soe",
"suffix": ""
},
{
"first": "Masao",
"middle": [],
"last": "Utiyama",
"suffix": ""
},
{
"first": "Eiichiro",
"middle": [],
"last": "Sumita",
"suffix": ""
}
],
"year": 2019,
"venue": "ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)",
"volume": "19",
"issue": "1",
"pages": "",
"other_ids": {
"DOI": [
"https://dl.acm.org/doi/pdf/10.1145/3325885"
]
},
"num": null,
"urls": [],
"raw_text": "Chenchen Ding, Hnin Thu Zar Aye, Win Pa Pa, Khin Thandar Nwet, Khin Mar Soe, Masao Utiyama, and Eiichiro Sumita. 2019. Towards Burmese (Myan- mar) morphological analysis: Syllable-based tok- enization and part-of-speech tagging. ACM Trans- actions on Asian and Low-Resource Language Infor- mation Processing (TALLIP), 19(1):5.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "A simple, fast, and effective reparameterization of IBM model 2",
"authors": [
{
"first": "Chris",
"middle": [],
"last": "Dyer",
"suffix": ""
},
{
"first": "Victor",
"middle": [],
"last": "Chahuneau",
"suffix": ""
},
{
"first": "Noah",
"middle": [
"A"
],
"last": "Smith",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
"volume": "",
"issue": "",
"pages": "644--648",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Chris Dyer, Victor Chahuneau, and Noah A. Smith. 2013. A simple, fast, and effective reparameter- ization of IBM model 2. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Hu- man Language Technologies, pages 644-648, At- lanta, Georgia. Association for Computational Lin- guistics.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Multi-hypothesis machine translation evaluation",
"authors": [
{
"first": "Marina",
"middle": [],
"last": "Fomicheva",
"suffix": ""
},
{
"first": "Lucia",
"middle": [],
"last": "Specia",
"suffix": ""
},
{
"first": "Francisco",
"middle": [],
"last": "Guzm\u00e1n",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "1218--1232",
"other_ids": {
"DOI": [
"10.18653/v1/2020.acl-main.113"
]
},
"num": null,
"urls": [],
"raw_text": "Marina Fomicheva, Lucia Specia, and Francisco Guzm\u00e1n. 2020. Multi-hypothesis machine transla- tion evaluation. In Proceedings of the 58th Annual Meeting of the Association for Computational Lin- guistics, pages 1218-1232, Online. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Microsoft translator at wmt 2019: Towards large-scale document-level neural machine translation",
"authors": [
{
"first": "Marcin",
"middle": [],
"last": "Junczys-Dowmunt",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the Fourth Conference on Machine Translation",
"volume": "2",
"issue": "",
"pages": "225--233",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marcin Junczys-Dowmunt. 2019. Microsoft translator at wmt 2019: Towards large-scale document-level neural machine translation. In Proceedings of the Fourth Conference on Machine Translation (Volume 2: Shared Task Papers, Day 1), pages 225-233, Flo- rence, Italy. Association for Computational Linguis- tics.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Marian: Fast neural machine translation in C++",
"authors": [
{
"first": "Marcin",
"middle": [],
"last": "Junczys-Dowmunt",
"suffix": ""
},
{
"first": "Roman",
"middle": [],
"last": "Grundkiewicz",
"suffix": ""
},
{
"first": "Tomasz",
"middle": [],
"last": "Dwojak",
"suffix": ""
},
{
"first": "Hieu",
"middle": [],
"last": "Hoang",
"suffix": ""
},
{
"first": "Kenneth",
"middle": [],
"last": "Heafield",
"suffix": ""
},
{
"first": "Tom",
"middle": [],
"last": "Neckermann",
"suffix": ""
},
{
"first": "Frank",
"middle": [],
"last": "Seide",
"suffix": ""
},
{
"first": "Ulrich",
"middle": [],
"last": "Germann",
"suffix": ""
},
{
"first": "Alham",
"middle": [],
"last": "Fikri Aji",
"suffix": ""
}
],
"year": 2018,
"venue": "Nikolay Bogoychev, Andr\u00e9 F. T. Martins, and Alexandra Birch",
"volume": "",
"issue": "",
"pages": "116--121",
"other_ids": {
"DOI": [
"10.18653/v1/P18-4020"
]
},
"num": null,
"urls": [],
"raw_text": "Marcin Junczys-Dowmunt, Roman Grundkiewicz, Tomasz Dwojak, Hieu Hoang, Kenneth Heafield, Tom Neckermann, Frank Seide, Ulrich Germann, Al- ham Fikri Aji, Nikolay Bogoychev, Andr\u00e9 F. T. Mar- tins, and Alexandra Birch. 2018a. Marian: Fast neural machine translation in C++. In Proceedings of ACL 2018, System Demonstrations, pages 116- 121, Melbourne, Australia. Association for Compu- tational Linguistics.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Marian: Cost-effective high-quality neural machine translation in C++",
