File size: 55,868 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 |
{
"paper_id": "2005",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:17:07.128726Z"
},
"title": "The MIT-LL/AFRL MT System",
"authors": [
{
"first": "Wade",
"middle": [],
"last": "Shen",
"suffix": "",
"affiliation": {},
"email": ""
},
{
"first": "Brian",
"middle": [],
"last": "Delaney",
"suffix": "",
"affiliation": {},
"email": "bdelaney@ll.mit.edu"
},
{
"first": "Tim",
"middle": [],
"last": "Anderson",
"suffix": "",
"affiliation": {},
"email": "timothy.anderson@wpafb.af.mil"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "The MIT-LL/AFRL MT system is a statistical phrase-based translation system that implements many modern SMT training and decoding techniques. Our system was designed with the long term goal of dealing with corrupted ASR input for Speech-to-Speech MT applications. This paper will discuss the architecture of the MIT-LL/AFRL MT system, and experiments with manual and ASR transcription data that were run as part of the IWSLT-2005 Chinese-to-English evaluation campaign. 1",
"pdf_parse": {
"paper_id": "2005",
"_pdf_hash": "",
"abstract": [
{
"text": "The MIT-LL/AFRL MT system is a statistical phrase-based translation system that implements many modern SMT training and decoding techniques. Our system was designed with the long term goal of dealing with corrupted ASR input for Speech-to-Speech MT applications. This paper will discuss the architecture of the MIT-LL/AFRL MT system, and experiments with manual and ASR transcription data that were run as part of the IWSLT-2005 Chinese-to-English evaluation campaign. 1",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "In recent years, the development of statistical methods for machine translation has resulted in high quality translations that can be used in real applications with increasing confidence. Specific advancements include:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "\u2022 Extracting word alignments from parallel corpora [1] [2] \u2022 Learning and modeling the translation of phrases [4] [5] \u2022 Combining and optimizing model parameters [6] [7] [8] \u2022 Decoding and rescoring techniques [9] [10] Our system draws from these advances and implements a number of these techniques including log-linear model combination and minimum error rate training to translate foreign language sentences. We developed our system during preparation for IWSLT-2005 to serve as a platform for future research. Most of the components of our system have been developed in-house in order to facilitate future experimentation.",
"cite_spans": [
{
"start": 51,
"end": 54,
"text": "[1]",
"ref_id": "BIBREF0"
},
{
"start": 55,
"end": 58,
"text": "[2]",
"ref_id": "BIBREF1"
},
{
"start": 110,
"end": 113,
"text": "[4]",
"ref_id": "BIBREF3"
},
{
"start": 114,
"end": 117,
"text": "[5]",
"ref_id": "BIBREF4"
},
{
"start": 162,
"end": 165,
"text": "[6]",
"ref_id": "BIBREF5"
},
{
"start": 166,
"end": 169,
"text": "[7]",
"ref_id": "BIBREF6"
},
{
"start": 170,
"end": 173,
"text": "[8]",
"ref_id": "BIBREF7"
},
{
"start": 210,
"end": 213,
"text": "[9]",
"ref_id": "BIBREF8"
},
{
"start": 214,
"end": 218,
"text": "[10]",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "In subsequent sections, we will discuss the details translation system including our alignment and language models and methods we've implemented for optimization and decoding. The basic translation training and decoding processes are shown in Figure 1 . We start with a word alignment extracted from a training set using GIZA++. These alignments are expanded and phrases are counted to form the phrase translation model. Language models are then trained from the English side of training set (and possibly with other English texts, if available).",
"cite_spans": [],
"ref_spans": [
{
"start": 243,
"end": 251,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "Using development bitexts separated from the training set, we then employ a minimum error rate training process to optimize model parameters utilizing a held out development set. These trained parameters and models can then be applied to test data during decoding and rescoring phases of the translation process. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "The basic phrase-translation model we employ is described in detail [4] and [5] . We use GIZA++ [1] , [3] and [4] to generate alignments between foreign language and English language words/tokens in both directions (f-to-e and e-to-f). An expanded alignment is then computed using heuristics that interpolate between the intersection and union of these bidirectional word alignments as detailed in [4] and [5] . Then phrases are extracted from the expanded alignments to build the phrase model. We introduced two minor modifications to this basic process in our system:",
"cite_spans": [
{
"start": 68,
"end": 71,
"text": "[4]",
"ref_id": "BIBREF3"
},
{
"start": 76,
"end": 79,
"text": "[5]",
"ref_id": "BIBREF4"
},
{
"start": 96,
"end": 99,
"text": "[1]",
"ref_id": "BIBREF0"
},
{
"start": 102,
"end": 105,
"text": "[3]",
"ref_id": "BIBREF2"
},
{
"start": 110,
"end": 113,
"text": "[4]",
"ref_id": "BIBREF3"
},
{
"start": 398,
"end": 401,
"text": "[4]",
"ref_id": "BIBREF3"
},
{
"start": 406,
"end": 409,
"text": "[5]",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Translation Models",
"sec_num": "2."
