File size: 93,421 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 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 | {
"paper_id": "P02-1024",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T09:30:40.589911Z"
},
"title": "Exploring Asymmetric Clustering for Statistical Language Modeling",
"authors": [
{
"first": "Jianfeng",
"middle": [],
"last": "Gao",
"suffix": "",
"affiliation": {},
"email": "jfgao@microsoft.com"
},
{
"first": "Joshua",
"middle": [
"T"
],
"last": "Goodman",
"suffix": "",
"affiliation": {},
"email": "joshuago@microsoft.com"
},
{
"first": "Guihong",
"middle": [],
"last": "Cao",
"suffix": "",
"affiliation": {},
"email": ""
},
{
"first": "Hang",
"middle": [],
"last": "Li",
"suffix": "",
"affiliation": {},
"email": "hangli@microsoft.com"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "The n-gram model is a stochastic model, which predicts the next word (predicted word) given the previous words (conditional words) in a word sequence. The cluster n-gram model is a variant of the n-gram model in which similar words are classified in the same cluster. It has been demonstrated that using different clusters for predicted and conditional words leads to cluster models that are superior to classical cluster models which use the same clusters for both words. This is the basis of the asymmetric cluster model (ACM) discussed in our study. In this paper, we first present a formal definition of the ACM. We then describe in detail the methodology of constructing the ACM. The effectiveness of the ACM is evaluated on a realistic application, namely Japanese Kana-Kanji conversion. Experimental results show substantial improvements of the ACM in comparison with classical cluster models and word n-gram models at the same model size. Our analysis shows that the high-performance of the ACM lies in the asymmetry of the model.",
"pdf_parse": {
"paper_id": "P02-1024",
"_pdf_hash": "",
"abstract": [
{
"text": "The n-gram model is a stochastic model, which predicts the next word (predicted word) given the previous words (conditional words) in a word sequence. The cluster n-gram model is a variant of the n-gram model in which similar words are classified in the same cluster. It has been demonstrated that using different clusters for predicted and conditional words leads to cluster models that are superior to classical cluster models which use the same clusters for both words. This is the basis of the asymmetric cluster model (ACM) discussed in our study. In this paper, we first present a formal definition of the ACM. We then describe in detail the methodology of constructing the ACM. The effectiveness of the ACM is evaluated on a realistic application, namely Japanese Kana-Kanji conversion. Experimental results show substantial improvements of the ACM in comparison with classical cluster models and word n-gram models at the same model size. Our analysis shows that the high-performance of the ACM lies in the asymmetry of the model.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "The n-gram model has been widely applied in many applications such as speech recognition, machine translation, and Asian language text input [Jelinek, 1990; Brown et al., 1990; Gao et al., 2002] . It is a stochastic model, which predicts the next word (predicted word) given the previous n-1 words (conditional words) in a word sequence.",
"cite_spans": [
{
"start": 141,
"end": 156,
"text": "[Jelinek, 1990;",
"ref_id": "BIBREF9"
},
{
"start": 157,
"end": 176,
"text": "Brown et al., 1990;",
"ref_id": "BIBREF2"
},
{
"start": 177,
"end": 194,
"text": "Gao et al., 2002]",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The cluster n-gram model is a variant of the word n-gram model in which similar words are classified in the same cluster. This has been demonstrated as an effective way to deal with the data sparseness problem and to reduce the memory sizes for realistic applications. Recent research [Yamamoto et al., 2001] shows that using different clusters for predicted and conditional words can lead to cluster models that are superior to classical cluster models, which use the same clusters for both words [Brown et al., 1992] . This is the basis of the asymmetric cluster model (ACM), which will be formally defined and empirically studied in this paper. Although similar models have been used in previous studies [Goodman and Gao, 2000; Yamamoto et al., 2001] , several issues have not been completely investigated. These include: (1) an effective methodology for constructing the ACM, (2) a thorough comparative study of the ACM with classical cluster models and word models when they are applied to a realistic application, and (3) an analysis of the reason why the ACM is superior.",
"cite_spans": [
{
"start": 285,
"end": 308,
"text": "[Yamamoto et al., 2001]",
"ref_id": "BIBREF18"
},
{
"start": 498,
"end": 518,
"text": "[Brown et al., 1992]",
"ref_id": "BIBREF3"
},
{
"start": 707,
"end": 730,
"text": "[Goodman and Gao, 2000;",
"ref_id": "BIBREF8"
},
{
"start": 731,
"end": 753,
"text": "Yamamoto et al., 2001]",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The goal of this study is to address the above three issues. We first present a formal definition of the ACM; then we describe in detail the methodology of constructing the ACM including (1) an asymmetric clustering algorithm in which different metrics are used for clustering the predicted and conditional words respectively; and (2) a method for model parameter optimization in which the optimal cluster numbers are found for different clusters. We evaluate the ACM on a real application, Japanese Kana-Kanji conversion, which converts phonetic Kana strings into proper Japanese orthography. The performance is measured in terms of character error rate (CER). Our results show substantial improvements of the ACM in comparison with classical cluster models and word n-gram models at the same model size. Our analysis shows that the high-performance of the ACM comes from better structure and better smoothing, both of which lie in the asymmetry of the model. This paper is organized as follows: Section 1 introduces our research topic, and then Section 2 reviews related work. Section 3 defines the ACM and describes in detail the method of model construction. Section 4 first introduces the Japanese Kana-Kanji conversion task; it then presents our main experiments and a discussion of our findings. Finally, conclusions are presented in Section 5.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "A large amount of previous research on clustering has been focused on how to find the best clusters [Brown et al., 1992; Kneser and Ney, 1993; Yamamoto and Sagisaka, 1999; Ueberla, 1996; Pereira et al., 1993; Bellegarda et al., 1996; Bai et al., 1998 ]. Only small differences have been observed, however, in the performance of the different techniques for constructing clusters. In this study, we focused our research on novel techniques for using clusters -the ACM, in which different clusters are used for predicted and conditional words respectively.",
