File size: 77,685 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 | {
"paper_id": "2022",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T15:24:44.246805Z"
},
"title": "Efficient Machine Translation Domain Adaptation",
"authors": [
{
"first": "Pedro",
"middle": [
"Henrique"
],
"last": "Martins",
"suffix": "",
"affiliation": {
"laboratory": "Instituto de Telecomunica\u00e7\u00f5es DeepMind Institute of Systems and Robotics LUMLIS (Lisbon ELLIS Unit)",
"institution": "Instituto Superior T\u00e9cnico Unbabel Lisbon",
"location": {
"country": "Portugal"
}
},
"email": "pedrohenriqueamartins@tecnico.ulisboa.pt"
},
{
"first": "Zita",
"middle": [],
"last": "Marinho",
"suffix": "",
"affiliation": {
"laboratory": "Instituto de Telecomunica\u00e7\u00f5es DeepMind Institute of Systems and Robotics LUMLIS (Lisbon ELLIS Unit)",
"institution": "Instituto Superior T\u00e9cnico Unbabel Lisbon",
"location": {
"country": "Portugal"
}
},
"email": "zmarinho@google.com"
},
{
"first": "Andr\u00e9",
"middle": [
"F T"
],
"last": "Martins",
"suffix": "",
"affiliation": {
"laboratory": "Instituto de Telecomunica\u00e7\u00f5es DeepMind Institute of Systems and Robotics LUMLIS (Lisbon ELLIS Unit)",
"institution": "Instituto Superior T\u00e9cnico Unbabel Lisbon",
"location": {
"country": "Portugal"
}
},
"email": "andre.t.martins@tecnico.ulisboa.pt."
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Machine translation models struggle when translating out-of-domain text, which makes domain adaptation a topic of critical importance. However, most domain adaptation methods focus on fine-tuning or training the entire or part of the model on every new domain, which can be costly. On the other hand, semi-parametric models have been shown to successfully perform domain adaptation by retrieving examples from an in-domain datastore (Khandelwal et al., 2021). A drawback of these retrievalaugmented models, however, is that they tend to be substantially slower. In this paper, we explore several approaches to speed up nearest neighbor machine translation. We adapt the methods recently proposed by He et al. (2021) for language modeling, and introduce a simple but effective caching strategy that avoids performing retrieval when similar contexts have been seen before. Translation quality and runtimes for several domains show the effectiveness of the proposed solutions. 1",
"pdf_parse": {
"paper_id": "2022",
"_pdf_hash": "",
"abstract": [
{
"text": "Machine translation models struggle when translating out-of-domain text, which makes domain adaptation a topic of critical importance. However, most domain adaptation methods focus on fine-tuning or training the entire or part of the model on every new domain, which can be costly. On the other hand, semi-parametric models have been shown to successfully perform domain adaptation by retrieving examples from an in-domain datastore (Khandelwal et al., 2021). A drawback of these retrievalaugmented models, however, is that they tend to be substantially slower. In this paper, we explore several approaches to speed up nearest neighbor machine translation. We adapt the methods recently proposed by He et al. (2021) for language modeling, and introduce a simple but effective caching strategy that avoids performing retrieval when similar contexts have been seen before. Translation quality and runtimes for several domains show the effectiveness of the proposed solutions. 1",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Modern neural machine translation models are mostly parametric (Bahdanau et al., 2015; Vaswani et al., 2017) , meaning that, for each input, the output depends only on a fixed number of model parameters, obtained using some training data, hopefully in the same domain. However, when running machine translation systems in the wild, it is often the case that the model is given input sentences or documents from domains that were not part of the training data, which frequently leads to subpar translations. One solution is training or fine-tuning the entire model or just part of it for each domain, but this can be expensive and may lead to catastrophic forgetting (Saunders, 2021) .",
"cite_spans": [
{
"start": 63,
"end": 86,
"text": "(Bahdanau et al., 2015;",
"ref_id": "BIBREF1"
},
{
"start": 87,
"end": 108,
"text": "Vaswani et al., 2017)",
"ref_id": "BIBREF16"
},
{
"start": 666,
"end": 682,
"text": "(Saunders, 2021)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Recently, an approach that has achieved promising results is augmenting parametric models with a retrieval component, leading to semi-parametric models (Gu et al., 2018; Zhang et al., 2018; Bapna and Firat, 2019; Khandelwal et al., 2021; Meng et al., 2021; Jiang et al., 2021) . These models construct a datastore based on a set of source / target sentences or word-level contexts (translation memories) and retrieve similar examples from this datastore, using this information in the generation process. This allows having only one model that can be used for every domain. However, the model's runtime increases with the size of the domain's datastore and searching for related examples on large datastores can be computationally very expensive: for example, when retrieving 64 neighbors from the datastore, the model may become two orders of magnitude slower (Khandelwal et al., 2021) . Due to this, some recent works have proposed methods that aim to make this process more efficient. Meng et al. (2021) proposed constructing a different datastore for each source sentence, by first searching for the neighbors of the source tokens; and He et al. (2021) proposed several techniques -datastore pruning, adaptive retrieval, dimension reduction -for nearest neighbor language modeling.",