"authors": [
{
"first": "Marcin",
"middle": [],
"last": "Junczys-Dowmunt",
"suffix": ""
},
{
"first": "Kenneth",
"middle": [],
"last": "Heafield",
"suffix": ""
},
{
"first": "Hieu",
"middle": [],
"last": "Hoang",
"suffix": ""
},
{
"first": "Roman",
"middle": [],
"last": "Grundkiewicz",
"suffix": ""
},
{
"first": "Anthony",
"middle": [],
"last": "Aue",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 2nd Workshop on Neural Machine Translation and Generation",
"volume": "",
"issue": "",
"pages": "129--135",
"other_ids": {
"DOI": [
"10.18653/v1/W18-2716"
]
},
"num": null,
"urls": [],
"raw_text": "Marcin Junczys-Dowmunt, Kenneth Heafield, Hieu Hoang, Roman Grundkiewicz, and Anthony Aue. 2018b. Marian: Cost-effective high-quality neural machine translation in C++. In Proceedings of the 2nd Workshop on Neural Machine Translation and Generation, pages 129-135, Melbourne, Australia. Association for Computational Linguistics.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Six challenges for neural machine translation",
"authors": [
{
"first": "Philipp",
"middle": [],
"last": "Koehn",
"suffix": ""
},
{
"first": "Rebecca",
"middle": [],
"last": "Knowles",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the First Workshop on Neural Machine Translation",
"volume": "",
"issue": "",
"pages": "28--39",
"other_ids": {
"DOI": [
"10.18653/v1/W17-3204"
]
},
"num": null,
"urls": [],
"raw_text": "Philipp Koehn and Rebecca Knowles. 2017. Six chal- lenges for neural machine translation. In Proceed- ings of the First Workshop on Neural Machine Trans- lation, pages 28-39, Vancouver. Association for Computational Linguistics.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Minimum Bayes-risk decoding for statistical machine translation",
"authors": [
{
"first": "Shankar",
"middle": [],
"last": "Kumar",
"suffix": ""
},
{
"first": "William",
"middle": [],
"last": "Byrne",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of the Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics: HLT-NAACL 2004",
"volume": "",
"issue": "",
"pages": "169--176",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shankar Kumar and William Byrne. 2004. Minimum Bayes-risk decoding for statistical machine transla- tion. In Proceedings of the Human Language Tech- nology Conference of the North American Chapter of the Association for Computational Linguistics: HLT-NAACL 2004, pages 169-176, Boston, Mas- sachusetts, USA. Association for Computational Linguistics.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Continuous space translation models with neural networks",
"authors": [
{
"first": "Alexandre",
"middle": [],
"last": "Hai Son Le",
"suffix": ""
},
{
"first": "Fran\u00e7ois",
"middle": [],
"last": "Allauzen",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Yvon",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
"volume": "",
"issue": "",
"pages": "39--48",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hai Son Le, Alexandre Allauzen, and Fran\u00e7ois Yvon. 2012. Continuous space translation models with neural networks. In Proceedings of the 2012 Con- ference of the North American Chapter of the As- sociation for Computational Linguistics: Human Language Technologies, pages 39-48, Montr\u00e9al, Canada. Association for Computational Linguistics.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "On the word alignment from neural machine translation",
"authors": [
{
"first": "Xintong",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Guanlin",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Lemao",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Max",
"middle": [],
"last": "Meng",
"suffix": ""
},
{
"first": "Shuming",
"middle": [],
"last": "Shi",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "1293--1303",
"other_ids": {
"DOI": [
"10.18653/v1/P19-1124"
]
},
"num": null,
"urls": [],
"raw_text": "Xintong Li, Guanlin Li, Lemao Liu, Max Meng, and Shuming Shi. 2019. On the word alignment from neural machine translation. In Proceedings of the 57th Annual Meeting of the Association for Com- putational Linguistics, pages 1293-1303, Florence, Italy. Association for Computational Linguistics.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "OpenSub-titles2016: Extracting large parallel corpora from movie and TV subtitles",