"cite_spans": [
{
"start": 100,
"end": 120,
"text": "[Brown et al., 1992;",
"ref_id": "BIBREF3"
},
{
"start": 121,
"end": 142,
"text": "Kneser and Ney, 1993;",
"ref_id": "BIBREF11"
},
{
"start": 143,
"end": 171,
"text": "Yamamoto and Sagisaka, 1999;",
"ref_id": "BIBREF19"
},
{
"start": 172,
"end": 186,
"text": "Ueberla, 1996;",
"ref_id": "BIBREF17"
},
{
"start": 187,
"end": 208,
"text": "Pereira et al., 1993;",
"ref_id": "BIBREF13"
},
{
"start": 209,
"end": 233,
"text": "Bellegarda et al., 1996;",
"ref_id": "BIBREF1"
},
{
"start": 234,
"end": 250,
"text": "Bai et al., 1998",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "The discussion of the ACM in this paper is an extension of several studies below. The first similar cluster model was presented by Goodman and Gao [2000] in which the clustering techniques were combined with Stolcke's [1998] pruning to reduce the language model (LM) size effectively. Goodman [2001] and Gao et al, [2001] give detailed descriptions of the asymmetric clustering algorithm. However, the impact of the asymmetric clustering on the performance of the resulting cluster model was not empirically studied there. Gao et al., [2001] presented a fairly thorough empirical study of clustering techniques for Asian language modeling. Unfortunately, all of the above work studied the ACM without applying it to an application; thus only perplexity results were presented. The first real application of the ACM was a simplified bigram ACM used in a Chinese text input system [Gao et al. 2002] . However, quite a few techniques (including clustering) were integrated to construct a Chinese language modeling system, and the contribution of using the ACM alone was by no means completely investigated.",
"cite_spans": [
{
"start": 131,
"end": 153,
"text": "Goodman and Gao [2000]",
"ref_id": "BIBREF8"
},
{
"start": 208,
"end": 224,
"text": "Stolcke's [1998]",
"ref_id": "BIBREF16"
},
{
"start": 285,
"end": 299,
"text": "Goodman [2001]",
"ref_id": "BIBREF7"
},
{
"start": 304,
"end": 321,
"text": "Gao et al, [2001]",
"ref_id": "BIBREF5"
},
{
"start": 523,
"end": 541,
"text": "Gao et al., [2001]",
"ref_id": "BIBREF5"
},
{
"start": 879,
"end": 896,
"text": "[Gao et al. 2002]",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Finally, there is one more point worth mentioning. Most language modeling improvements reported previously required significantly more space than word trigram models [Rosenfeld, 2000] . Their practical value is questionable since all realistic applications have memory constraints. In this paper, our goal is to achieve a better tradeoff between LM performance (perplexity and CER) and model size. Thus, whenever we compare the performance of different models (i.e. ACM vs. word trigram model), Stolcke's pruning is employed to bring the models compared to similar sizes.",
"cite_spans": [
{
"start": 166,
"end": 183,
"text": "[Rosenfeld, 2000]",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "The LM predicts the next word w i given its history h by estimating the conditional probability P(w i |h). Using the trigram approximation, we have P(w i |h)\u2248P(w i |w i-2 w i-1 ), assuming that the next word depends only on the two preceding words.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Model",
"sec_num": "3.1"
},
{
"text": "In the ACM, we will use different clusters for words in different positions. For the predicted word, w i , we will denote the cluster of the word by PW i , and we will refer to this as the predictive cluster. . For the words w i-2 and w i-1 that we are conditioning on, we will denote their clusters by CW i-2 and CW i-1 which we call conditional clusters. When we which to refer to a cluster of a word w in general we will use the notation W. The ACM estimates the probability of w i given the two preceeding words w i-2 and w i-1 as the product of the following two probabilities:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Model",
"sec_num": "3.1"
},
{
"text": "(1) The probability of the predicted cluster PW i given the preceding conditional clusters CW i-2 and CW i-1 , P(PW i |CW i-2 CW i-1 ), and (2) The probability of the word given its cluster PW i and the preceding conditional clusters CW i-2 and CW i-1 , P(w i |CW i-2 CW i-1 PW i ). Thus, the ACM can be parameterized by",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Model",
"sec_num": "3.1"
},
{
"text": ") | ( ) | ( ) | ( 1 2 1 2 i i i i i i i i PW CW CW w P CW CW PW P h w P \u2212 \u2212 \u2212 \u2212 \u00d7 \u2248 (1)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Model",
"sec_num": "3.1"
},
{
"text": "The ACM consists of two sub-models: (1) the cluster sub-model P(PW i |CW i-2 CW i-1 ), and (2) the word sub-model P(w i |CW i-2 CW i-1 PW i ). To deal with the data sparseness problem, we used a backoff scheme (Katz, 1987) for the parameter estimation of each sub-model. The backoff scheme recursively estimates the probability of an unseen n-gram by utilizing (n-1)-gram estimates.",
"cite_spans": [
{
"start": 210,
"end": 222,
"text": "(Katz, 1987)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Model",
"sec_num": "3.1"
},
{
"text": "The basic idea underlying the ACM is the use of different clusters for predicted and conditional words respectively. Classical cluster models are symmetric in that the same clusters are employed for both predicted and conditional words. However, the symmetric cluster model is suboptimal in practice. For example, consider a pair of words like \"a\" and \"an\". In general, \"a\" and \"an\" can follow the same words, and thus, as predicted words, belong in the same cluster. But, there are very few words that can follow both \"a\" and \"an\". So as conditional words, they belong in different clusters.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Model",
"sec_num": "3.1"
},
{
"text": "In generating clusters, two factors need to be considered: (1) clustering metrics, and (2) cluster numbers. In what follows, we will investigate the impact of each of the factors.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Model",
"sec_num": "3.1"
},
{
"text": "The basic criterion for statistical clustering is to maximize the resulting probability (or minimize the resulting perplexity) of the training data. Many traditional clustering techniques [Brown et al., 1992] attempt to maximize the average mutual information of adjacent clusters",
"cite_spans": [
{
"start": 188,
"end": 208,
"text": "[Brown et al., 1992]",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "\u2211 = 2 1 , 2 1 2 2 1 2 1 ) ( ) | ( log ) ( ) , ( W W W P W W P W W P W W I ,",
"eq_num": "(2)"
}
],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "where the same clusters are used for both predicted and conditional words. We will call these clustering techniques symmetric clustering, and the resulting clusters both clusters. In constructing the ACM, we used asymmetric clustering, in which different clusters are used for predicted and conditional words. In particular, for clustering conditional words, we try to minimize the perplexity of training data for a bigram of the form",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "P(w i |W i-1 ), which is equivalent to maximizing \u220f = \u2212 N i i i W w P 1 1 ) | ( .",