"cite_spans": [
{
"start": 152,
"end": 169,
"text": "(Gu et al., 2018;",
"ref_id": "BIBREF3"
},
{
"start": 170,
"end": 189,
"text": "Zhang et al., 2018;",
"ref_id": "BIBREF17"
},
{
"start": 190,
"end": 212,
"text": "Bapna and Firat, 2019;",
"ref_id": "BIBREF2"
},
{
"start": 213,
"end": 237,
"text": "Khandelwal et al., 2021;",
"ref_id": "BIBREF7"
},
{
"start": 238,
"end": 256,
"text": "Meng et al., 2021;",
"ref_id": "BIBREF9"
},
{
"start": 257,
"end": 276,
"text": "Jiang et al., 2021)",
"ref_id": "BIBREF5"
},
{
"start": 861,
"end": 886,
"text": "(Khandelwal et al., 2021)",
"ref_id": "BIBREF7"
},
{
"start": 988,
"end": 1006,
"text": "Meng et al. (2021)",
"ref_id": "BIBREF9"
},
{
"start": 1140,
"end": 1156,
"text": "He et al. (2021)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper, we adapt several methods proposed by He et al. (2021) to machine translation, and we further propose a new approach that increases the model's efficiency: the use of a retrieval distributions cache. By caching the kNN probability distributions, together with the corresponding decoder representations, for the previous steps of the generation of the current translation(s), the model can quickly retrieve the retrieval distribution when the current representation is similar to a cached one, instead of having to search for neighbors in the datastore at every single step.",
"cite_spans": [
{
"start": 52,
"end": 68,
"text": "He et al. (2021)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We perform a thorough analysis of the model's efficiency on a controlled setting, which shows that the combination of our proposed techniques results in a model, the efficient kNN-MT, which is approx-imately twice as fast as the vanilla kNN-MT. This comes without harming translation performance, which is, on average, more than 8 BLEU points and 5 COMET points better than the base MT model.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In sum, this paper presents the following contributions:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 We adapt the methods proposed by He et al. (2021) for efficient nearest neighbor language modeling to machine translation.",
"cite_spans": [
{
"start": 35,
"end": 51,
"text": "He et al. (2021)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 We propose a caching strategy to store the retrieval probability distributions, improving the translation speed.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 We compare the efficiency and translation quality of the different methods, which show the benefits of the proposed and adapted techniques.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "When performing machine translation, the model is given a source sentence or document,",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": "x = [x 1 , . . . , x L ]",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": ", on one language, and the goal is to output a translation of the sentence in the desired language, y = [y 1 , . . . , y N ]. This is usually done using a parametric sequence-to-sequence model (Bahdanau et al., 2015; Vaswani et al., 2017) , in which the encoder receives the source sentence as input and outputs a set of hidden states. Then, at each step t, the decoder attends to these hidden states and outputs a probability distribution p NMT (y t |y <t , x) over the vocabulary. Finally, these probability distributions are used to predict the output tokens, typically with beam search.",
"cite_spans": [
{
"start": 193,
"end": 216,
"text": "(Bahdanau et al., 2015;",
"ref_id": "BIBREF1"
},
{
"start": 217,
"end": 238,
"text": "Vaswani et al., 2017)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": "Khandelwal et al. (2021) introduced a nearest neighbor machine translation model, kNN-MT, which is a semi-parametric model. This means that besides having a parametric component that outputs a probability distribution over the vocabulary, p NMT (y t |y <t , x), the model also has a nearest neighbor retrieval mechanism, which allows direct access to a datastore of examples. More specifically, we build a datastore D which consists of a key-value memory, where each entry key is the decoder's output representation, f (x, y <t ), and the value is the target token y t :",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Nearest Neighbor Machine Translation",
"sec_num": "2.1"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "D = {(f (x, y <t ) , y t ) \u2200y t \u2208 y | (x, y) \u2208 (X , Y)},",
"eq_num": "(1)"
}
],
"section": "Nearest Neighbor Machine Translation",
"sec_num": "2.1"
},
{
"text": "where (X , Y) corresponds to a set of parallel source and target sequences. Then, at inference time, the model searches the datastore to retrieve the set of k nearest neighbors N . Using their distances d(\u2022) to the current decoder's output representation, we can compute the retrieval distribution p kNN (y t |y <t , x) as:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Nearest Neighbor Machine Translation",
"sec_num": "2.1"
},
{
