File size: 89,501 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 | {
"paper_id": "P11-1027",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T08:46:27.980235Z"
},
"title": "Faster and Smaller N -Gram Language Models",
"authors": [
{
"first": "Adam",
"middle": [],
"last": "Pauls",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of California",
"location": {
"settlement": "Berkeley"
}
},
"email": "adpauls@cs.berkeley.edu"
},
{
"first": "Dan",
"middle": [],
"last": "Klein",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of California",
"location": {
"settlement": "Berkeley"
}
},
"email": "klein@cs.berkeley.edu"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "N-gram language models are a major resource bottleneck in machine translation. In this paper, we present several language model implementations that are both highly compact and fast to query. Our fastest implementation is as fast as the widely used SRILM while requiring only 25% of the storage. Our most compact representation can store all 4 billion n-grams and associated counts for the Google n-gram corpus in 23 bits per n-gram, the most compact lossless representation to date, and even more compact than recent lossy compression techniques. We also discuss techniques for improving query speed during decoding, including a simple but novel language model caching technique that improves the query speed of our language models (and SRILM) by up to 300%.",
"pdf_parse": {
"paper_id": "P11-1027",
"_pdf_hash": "",
"abstract": [
{
"text": "N-gram language models are a major resource bottleneck in machine translation. In this paper, we present several language model implementations that are both highly compact and fast to query. Our fastest implementation is as fast as the widely used SRILM while requiring only 25% of the storage. Our most compact representation can store all 4 billion n-grams and associated counts for the Google n-gram corpus in 23 bits per n-gram, the most compact lossless representation to date, and even more compact than recent lossy compression techniques. We also discuss techniques for improving query speed during decoding, including a simple but novel language model caching technique that improves the query speed of our language models (and SRILM) by up to 300%.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "For modern statistical machine translation systems, language models must be both fast and compact. The largest language models (LMs) can contain as many as several hundred billion n-grams (Brants et al., 2007) , so storage is a challenge. At the same time, decoding a single sentence can trigger hundreds of thousands of queries to the language model, so speed is also critical. As always, trade-offs exist between time, space, and accuracy, with many recent papers considering smallbut-approximate noisy LMs (Chazelle et al., 2004; Guthrie and Hepple, 2010) or small-but-slow compressed LMs (Germann et al., 2009) .",
"cite_spans": [
{
"start": 188,
"end": 209,
"text": "(Brants et al., 2007)",
"ref_id": "BIBREF2"
},
{
"start": 509,
"end": 532,
"text": "(Chazelle et al., 2004;",
"ref_id": "BIBREF3"
},
{
"start": 533,
"end": 558,
"text": "Guthrie and Hepple, 2010)",
"ref_id": "BIBREF10"
},
{
"start": 592,
"end": 614,
"text": "(Germann et al., 2009)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper, we present several lossless methods for compactly but efficiently storing large LMs in memory. As in much previous work (Whittaker and Raj, 2001; Hsu and Glass, 2008) , our methods are conceptually based on tabular trie encodings wherein each n-gram key is stored as the concatenation of one word (here, the last) and an offset encoding the remaining words (here, the context). After presenting a bit-conscious basic system that typifies such approaches, we improve on it in several ways. First, we show how the last word of each entry can be implicitly encoded, almost entirely eliminating its storage requirements. Second, we show that the deltas between adjacent entries can be efficiently encoded with simple variable-length encodings. Third, we investigate block-based schemes that minimize the amount of compressed-stream scanning during lookup.",
"cite_spans": [
{
"start": 135,
"end": 160,
"text": "(Whittaker and Raj, 2001;",
"ref_id": "BIBREF18"
},
{
"start": 161,
"end": 181,
"text": "Hsu and Glass, 2008)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "To speed up our language models, we present two approaches. The first is a front-end cache. Caching itself is certainly not new to language modeling, but because well-tuned LMs are essentially lookup tables to begin with, naive cache designs only speed up slower systems. We present a direct-addressing cache with a fast key identity check that speeds up our systems (or existing fast systems like the widelyused, speed-focused SRILM) by up to 300%.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Our second speed-up comes from a more fundamental change to the language modeling interface. Where classic LMs take word tuples and produce counts or probabilities, we propose an LM that takes a word-and-context encoding (so the context need not be re-looked up) and returns both the probability and also the context encoding for the suffix of the original query. This setup substantially accelerates the scrolling queries issued by decoders, and also exploits language model state equivalence (Li and Khudanpur, 2008) .",
"cite_spans": [
{
"start": 494,
"end": 518,
"text": "(Li and Khudanpur, 2008)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Overall, we are able to store the 4 billion n-grams of the Google Web1T (Brants and Franz, 2006) cor-pus, with associated counts, in 10 GB of memory, which is smaller than state-of-the-art lossy language model implementations (Guthrie and Hepple, 2010) , and significantly smaller than the best published lossless implementation (Germann et al., 2009) . We are also able to simultaneously outperform SRILM in both total size and speed. Our LM toolkit, which is implemented in Java and compatible with the standard ARPA file formats, is available on the web. 1",
"cite_spans": [
{
"start": 72,
"end": 96,
"text": "(Brants and Franz, 2006)",
"ref_id": "BIBREF1"
},
{
"start": 226,
"end": 252,
"text": "(Guthrie and Hepple, 2010)",
"ref_id": "BIBREF10"
},
{
"start": 329,
"end": 351,
"text": "(Germann et al., 2009)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Our goal in this paper is to provide data structures that map n-gram keys to values, i.e. probabilities or counts. Maps are fundamental data structures and generic implementations of mapping data structures are readily available. However, because of the sheer number of keys and values needed for n-gram language modeling, generic implementations do not work efficiently \"out of the box.\" In this section, we will review existing techniques for encoding the keys and values of an n-gram language model, taking care to account for every bit of memory required by each implementation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Preliminaries",
"sec_num": "2"
},
{