},
{
"text": "we first add points that are unaligned in both the source and target language sentences that otherwise fit the standard inclusion criteria. The combination of these two enhancements provided a two point improvement in BLEU score on the IWSLT-2004 development set in the supplied data condition.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Prior to iterative expansion of intersection alignment,",
"sec_num": "1."
},
{
"text": "Alignment",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Model Training Translation",
"sec_num": null
},
{
"text": "We extract translation models for both translation directions (i.e. P(f|e) and P(e|f)). In addition to these models, we add lexical weights extracted from the expanded alignment process in both translation directions [5] and a fixed phrasepenalty [11] .",
"cite_spans": [
{
"start": 217,
"end": 220,
"text": "[5]",
"ref_id": "BIBREF4"
},
{
"start": 247,
"end": 251,
"text": "[11]",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Model Training Translation",
"sec_num": null
},
{
"text": "For this evaluation we used a simple distortion model that was described by Koehn et al in [5] and shown here:",
"cite_spans": [
{
"start": 91,
"end": 94,
"text": "[5]",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Distortion Model",
"sec_num": "3."
},
{
"text": "\u2211 \u2212 + \u2212 = \u2212 i i i D FirstW FinalW f e P ) ) 1 ( exp( ) | ( 1 (1) where 1 \u2212 i",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Distortion Model",
"sec_num": "3."
},
{
"text": "FinalW is the position of the final word of the previous phrase and i FirstW is position of the first word of the current phrase. Although simple, we found this model to be very effective for the Chinese-to-English task. Additional constraints on distortion can be chosen to limit the phrase reordering during search, however we chose not to limit distortion when decoding on the IWSLT05 test data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Distortion Model",
"sec_num": "3."
},
{
"text": "We used the SRI language modeling toolkit to build language models [12] . A trigram language model is used during initial decoding. Two additional language models are introduced during rescoring and minimum error rate training: a four-gram language model, and a 5-gram class-based language model. All of these models were trained with modified Knesser-Ney interpolation as suggested by Goodman and Chen in [13] and [14] .",
"cite_spans": [
{
"start": 67,
"end": 71,
"text": "[12]",
"ref_id": "BIBREF11"
},
{
"start": 406,
"end": 410,
"text": "[13]",
"ref_id": "BIBREF12"
},
{
"start": 415,
"end": 419,
"text": "[14]",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Language Models",
"sec_num": "4."