"text": "p kNN (y t |y <t , x) = (2) (k j ,v j )\u2208N 1 yt=v j exp (\u2212d (k j , f (x, y <t )) /T ) (k j ,v j )\u2208N exp (\u2212d (k j , f (x, y <t )) /T ) ,",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Nearest Neighbor Machine Translation",
"sec_num": "2.1"
},
{
"text": "where T is the softmax temperature, k j denotes the key of the j th neighbor and v j its value. Finally, p NMT (y t |y <t , x) and p kNN (y t |y <t , x) are combined to obtain the final distribution, which is used to generate the translation through beam search, by performing interpolation:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Nearest Neighbor Machine Translation",
"sec_num": "2.1"
},
{
"text": "p(y t |y <t , x) =(1 \u2212 \u03bb) p NMT (y t |y <t , x) (3) + \u03bb p kNN (y t |y <t , x),",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Nearest Neighbor Machine Translation",
"sec_num": "2.1"
},
{
"text": "where \u03bb is a hyper-parameter that controls the weights given to the two distributions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Nearest Neighbor Machine Translation",
"sec_num": "2.1"
},
{
"text": "In this section, we describe the approaches introduced by He et al. (2021) to speed-up the inference time for nearest neighbor language modeling, such as pruning the datastore ( \u00a73.1) and reducing the representations dimension ( \u00a73.2), which we adapt to machine translation. We further describe a novel method that allows the model to have access to examples without having to search them in the datastore at every step, by maintaining a cache of the past retrieval distributions, for the current translation(s) ( \u00a73.3).",
"cite_spans": [
{
"start": 58,
"end": 74,
"text": "He et al. (2021)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Efficient kNN-MT",
"sec_num": "3"
},
{
"text": "has the same value, we merge the two entries, by simply removing the neighboring one.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Datastore Pruning",
"sec_num": "3.1"
},
{
"text": "The decoder's output representations, f (x, y <t ) are, usually, high-dimensional (1024, in our case). This leads to a high computational cost when computing vector distances, which are needed for retrieving neighbors from the datastore. To alleviate this, we follow He et al. 2021, and use principal component analysis (PCA), an efficient dimension reduction method, to reduce the dimension of the decoder's output representation to a pre-defined dimension, d, and generate a compressed datastore.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dimension Reduction",
"sec_num": "3.2"
},
{
"text": "The model does not need to search the datastore at every step of the translation generation in order to do it correctly. Here, we aim to predict when it needs to retrieve neighbors from the datastore, so that, by only searching the datastore in the necessary steps, we can increase the generation speed.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cache",
"sec_num": "3.3"
},
{
"text": "Adaptive retrieval. To do so, first we follow He et al. 2021, and use a simple MLP to predict the value of the interpolation coefficient \u03bb at each step. Then, we define a threshold, \u03b1, so that the model only performs retrieval when \u03bb > \u03b1. However, we observed that this leads to results ( \u00a7A.3) similar to randomly selecting when to search the datastore. We posit that this occurs because it is difficult to predict when the model should perform retrieval, for domain adaptation (He et al., 2021) , and because in machine translation error propagation occurs more prominently than in language modeling.",
"cite_spans": [
{
"start": 479,
"end": 496,
"text": "(He et al., 2021)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Cache",
"sec_num": "3.3"
},
{
"text": "Cache. Because it is common to have similar contexts along the generation process, when using beam search, the model can be often retrieving similar neighbors at different steps, which is not efficient. To avoid repeating searches on the datastore for similar context vectors, f (x, y <t ), we propose keeping a cache of the previous retrieval distributions, of the current translation(s). More specifically, at each step of the generation of y, we add the decoder's representation vector along with the retrieval distribution p kNN (y t |y <t , x), corresponding to all beams, B, to the cache C:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cache",
"sec_num": "3.3"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "C={(f (x, y <t ), p kNN (y t |y <t ,x))\u2200y t \u2208 y | y \u2208 B}.",
"eq_num": "(4)"
}
],
"section": "Cache",
"sec_num": "3.3"
},
{
"text": "Then, at each step of the generation, we compute the Euclidean distance between the current decoder's representation and the keys on the cache. If all distances are bigger than a threshold \u03c4 , the model searches the datastore to find the nearest neighbors. Otherwise, the model retrieves, from the cache, the retrieval distribution that corresponds to the closest key.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cache",
"sec_num": "3.3"
},
{
"text": "Dataset and metrics. We perform experiments on the Medical, Law, IT, and Koran domain data of the multi-domains dataset (Koehn and Knowles, 2017) re-splitted by Aharoni and Goldberg (2020) .",
"cite_spans": [
{
"start": 120,
"end": 145,
"text": "(Koehn and Knowles, 2017)",