"text": "To provide absolute numbers for the storage requirements of different implementations, we will use the Google Web1T corpus as a benchmark. This corpus, which is on the large end of corpora typically employed in language modeling, is a collection of nearly 4 billion n-grams extracted from over a trillion tokens of English text, and has a vocabulary of about 13.5 million words.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Preliminaries",
"sec_num": "2"
},
{
"text": "In the Web1T corpus, the most frequent n-gram occurs about 95 billion times. Storing this count explicitly would require 37 bits, but, as noted by Guthrie and Hepple (2010) , the corpus contains only about 770 000 unique counts, so we can enumerate all counts using only 20 bits, and separately store an array called the value rank array which converts the rank encoding of a count back to its raw count. The additional array is small, requiring only about 3MB, but we save 17 bits per n-gram, reducing value storage from around 16GB to about 9GB for Web1T.",
"cite_spans": [
{
"start": 147,
"end": 172,
"text": "Guthrie and Hepple (2010)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Encoding Values",
"sec_num": "2.1"
},
{
"text": "We can rank encode probabilities and back-offs in the same way, allowing us to be agnostic to whether we encode counts, probabilities and/or back-off weights in our model. In general, the number of bits per value required to encode all value ranks for a given language model will vary -we will refer to this variable as v .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Encoding Values",
"sec_num": "2.1"
},
{
"text": "The data structure of choice for the majority of modern language model implementations is a trie (Fredkin, 1960) . Tries or variants thereof are implemented in many LM tool kits, including SRILM (Stolcke, 2002) , IRSTLM (Federico and Cettolo, 2007) , CMU SLM (Whittaker and Raj, 2001) , and MIT LM (Hsu and Glass, 2008) . Tries represent collections of n-grams using a tree. Each node in the tree encodes a word, and paths in the tree correspond to n-grams in the collection. Tries ensure that each n-gram prefix is represented only once, and are very efficient when n-grams share common prefixes. Values can also be stored in a trie by placing them in the appropriate nodes.",
"cite_spans": [
{
"start": 97,
"end": 112,
"text": "(Fredkin, 1960)",
"ref_id": "BIBREF7"
},
{
"start": 195,
"end": 210,
"text": "(Stolcke, 2002)",
"ref_id": "BIBREF17"
},
{
"start": 220,
"end": 248,
"text": "(Federico and Cettolo, 2007)",
"ref_id": "BIBREF6"
},
{
"start": 259,
"end": 284,
"text": "(Whittaker and Raj, 2001)",
"ref_id": "BIBREF18"
},
{
"start": 298,
"end": 319,
"text": "(Hsu and Glass, 2008)",
"ref_id": "BIBREF12"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Trie-Based Language Models",
"sec_num": "2.2"
},
{
"text": "Conceptually, trie nodes can be implemented as records that contain two entries: one for the word in the node, and one for either a pointer to the parent of the node or a list of pointers to children. At a low level, however, naive implementations of tries can waste significant amounts of space. For example, the implementation used in SRILM represents a trie node as a C struct containing a 32-bit integer representing the word, a 64-bit memory 2 pointer to the list of children, and a 32-bit floating point number representing the value stored at a node. The total storage for a node alone is 16 bytes, with additional overhead required to store the list of children. In total, the most compact implementation in SRILM uses 33 bytes per n-gram of storage, which would require around 116 GB of memory to store Web1T.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Trie-Based Language Models",
"sec_num": "2.2"
},
{
"text": "While it is simple to implement a trie node in this (already wasteful) way in programming languages that offer low-level access to memory allocation like C/C++, the situation is even worse in higher level programming languages. In Java, for example, Cstyle structs are not available, and records are most naturally implemented as objects that carry an additional 64 bits of overhead.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Trie-Based Language Models",
"sec_num": "2.2"
},
{
"text": "Despite its relatively large storage requirements, the implementation employed by SRILM is still widely in use today, largely because of its speed -to our knowledge, SRILM is the fastest freely available language model implementation. We will show that we can achieve access speeds comparable to SRILM but using only 25% of the storage.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Trie-Based Language Models",
"sec_num": "2.2"
},
{
"text": "A more compact implementation of a trie is described in Whittaker and Raj (2001) . In their implementation, nodes in a trie are represented implicitly as entries in an array. Each entry encodes a word with enough bits to index all words in the language model (24 bits for Web1T), a quantized value, and a 32-bit 3 offset that encodes the contiguous block of the array containing the children of the node. Note that 32 bits is sufficient to index all n-grams in Web1T; for larger corpora, we can always increase the size of the offset.",
"cite_spans": [
{
"start": 56,
"end": 80,
"text": "Whittaker and Raj (2001)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Implicit Tries",
"sec_num": "2.3"
},
{
"text": "Effectively, this representation replaces systemlevel memory pointers with offsets that act as logical pointers that can reference other entries in the array, rather than arbitrary bytes in RAM. This representation saves space because offsets require fewer bits than memory pointers, but more importantly, it permits straightforward implementation in any higherlevel language that provides access to arrays of integers. 4",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Implicit Tries",
"sec_num": "2.3"
},
{
"text": "Hsu and Glass (2008) describe a variant of the implicit tries of Whittaker and Raj (2001) in which each node in the trie stores the prefix (i.e. parent). This representation has the property that we can refer to each n-gram w n 1 by its last word w n and the offset c(w n\u22121 1 ) of its prefix w n\u22121 1 , often called the context. At a low-level, we can efficiently encode this pair (w n , c(w n\u22121 1 )) as a single 64-bit integer, where the first 24 bits refer to w n and the last 40 bits encode c(w n\u22121 1 ). We will refer to this encoding as a context encoding.",
"cite_spans": [
{
"start": 65,
"end": 89,
"text": "Whittaker and Raj (2001)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Encoding n-grams",
"sec_num": "2.4"
},
{
"text": "Note that typically, n-grams are encoded in tries in the reverse direction (first-rest instead of lastrest), which enables a more efficient computation of back-offs. In our implementations, we found that the speed improvement from switching to a first-rest encoding and implementing more efficient queries was modest. However, as we will see in Section 4.2, the last-rest encoding allows us to exploit the scrolling nature of queries issued by decoders, which results in speedups that far outweigh those achieved by reversing the trie.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Encoding n-grams",
"sec_num": "2.4"
},
{