"authors": [
{
"first": "Pierre",
"middle": [],
"last": "Lison",
"suffix": ""
},
{
"first": "J\u00f6rg",
"middle": [],
"last": "Tiedemann",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)",
"volume": "",
"issue": "",
"pages": "923--929",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Pierre Lison and J\u00f6rg Tiedemann. 2016. OpenSub- titles2016: Extracting large parallel corpora from movie and TV subtitles. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16), pages 923-929, Por- toro\u017e, Slovenia. European Language Resources As- sociation (ELRA).",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Domain robustness in neural machine translation",
"authors": [
{
"first": "Mathias",
"middle": [],
"last": "M\u00fcller",
"suffix": ""
},
{
"first": "Annette",
"middle": [],
"last": "Rios",
"suffix": ""
},
{
"first": "Rico",
"middle": [],
"last": "Sennrich",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 14th Conference of the Association for Machine Translation in the Americas (AMTA 2020)",
"volume": "",
"issue": "",
"pages": "151--164",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mathias M\u00fcller, Annette Rios, and Rico Sennrich. 2020. Domain robustness in neural machine trans- lation. In Proceedings of the 14th Conference of the Association for Machine Translation in the Americas (AMTA 2020), pages 151-164, Virtual. Association for Machine Translation in the Americas.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Sequence level training with recurrent neural networks",
"authors": [
{
"first": "Aurelio",
"middle": [],
"last": "Marc",
"suffix": ""
},
{
"first": "Sumit",
"middle": [],
"last": "Ranzato",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Chopra",
"suffix": ""
},
{
"first": "Wojciech",
"middle": [],
"last": "Auli",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Zaremba",
"suffix": ""
}
],
"year": 2015,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marc'Aurelio Ranzato, Sumit Chopra, Michael Auli, and Wojciech Zaremba. 2015. Sequence level train- ing with recurrent neural networks.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Smooth bilingual n-gram translation",
"authors": [
{
"first": "Holger",
"middle": [],
"last": "Schwenk",
"suffix": ""
},
{
"first": "Marta",
"middle": [
"R"
],
"last": "Costa-Juss\u00e0",
"suffix": ""
},
{
"first": "Jose",
"middle": [
"A R"
],
"last": "Fonollosa",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP-CoNLL)",
"volume": "",
"issue": "",
"pages": "430--438",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Holger Schwenk, Marta R. Costa-juss\u00e0, and Jose A. R. Fonollosa. 2007. Smooth bilingual n-gram trans- lation. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Process- ing and Computational Natural Language Learning (EMNLP-CoNLL), pages 430-438, Prague, Czech Republic. Association for Computational Linguis- tics.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Neural machine translation of rare words with subword units",
"authors": [
{
"first": "Rico",
"middle": [],
"last": "Sennrich",
"suffix": ""
},
{
"first": "Barry",
"middle": [],
"last": "Haddow",
"suffix": ""
},
{
"first": "Alexandra",
"middle": [],
"last": "Birch",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "1715--1725",
"other_ids": {
"DOI": [
"10.18653/v1/P16-1162"
]
},
"num": null,
"urls": [],
"raw_text": "Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Neural machine translation of rare words with subword units. In Proceedings of the 54th An- nual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1715- 1725, Berlin, Germany. Association for Computa- tional Linguistics.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Minimum risk training for neural machine translation",
"authors": [
{
"first": "Shiqi",
"middle": [],
"last": "Shen",
"suffix": ""
},
{
"first": "Yong",
"middle": [],
"last": "Cheng",
"suffix": ""
},
{
"first": "Zhongjun",
"middle": [],
"last": "He",
"suffix": ""
},
{
"first": "Wei",
"middle": [],
"last": "He",
"suffix": ""
},
{
"first": "Hua",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Maosong",