"ref_id": "BIBREF8"
},
{
"start": 161,
"end": 188,
"text": "Aharoni and Goldberg (2020)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "4"
},
{
"text": "To build the datastores we use the in-domain training sets which have from 17,982 to 467,309 sentences. The validation and test sets have 2,000 sentences.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "4"
},
{
"text": "To evaluate the models we use BLEU (Papineni et al., 2002; Post, 2018) and COMET (Rei et al., 2020) .",
"cite_spans": [
{
"start": 35,
"end": 58,
"text": "(Papineni et al., 2002;",
"ref_id": "BIBREF12"
},
{
"start": 59,
"end": 70,
"text": "Post, 2018)",
"ref_id": "BIBREF13"
},
{
"start": 81,
"end": 99,
"text": "(Rei et al., 2020)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "4"
},
{
"text": "Settings. We use the WMT'19 German-English news translation task winner ) (with 269 M parameters), available on the Fairseq library , as the base MT model.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "4"
},
{
"text": "As baselines, we consider the base MT model, the vanilla kNN-MT model (Khandelwal et al., 2021) , and the Fast kNN-MT model (Meng et al., 2021) . For all models, which perform retrieval, we select the hyper-parameters, for each method and each domain, by performing grid search on k \u2208 {8, 16, 32, 64} and \u03bb \u2208 {0.5, 0.6, 0.7, 0.8}. The selected values are stated in Table 9 of App. B.",
"cite_spans": [
{
"start": 70,
"end": 95,
"text": "(Khandelwal et al., 2021)",
"ref_id": "BIBREF7"
},
{
"start": 124,
"end": 143,
"text": "(Meng et al., 2021)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [
{
"start": 365,
"end": 372,
"text": "Table 9",
"ref_id": null
}
],
"eq_spans": [],
"section": "Experiments",
"sec_num": "4"
},
{
"text": "For the vanilla kNN-MT model and the efficient kNN-MT we follow Khandelwal et al. (2021) and use the Euclidean distance to perform retrieval and the proposed softmax temperature. For the Fast kNN-MT, we use the cosine distance and the softmax temperature proposed by Meng et al. (2021) . For the efficient kNN-MT we selected parameters that ensure a good speed/quality trade-off: k = 2 for datastore pruning, d = 256 for PCA, and \u03c4 = 6 as the cache threshold. Results for each methods using different parameters are reported in App. A.",
"cite_spans": [
{
"start": 64,
"end": 88,
"text": "Khandelwal et al. (2021)",
"ref_id": "BIBREF7"
},
{
"start": 267,
"end": 285,
"text": "Meng et al. (2021)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "4"
},
{
"text": "The translation scores are reported on Figure 1 : Plots of the generation speed (tokens/s) for the different models on the medical, law, IT, and Koran domains, for different batch sizes (1, 8, 16) . The generation speed (y-axis) is in log scale. When using the Fast kNN-MT model, the maximum batch size that we are able to use is 2, due to out of memory errors. points and 5 COMET points more than the base MT model.",
"cite_spans": [
{
"start": 186,
"end": 189,
"text": "(1,",
"ref_id": null
},
{
"start": 190,
"end": 192,
"text": "8,",
"ref_id": null
},
{
"start": 193,
"end": 196,
"text": "16)",
"ref_id": null
}
],
"ref_spans": [
{
"start": 39,
"end": 47,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "4.1"
},
{
"text": "Computational infrastructure. All experiments were performed on a server with 3 RTX 2080 Ti (11 GB), 12 AMD Ryzen 2920X CPUs (24 cores), and 128 Gb of RAM. For the generation speed measurements, we ran each model on a single GPU while no other process was running on the server, to have a controlled environment. To search the datastore, we used the FAISS library (Johnson et al., 2019) . When using the vanilla kNN-MT and efficient kNN-MT, the nearest neighbor search is performed on the CPUs, since not all datastores fit into memory, while when using the Fast kNN-MT this is done on the GPU.",
"cite_spans": [
{
"start": 364,
"end": 386,
"text": "(Johnson et al., 2019)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Generation speed",
"sec_num": "4.2"
},
{
"text": "Analysis. As can be seen on the plots of Figure 1 , for a batch size of 1 Fast kNN-MT leads to a generation speed higher than our proposed method and vanilla kNN-MT. However, because of its high memory requirements, we are not able to run Fast kNN-MT for batch sizes larger than 2, on the computational infrastructure stated above. On the contrary, when using the proposed methods (efficient kNN-MT) we are able to run the model with higher batch sizes, achieving superior generation speeds to Fast kNN-MT and vanilla kNN-MT, and reducing the gap to the base MT model. Ablation. We plot the generation speed for different combinations of the proposed methods (averaged across domains), for several batch sizes, on Figure 2 . On this plot, we can clearly see that every method contributes to the speed-up achieved by the model that combines all approaches. Moreover, we can observe that the method which leads to the largest speed-up is the use of a cache of retrieval distributions, by saving, on average 57% of the retrieval searches.",
"cite_spans": [],
"ref_spans": [
{
"start": 41,
"end": 49,
"text": "Figure 1",
"ref_id": null
},
{
"start": 714,
"end": 722,
"text": "Figure 2",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Generation speed",
"sec_num": "4.2"
},
{