"eq_num": "(3)"
}
],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "where N is the total number of words in the training data. We will call the resulting clusters conditional clusters denoted by CW. For clustering predicted words, we try to minimize the perplexity of training data of P(W i |w i-1 )\u00d7P(w i |W i ). We will call the resulting clusters predicted clusters denoted by PW.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "We have 2",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "\u220f \u220f = \u2212 \u2212 = \u2212 \u00d7 = \u00d7 N i i i i i i i N i i i i i W P w W P w P W w P W w P w W P 1 1 1 1 1 ) ( ) ( ) ( ) ( ) | ( ) | ( \u220f = \u2212 \u2212 \u00d7 = N i i i i i i i W P W w P w P w W P 1 1 1 ) ( ) ( ) ( ) ( \u220f = \u2212 \u2212 \u00d7 = N i i i i i W w P w P w P 1 1 1 ) | ( ) ( ) ( . Now, ) ( ) ( 1 \u2212 i i w P w P",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "is independent of the clustering used.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "Therefore, for the selection of the best clusters, it is sufficient to try to maximize",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "\u220f = \u2212 N i i i W w P 1 1 ) | ( .",
"eq_num": "(4)"
}
],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "This is very convenient since it is exactly the opposite of what was done for conditional clustering. It means that we can use the same clustering tool for both, and simply switch the order used by the program used to get the raw counts for clustering. The clustering technique we used creates a binary branching tree with words at the leaves. The ACM in this study is a hard cluster model, meaning that each word belongs to only one cluster. So in the clustering tree, each word occurs in a single leaf. In the ACM, we actually use two different clustering trees. One is optimized for predicted words, and the other for conditional words.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "The basic approach to clustering we used is a top-down, splitting clustering algorithm. In each iteration, a cluster is split into two clusters in the way that the splitting achieves the maximal entropy decrease (estimated by Equations (3) or (4)). Finally, we can also perform iterations of swapping all words between all clusters until convergence i.e. no more entropy decrease can be found 3 . We find that our algorithm is much more efficient than agglomerative clustering algorithms -those which merge words bottom up.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Asymmetric clustering",
"sec_num": "3.2"
},
{
"text": "Asymmetric clustering results in two binary clustering trees. By cutting the trees at a certain level, it is possible to achieve a wide variety of different numbers of clusters. For instance, if the tree is cut after the 8 th level, there will be roughly 2 8 =256 clusters. Since the tree is not balanced, the actual number of clusters may be somewhat smaller. We use W l to represent the cluster of a word w using a tree cut at level l. In particular, if we set l to the value \"all\", it means that the tree is cut at infinite depth, i.e. each cluster contains a single word. The ACM model of Equation (1) can be rewritten as",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "P(PW i l |CW i-2 j CW i-1 j )\u00d7P(w i |PW i-2 k CW i-1 k CW i l ). (5)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "To optimally apply the ACM to realistic applications with memory constraints, we are always seeking the correct balance between model size and performance. We used Stolcke's pruning method to produce many ACMs with different model sizes. In our experiments, whenever we compare techniques, we do so by comparing the performance (perplexity and CER) of the LM techniques at the same model sizes. Stolcke's pruning is an entropy-based cutoff method, which can be described as follows: all n-grams that change perplexity by less than a threshold are removed from the model. For pruning the ACM, we have two thresholds: one for the cluster sub-model",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "P(PW i l |CW i-2 j CW i-1 j",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": ") and one for the word sub-model",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "P(w i |CW i-2 k CW i-1 k PW i l )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "respectively, denoted by t c and t w below.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "In this way, we have 5 different parameters that need to be simultaneously optimized: l, j, k, t c , and t w , where j, k, and l are the numbers of clusters, and t c and t w are the pruning thresholds.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "A brute-force approach to optimizing such a large number of parameters is prohibitively expensive. Rather than trying a large number of combinations of all 5 parameters, we give an alternative technique that is significantly more efficient. Simple math shows that the perplexity of the overall model",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "P(PW i l |CW i-2 j CW i-1 j )\u00d7 P(w i |CW i-2 k CW i-1 k PW i l )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "is equal to the perplexity of the cluster sub-model",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "P(PW i l |CW i-2 j CW i-1 j ) times the perplexity of the word sub-model P(w i |CW i-2 k CW i-1 k PW i l )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": ". The size of the overall model is clearly the sum of the sizes of the two sub-models. Thus, we try a large number of values of j, l, and a pruning threshold t c for P",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "(PW i l |CW i-2 j CW i-1 j )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": ", computing sizes and perplexities of each, and a similarly large number of values of l, k, and a separate threshold t w for",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "P(w i |CW i-2 k CW i-1 k PW i l )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": ". We can then look at all compatible pairs of these models (those with the same value of l) and quickly compute the perplexity and size of the overall models. This allows us to relatively quickly search through what would otherwise be an overwhelmingly large search space.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter optimization",
"sec_num": "3.3"
},
{
"text": "Japanese Kana-Kanji conversion is the standard method of inputting Japanese text by converting a syllabary-based Kana string into the appropriate combination of ideographic Kanji and Kana. This is a similar problem to speech recognition, except that it does not include acoustic ambiguity. The performance is generally measured in terms of character error rate (CER), which is the number of characters wrongly converted from the phonetic string divided by the number of characters in the correct transcript. The role of the language model is, for all possible word strings that match the typed phonetic symbol string, to select the word string with the highest language model probability. Current products make about 5-10% errors in conversion of real data in a wide variety of domains.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Japanese Kana-Kanji Conversion Task",