},
{
"text": "Our system employs minimum error rate training to optimize parameter weights for each of the individual model components that are summarized in table 1. Many different optimization strategies have been proposed for combination of log-linear models (see [6] , [7] , [8] and [15] for further details). The algorithm we used is a variation of the method described by Och in [7] . In particular, we optimize each of these parameters assuming log-linear combination as given by The optimization process attempts to sets parameter weights that minimize the overall error rate. To do this, sentences from a development set are used to sample the error surface defined by the model parameters listed above and the loss function to be minimized. Each sentence is decoded to produce n-best translation output samples and a line search is then performed per sentence/model parameter to minimize the overall error rate. This process yields a set of parameter weights for each model that can be then used for decoding and rescoring. The training is iterative in that n-best lists are created using the new parameters and merged with the existing n-best lists to increase the resolution of our sampling for each iteration. As reported by Och [7] , this iterative procedure converges after 6-8 iterations. ClassLM -Five-gram class-based LM ",
"cite_spans": [
{
"start": 253,
"end": 256,
"text": "[6]",
"ref_id": "BIBREF5"
},
{
"start": 259,
"end": 262,
"text": "[7]",
"ref_id": "BIBREF6"
},
{
"start": 265,
"end": 268,
"text": "[8]",
"ref_id": "BIBREF7"
},
{
"start": 273,
"end": 277,
"text": "[15]",
"ref_id": "BIBREF14"
},
{
"start": 371,
"end": 374,
"text": "[7]",
"ref_id": "BIBREF6"
},
{
"start": 1228,
"end": 1231,
"text": "[7]",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Minimum Error Rate Training",
"sec_num": "5."
},
{
"text": "\u2211 \u2211 \u2211 \u2200 \u2200 \u2200 = ' )) ' , ( exp( )) , ( exp( ) | (",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Minimum Error Rate Training",
"sec_num": "5."
},
{
"text": "For the IWSLT 2005 results that were submitted, we used Philip Koehn's Pharaoh Decoder [9] . Pharaoh is stack decoder with an A-star search heuristic. We have recreated similar results with our own in-house decoder, which was not ready before the submission deadline.",
"cite_spans": [
{
"start": 87,
"end": 90,
"text": "[9]",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Decoding",
"sec_num": "6."
},
{
"text": "Our in-house decoder utilizes a Viterbi graph search algorithm, where the graph is built from left to right a word at a time. Each node in the search graph contains a list of backpointers to previous nodes, the probabilities for each translation option, the trigram/bigram context needed for node expansion, and the best path so far. When creating new nodes, the number of nodes can be greatly reduced by using bigrams whenever a particular trigram does not exist in the language model. After all possible phrases are added for each word, both beam pruning and histrogram pruning are used to rid the search graph of unlikely candidate nodes based on the best path.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Decoding",
"sec_num": "6."
},
{
"text": "This search algorithm offers fast decoding and easy generation of output word lattices by simply traversing the final data structure. In the case of monotone decoding, this search algorithm is capable of real-time decoding one word at a time (as from a speech recognizer).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Decoding",
"sec_num": "6."
},
{
"text": "In order to include distortion in the search, additional information needs to be kept for each node including a word coverage vector, the distortion probability, and an estimate of the future cost. Nodes are connected to previous nodes only if the intersection of the word coverage vectors is empty. The future cost estimate is used so that all nodes can be pruned together (A-star search). This limits the search space enough so that unconstrained reordering can be used without running out of memory, but there is a possibility that the search will not be able to select a final path that covers all input words. In the case of a search failure, the search is restarted using improved future cost heuristics from the previous pass.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Decoding",
"sec_num": "6."
},
{
"text": "In this section, we present results from applying the system described above to the IWSLT 2005 evaluation. We present results from both manual and ASR transcription conditions in the supplied data track for Chinese-to-English translation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "7."
},
{
"text": "Because the training data in this track is limited to 20,000 sentence pairs, we employed a number of small modifications to the basic phrase-model extraction procedure to minimize the number of out-of-vocabulary tokens during decoding. These techniques are described in the section below.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "7."
},
{
"text": "For both ASR and manual transcription conditions we did not limit distortion and the four-gram and class-based language models were applied during rescoring.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "7."
},
{
"text": "For efficiency reasons we limited the decoder stack to 350 hypotheses.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "7."