"text": "In this paper we propose the efficient kNN-MT, in which we combine several methods to improve the kNN-MT generation speed. First, we adapted to machine translation methods that improve retrieval efficiency in language modeling (He et al., 2021) . Then we proposed a new method which consists on keeping in cache the previous retrieval distributions so that the model does not need to search for neighbors in the datastore at every step. Through experiments on domain adaptation, we show that the combination of the proposed methods leads to a considerable speed-up (up to 2x) without harming the translation performance substantially.",
"cite_spans": [
{
"start": 227,
"end": 244,
"text": "(He et al., 2021)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "In this section we report the BLEU scores as well as additional statistics for the different methods, when varying their hyper-parameters.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "A Additional results",
"sec_num": null
},
{
"text": "We report on Table 2 the BLEU scores for datastore pruning, when varying the number of neighbors used for greedy merging, k. The resulting datastore sizes are presented on Table 3 4,039,432 11,103,775 2,303,808 353,007 k = 5 3,084,106 8,486,551 1,852,191 290,192 ",
"cite_spans": [
{
"start": 180,
"end": 262,
"text": "4,039,432 11,103,775 2,303,808 353,007 k = 5 3,084,106 8,486,551 1,852,191 290,192",
"ref_id": null
}
],
"ref_spans": [
{
"start": 13,
"end": 20,
"text": "Table 2",
"ref_id": "TABREF2"
},
{
"start": 172,
"end": 179,
"text": "Table 3",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "A.1 Datastore pruning",
"sec_num": null
},
{
"text": "We report on Table 4 the BLEU scores for dimension reduction, when varying the output dimension d. ",
"cite_spans": [],
"ref_spans": [
{
"start": 13,
"end": 20,
"text": "Table 4",
"ref_id": "TABREF5"
}
],
"eq_spans": [],
"section": "A.2 Dimension reduction",
"sec_num": null
},
{
"text": "We report on Table 5 the BLEU scores for adaptive retrieval, when varying the threshold \u03b1. The percentage of times the model performs retrieval is stated on Table 6 . Table 6 : Percentage of times the model searches for neighbors on the datastore when performing adaptive retrieval for different values of the threshold \u03b1, for a batch size of 8.",
"cite_spans": [],
"ref_spans": [
{
"start": 13,
"end": 20,
"text": "Table 5",
"ref_id": "TABREF7"
},
{
"start": 157,
"end": 164,
"text": "Table 6",
"ref_id": null
},
{
"start": 167,
"end": 174,
"text": "Table 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "A.3 Adaptive retrieval",
"sec_num": null
},
{
"text": "We report on Table 7 the BLEU scores for a model using a cache of the retrieval distributions, when varying the threshold \u03c4 . The percentage of times the model performs retrieval is stated on Table 8 : Percentage of times the model searches for neighbors on the datastore when using a retrieval distributions' cache for different values of the threshold \u03c4 , for a batch size of 8.",
"cite_spans": [],
"ref_spans": [
{
"start": 13,
"end": 20,
"text": "Table 7",
"ref_id": "TABREF9"
},
{
"start": 192,
"end": 199,
"text": "Table 8",
"ref_id": "TABREF8"
}
],
"eq_spans": [],
"section": "A.4 Cache",
"sec_num": null
},
{
"text": "On Table 9 we report the values for the hyperparameters: number of neighbors to be retrieved",
"cite_spans": [],
"ref_spans": [
{
"start": 3,
"end": 10,
"text": "Table 9",
"ref_id": null
}
],
"eq_spans": [],
"section": "B Hyper-parameters",
"sec_num": null
},
{
"text": "The code is available at https://github.com/ deep-spin/efficient_kNN_MT.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "The goal of datastore pruning is to reduce the size of the datastore, so that the model is able to search for the nearest neighbors faster, without severely compromising the translation performance. To do so, we followHe et al. (2021), and use greedy merging. In greedy merging, we aim to merge datastore entries that share the same value (target token) while their keys are close to each other in vector space. To do this, we first need to find the k nearest neighbors of every entry of the datastore, where k is a hyper-parameter. Then, if in the set of neighbors, retrieved for a given entry, there is an entry which has not been merged before and",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "This work was supported by the European Research Council (ERC StG DeepSPIN 758969), by the P2020 project MAIA (contract 045909), by the Funda\u00e7\u00e3o para a Ci\u00eancia e Tecnologia through project PTDC/CCI-INF/4703/2021 (PRE-LUNA, contract UIDB/50008/2020), and by contract PD/BD/150633/2020 in the scope of the Doctoral Program FCT -PD/00140/2013 NETSyS. We thank Junxian He, Graham Neubig, the SARDINE team members, and the reviewers for helpful discussion and feedback.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgments",
"sec_num": null
},
{
"text": "IT Koran k \u03bb T k \u03bb T k \u03bb T k \u03bb T kNN-MT 8 0.7 10 8 0.8 10 8 0.7 10 8 0.6 100 Fast kNN-MT 16 0.7 .015 32 0.6 .015 8 0.6 .02 16 0.6 .05 cache 8 0.7 10 8 0.8 10 8 0.7 10 8 0.6 100 PCA + cache 8 0.8 10 8 0.8 10 8 0.7 10 8 0.7 100 PCA + pruning 8 0.7 10 8 0.8 10 8 0.7 10 8 0.7 100 PCA + cache + pruning 8 0.7 10 8 0.8 10 8 0.7 10 8 0.7 100 Table 9 : Values of the hyper-parameters: number of neighbors to be retrieved k, interpolation coefficient \u03bb, and retrieval softmax temperature T .k \u2208 {8, 16, 32, 64}, the interpolation coefficient \u03bb \u2208 {0.5, 0.6, 0.7, 0.8}, and retrieval softmax temperature T . For decoding we use beam search with a beam size of 5.",