"sec_num": "4.1"
},
{
"text": "In the experiments, we used two Japanese newspaper corpora: the Nikkei Newspaper corpus, and the Yomiuri Newspaper corpus. Both text corpora have been word-segmented using a lexicon containing 167,107 entries.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Settings",
"sec_num": "4.2"
},
{
"text": "We performed two sets of experiments: (1) pilot experiments, in which model performance is measured in terms of perplexity and (2) Japanese Kana-Kanji conversion experiments, in which the performance of which is measured in terms of CER.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Settings",
"sec_num": "4.2"
},
{
"text": "In the pilot experiments, we used a subset of the Nikkei newspaper corpus: ten million words of the Nikkei corpus for language model training, 10,000 words for held-out data, and 20,000 words for testing data. None of the three data sets overlapped. In the Japanese Kana-Kanji conversion experiments, we built language models on a subset of the Nikkei Newspaper corpus, which contains 36 million words. We performed parameter optimization on a subset of held-out data from the Yomiuri Newspaper corpus, which contains 100,000 words. We performed testing on another subset of the Yomiuri Newspaper corpus, which contains 100,000 words. In both sets of experiments, word clusters were derived from bigram counts generated from the training corpora. Out-of-vocabulary words were not included in perplexity and error rate computations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Settings",
"sec_num": "4.2"
},
{
"text": "As described in Section 3.2, depending on the clustering metrics we chose for generating clusters, we obtained three types of clusters: both clusters (the metric of Equation (2)), conditional clusters (the metric of Equation (3)), and predicted clusters (the metric of Equation (4)). We then performed a series of experiments to investigate the impact of different types of clusters on the ACM. We used three variants of the trigram ACM: (1) the predictive cluster model",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of asymmetric clustering",
"sec_num": "4.3"
},
{
"text": "P(w i |w i-2 w i-1 W i )\u00d7 P(W i |w i-2 w i-1 )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of asymmetric clustering",
"sec_num": "4.3"
},
{
"text": "where only predicted words are clustered, (2) the conditional cluster model P(w i |W i-2 W i-1 ) where only conditional words are clustered, and (3) the IBM model P(w i |W i )\u00d7 P(W i |W i-2 W i-1 ) which can be treated as a special case of the ACM of Equation (5) by using the same type of cluster for both predicted and conditional words, and setting k = 0, and l = j. For each cluster trigram model, we compared their perplexities and CER results on Japanese Kana-Kanji conversion using different types of clusters. For each cluster type, the number of clusters were fixed to the same value 2^6 just for comparison. The results are shown in Table 1 . It turns out that the benefit of using different clusters in different positions is obvious. For each cluster trigram model, the best results were achieved by using the \"matched\" clusters, e.g. the predictive cluster model P(w i |w i-2 w i-1 W i )\u00d7 P(W i |w i-2 w i-1 ) has the best performance when the cluster W i is the predictive cluster PW i generated by using the metric of Equation (4). In particular, the IBM model achieved the best results when predicted and conditional clusters were used for predicted and conditional words respectively. That is, the IBM model is of the form P(w i |PW i )\u00d7 P(PW i |CW i-2 CW i-1 ). ",
"cite_spans": [],
"ref_spans": [
{
"start": 643,
"end": 650,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Impact of asymmetric clustering",
"sec_num": "4.3"
},
{
"text": "In this section, we first present our pilot experiments of finding the optimal parameter set of the ACM (l, j, k, t c , t w ) described in Section 2.3. Then, we compare the ACM to the IBM model, showing that the superiority of the ACM results from its better structure. In this section, the performance of LMs was measured in terms of perplexity, and the size was measured as the total number of parameters of the LM: one parameter for each bigram and trigram, one parameter for each normalization parameter \u03b1 that was needed, and one parameter for each unigram.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "We first used the conditional cluster model of the form",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "P(w i |CW i-2 j CW i-1 j )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": ". Some sample settings of parameters (j, t w ) are shown in Figure 1 . The performance was consistently improved by increasing the number of clusters j, except at the smallest sizes. The word trigram model was consistently the best model, except at the smallest sizes, and even then was only marginally worse than the conditional cluster models. This is not surprising because the conditional cluster model always discards information for predicting words.",
"cite_spans": [],
"ref_spans": [
{
"start": 60,
"end": 68,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "We then used the predictive cluster model of the form P",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "(PW i l |w i-2 w i-1 )\u00d7P(w i |w i-2 w i-1 PW i l )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": ", where only predicted words are clustered. Some sample settings of the parameters (l, t c , t w ) are shown in Figure 2 . For simplicity, we assumed t c =t w , meaning that the same pruning threshold values were used for both sub-models. It turns out that predictive cluster models achieve the best perplexity results at about 2^6 or 2^8 clusters. The models consistently outperform the baseline word trigram models.",
"cite_spans": [],
"ref_spans": [
{
"start": 112,
"end": 120,
"text": "Figure 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "We finally returned to the ACM of Equation 5, where both conditional words and the predicted word are clustered (with different numbers of clusters), and which is referred to as the combined cluster model below. In addition, we allow different values of the threshold for different sub-models. Therefore, we need to optimize the model parameter set l, j, k, t c , t w .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "Based on the pilot experiment results using conditional and predictive cluster models, we tried combined cluster models for values l \u2208 [4, 10], j, k\u2208 [8, 16] . We also allow j, k=all. Rather than plot all points of all models together, we show only the outer envelope of the points. That is, if for a given model type and a given point there is some other point of the same type with both lower perplexity and smaller size than the first point, then we do not plot the first, worse point.",