},
{
"text": "Two development sets were provided as part of the IWSLT 2005 evaluation campaign: For minimum error rate training we held out one of devsets 1-4 and tested against another devset (without overlap). Interestingly, devset sentences were generally longer in set 4 and, not surprisingly, scores were generally worse. Devset 5 was used to optimize final parameters for test set decoding. Tables 2 and 3 Table 3 : Devset 3 and 4 Results",
"cite_spans": [],
"ref_spans": [
{
"start": 383,
"end": 397,
"text": "Tables 2 and 3",
"ref_id": "TABREF4"
},
{
"start": 398,
"end": 405,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Dev Set Experiments",
"sec_num": "7.1."
},
{
"text": "Using devset 2 for development testing, we experimented with a number of different configurations of our basic SMT system. Table 4 shows the effect of different preprocessing options and training parameters on BLEU scores for devset 2. Applying these settings and the optimization weights used for devset 2, we tested against each of the ASR outputs for devset 2. Table 5 shows results from each of ASR engine and their corresponding ASR for both 1- For each of the n-best > 1 conditions we simply decode the entire ASR n-best list and merge results before rescoring. We did not weight ASR acoustic or language models for this experiment, but this remains to be explored in future experiments.",
"cite_spans": [
{
"start": 447,
"end": 449,
"text": "1-",
"ref_id": null
}
],
"ref_spans": [
{
"start": 123,
"end": 130,
"text": "Table 4",
"ref_id": "TABREF6"
},
{
"start": 364,
"end": 371,
"text": "Table 5",
"ref_id": "TABREF8"
}
],
"eq_spans": [],
"section": "Dev Set Experiments",
"sec_num": "7.1."
},
{
"text": "We submitted results using multiple optimization parameters from different devsets (devset 5 parameters being primary). These results are shown in Table 6 . As expected, optimizing with devset 5 yielded the best performance. Although our optimization loss function was based on BLEU-4, our system performed reasonably well with other metrics as well (METEOR, PER, NIST). It is also interesting to note that the additional 506 sentences of the CSTAR-03 development set did not provide us much gain over the IWSLT-04 set alone.",
"cite_spans": [],
"ref_spans": [
{
"start": 147,
"end": 154,
"text": "Table 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "IWSLT 2005 Manual Transcription",
"sec_num": "7.2."
},
{
"text": "Better language model rescoring seems to have made the biggest difference. As we did not fully examine language model training possibilities, it is possible that further improvements in this area may yield even better performance. Table 6 : Test set scores with various optimization settings",
"cite_spans": [],
"ref_spans": [
{
"start": 231,
"end": 238,
"text": "Table 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "IWSLT 2005 Manual Transcription",
"sec_num": "7.2."
},
{
"text": "Our primary submission for this track used n-best output from the ASR system. As we saw with our devset 2 experiments, n-best ASR output gives a gain of 2-3 BLEU points over the 1-best hypothesis.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "IWSLT 2005 ASR Transcription",
"sec_num": "7.3."
},
{
"text": "NIST BLEU Table 7 : Test set scores with various optimization settings Again in this condition, our system's performance was quite reasonable, especially with respect to non-BLEU metrics. We expect that more efficient use of ASR model parameters (acoustic and language model scores, and perhaps lattice posteriors) we could yield even better performance.",
"cite_spans": [],
"ref_spans": [
{
"start": 10,
"end": 17,
"text": "Table 7",
"ref_id": null
}
],
"eq_spans": [],
"section": "Configurations N-best Length",
"sec_num": null
},
{
"text": "From experiments with both the development and test data, it is clear that better language modeling can increase the performance of phrase-based statistical machine translation systems as evidence from [16] and [17] also suggests. It is interesting to note that in the case of Chinese translation, distortion models seems to play a major role and, as such, further work is needed to design models that account for the reordering effects we see in Chinese. In past experiments, this has not necessarily been true, for instance with French to English translation monotone decoding often yields better performance.",
"cite_spans": [
{
"start": 202,
"end": 206,
"text": "[16]",
"ref_id": "BIBREF15"
},
{
"start": 211,
"end": 215,
"text": "[17]",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion",
"sec_num": "8."