"middle": [],
"last": "Sun",
"suffix": ""
},
{
"first": "Yang",
"middle": [],
"last": "Liu",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "1683--1692",
"other_ids": {
"DOI": [
"10.18653/v1/P16-1159"
]
},
"num": null,
"urls": [],
"raw_text": "Shiqi Shen, Yong Cheng, Zhongjun He, Wei He, Hua Wu, Maosong Sun, and Yang Liu. 2016. Minimum risk training for neural machine translation. In Pro- ceedings of the 54th Annual Meeting of the Associa- tion for Computational Linguistics (Volume 1: Long Papers), pages 1683-1692, Berlin, Germany. Asso- ciation for Computational Linguistics.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Neural machine translation by minimising the Bayes-risk with respect to syntactic translation lattices",
"authors": [
{
"first": "Felix",
"middle": [],
"last": "Stahlberg",
"suffix": ""
},
{
"first": "Adri\u00e0",
"middle": [],
"last": "De Gispert",
"suffix": ""
},
{
"first": "Eva",
"middle": [],
"last": "Hasler",
"suffix": ""
},
{
"first": "Bill",
"middle": [],
"last": "Byrne",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 15th Conference of the European Chapter",
"volume": "2",
"issue": "",
"pages": "362--368",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Felix Stahlberg, Adri\u00e0 de Gispert, Eva Hasler, and Bill Byrne. 2017. Neural machine translation by minimising the Bayes-risk with respect to syntactic translation lattices. In Proceedings of the 15th Con- ference of the European Chapter of the Association for Computational Linguistics: Volume 2, Short Pa- pers, pages 362-368, Valencia, Spain. Association for Computational Linguistics.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Attention is all you need",
"authors": [
{
"first": "Ashish",
"middle": [],
"last": "Vaswani",
"suffix": ""
},
{
"first": "Noam",
"middle": [],
"last": "Shazeer",
"suffix": ""
},
{
"first": "Niki",
"middle": [],
"last": "Parmar",
"suffix": ""
},
{
"first": "Jakob",
"middle": [],
"last": "Uszkoreit",
"suffix": ""
},
{
"first": "Llion",
"middle": [],
"last": "Jones",
"suffix": ""
},
{
"first": "Aidan",
"middle": [
"N"
],
"last": "Gomez",
"suffix": ""
},
{
"first": "\u0141ukasz",
"middle": [],
"last": "Kaiser",
"suffix": ""
},
{
"first": "Illia",
"middle": [],
"last": "Polosukhin",
"suffix": ""
}
],
"year": 2017,
"venue": "Advances in Neural Information Processing Systems",
"volume": "30",
"issue": "",
"pages": "5998--6008",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, \u0141ukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Gar- nett, editors, Advances in Neural Information Pro- cessing Systems 30, pages 5998-6008. Curran Asso- ciates, Inc.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "On exposure bias, hallucination and domain shift in neural machine translation",
"authors": [
{
"first": "Chaojun",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Rico",
"middle": [],
"last": "Sennrich",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "3544--3552",
"other_ids": {
"DOI": [
"10.18653/v1/2020.acl-main.326"
]
},
"num": null,
"urls": [],
"raw_text": "Chaojun Wang and Rico Sennrich. 2020. On exposure bias, hallucination and domain shift in neural ma- chine translation. In Proceedings of the 58th Annual Meeting of the Association for Computational Lin- guistics, pages 3544-3552, Online. Association for Computational Linguistics.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Sequence-to-sequence learning as beam-search optimization",
"authors": [
{
"first": "Sam",
"middle": [],
"last": "Wiseman",
"suffix": ""
},
{
"first": "Alexander",
"middle": [
"M"
],
"last": "Rush",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "1296--1306",
"other_ids": {
"DOI": [
"10.18653/v1/D16-1137"
]
},
"num": null,
"urls": [],
"raw_text": "Sam Wiseman and Alexander M. Rush. 2016. Sequence-to-sequence learning as beam-search op- timization. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Process- ing, pages 1296-1306, Austin, Texas. Association for Computational Linguistics.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Bridging the gap between training and inference for neural machine translation",
"authors": [
{
"first": "Wen",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Yang",
"middle": [],
"last": "Feng",
"suffix": ""
},
{
"first": "Fandong",