"cite_spans": [],
"ref_spans": [
{
"start": 3,
"end": 44,
"text": "Koran k \u03bb T k \u03bb T k \u03bb T k \u03bb T",
"ref_id": null
},
{
"start": 348,
"end": 355,
"text": "Table 9",
"ref_id": null
}
],
"eq_spans": [],
"section": "Law",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Unsupervised domain clusters in pretrained language models",
"authors": [
{
"first": "Roee",
"middle": [],
"last": "Aharoni",
"suffix": ""
},
{
"first": "Yoav",
"middle": [],
"last": "Goldberg",
"suffix": ""
}
],
"year": 2020,
"venue": "Proc. ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Roee Aharoni and Yoav Goldberg. 2020. Unsupervised domain clusters in pretrained language models. In Proc. ACL.",
"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": "Kyung",
"middle": [
"Hyun"
],
"last": "Cho",
"suffix": ""
},
{
"first": "Yoshua",
"middle": [],
"last": "Bengio",
"suffix": ""
}
],
"year": 2015,
"venue": "Proc. ICLR",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dzmitry Bahdanau, Kyung Hyun Cho, and Yoshua Ben- gio. 2015. Neural machine translation by jointly learning to align and translate. In Proc. ICLR.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Non-Parametric Adaptation for Neural Machine Translation",
"authors": [
{
"first": "Ankur",
"middle": [],
"last": "Bapna",
"suffix": ""
},
{
"first": "Orhan",
"middle": [],
"last": "Firat",
"suffix": ""
}
],
"year": 2019,
"venue": "Proc. NAACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ankur Bapna and Orhan Firat. 2019. Non-Parametric Adaptation for Neural Machine Translation. In Proc. NAACL.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Search engine guided neural machine translation",
"authors": [
{
"first": "Jiatao",
"middle": [],
"last": "Gu",
"suffix": ""
},
{
"first": "Yong",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Kyunghyun",
"middle": [],
"last": "Cho",
"suffix": ""
},
{
"first": "O",
"middle": [
"K"
],
"last": "Victor",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Li",
"suffix": ""
}
],
"year": 2018,
"venue": "Proc. AAAI",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jiatao Gu, Yong Wang, Kyunghyun Cho, and Victor OK Li. 2018. Search engine guided neural machine trans- lation. In Proc. AAAI.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Efficient Nearest Neighbor Language Models",
"authors": [
{
"first": "Junxian",
"middle": [],
"last": "He",
"suffix": ""
},
{
"first": "Graham",
"middle": [],
"last": "Neubig",
"suffix": ""
},
{
"first": "Taylor",
"middle": [],
"last": "Berg-Kirkpatrick",
"suffix": ""
}
],
"year": 2021,
"venue": "Proc. EMNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Junxian He, Graham Neubig, and Taylor Berg- Kirkpatrick. 2021. Efficient Nearest Neighbor Lan- guage Models. In Proc. EMNLP.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Learning Kernel-Smoothed Machine Translation with Retrieved Examples",
"authors": [
{
"first": "Qingnan",
"middle": [],
"last": "Jiang",
"suffix": ""
},
{
"first": "Mingxuan",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Jun",
"middle": [],
"last": "Cao",
"suffix": ""
},
{
"first": "Shanbo",
"middle": [],
"last": "Cheng",
"suffix": ""
},
{
"first": "Shujian",
"middle": [],
"last": "Huang",
"suffix": ""
},
{
"first": "Lei",
"middle": [],
"last": "Li",
"suffix": ""
}
],
"year": 2021,
"venue": "Proc. EMNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Qingnan Jiang, Mingxuan Wang, Jun Cao, Shanbo Cheng, Shujian Huang, and Lei Li. 2021. Learn- ing Kernel-Smoothed Machine Translation with Re- trieved Examples. In Proc. EMNLP.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Billion-scale similarity search with gpus",
"authors": [
{
"first": "Jeff",
"middle": [],
"last": "Johnson",
"suffix": ""
},
{
"first": "Matthijs",
"middle": [],
"last": "Douze",
"suffix": ""
},
{
"first": "Herv\u00e9",
"middle": [],
"last": "J\u00e9gou",
"suffix": ""
}
],
"year": 2019,
"venue": "IEEE Transactions on Big Data",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jeff Johnson, Matthijs Douze, and Herv\u00e9 J\u00e9gou. 2019. Billion-scale similarity search with gpus. IEEE Transactions on Big Data.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Nearest neighbor machine translation",
"authors": [
{
"first": "Urvashi",
"middle": [],
"last": "Khandelwal",
"suffix": ""
},
{
"first": "Angela",
"middle": [],
"last": "Fan",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Jurafsky",
"suffix": ""
},
{
"first": "Luke",
"middle": [],
"last": "Zettlemoyer",
"suffix": ""
},
{
"first": "Mike",
"middle": [],
"last": "Lewis",
"suffix": ""
}
],
"year": 2021,
"venue": "Proc. ICLR",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Urvashi Khandelwal, Angela Fan, Dan Jurafsky, Luke Zettlemoyer, and Mike Lewis. 2021. Nearest neigh- bor machine translation. In Proc. ICLR.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"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": "",
"other_ids": {},
"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.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Fast Nearest Neighbor Machine Translation",
"authors": [
{
"first": "Yuxian",
"middle": [],
"last": "Meng",
"suffix": ""
},
{
"first": "Xiaoya",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Xiayu",