"cite_spans": [
{
"start": 150,
"end": 153,
"text": "[8,",
"ref_id": null
},
{
"start": 154,
"end": 157,
"text": "16]",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "The results are shown in Figure 3 , where the cluster number of IBM models is 2^14 which achieves the best performance for IBM models in our experiments. It turns out that when l\u2208 [6, 8] and j, k>12, combined cluster models yield the best results. We also found that the predictive cluster models give as good performance as the best combined ones while combined models outperformed very slightly only when model sizes are small. This is not difficult to explain. Recall that the predictive cluster model is a special case of the combined model where words are used in conditional positions, i.e. j=k=all. Our experiments show that combined models achieved good performance when large numbers of clusters are used for conditional words, i.e. large j, k>12, which are similar to words.",
"cite_spans": [
{
"start": 180,
"end": 183,
"text": "[6,",
"ref_id": null
},
{
"start": 184,
"end": 186,
"text": "8]",
"ref_id": null
}
],
"ref_spans": [
{
"start": 25,
"end": 33,
"text": "Figure 3",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "The most interesting analysis is to look at some sample settings of the parameters of the combined cluster models in Figure 3 . In Table 2 , we show the best parameter settings at several levels of model size. Notice that in larger model sizes, predictive cluster models (i.e. j=k=all) perform the best in some cases. The 'prune' columns (i.e. columns 6 and 7) indicate the Stolcke pruning parameter we used.",
"cite_spans": [],
"ref_spans": [
{
"start": 117,
"end": 125,
"text": "Figure 3",
"ref_id": "FIGREF0"
},
{
"start": 131,
"end": 138,
"text": "Table 2",
"ref_id": "TABREF4"
}
],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "First, notice that the two pruning parameters (in columns 6 and 7) tend to be very similar. This is desirable since applying the theory of relative entropy pruning predicts that the two pruning parameters should actually have the same value.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "Next, let us compare the ACM same type of cluster is used for both predictive and conditional words). Our results in Figure 3 show that the performance of IBM models is roughly an order of magnitude worse than that of ACMs. This is because in addition to the use of the symmetric cluster model, the traditional IBM model makes two more assumptions that we consider suboptimal. First, it assumes that j=l. We see that the best results come from unequal settings of j and l. Second, more importantly, IBM clustering assumes that k=0. We see that not only is the optimal setting for k not 0, but also typically the exact opposite is the optimal: k=all in which case",
"cite_spans": [],
"ref_spans": [
{
"start": 117,
"end": 125,
"text": "Figure 3",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "P(PW i l |CW i-2 j CW i-1 j )\u00d7P(w i |CW i-2 k CW i-1 k PW i l ) to traditional IBM clustering of the form P(W i l |W i-2 l W i-1 l )\u00d7P(w i |W i l ), which is equal to P(W i l |W i-2 l W i-1 l )\u00d7P(w i |W i-2 0 W i-1 0 W i l )",
"eq_num": "(assuming"
}
],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "P(w i |CW i-2 k CW i-1 k PW i l )= P(w i |w i-2 w i-1 PW i l )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": ", or k=14, 16, which is very similar. That is, we see that words depend on the previous words and that an independence assumption is a poor one. Of course, many of these word dependencies are pruned away -but when a word does depend on something, the previous words are better predictors than the previous clusters.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "Another important finding here is that for most of these settings, the unpruned model is actually larger than a normal trigram model -whenever k=all or 14, 16, the unpruned model",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "P(PW i l |CW i-2 j CW i-1 j ) \u00d7 P(w i |CW i-2 k CW i-1 k PW i l",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": ") is actually larger than an unpruned model P(w i |w i-2 w i-1 ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "This analysis of the data is very interesting -it implies that the gains from clustering are not from compression, but rather from capturing structure. Factoring the model into two models, in which the cluster is predicted first, and then the word is predicted given the cluster, allows the structure and regularities of the model to be found. This larger, better structured model can be pruned more effectively, and it achieved better performance than a word trigram model at the same model size. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Impact of parameter optimization",
"sec_num": "4.4"
},
{
"text": "Before we present CER results of the Japanese Kana-Kanji conversion system, we briefly describe our method for storing the ACM in practice. One of the most common methods for storing backoff n-gram models is to store n-gram probabilities (and backoff weights) in a tree structure, which begins with a hypothetical root node that branches out into unigram nodes at the first level of the tree, and each of those unigram nodes in turn branches out into bigram nodes at the second level and so on. To save storage, n-gram probabilities such as P(w i |w i-1 ) and backoff weights such as \u03b1(w i-2 w i-1 ) are stored in a single (bigram) node array (Clarkson and Rosenfeld, 1997) .",
"cite_spans": [
{
"start": 643,
"end": 673,
"text": "(Clarkson and Rosenfeld, 1997)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "CER results",
"sec_num": "4.5"
},
{
"text": "Applying the above tree structure to storing the ACM is a bit complicated -there are some representation issues. ) cannot be stored in a single (bigram) node array, because l \u2260 j and PW\u2260CW. Therefore, we used two separate trees to store probabilities and backoff weights, respectively. As a result, we used four tree structures to store ACMs in practice: two for the cluster sub-model P(PW i",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "CER results",
"sec_num": "4.5"
},
{
"text": "l |CW i-2 j CW i-1 j )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "CER results",
"sec_num": "4.5"
},
{
"text": ", and two for the word sub-model",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "CER results",
"sec_num": "4.5"
},
{
"text": "P(w i |CW i-2 k CW i-1 k PW i l )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "CER results",
"sec_num": "4.5"
},
{
"text": ". We found that the effect of the storage structure cannot be ignored in a real application.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "CER results",
"sec_num": "4.5"
},
{