"middle": [],
"last": "Meng",
"suffix": ""
},
{
"first": "Di",
"middle": [],
"last": "You",
"suffix": ""
},
{
"first": "Qun",
"middle": [],
"last": "Liu",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "4334--4343",
"other_ids": {
"DOI": [
"10.18653/v1/P19-1426"
]
},
"num": null,
"urls": [],
"raw_text": "Wen Zhang, Yang Feng, Fandong Meng, Di You, and Qun Liu. 2019. Bridging the gap between training and inference for neural machine translation. In Pro- ceedings of the 57th Annual Meeting of the Asso- ciation for Computational Linguistics, pages 4334- 4343, Florence, Italy. Association for Computational Linguistics.",
"links": null
}
},
"ref_entries": {
"TABREF0": {
"content": "<table><tr><td/><td/><td>60.0</td><td>59.5</td><td>60.3</td><td>59.1</td><td>61.4</td><td>58.2</td><td>57.6</td><td>60.4</td></tr><tr><td>Koran</td><td/><td>0.9</td><td>1.0</td><td>0.7</td><td>1.1</td><td>0.8</td><td>0.9</td><td>0.9</td><td>1.0</td></tr><tr><td>law</td><td/><td>19.6</td><td>20.6</td><td>16.6</td><td>17.8</td><td>17.8</td><td>19.3</td><td>19.8</td><td>20.8</td></tr><tr><td>IT</td><td/><td>15.0</td><td>16.3</td><td>10.1</td><td>11.5</td><td>15.7</td><td>18.0</td><td>15.3</td><td>17.8</td></tr><tr><td colspan=\"2\">subtitles</td><td>2.8</td><td>3.1</td><td>1.4</td><td>1.9</td><td>2.6</td><td>2.8</td><td>2.4</td><td>2.8</td></tr><tr><td colspan=\"3\">Domain System</td><td colspan=\"3\">1-to 4-gram precisions</td><td>Brevity penalty</td><td colspan=\"3\">BLEU ( ) METEOR ( )</td></tr><tr><td/><td colspan=\"6\">baseline 53.0 27.5 16.9 11.0 0.778</td><td>17.8</td><td>0.36</td></tr><tr><td>law</td><td colspan=\"6\">shortlist 56.1 29.4 17.9 11.4 0.804 re-rank 51.4 26.4 16.1 10.5 0.906</td><td>19.3 (+1.5) 19.8 (+2.0)</td><td colspan=\"2\">0.39 (+0.03) 0.31 (\u22120.05)</td></tr><tr><td/><td/><td>both</td><td colspan=\"4\">53.1 27.6 16.7 10.7 0.919</td><td>20.8 (+3.0)</td><td colspan=\"2\">0.35 (\u22120.01)</td></tr><tr><td/><td colspan=\"5\">baseline 34.6 18.8 13.1 9.5</td><td>0.930</td><td>15.7</td><td>0.16</td></tr><tr><td>IT</td><td colspan=\"6\">shortlist 43.9 24.7 17.1 12.1 0.828 re-rank 33.5 17.2 11.7 8.1 1.000</td><td>18.0 (+2.3) 15.3 (\u22120.4)</td><td colspan=\"2\">0.18 (+0.02) 0.12 (\u22120.04)</td></tr><tr><td/><td/><td>both</td><td colspan=\"3\">38.0 20.1 13.6 9.7</td><td>1.000</td><td>17.8 (+2.1)</td><td colspan=\"2\">0.09 (\u22120.07)</td></tr></table>",
"num": null,
"html": null,
"text": "DomainBPE trained on medical only BPE trained on all except subtitles baseline shortlist re-rank both baseline shortlist re-rank both medicalTable 1: BLEU results on German-English systems trained on the medical domain, and tested on in-domain and out-domain datasets. In-domain results are in italics and the best BLEU on each domain dataset are in bold.",
"type_str": "table"
},
"TABREF1": {
"content": "<table/>",
"num": null,
"html": null,
"text": "",
"type_str": "table"
},
"TABREF2": {
"content": "<table><tr><td>Domain</td><td colspan=\"3\">law medical subtitles \u2020</td><td colspan=\"2\">IT Koran</td></tr><tr><td>Number of sentences</td><td>695k</td><td>1M</td><td colspan=\"2\">1M 372k</td><td>529k</td></tr><tr><td>Avg. original sentence length</td><td>22.1</td><td>12.5</td><td>8.0</td><td>7.5</td><td>20.4</td></tr><tr><td>Avg. BPE sentence length</td><td>30.4</td><td>14.3</td><td colspan=\"2\">11.1 12.7</td><td>24.1</td></tr><tr><td>Vocab size, appearing >20 times</td><td>34k</td><td>36k</td><td>30k</td><td>15k</td><td>20k</td></tr><tr><td>Vocab overlap with medical</td><td>11.5k</td><td>36k</td><td colspan=\"2\">9.0k 5.8k</td><td>5.1k</td></tr><tr><td>\u2020</td><td/><td/><td/><td/><td/></tr></table>",
"num": null,
"html": null,
"text": ", on which we have seen the largest leap The subtitles corpus was sampled down from 20M to 1M sentence pairs.",
"type_str": "table"
},
"TABREF3": {
"content": "<table/>",
"num": null,
"html": null,
"text": "Corpus statistics for the different domains.",
"type_str": "table"
},
"TABREF4": {
"content": "<table/>",
"num": null,
"html": null,
"text": "Two random German-English sentence pairs from the subtitles dataset after BPE.",
"type_str": "table"
},
"TABREF6": {
"content": "<table/>",
"num": null,
"html": null,
"text": "High-resource German-English results.",
"type_str": "table"
}
}
}
} |