"text": "In the previous section, we reviewed well-known techniques in language model implementation. In this section, we combine these techniques to build simple data structures in ways that are to our knowledge novel, producing language models with stateof-the-art memory requirements and speed. We will also show that our data structures can be very effectively compressed by implicitly encoding the word w n , and further compressed by applying a variablelength encoding on context deltas.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Language Model Implementations",
"sec_num": "3"
},
{
"text": "A standard way to implement a map is to store an array of key/value pairs, sorted according to the key. Lookup is carried out by performing binary search on a key. For an n-gram language model, we can apply this implementation with a slight modification: we need n sorted arrays, one for each n-gram order. We construct keys (w n , c(w n\u22121 1 )) using the context encoding described in the previous section, where the context offsets c refer to entries in the sorted array of (n \u2212 1)-grams. This data structure is shown graphically in Figure 1 .",
"cite_spans": [],
"ref_spans": [
{
"start": 534,
"end": 542,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Sorted Array",
"sec_num": "3.1"
},
{
"text": "Because our keys are sorted according to their context-encoded representation, we cannot straightforwardly answer queries about an n-gram w without first determining its context encoding. We can do this efficiently by building up the encoding incrementally: we start with the context offset of the unigram w 1 , which is simply its integer representation, and use that to form the context encoding of the bigram w 2 1 = (w 2 , c(w 1 )). We can find the offset of ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sorted Array",
"sec_num": "3.1"
},
{
"text": "\"slept\" 1933 1933 1933 1933 1933 1933 1935 1935 1935 . .",
"cite_spans": [
{
"start": 8,
"end": 52,
"text": "1933 1933 1933 1933 1933 1933 1935 1935 1935",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": ". .",
"sec_num": null
},
{
"text": "v bits . . \"dog\"",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "val val val",
"sec_num": null
},
{
"text": "3-grams 2-grams 1-grams w Figure 1 : Our SORTED implementation of a trie. The dotted paths correspond to \"the cat slept\", \"the cat ran\", and \"the dog ran\". Each node in the trie is an entry in an array with 3 parts: w represents the word at the node; val represents the (rank encoded) value; and c is an offset in the array of n \u2212 1 grams that represents the parent (prefix) of a node. Words are represented as offsets in the unigram array.",
"cite_spans": [],
"ref_spans": [
{
"start": 26,
"end": 34,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "val val val",
"sec_num": null
},
{
"text": "the bigram using binary search, and form the context encoding of the trigram, and so on. Note, however, that if our queries arrive in context-encoded form, queries are faster since they involve only one binary search in the appropriate array. We will return to this later in Section 4.2 This implementation, SORTED, uses 64 bits for the integer-encoded keys and v bits for the values. Lookup is linear in the length of the key and logarithmic in the number of n-grams. For Web1T (v = 20), the total storage is 10.5 bytes/n-gram or about 37GB.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "val val val",
"sec_num": null
},
{
"text": "Hash tables are another standard way to implement associative arrays. To enable the use of our context encoding, we require an implementation in which we can refer to entries in the hash table via array offsets. For this reason, we use an open address hash map that uses linear probing for collision resolution.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hash Table",
"sec_num": "3.2"
},
{
"text": "As in the sorted array implementation, in order to insert an n-gram w n 1 into the hash table, we must form its context encoding incrementally from the offset of w 1 . However, unlike the sorted array implementation, at query time, we only need to be able to check equality between the query key w n 1 = (w n , c(w n\u22121 1 )) and a key w n 1 = (w n , c(w n\u22121 1 )) in the table. Equality can easily be checked by first checking if w n = w n , then recursively checking equality between w n\u22121 1 and w n\u22121 1 , though again, equality is even faster if the query is already contextencoded.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hash Table",
"sec_num": "3.2"
},
{
"text": "This HASH data structure also uses 64 bits for integer-encoded keys and v bits for values. However, to avoid excessive hash collisions, we also allocate additional empty space according to a userdefined parameter that trades off speed and timewe used about 40% extra space in our experiments. For Web1T, the total storage for this implementation is 15 bytes/n-gram or about 53 GB total.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hash Table",
"sec_num": "3.2"
},
{
"text": "Look up in a hash map is linear in the length of an n-gram and constant with respect to the number of n-grams. Unlike the sorted array implementation, the hash table implementation also permits efficient insertion and deletion, making it suitable for stream-based language models (Levenberg and Osborne, 2009) .",
"cite_spans": [
{
"start": 280,
"end": 309,
"text": "(Levenberg and Osborne, 2009)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Hash Table",
"sec_num": "3.2"
},
{
"text": "The context encoding we have used thus far still wastes space. This is perhaps most evident in the sorted array representation (see Figure 1 ): all ngrams ending with a particular word w i are stored contiguously. We can exploit this redundancy by storing only the context offsets in the main array, using as many bits as needed to encode all context offsets (32 bits for Web1T). In auxiliary arrays, one for each n-gram order, we store the beginning and end of the range of the trie array in which all (w i , c) keys are stored for each w i . These auxiliary arrays are negligibly small -we only need to store 2n offsets for each word.",
"cite_spans": [],
"ref_spans": [
{
"start": 132,
"end": 140,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Implicitly Encoding w n",
"sec_num": "3.3"
},
{
"text": "The same trick can be applied in the hash table implementation. We allocate contiguous blocks of the main array for n-grams which all share the same last word w i , and distribute keys within those ranges using the hashing function.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Implicitly Encoding w n",
"sec_num": "3.3"
},
{
"text": "This representation reduces memory usage for keys from 64 bits to 32 bits, reducing overall storage for Web1T to 6.5 bytes/n-gram for the sorted implementation and 9.1 bytes for the hashed implementation, or about 23GB and 32GB in total. It also increases query speed in the sorted array case, since to find (w i , c), we only need to search the range of the array over which w i applies. Because this implicit encoding reduces memory usage without a performance cost, we will assume its use for the rest of this paper.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Implicitly Encoding w n",
"sec_num": "3.3"
},
{