"text": "In addition, we used several techniques to compress model parameters (i.e. word id, n-gram probability, and backoff weight, etc.) and reduce the storage space of models significantly. For example, rather than store 4-byte floating point values for all n-gram probabilities and backoff weights, the values are quantized to a small number of quantization levels. Quantization is performed separately on each of the n-gram probability and backoff weight lists, and separate quantization level look-up tables are generated for each of these sets of parameters. We used 8-bit quantization, which shows no performance decline in our experiments.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "CER results",
"sec_num": "4.5"
},
{
"text": "Our goal is to achieve the best tradeoff between performance and model size. Therefore, we would like to compare the ACM with the word trigram model at the same model size. Unfortunately, the ACM contains four sub-models and this makes it difficult to be pruned to a specific size. Thus for comparison, we always choose the ACM with smaller size than its competing word trigram model to guarantee that our evaluation is under-estimated. Experiments show that the ACMs achieve statistically significant improvements over word trigram models at even smaller model sizes (p-value =8.0E-9). Some results are shown in Table 3 Table 3 : CER results of ACMs and word trigram models at different model sizes Now we discuss why the ACM is superior to simple word trigrams. In addition to the better structure as shown in Section 3.3, we assume here that the benefit of our model also comes from its better smoothing. Consider a probability such as P(Tuesday| party on). If we put the word \"Tuesday\" into the cluster WEEKDAY, we decompose the probability When each word belongs to one class, simple math shows that this decomposition is a strict equality. However, when smoothing is taken into consideration, using the clustered probability will be more accurate than using the non-clustered probability. For instance, even if we have never seen an example of \"party on Tuesday\", perhaps we have seen examples of other phrases, such as \"party on Wednesday\"; thus, the probability P(WEEKDAY | party on) will be relatively high. Furthermore, although we may never have seen an example of \"party on WEEKDAY Tuesday\", after we backoff or interpolate with a lower order model, we may able to accurately estimate P(Tuesday | on WEEKDAY). Thus, our smoothed clustered estimate may be a good one.",
"cite_spans": [],
"ref_spans": [
{
"start": 613,
"end": 620,
"text": "Table 3",
"ref_id": null
},
{
"start": 621,
"end": 628,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "CER results",
"sec_num": "4.5"
},
{
"text": "Our assumption can be tested empirically by following experiments. We first constructed several test sets with different backoff rates 4 . The backoff rate of a test set, when presented to a trigram model, is defined as the number of words whose trigram probabilities are estimated by backoff bigram probabilities divided by the number of words in the test set. Then for each test set, we obtained a pair of CER results using the ACM and the word trigram model respectively. As shown in Figure 4 , in both cases, CER increases as the backoff rate increases from 28% to 40%. But the curve of the word trigram model has a steeper upward trend. The difference of the upward trends of the two curves can be shown more clearly by plotting the CER difference between them, as shown in Figure 5 . The results indicate that because of its better smoothing, when the backoff rate increases, the CER using the ACM does not increase as fast as that using the word trigram model. Therefore, we are reasonably confident that some portion of the benefit of the ACM comes from its better smoothing. Figure 5 : CER difference vs. backoff rate.",
"cite_spans": [],
"ref_spans": [
{
"start": 487,
"end": 495,
"text": "Figure 4",
"ref_id": null
},
{
"start": 779,
"end": 787,
"text": "Figure 5",
"ref_id": null
},
{
"start": 1084,
"end": 1092,
"text": "Figure 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "CER results",
"sec_num": "4.5"
},
{
"text": "There are three main contributions of this paper. First, after presenting a formal definition of the ACM, we described in detail the methodology of constructing the ACM effectively. We showed empirically that both the asymmetric clustering and the parameter optimization (i.e. optimal cluster numbers) have positive impacts on the performance of the resulting ACM. The finding demonstrates partially the effectiveness of our research focus: techniques for using clusters (i.e. the ACM) rather than techniques for finding clusters (i.e. clustering algorithms). Second, we explored the actual representation of the ACM and evaluate it on a realistic application -Japanese Kana-Kanji conversion. Results show approximately 6-10% CER reduction of the ACMs in comparison with the word trigram models, even when the ACMs are slightly smaller. Third, the reasons underlying the superiority of the ACM are analyzed. For instance, our analysis suggests the benefit of the ACM comes partially from its better structure and its better smoothing.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "All cluster models discussed in this paper are based on hard clustering, meaning that each word belongs to only one cluster. One area we have not explored is the use of soft clustering, where a word w can be assigned to multiple clusters W with a probability P(W|w) [Pereira et al., 1993] . Saul and Pereira [1997] demonstrated the utility of soft clustering and concluded that any method that assigns each word to a single cluster would lose information. It is an interesting question whether our techniques for hard clustering can be extended to soft clustering. On the other hand, soft clustering models tend to be larger than hard clustering models because a given word can belong to multiple clusters, and thus a training instance P(w i |w i-2 w i-1 ) can lead to multiple counts instead of just 1.",
"cite_spans": [
{
"start": 266,
"end": 288,
"text": "[Pereira et al., 1993]",
"ref_id": "BIBREF13"
},
{
"start": 291,
"end": 314,
"text": "Saul and Pereira [1997]",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "Thanks to Lillian Lee for suggesting this justification of predictive clusters.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Notice that for experiments reported in this paper, we used the basic top-down algorithm without swapping. Although the resulting clusters without swapping are not even locally optimal, our experiments show that the quality of clusters (in terms of the perplexity of the resulting ACM) is not inferior to that of clusters with swapping.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "The backoff rates are estimated using the baseline trigram model, so the choice could be biased against the word trigram model.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Building class-based language models with contextual statistics",
"authors": [
{
"first": "S",
"middle": [],