"middle": [],
"last": "Zheng",
"suffix": ""
},
{
"first": "Fei",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Xiaofei",
"middle": [],
"last": "Sun",
"suffix": ""
},
{
"first": "Tianwei",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Jiwei",
"middle": [],
"last": "Li",
"suffix": ""
}
],
"year": 2021,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yuxian Meng, Xiaoya Li, Xiayu Zheng, Fei Wu, Xi- aofei Sun, Tianwei Zhang, and Jiwei Li. 2021. Fast Nearest Neighbor Machine Translation.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Facebook FAIR's WMT19 News Translation Task Submission",
"authors": [
{
"first": "Nathan",
"middle": [],
"last": "Ng",
"suffix": ""
},
{
"first": "Kyra",
"middle": [],
"last": "Yee",
"suffix": ""
},
{
"first": "Alexei",
"middle": [],
"last": "Baevski",
"suffix": ""
},
{
"first": "Myle",
"middle": [],
"last": "Ott",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Auli",
"suffix": ""
},
{
"first": "Sergey",
"middle": [],
"last": "Edunov",
"suffix": ""
}
],
"year": 2019,
"venue": "Proc. of the Fourth Conference on Machine Translation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nathan Ng, Kyra Yee, Alexei Baevski, Myle Ott, Michael Auli, and Sergey Edunov. 2019. Facebook FAIR's WMT19 News Translation Task Submission. In Proc. of the Fourth Conference on Machine Trans- lation.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "fairseq: A Fast, Extensible Toolkit for Sequence Modeling",
"authors": [
{
"first": "Myle",
"middle": [],
"last": "Ott",
"suffix": ""
},
{
"first": "Sergey",
"middle": [],
"last": "Edunov",
"suffix": ""
},
{
"first": "Alexei",
"middle": [],
"last": "Baevski",
"suffix": ""
},
{
"first": "Angela",
"middle": [],
"last": "Fan",
"suffix": ""
},
{
"first": "Sam",
"middle": [],
"last": "Gross",
"suffix": ""
},
{
"first": "Nathan",
"middle": [],
"last": "Ng",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Grangier",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Auli",
"suffix": ""
}
],
"year": 2019,
"venue": "Proc. NAACL (Demonstrations)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, and Michael Auli. 2019. fairseq: A Fast, Extensible Toolkit for Sequence Modeling. In Proc. NAACL (Demonstra- tions).",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Bleu: a method for automatic evaluation of machine translation",
"authors": [
{
"first": "Kishore",
"middle": [],
"last": "Papineni",
"suffix": ""
},
{
"first": "Salim",
"middle": [],
"last": "Roukos",
"suffix": ""
},
{
"first": "Todd",
"middle": [],
"last": "Ward",
"suffix": ""
},
{
"first": "Wei-Jing",
"middle": [],
"last": "Zhu",
"suffix": ""
}
],
"year": 2002,
"venue": "Proc. ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kishore Papineni, Salim Roukos, Todd Ward, and Wei- Jing Zhu. 2002. Bleu: a method for automatic evalu- ation of machine translation. In Proc. ACL.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "A Call for Clarity in Reporting BLEU Scores",
"authors": [
{
"first": "Matt",
"middle": [],
"last": "Post",
"suffix": ""
}
],
"year": 2018,
"venue": "Proc. Third Conference on Machine Translation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Matt Post. 2018. A Call for Clarity in Reporting BLEU Scores. In Proc. Third Conference on Machine Trans- lation.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "COMET: A Neural Framework for MT Evaluation",
"authors": [
{
"first": "Ricardo",
"middle": [],
"last": "Rei",
"suffix": ""
},
{
"first": "Craig",
"middle": [],
"last": "Stewart",
"suffix": ""
},
{
"first": "Ana",
"middle": [
"C"
],
"last": "Farinha",
"suffix": ""
},
{
"first": "Alon",
"middle": [],
"last": "Lavie",
"suffix": ""
}
],
"year": 2020,
"venue": "Proc. EMNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ricardo Rei, Craig Stewart, Ana C Farinha, and Alon Lavie. 2020. COMET: A Neural Framework for MT Evaluation. In Proc. EMNLP.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Domain Adaptation and Multi-Domain Adaptation for Neural Machine Translation: A Survey",
"authors": [
{
"first": "Danielle",
"middle": [],
"last": "Saunders",
"suffix": ""
}
],
"year": 2021,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Danielle Saunders. 2021. Domain Adaptation and Multi-Domain Adaptation for Neural Machine Trans- lation: A Survey.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"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": "Proc. NeurIPS",
"volume": "",
"issue": "",
"pages": "",
"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 Proc. NeurIPS.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Guiding Neural Machine Translation with Retrieved Translation Pieces",
"authors": [
{
"first": "Jingyi",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Masao",
"middle": [],
"last": "Utiyama",
"suffix": ""
},
{
"first": "Eiichiro",
"middle": [],
"last": "Sumita",
"suffix": ""
}
],
"year": 2018,
"venue": "Proc. NAACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jingyi Zhang, Masao Utiyama, Eiichiro Sumita, Gra- ham Neubig, and Satoshi Nakamura. 2018. Guiding Neural Machine Translation with Retrieved Transla- tion Pieces. In Proc. NAACL.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Adaptive Nearest Neighbor Machine Translation",
"authors": [
{
"first": "Xin",
"middle": [],
"last": "Zheng",
"suffix": ""
},
{
"first": "Zhirui",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Junliang",
"middle": [],