"text": "The distribution of value ranks in language modeling is Zipfian, with far more n-grams having low counts than high counts. If we ensure that the value rank array sorts raw values by descending order of frequency, then we expect that small ranks will occur much more frequently than large ones, which we can exploit with a variable-length encoding.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Variable-Length Coding",
"sec_num": "3.4.1"
},
{
"text": "To compress n-grams, we can exploit the context encoding of our keys. In Figure 2, of the key array used in our sorted array implementation. While we have already exploited the fact that the 24 word bits repeat in the previous section, we note here that consecutive context offsets tend to be quite close together. We found that for 5-grams, the median difference between consecutive offsets was about 50, and 90% of offset deltas were smaller than 10000. By using a variable-length encoding to represent these deltas, we should require far fewer than 32 bits to encode context offsets. We used a very simple variable-length coding to encode offset deltas, word deltas, and value ranks. Our encoding, which is referred to as \"variablelength block coding\" in Boldi and Vigna (2005) , works as follows: we pick a (configurable) radix r = 2 k . To encode a number m, we determine the number of digits d required to express m in base r. We write d in unary, i.e. d \u2212 1 zeroes followed by a one. We then write the d digits of m in base r, each of which requires k bits. For example, using k = 2, we would encode the decimal number 7 as 010111. We can choose k separately for deltas and value indices, and also tune these parameters to a given language model.",
"cite_spans": [
{
"start": 758,
"end": 780,
"text": "Boldi and Vigna (2005)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [
{
"start": 73,
"end": 82,
"text": "Figure 2,",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Variable-Length Coding",
"sec_num": "3.4.1"
},
{
"text": "We found this encoding outperformed other standard prefix codes, including Golomb codes (Golomb, 1966; Church et al., 2007) and Elias \u03b3 and \u03b4 codes. We also experimented with the \u03b6 codes of Boldi and Vigna (2005) , which modify variable-length block codes so that they are optimal for certain power law distributions. We found that \u03b6 codes performed no better than variable-length block codes and were slightly more complex. Finally, we found that Huffman codes outperformed our encoding slightly, but came at a much higher computational cost.",
"cite_spans": [
{
"start": 88,
"end": 102,
"text": "(Golomb, 1966;",
"ref_id": "BIBREF9"
},
{
"start": 103,
"end": 123,
"text": "Church et al., 2007)",
"ref_id": "BIBREF5"
},
{
"start": 190,
"end": 212,
"text": "Boldi and Vigna (2005)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Variable-Length Coding",
"sec_num": "3.4.1"
},
{
"text": "We could in principle compress the entire array of key/value pairs with the encoding described above, but this would render binary search in the array impossible: we cannot jump to the mid-point of the array since in order to determine what key lies at a particular point in the compressed bit stream, we would need to know the entire history of offset deltas.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Block Compression",
"sec_num": "3.4.2"
},
{
"text": "Instead, we employ block compression, a technique also used by Harb et al. (2009) for smaller language models. In particular, we compress the key/value array in blocks of 128 bytes. At the beginning of the block, we write out a header consisting of: an explicit 64-bit key that begins the block; a 32-bit integer representing the offset of the header key in the uncompressed array; 5 the number of bits of compressed data in the block; and the variablelength encoding of the value rank of the header key. The remainder of the block is filled with as many compressed key/value pairs as possible. Once the block is full, we start a new block. See Figure 2 for a depiction.",
"cite_spans": [
{
"start": 63,
"end": 81,
"text": "Harb et al. (2009)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [
{
"start": 645,
"end": 653,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Block Compression",
"sec_num": "3.4.2"
},
{
"text": "When we encode an offset delta, we store the delta of the word portion of the key separately from the delta of the context offset. When an entire block shares the same word portion of the key, we set a single bit in the header that indicates that we do not encode any word deltas.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Block Compression",
"sec_num": "3.4.2"
},
{
"text": "To find a key in this compressed array, we first perform binary search over the header blocks (which are predictably located every 128 bytes), followed by a linear search within a compressed block.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Block Compression",
"sec_num": "3.4.2"
},
{
"text": "Using k = 6 for encoding offset deltas and k = 5 for encoding value ranks, this COMPRESSED implementation stores Web1T in less than 3 bytes per n-gram, or about 10.2GB in total. This is about 6GB less than the storage required by Germann et al. (2009) , which is the best published lossless compression to date.",
"cite_spans": [
{
"start": 230,
"end": 251,
"text": "Germann et al. (2009)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Block Compression",
"sec_num": "3.4.2"
},
{
"text": "In the previous section, we provided compact and efficient implementations of associative arrays that allow us to query a value for an arbitrary n-gram. However, decoders do not issue language model requests at random. In this section, we show that language model requests issued by a standard decoder exhibit two patterns we can exploit: they are highly repetitive, and also exhibit a scrolling effect.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Speeding up Decoding",
"sec_num": "4"
},
{
"text": "In a simple experiment, we recorded all of the language model queries issued by the Joshua decoder (Li et al., 2009) on a 100 sentence test set. Of the 31 million queries, only about 1 million were unique. Therefore, we expect that keeping the results of language model queries in a cache should be effective at reducing overall language model latency.",
"cite_spans": [
{
"start": 99,
"end": 116,
"text": "(Li et al., 2009)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Exploiting Repetitive Queries",
"sec_num": "4.1"
},
{
"text": "To this end, we added a very simple cache to our language model. Our cache uses an array of key/value pairs with size fixed to 2 b \u2212 1 for some integer b (we used 24). We use a b-bit hash function to compute the address in an array where we will always place a given n-gram and its fully computed language model score. Querying the cache is straightforward: we check the address of a key given by its b-bit hash. If the key located in the cache array matches the query key, then we return the value stored in the cache. Otherwise, we fetch the language model probability from the language model and place the new key and value in the cache, evicting the old key in the process. This scheme is often called a direct-mapped cache because each key has exactly one possible address.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Exploiting Repetitive Queries",