"last": "Bai",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Z",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Yuan",
"suffix": ""
}
],
"year": 1998,
"venue": "ICASSP-98",
"volume": "",
"issue": "",
"pages": "173--176",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bai, S., Li, H., Lin, Z., and Yuan, B. (1998). Building class-based language models with contextual statistics. In ICASSP-98, pp. 173-176.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "A novel word clustering algorithm based on latent semantic analysis",
"authors": [
{
"first": "J",
"middle": [
"R"
],
"last": "Bellegarda",
"suffix": ""
},
{
"first": "J",
"middle": [
"W"
],
"last": "Butzberger",
"suffix": ""
},
{
"first": "Y",
"middle": [
"L"
],
"last": "Chow",
"suffix": ""
},
{
"first": "N",
"middle": [
"B"
],
"last": "Coccaro",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Naik",
"suffix": ""
}
],
"year": 1996,
"venue": "ICASSP-96",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bellegarda, J. R., Butzberger, J. W., Chow, Y. L., Coccaro, N. B., and Naik, D. (1996). A novel word clustering algorithm based on latent semantic analysis. In ICASSP-96.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "A statistical approach to machine translation",
"authors": [
{
"first": "P",
"middle": [
"F"
],
"last": "Brown",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Cocke",
"suffix": ""
},
{
"first": "S",
"middle": [
"A"
],
"last": "Dellapietra",
"suffix": ""
},
{
"first": "V",
"middle": [
"J"
],
"last": "Dellapietra",
"suffix": ""
},
{
"first": "F",
"middle": [],
"last": "Jelinek",
"suffix": ""
},
{
"first": "J",
"middle": [
"D"
],
"last": "Lafferty",
"suffix": ""
},
{
"first": "R",
"middle": [
"L"
],
"last": "Mercer",
"suffix": ""
},
{
"first": "P",
"middle": [
"S"
],
"last": "Roossin",
"suffix": ""
}
],
"year": 1990,
"venue": "Computational Linguistics",
"volume": "16",
"issue": "2",
"pages": "79--85",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Brown, P. F., Cocke, J., DellaPietra, S. A., DellaPietra, V. J., Jelinek, F., Lafferty, J. D., Mercer, R. L., and Roossin, P. S. (1990). A statistical approach to machine translation. Computational Linguistics, 16(2), pp. 79-85.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Class-based n-gram models of natural language",
"authors": [
{
"first": "P",
"middle": [
"F"
],
"last": "Brown",
"suffix": ""
},
{
"first": "V",
"middle": [
"J"
],
"last": "Dellapietra",
"suffix": ""
},
{
"first": "P",
"middle": [
"V"
],
"last": "Desouza",
"suffix": ""
},
{
"first": "J",
"middle": [
"C"
],
"last": "Lai",
"suffix": ""
},
{
"first": "R",
"middle": [
"L"
],
"last": "Mercer",
"suffix": ""
}
],
"year": 1992,
"venue": "Computational Linguistics",
"volume": "18",
"issue": "4",
"pages": "467--479",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Brown, P. F., DellaPietra V. J., deSouza, P. V., Lai, J. C., and Mercer, R. L. (1992). Class-based n-gram models of natural language. Computational Linguistics, 18(4), pp. 467-479.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Statistical language modeling using the CMU-Cambridge toolkit",
"authors": [
{
"first": "P",
"middle": [
"R"
],
"last": "Clarkson",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Rosenfeld",
"suffix": ""
}
],
"year": 1997,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Clarkson, P. R., and Rosenfeld, R. (1997). Statistical language modeling using the CMU-Cambridge toolkit. In Eurospeech 1997, Rhodes, Greece.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "The use of clustering techniques for language model -application to Asian language",
"authors": [
{
"first": "J",
"middle": [],
"last": "Gao",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Goodman",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Miao",
"suffix": ""
}
],
"year": 2001,
"venue": "Computational Linguistics and Chinese Language Processing",
"volume": "6",
"issue": "1",
"pages": "27--60",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gao, J. Goodman, J. and Miao, J. (2001). The use of clustering techniques for language model -application to Asian language. Computational Linguistics and Chinese Language Processing. Vol. 6, No. 1, pp 27-60.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Toward a unified approach to statistical language modeling for Chinese",
"authors": [
{
"first": "J",
"middle": [],
"last": "Gao",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Goodman",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "K",
"middle": [
"F"
],
"last": "Lee",
"suffix": ""
}
],
"year": 2002,
"venue": "ACM Transactions on Asian Language Information Processing",
"volume": "1",
"issue": "1",
"pages": "3--33",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gao, J., Goodman, J., Li, M., and Lee, K. F. (2002). Toward a unified approach to statistical language modeling for Chinese. ACM Transactions on Asian Language Information Processing. Vol. 1, No. 1, pp 3-33.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "A bit of progress in language modeling",
"authors": [
{
"first": "J",
"middle": [],
"last": "Goodman",
"suffix": ""
}
],
"year": 2001,
"venue": "Computer Speech and Language",
"volume": "",
"issue": "",
"pages": "403--434",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Goodman, J. (2001). A bit of progress in language modeling. In Computer Speech and Language, October 2001, pp 403-434.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Language model size reduction by predictive clustering",
"authors": [
{
"first": "J",
"middle": [],
"last": "Goodman",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Gao",
"suffix": ""
}
],
"year": 2000,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Goodman, J., and Gao, J. (2000). Language model size reduction by predictive clustering. ICSLP-2000, Beijing.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Self-organized language modeling for speech recognition",
"authors": [
{
"first": "F",
"middle": [],
"last": "Jelinek",
"suffix": ""
}
],
"year": 1990,
"venue": "Readings in Speech Recognition",
"volume": "",
"issue": "",
"pages": "450--506",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jelinek, F. (1990). Self-organized language modeling for speech recognition. In Readings in Speech Recognition, A. Waibel and K. F. Lee, eds., Morgan-Kaufmann, San Mateo, CA, pp. 450-506.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Estimation of probabilities from sparse data for the language model component of a speech recognizer",
"authors": [
{
"first": "S",
"middle": [
"M"
],
"last": "Katz",
"suffix": ""
}
],
"year": 1987,
"venue": "IEEE Transactions on Acoustics, Speech and Signal Processing",
"volume": "35",
"issue": "3",
"pages": "400--401",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Katz, S. M. (1987). Estimation of probabilities from sparse data for the language model component of a speech recognizer. IEEE Transactions on Acoustics, Speech and Signal Processing, ASSP-35(3):400-401, March.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Improved clustering techniques for class-based statistical language modeling",
"authors": [
{
"first": "R",