"last": "Guo",
"suffix": ""
},
{
"first": "Shujian",
"middle": [],
"last": "Huang",
"suffix": ""
},
{
"first": "Boxing",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Weihua",
"middle": [],
"last": "Luo",
"suffix": ""
},
{
"first": "Jiajun",
"middle": [],
"last": "Chen",
"suffix": ""
}
],
"year": 2021,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xin Zheng, Zhirui Zhang, Junliang Guo, Shujian Huang, Boxing Chen, Weihua Luo, and Jiajun Chen. 2021. Adaptive Nearest Neighbor Machine Translation.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"text": "Plot of the generation speed (tokens/s), averaged across domains, for different combinations of the proposed methods.",
"uris": null,
"num": null,
"type_str": "figure"
},
"TABREF0": {
"num": null,
"type_str": "table",
"text": "We can clearly see that both Fast kNN-MT and the efficient kNN-MT (combining the different methods) do not hurt the translation performance substantially, still leading to, on average, 8 BLEU BLEU and COMET scores on the multi-domains test set, for a batch size of 8.",
"html": null,
"content": "<table><tr><td/><td/><td/><td colspan=\"2\">Medical Law</td><td>BLEU IT</td><td colspan=\"2\">Koran Average</td><td colspan=\"2\">Medical Law</td><td>COMET IT</td><td colspan=\"2\">Koran Average</td></tr><tr><td colspan=\"3\">Baselines Base MT kNN-MT Fast kNN-MT</td><td>40.01 54.47 52.90</td><td colspan=\"3\">45.64 37.91 16.35 61.23 45.96 21.02 55.71 44.73 21.29</td><td>34.98 45.67 43.66</td><td>.4702 .5760 .5293</td><td colspan=\"3\">.5770 .3942 -.0097 .6781 .5163 .0480 .5944 .5445 -.0455</td><td>.3579 .4546 .4057</td></tr><tr><td colspan=\"3\">Efficient kNN-MT cache PCA + cache PCA + pruning PCA + cache + pruning</td><td>53.30 53.58 53.23 51.90</td><td colspan=\"3\">59.12 45.39 20.67 58.57 46.29 20.67 60.38 45.16 20.52 57.82 44.44 20.11</td><td>44.62 44.78 44.82 43.57</td><td>.5625 .5457 .5658 .5513</td><td colspan=\"3\">.6403 .5085 .0346 .6379 .5311 -.0021 .6639 .4981 .0298 .6260 .4909 -.0052</td><td>.4365 .4282 .4394 .4158</td></tr><tr><td>Generation speed</td><td>10 2 10 3</td><td>base kNN-MT fast kNN-MT efficient kNN-MT Medical</td><td>10 2 10 3</td><td/><td>Law</td><td/><td>10 2 10 3</td><td>IT</td><td/><td>10 2 10 3</td><td>Koran</td></tr><tr><td/><td>1</td><td>8 Batch size</td><td>16</td><td>1</td><td>8 Batch size</td><td>16</td><td>1</td><td>8 Batch size</td><td>16</td><td>1</td><td>8 Batch size</td><td>16</td></tr></table>"
},
"TABREF1": {
"num": null,
"type_str": "table",
"text": ".",
"html": null,
"content": "<table><tr><td/><td colspan=\"2\">Medical Law</td><td>IT</td><td>Koran Average</td></tr><tr><td>kNN-MT</td><td>54.47</td><td colspan=\"3\">61.23 45.96 21.02</td><td>45.67</td></tr><tr><td>k = 1 k = 2 k = 5</td><td>53.60 52.95 51.63</td><td colspan=\"3\">60.23 45.03 20.81 59.40 44.76 20.12 57.55 44.07 19.29</td><td>44.92 44.31 43.14</td></tr></table>"
},
"TABREF2": {
"num": null,
"type_str": "table",
"text": "BLEU scores on the multi-domains test set when performing datastore pruning with several values of k, for a batch size of 8.",
"html": null,
"content": "<table><tr><td/><td>Medical</td><td>Law</td><td>IT</td><td>Koran</td></tr><tr><td colspan=\"5\">kNN-MT 6,903,141 19,061,382 3,602,862 524,374</td></tr><tr><td>k = 1</td><td colspan=\"4\">4,780,514 13,130,326 2,641,709 400,385</td></tr><tr><td>k = 2</td><td/><td/><td/></tr></table>"
},
"TABREF3": {
"num": null,
"type_str": "table",
"text": "Sizes of the in-domain datastores when performing datastore pruning with several values of k, for a batch size of 8.",
"html": null,
"content": "<table/>"
},
"TABREF5": {
"num": null,
"type_str": "table",
"text": "BLEU scores on the multi-domains test set when performing PCA with different dimension, d, values, for a batch size of 8.",
"html": null,
"content": "<table/>"
},
"TABREF7": {
"num": null,
"type_str": "table",
"text": "BLEU scores on the multi-domains test set when performing adaptive retrieval for different values of the threshold \u03b1, for a batch size of 8.",
"html": null,
"content": "<table><tr><td/><td>Medical</td><td>Law</td><td>IT</td><td>Koran</td></tr><tr><td>kNN-MT</td><td>100%</td><td colspan=\"3\">100% 100% 100%</td></tr><tr><td>\u03b1 = 0.25 \u03b1 = 0.5 \u03b1 = 0.75</td><td>78% 96% 98%</td><td>73% 96% 99%</td><td>38% 60% 92%</td><td>4% 61% 91%</td></tr></table>"
},
"TABREF8": {
"num": null,
"type_str": "table",
"text": "",
"html": null,
"content": "<table><tr><td/><td colspan=\"2\">Medical Law</td><td>IT</td><td>Koran Average</td></tr><tr><td>kNN-MT</td><td>54.47</td><td colspan=\"3\">61.23 45.96 21.02</td><td>45.67</td></tr><tr><td>\u03c4 = 2 \u03c4 = 4 \u03c4 = 6 \u03c4 = 8</td><td>54.47 54.17 53.30 30.06</td><td colspan=\"3\">61.23 45.93 20.98 61.10 46.07 21.00 59.12 45.39 20.67 23.01 25.53 16.08</td><td>45.65 45.58 44.62 23.67</td></tr></table>"
},
"TABREF9": {
"num": null,
"type_str": "table",
"text": "BLEU scores on the multi-domains test set when using a retrieval distributions' cache for different values of the threshold \u03c4 , for a batch size of 8.",
"html": null,
"content": "<table><tr><td/><td>Medical</td><td>Law</td><td>IT</td><td>Koran</td></tr><tr><td>kNN-MT</td><td>100%</td><td colspan=\"3\">100% 100% 100%</td></tr><tr><td>\u03c4 = 2 \u03c4 = 4 \u03c4 = 6 \u03c4 = 8</td><td>59% 50% 43% 26%</td><td>51% 42% 35% 16%</td><td>67% 57% 49% 29%</td><td>64% 53% 45% 31%</td></tr></table>"
}
}
}
} |