"sec_num": "4.1"
},
{
"text": "Caching n-grams in this way reduces overall latency for two reasons: first, lookup in the cache is extremely fast, requiring only a single evaluation of the hash function, one memory lookup to find the cache key, and one equality check on the key. In contrast, even our fastest (HASH) implementation may have to perform multiple memory lookups and equality checks in order to resolve collisions. Second, when calculating the probability for an n-gram Figure 3 : Queries issued when scoring trigrams that are created when a state with LM context \"the cat\" combines with \"fell down\". In the standard explicit representation of an n-gram as list of words, queries are issued atomically to the language model. When using a contextencoding, a query from the n-gram \"the cat fell\" returns the context offset of \"cat fell\", which speeds up the query of \"cat fell down\".",
"cite_spans": [],
"ref_spans": [
{
"start": 451,
"end": 459,
"text": "Figure 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Exploiting Repetitive Queries",
"sec_num": "4.1"
},
{
"text": "not in the language model, language models with back-off schemes must in general perform multiple queries to fetch the necessary back-off information.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Exploiting Repetitive Queries",
"sec_num": "4.1"
},
{
"text": "Our cache retains the full result of these calculations and thus saves additional computation. Federico and Cettolo (2007) also employ a cache in their language model implementation, though based on traditional hash table cache with linear probing. Unlike our cache, which is of fixed size, their cache must be cleared after decoding a sentence. We would not expect a large performance increase from such a cache for our faster models since our HASH implementation is already a hash table with linear probing. We found in our experiments that a cache using linear probing provided marginal performance increases of about 40%, largely because of cached back-off computation, while our simpler cache increases performance by about 300% even over our HASH LM implementation. More timing results are presented in Section 5.",
"cite_spans": [
{
"start": 95,
"end": 122,
"text": "Federico and Cettolo (2007)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Exploiting Repetitive Queries",
"sec_num": "4.1"
},
{
"text": "Decoders with integrated language models (Och and Ney, 2004; Chiang, 2005 ) score partial translation hypotheses in an incremental way. Each partial hypothesis maintains a language model context consisting of at most n \u2212 1 target-side words. When we combine two language model contexts, we create several new n-grams of length of n, each of which generate a query to the language model. These new WMT2010 Order #n-grams 1gm 4,366,395 2gm 61,865,588 3gm 123,158,761 4gm 217,869,981 5gm 269,614,330 Total 676,875,055 WEB1T Order #n-grams 1gm 13,588,391 2gm 314,843,401 3gm 977,069,902 4gm 1,313,818,354 5gm 1,176,470,663 Total 3,795,790,711 n-grams exhibit a scrolling effect, shown in Figure 3: the n \u2212 1 suffix words of one n-gram form the n \u2212 1 prefix words of the next.",
"cite_spans": [
{
"start": 41,
"end": 60,
"text": "(Och and Ney, 2004;",
"ref_id": "BIBREF16"
},
{
"start": 61,
"end": 73,
"text": "Chiang, 2005",
"ref_id": "BIBREF4"
}
],
"ref_spans": [
{
"start": 397,
"end": 655,
"text": "WMT2010 Order #n-grams 1gm 4,366,395 2gm 61,865,588 3gm 123,158,761 4gm 217,869,981 5gm 269,614,330 Total 676,875,055 WEB1T Order #n-grams 1gm 13,588,391 2gm 314,843,401 3gm 977,069,902 4gm 1,313,818,354 5gm 1,176,470,663 Total",
"ref_id": "TABREF2"
},
{
"start": 715,
"end": 721,
"text": "Figure",
"ref_id": null
}
],
"eq_spans": [],
"section": "Exploiting Scrolling Queries",
"sec_num": "4.2"
},
{
"text": "As discussed in Section 3, our LM implementations can answer queries about context-encoded ngrams faster than explicitly encoded n-grams. With this in mind, we augment the values stored in our language model so that for a key (w n , c(w n\u22121 1 )), we store the offset of the suffix c(w n 2 ) as well as the normal counts/probabilities. Then, rather than represent the LM context in the decoder as an explicit list of words, we can simply store context offsets. When we query the language model, we get back both a language model score and context offset c(\u0175 n\u22121 1 ), where\u0175 n\u22121 1 is the the longest suffix of w n\u22121 1 contained in the language model. We can then quickly form the context encoding of the next query by simply concatenating the new word with the offset c(\u0175 n\u22121 1 ) returned from the previous query.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Exploiting Scrolling Queries",
"sec_num": "4.2"
},
{
"text": "In addition to speeding up language model queries, this approach also automatically supports an equivalence of LM states (Li and Khudanpur, 2008) : in standard back-off schemes, whenever we compute the probability for an n-gram (w n , c(w n\u22121 1 )) when w n\u22121 1 is not in the language model, the result will be the same as the result of the query (w n , c(\u0175 n\u22121 1 ). It is therefore only necessary to store as much of the context as the language model contains instead of all n \u2212 1 words in the context. If a decoder maintains LM states using the context offsets returned by our language model, then the decoder will automatically exploit this equivalence and the size of the search space will be reduced. This same effect is exploited explicitly by some decoders (Li and Khudanpur, 2008 ",
"cite_spans": [
{
"start": 121,
"end": 145,
"text": "(Li and Khudanpur, 2008)",
"ref_id": "BIBREF14"
},
{
"start": 763,
"end": 786,
"text": "(Li and Khudanpur, 2008",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Exploiting Scrolling Queries",
"sec_num": "4.2"
},
{
"text": "To test our LM implementations, we performed experiments with two different language models. Our first language model, WMT2010, was a 5gram Kneser-Ney language model which stores probability/back-off pairs as values. We trained this language model on the English side of all French-English corpora provided 6 for use in the WMT 2010 workshop, about 2 billion tokens in total. This data was tokenized using the tokenizer.perl script provided with the data. We trained the language model using SRILM. We also extracted a countbased language model, WEB1T, from the Web1T corpus (Brants and Franz, 2006) . Since this data is provided as a collection of 1-to 5-grams and associated counts, we used this data without further preprocessing. The make up of these language models is shown in Table 1 .",
"cite_spans": [
{
"start": 575,
"end": 599,
"text": "(Brants and Franz, 2006)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [
{
"start": 783,
"end": 790,
"text": "Table 1",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Experiments 5.1 Data",
"sec_num": "5"
},
{