"middle": [],
"last": "Kneser",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Ney",
"suffix": ""
}
],
"year": 1993,
"venue": "Eurospeech",
"volume": "2",
"issue": "",
"pages": "973--976",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kneser, R. and Ney, H. (1993). Improved clustering techniques for class-based statistical language modeling. In Eurospeech, Vol. 2, pp. 973-976, Berlin, Germany.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "On structuring probabilistic dependences in stochastic language modeling",
"authors": [
{
"first": "H",
"middle": [],
"last": "Ney",
"suffix": ""
},
{
"first": "U",
"middle": [],
"last": "Essen",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Kneser",
"suffix": ""
}
],
"year": 1994,
"venue": "Computer, Speech, and Language",
"volume": "8",
"issue": "",
"pages": "1--38",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ney, H., Essen, U., and Kneser, R. (1994). On structuring probabilistic dependences in stochastic language modeling. Computer, Speech, and Language, 8:1-38.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Distributional clustering of English words",
"authors": [
{
"first": "F",
"middle": [],
"last": "Pereira",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Tishby",
"suffix": ""
},
{
"first": "Lee",
"middle": [
"L"
],
"last": "",
"suffix": ""
}
],
"year": 1993,
"venue": "Proceedings of the 31 st Annual Meeting of the ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Pereira, F., Tishby, N., and Lee L. (1993). Distributional clustering of English words. In Proceedings of the 31 st Annual Meeting of the ACL.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Two decades of statistical language modeling: where do we go from here",
"authors": [
{
"first": "R",
"middle": [],
"last": "Rosenfeld",
"suffix": ""
}
],
"year": 2000,
"venue": "Proceeding of the IEEE",
"volume": "88",
"issue": "",
"pages": "1270--1278",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rosenfeld, R. (2000). Two decades of statistical language modeling: where do we go from here. In Proceeding of the IEEE, 88:1270-1278, August.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Aggregate and mixed-order Markov models for statistical language processing",
"authors": [
{
"first": "L",
"middle": [],
"last": "Saul",
"suffix": ""
},
{
"first": "F",
"middle": [
"C N"
],
"last": "Pereira",
"suffix": ""
}
],
"year": 1997,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Saul, L., and Pereira, F.C.N. (1997). Aggregate and mixed-order Markov models for statistical language processing. In EMNLP-1997.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Entropy-based Pruning of Backoff Language Models",
"authors": [
{
"first": "A",
"middle": [],
"last": "Stolcke",
"suffix": ""
}
],
"year": 1998,
"venue": "Proc. DARPA News Transcription and Understanding Workshop",
"volume": "",
"issue": "",
"pages": "270--274",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Stolcke, A. (1998). Entropy-based Pruning of Backoff Language Models. Proc. DARPA News Transcription and Understanding Workshop, 1998, pp. 270-274.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "An extended clustering algorithm for statistical language models",
"authors": [
{
"first": "J",
"middle": [
"P"
],
"last": "Ueberla",
"suffix": ""
}
],
"year": 1996,
"venue": "IEEE Transactions on Speech and Audio Processing",
"volume": "4",
"issue": "4",
"pages": "313--316",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ueberla, J. P. (1996). An extended clustering algorithm for statistical language models. IEEE Transactions on Speech and Audio Processing, 4(4): 313-316.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Multi-Class Composite N-gram Language Model for Spoken Language Processing Using Multiple Word Clusters",
"authors": [
{
"first": "H",
"middle": [],
"last": "Yamamoto",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Isogai",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Sagisaka",
"suffix": ""
}
],
"year": 2001,
"venue": "39 th Annual meetings of the Association for Computational Linguistics (ACL'01)",
"volume": "",
"issue": "",
"pages": "6--11",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yamamoto, H., Isogai, S., and Sagisaka, Y. (2001). Multi-Class Composite N-gram Language Model for Spoken Language Processing Using Multiple Word Clusters. 39 th Annual meetings of the Association for Computational Linguistics (ACL'01), Toulouse, 6-11 July 2001.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Multi-class Composite N-gram based on Connection Direction",
"authors": [
{
"first": "H",
"middle": [],
"last": "Yamamoto",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Sagisaka",
"suffix": ""
}
],
"year": 1999,
"venue": "Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yamamoto, H., and Sagisaka, Y. (1999). Multi-class Composite N-gram based on Connection Direction, In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, May, Phoenix, Arizona.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"type_str": "figure",
"uris": null,
"num": null,
"text": "Comparison of ACMs, predictive cluster model, IBM model, and word trigram model"
},
"FIGREF1": {
"type_str": "figure",
"uris": null,
"num": null,
"text": "For example, consider the cluster sub-model P(PW i l |CW i-2 j CW i-1 j ). N-gram probabilities such as P(PW i l |CW i-1 j ) and backoff weights such as \u03b1(CW i-2 j CW i-1 j"
},
"TABREF1": {
"type_str": "table",
"html": null,
"text": "",
"num": null,
"content": "<table/>"
},
"TABREF4": {
"type_str": "table",
"html": null,
"text": "Sample parameter settings for the ACM",
"num": null,
"content": "<table/>"
},
"TABREF5": {
"type_str": "table",
"html": null,
"text": ".",
"num": null,
"content": "<table><tr><td colspan=\"2\">Word trigram model</td><td/><td>ACM</td><td/></tr><tr><td>Size</td><td>CER</td><td>Size</td><td>CER</td><td>CER</td></tr><tr><td>(MB)</td><td/><td>(MB)</td><td/><td>Reduction</td></tr><tr><td>1.8</td><td>4.56%</td><td>1.7</td><td>4.25%</td><td>6.8%</td></tr><tr><td>5.8</td><td>4.08%</td><td>4.5</td><td>3.83%</td><td>6.1%</td></tr><tr><td>11.7</td><td>4.04%</td><td>10.7</td><td>3.73%</td><td>7.7%</td></tr><tr><td>23.5</td><td>4.00%</td><td>21.7</td><td>3.63%</td><td>9.3%</td></tr><tr><td>42.4</td><td>3.98%</td><td>40.4</td><td>3.63%</td><td>8.8%</td></tr></table>"
},
"TABREF6": {
"type_str": "table",
"html": null,
"text": "CER vs. backoff rate.",
"num": null,
"content": "<table><tr><td/><td>0.41</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>0.39</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>error rate difference</td><td>0.29 0.31 0.33 0.35 0.37</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>0.27</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>2.1 2.3 2.5 2.7 2.9 3.1 3.3 3.5 3.7 3.9 0.28 Figure 4: 0.25 0.29 error rate 0.28 0.3 0.32 0.34 0.36 0.38 0.4 0.42 backoff rate</td><td>0.3</td><td>0.31 word trigram model 0.32 0.33 0.34 backoff rate 0.35 ACM</td><td>0.36</td><td>0.37</td><td>0.38</td><td>0.39</td><td>0.4</td><td>0.41</td></tr></table>"
}
}
}
} |