"text": "We tested our three implementations (HASH, SORTED, and COMPRESSED) on the WMT2010 language model. For this language model, there are about 80 million unique probability/back-off pairs, so v \u2248 36. Note that here v includes both the cost per key of storing the value rank as well as the (amortized) cost of storing two 32 bit floating point numbers (probability and back-off) for each unique value. The results are shown in We compare against three baselines. The first two, SRILM-H and SRILM-S, refer to the hash tableand sorted array-based trie implementations provided by SRILM. The third baseline is the Tightly-Packed Trie (TPT) implementation of Germann et al. (2009) . Because this implementation is not freely available, we use their published memory usage in bytes per n-gram on a language model of similar size and project total usage.",
"cite_spans": [
{
"start": 650,
"end": 671,
"text": "Germann et al. (2009)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Compression Experiments",
"sec_num": "5.2"
},
{
"text": "The memory usage of all of our models is considerably smaller than SRILM -our HASH implementation is about 25% the size of SRILM-H, and our SORTED implementation is about 25% the size of SRILM-S. Our COMPRESSED implementation is also smaller than the state-of-the-art compressed TPT implementation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Compression Experiments",
"sec_num": "5.2"
},
{
"text": "In Table 3 , we show the results of our COM-PRESSED implementation on WEB1T and against two baselines. The first is compression of the ASCII text count files using gzip, and the second is the Tiered Minimal Perfect Hash (T-MPHR) of Guthrie and Hepple (2010) . The latter is a lossy compression technique based on Bloomier filters (Chazelle et al., 2004) and additional variable-length encoding that achieves the best published compression of WEB1T to date. Our COMPRESSED implementation is even smaller than T-MPHR, despite using a lossless compression technique. Note that since T-MPHR uses a lossy encoding, it is possible to reduce the storage requirements arbitrarily at the cost of additional errors in the model. We quote here the storage required when keys 7 are encoded using 12bit hash codes, which gives a false positive rate of about 2 \u221212 =0.02%. Table 5 : Full decoding times for various language model implementations. Our HASH LM is as fast as SRILM while using 25% of the memory. Our caching also reduces total decoding time by about 20% for our fastest models and speeds up COMPRESSED by a factor of 6. Times were averaged over 3 runs on the same machine.",
"cite_spans": [
{
"start": 232,
"end": 257,
"text": "Guthrie and Hepple (2010)",
"ref_id": "BIBREF10"
},
{
"start": 330,
"end": 353,
"text": "(Chazelle et al., 2004)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [
{
"start": 3,
"end": 10,
"text": "Table 3",
"ref_id": "TABREF6"
},
{
"start": 859,
"end": 866,
"text": "Table 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "Compression Experiments",
"sec_num": "5.2"
},
{
"text": "We first measured pure query speed by logging all LM queries issued by a decoder and measuring the time required to query those n-grams in isolation. We used the the Joshua decoder 8 with the WMT2010 model to generate queries for the first 100 sentences of the French 2008 News test set. This produced about 30 million queries. We measured the time 9 required to perform each query in order with and without our direct-mapped caching, not including any time spent on file I/O. The results are shown in Table 4 . As expected, HASH is the fastest of our implementations, and comparable 10 in speed to SRILM-H, but using sig- 8 We used a grammar trained on all French-English data provided for WMT 2010 using the make scripts provided at http://sourceforge.net/projects/joshua/files /joshua/1.3/wmt2010-experiment.tgz/download 9 All experiments were performed on an Amazon EC2 High-Memory Quadruple Extra Large instance, with an Intel Xeon X5550 CPU running at 2.67GHz and 8 MB of cache.",
"cite_spans": [
{
"start": 623,
"end": 624,
"text": "8",
"ref_id": null
}
],
"ref_spans": [
{
"start": 502,
"end": 509,
"text": "Table 4",
"ref_id": "TABREF8"
}
],
"eq_spans": [],
"section": "Timing Experiments",
"sec_num": "5.3"
},
{
"text": "10 Because we implemented our LMs in Java, we issued queries to SRILM via Java Native Interface (JNI) calls, which introduces a performance overhead. When called natively, we found that SRILM was about 200 ns/query faster. Unfortu-nificantly less space. SORTED is slower but of course more memory efficient, and COMPRESSED is the slowest but also the most compact representation. In HASH+SCROLL, we issued queries to the language model using the context encoding, which speeds up queries substantially. Finally, we note that our direct-mapped cache is very effective. The query speed of all models is boosted substantially. In particular, our COMPRESSED implementation with caching is nearly as fast as SRILM-H without caching, and even the already fast HASH implementation is 300% faster in raw query speed with caching enabled.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Timing Experiments",
"sec_num": "5.3"
},
{
"text": "We also measured the effect of LM performance on overall decoder performance. We modified Joshua to optionally use our LM implementations during decoding, and measured the time required to decode all 2051 sentences of the 2008 News test set. The results are shown in Table 5 . Without caching, SRILM-H and HASH were comparable in speed, while COMPRESSED introduces a performance penalty. With caching enabled, overall decoder speed is improved for both HASH and SRILM-H, while the COMPRESSED implementation is only about 50% slower that the others.",
"cite_spans": [],
"ref_spans": [
{
"start": 267,
"end": 274,
"text": "Table 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "Timing Experiments",
"sec_num": "5.3"
},
{
"text": "We have presented several language model implementations which are state-of-the-art in both size and speed. Our experiments have demonstrated improvements in query speed over SRILM and compression rates against state-of-the-art lossy compression. We have also described a simple caching technique which leads to performance increases in overall decoding time.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "http://code.google.com/p/berkeleylm/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "While 32-bit architectures are still in use today, their limited address space is insufficient for modern language models and we will assume all machines use a 64-bit architecture.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "The implementation described in the paper represents each 32-bit integer compactly using only 16 bits, but this representation is quite inefficient, because determining the full 32-bit offset requires a binary search in a look up table.4 Typically, programming languages only provide support for arrays of bytes, not bits, but it is of course possible to simulate arrays with arbitrary numbers of bits using byte arrays and bit manipulation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "We need this because n-grams refer to their contexts using array offsets.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Guthrie and Hepple (2010) also report additional savings by quantizing values, though we could perform the same quantization in our storage scheme.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "This work was supported by a Google Fellowship for the first author and by BBN under DARPA contract HR0011-06-C-0022. We would like to thank David Chiang, Zhifei Li, and the anonymous reviewers for their helpful comments.nately, it is not completely fair to compare our LMs against either of these numbers: although the JNI overhead slows down SRILM, implementing our LMs in Java instead of C++ slows down our LMs. In the tables, we quote times which include the JNI overhead, since this reflects the true cost to a decoder written in Java (e.g. Joshua).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Codes for the world wide web",
"authors": [
{
"first": "Paolo",
"middle": [],
"last": "Boldi",
"suffix": ""
},
{
"first": "Sebastiano",
"middle": [],
"last": "Vigna",
"suffix": ""
}
],
"year": 2005,
"venue": "Internet Mathematics",
"volume": "2",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Paolo Boldi and Sebastiano Vigna. 2005. Codes for the world wide web. Internet Mathematics, 2.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Google web1t 5-gram corpus, version 1",
"authors": [
{
"first": "Thorsten",
"middle": [],
"last": "Brants",
"suffix": ""
},
{
"first": "Alex",
"middle": [],
"last": "Franz",
"suffix": ""
}
],
"year": 2006,
"venue": "Linguistic Data Consortium, Philadelphia, Catalog Number LDC2006T13",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Thorsten Brants and Alex Franz. 2006. Google web1t 5-gram corpus, version 1. In Linguistic Data Consor- tium, Philadelphia, Catalog Number LDC2006T13.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Large language models in machine translation",
"authors": [
{
"first": "Thorsten",
"middle": [],
"last": "Brants",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Ashok",
"suffix": ""
},
{
"first": "Peng",
"middle": [],
"last": "Popat",
"suffix": ""
},
{
"first": "Franz",
"middle": [
"J"
],
"last": "Xu",
"suffix": ""
},
{
"first": "Jeffrey",
"middle": [],
"last": "Och",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Dean",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Thorsten Brants, Ashok C. Popat, Peng Xu, Franz J. Och, and Jeffrey Dean. 2007. Large language models in machine translation. In Proceedings of the Conference on Empirical Methods in Natural Language Process- ing.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "The Bloomier filter: an efficient data structure for static support lookup tables",
"authors": [
{
"first": "Bernard",
"middle": [],
"last": "Chazelle",
"suffix": ""
},
{
"first": "Joe",
"middle": [],
"last": "Kilian",
"suffix": ""
},
{
"first": "Ronitt",
"middle": [],
"last": "Rubinfeld",
"suffix": ""
},
{
"first": "Ayellet",
"middle": [],
"last": "Tal",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of the fifteenth annual ACM-SIAM symposium on Discrete algorithms",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bernard Chazelle, Joe Kilian, Ronitt Rubinfeld, and Ayellet Tal. 2004. The Bloomier filter: an efficient data structure for static support lookup tables. In Pro- ceedings of the fifteenth annual ACM-SIAM sympo- sium on Discrete algorithms.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "A hierarchical phrase-based model for statistical machine translation",
"authors": [
{
"first": "David",
"middle": [],
"last": "Chiang",
"suffix": ""
}
],
"year": 2005,
"venue": "The Annual Conference of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "David Chiang. 2005. A hierarchical phrase-based model for statistical machine translation. In The Annual Con- ference of the Association for Computational Linguis- tics.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Compressing trigram language models with golomb coding",
"authors": [
{
"first": "Kenneth",
"middle": [],
"last": "Church",
"suffix": ""
},
{
"first": "Ted",
"middle": [],
"last": "Hart",
"suffix": ""
},
{
"first": "Jianfeng",
"middle": [],
"last": "Gao",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kenneth Church, Ted Hart, and Jianfeng Gao. 2007. Compressing trigram language models with golomb coding. In Proceedings of the Conference on Empiri- cal Methods in Natural Language Processing.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Efficient handling of n-gram language models for statistical machine translation",
"authors": [
{
"first": "Marcello",
"middle": [],
"last": "Federico",
"suffix": ""
},
{
"first": "Mauro",
"middle": [],
"last": "Cettolo",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the Second Workshop on Statistical Machine Translation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marcello Federico and Mauro Cettolo. 2007. Efficient handling of n-gram language models for statistical ma- chine translation. In Proceedings of the Second Work- shop on Statistical Machine Translation.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Trie memory",
"authors": [
{
"first": "Edward",
"middle": [],
"last": "Fredkin",
"suffix": ""
}
],
"year": 1960,
"venue": "Communications of the ACM",
"volume": "3",
"issue": "",
"pages": "490--499",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Edward Fredkin. 1960. Trie memory. Communications of the ACM, 3:490-499, September.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Tightly packed tries: how to fit large models into memory, and make them load fast, too",
"authors": [
{
"first": "Ulrich",
"middle": [],
"last": "Germann",
"suffix": ""
},
{
"first": "Eric",
"middle": [],
"last": "Joanis",
"suffix": ""
},
{
"first": "Samuel",
"middle": [],
"last": "Larkin",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the Workshop on Software Engineering, Testing, and Quality Assurance for Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ulrich Germann, Eric Joanis, and Samuel Larkin. 2009. Tightly packed tries: how to fit large models into mem- ory, and make them load fast, too. In Proceedings of the Workshop on Software Engineering, Testing, and Quality Assurance for Natural Language Processing.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Run-length encodings",
"authors": [
{
"first": "S",
"middle": [
"W"
],
"last": "Golomb",
"suffix": ""
}
],
"year": 1966,
"venue": "IEEE Transactions on Information Theory",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "S. W. Golomb. 1966. Run-length encodings. IEEE Transactions on Information Theory, 12.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Storing the web in memory: space efficient language models with constant time retrieval",
"authors": [
{
"first": "David",
"middle": [],
"last": "Guthrie",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Hepple",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "David Guthrie and Mark Hepple. 2010. Storing the web in memory: space efficient language models with con- stant time retrieval. In Proceedings of the Conference on Empirical Methods in Natural Language Process- ing.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Back-off language model compression",
"authors": [
{
"first": "Boulos",
"middle": [],
"last": "Harb",
"suffix": ""
},
{
"first": "Ciprian",
"middle": [],
"last": "Chelba",
"suffix": ""
},
{
"first": "Jeffrey",
"middle": [],
"last": "Dean",
"suffix": ""
},
{
"first": "Sanjay",
"middle": [],
"last": "Ghemawat",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of Interspeech",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Boulos Harb, Ciprian Chelba, Jeffrey Dean, and Sanjay Ghemawat. 2009. Back-off language model compres- sion. In Proceedings of Interspeech.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Iterative language model estimation: Efficient data structure and algorithms",
"authors": [
{
"first": "Bo-June",
"middle": [],
"last": "Hsu",
"suffix": ""
},
{
"first": "James",
"middle": [],
"last": "Glass",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of Interspeech",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bo-June Hsu and James Glass. 2008. Iterative language model estimation: Efficient data structure and algo- rithms. In Proceedings of Interspeech.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Streambased randomised language models for smt",
"authors": [
{
"first": "Abby",
"middle": [],
"last": "Levenberg",
"suffix": ""
},
{
"first": "Miles",
"middle": [],
"last": "Osborne",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Abby Levenberg and Miles Osborne. 2009. Stream- based randomised language models for smt. In Pro- ceedings of the Conference on Empirical Methods in Natural Language Processing.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "A scalable decoder for parsing-based machine translation with equivalent language model state maintenance",
"authors": [
{
"first": "Zhifei",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Sanjeev",
"middle": [],
"last": "Khudanpur",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the Second Workshop on Syntax and Structure in Statistical Translation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Zhifei Li and Sanjeev Khudanpur. 2008. A scalable decoder for parsing-based machine translation with equivalent language model state maintenance. In Pro- ceedings of the Second Workshop on Syntax and Struc- ture in Statistical Translation.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Joshua: an open source toolkit for parsingbased machine translation",
"authors": [
{
"first": "Zhifei",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Callison-Burch",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Dyer",
"suffix": ""
},
{
"first": "Juri",
"middle": [],
"last": "Ganitkevitch",
"suffix": ""
},
{
"first": "Sanjeev",
"middle": [],
"last": "Khudanpur",
"suffix": ""
},
{
"first": "Lane",
"middle": [],
"last": "Schwartz",
"suffix": ""
},
{
"first": "N",
"middle": [
"G"
],
"last": "Wren",
"suffix": ""
},
{
"first": "Jonathan",
"middle": [],
"last": "Thornton",
"suffix": ""
},
{
"first": "Omar",
"middle": [
"F"
],
"last": "Weese",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Zaidan",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the Fourth Workshop on Statistical Machine Translation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Zhifei Li, Chris Callison-Burch, Chris Dyer, Juri Gan- itkevitch, Sanjeev Khudanpur, Lane Schwartz, Wren N. G. Thornton, Jonathan Weese, and Omar F. Zaidan. 2009. Joshua: an open source toolkit for parsing- based machine translation. In Proceedings of the Fourth Workshop on Statistical Machine Translation.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "The alignment template approach to statistical machine translation",
"authors": [
{
"first": "Josef",
"middle": [],
"last": "Franz",
"suffix": ""
},
{
"first": "Hermann",
"middle": [],
"last": "Och",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Ney",
"suffix": ""
}
],
"year": 2004,
"venue": "Computationl Linguistics",
"volume": "30",
"issue": "",
"pages": "417--449",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Franz Josef Och and Hermann Ney. 2004. The align- ment template approach to statistical machine transla- tion. Computationl Linguistics, 30:417-449, Decem- ber.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "SRILM: An extensible language modeling toolkit",
"authors": [
{
"first": "Andreas",
"middle": [],
"last": "Stolcke",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of Interspeech",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Andreas Stolcke. 2002. SRILM: An extensible language modeling toolkit. In Proceedings of Interspeech.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Quantizationbased language model compression",
"authors": [
{
"first": "E",
"middle": [
"W D"
],
"last": "Whittaker",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Raj",
"suffix": ""
}
],
"year": 2001,
"venue": "Proceedings of Eurospeech",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "E. W. D. Whittaker and B. Raj. 2001. Quantization- based language model compression. In Proceedings of Eurospeech.",
"links": null
}
},
"ref_entries": {
"FIGREF1": {
"num": null,
"uris": null,
"type_str": "figure",
"text": "Compression using variable-length encoding. (a) A snippet of an (uncompressed) context-encoded array. (b) The context and word deltas. (c) The number of bits required to encode the context and word deltas as well as the value ranks. Word deltas use variable-length block coding with k = 1, while context deltas and value ranks use k = 2. (d) A snippet of the compressed encoding array. The header is outlined in bold."
},
"TABREF2": {
"content": "<table/>",
"text": "Sizes of the two language models used in our experiments.",
"num": null,
"html": null,
"type_str": "table"
},
"TABREF4": {
"content": "<table><tr><td>: Memory usages of several language model im-</td></tr><tr><td>plementations on the WMT2010 language model. A</td></tr><tr><td>* * indicates that the storage in bytes per n-gram is re-</td></tr><tr><td>ported for a different language model of comparable size,</td></tr><tr><td>and the total size is thus a rough projection.</td></tr></table>",
"text": "",
"num": null,
"html": null,
"type_str": "table"
},
"TABREF5": {
"content": "<table><tr><td/><td/><td>WEB1T</td><td/><td/></tr><tr><td>LM Type</td><td colspan=\"3\">bytes/ bytes/ bytes/</td><td>Total</td></tr><tr><td/><td>key</td><td colspan=\"3\">value n-gram Size</td></tr><tr><td>Gzip</td><td>-</td><td>-</td><td>7.0</td><td>24.7G</td></tr><tr><td>T-MPHR \u2020</td><td>-</td><td>-</td><td>3.0</td><td>10.5G</td></tr><tr><td colspan=\"2\">COMPRESSED 1.3</td><td>1.6</td><td>2.9</td><td>10.2G</td></tr></table>",
"text": "",
"num": null,
"html": null,
"type_str": "table"
},
"TABREF6": {
"content": "<table/>",
"text": "Memory usages of several language model implementations on the WEB1T. A \u2020 indicates lossy compression.",
"num": null,
"html": null,
"type_str": "table"
},
"TABREF8": {
"content": "<table><tr><td>LM Type</td><td>No Cache</td><td>Cache</td><td>Size</td></tr><tr><td colspan=\"3\">COMPRESSED 9880\u00b182s 1547\u00b17s</td><td>3.7G</td></tr><tr><td>SRILM-H</td><td>1120\u00b126s</td><td colspan=\"2\">938\u00b111s 26.6G</td></tr><tr><td>HASH</td><td>1146\u00b18s</td><td>943\u00b116s</td><td>7.5G</td></tr></table>",
"text": "Raw query speeds of various language model implementations. Times were averaged over 3 runs on the same machine. For HASH+SCROLL, all queries were issued to the decoder in context-encoded form, which speeds up queries that exhibit scrolling behaviour. Note that memory usage is higher than for HASH because we store suffix offsets along with the values for an n-gram.",
"num": null,
"html": null,
"type_str": "table"
}
}
}
} |