},
{
"text": "We are just beginning to explore methods for fusing ASR and MT systems. From an anecdotal evaluation of our ASR MT results by a native speaker, we found that only 2 of 30 sampled sentences contained additional errors (relative to MT output from manual transcription). Both this and objective evaluation results from this evaluation are encouraging, but it is clear that much more work is needed to make the output of Speech MT usable. We expect that joint optimization of ASR model parameters and MT model parameters could yield better results but careful research is needed to integrate ASR and MT parameter appropriately.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion",
"sec_num": "8."
}
],
"back_matter": [
{
"text": "We would like to thank John Luu at AFRL for his error analysis efforts after the evaluation. We would also like to thank Doug Jones for setting up an in-house MT evaluation test suite for benchmarking our development progress. Finally, thanks to the members of Lincoln Lab IST group for leaving us plenty of machines to run our test data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": "9."
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "The mathematics of statistical machine translation: Parameter estimation",
"authors": [
{
"first": "P",
"middle": [],
"last": "Brown",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Della Pietra",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Della Pietra",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Mercer",
"suffix": ""
}
],
"year": 1993,
"venue": "Computational Linguistics",
"volume": "19",
"issue": "2",
"pages": "263--311",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Brown, P., Della Pietra, V., Della Pietra, S. and Mercer, R. \"The mathematics of statistical machine translation: Parameter estimation\", Computational Linguistics 19(2):263--311. 1993.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "HMM-based word alignment in statistical translation",
"authors": [
{
"first": "S",
"middle": [],
"last": "Vogel",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Ney",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Tillmann",
"suffix": ""
}
],
"year": 1996,
"venue": "Proceedings of the 16th Conference on Computational Linguistics",
"volume": "2",
"issue": "",
"pages": "836--841",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Vogel, S., Ney, H., and Tillmann, C. \"HMM-based word alignment in statistical translation\", In Proceedings of the 16th Conference on Computational Linguistics -Volume 2, pp. 836-841, Copenhagen, Denmark, August, 1996.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Statistical machine translation: Final report",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Al-Onaizan",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Curin",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Jahr",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Knight",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Lafferty",
"suffix": ""
},
{
"first": "I",
"middle": [
"D"
],
"last": "Melamed",
"suffix": ""
},
{
"first": "F",
"middle": [
"J"
],
"last": "Och",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Purdy",
"suffix": ""
},
{
"first": "N",
"middle": [
"A"
],
"last": "Smith",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Yarowsky",
"suffix": ""
}
],
"year": 1999,
"venue": "Proceedings of the Summer Workshop on Language Engineering",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Al-Onaizan, Y., Curin, J., Jahr, M., Knight, K., Lafferty, J., Melamed, I.D., Och, F.J., Purdy, D., Smith, N.A., Yarowsky, D., \"Statistical machine translation: Final report\", In Proceedings of the Summer Workshop on Language Engineering. John Hopkins University Center for Language and Speech Processing, Baltimore, MD 1999.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Improved Statistical Alignment Models",
"authors": [
{
"first": "F",
"middle": [
"J"
],
"last": "Och",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Hermann",
"suffix": ""
}
],
"year": 2000,
"venue": "Proc. of the 38th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "440--447",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Och, F. J. and Hermann, N. \"Improved Statistical Alignment Models\", In Proc. of the 38th Annual Meeting of the Association for Computational Linguistics, pp. 440-447, Hong Kong, October, 2000.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Statistical Phrase-Based Translation",
"authors": [
{
"first": "P",
"middle": [],
"last": "Koehn",
"suffix": ""
},
{
"first": "F",
"middle": [
"J"
],
"last": "Och",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Marcu",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of the Human Language Technology Conference 2003 (HLT-NAACL 2003)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Koehn, P., Och, F. J. and Marcu, D., \"Statistical Phrase- Based Translation\", In Proceedings of the Human Language Technology Conference 2003 (HLT-NAACL 2003), Edmonton, Canada, May 2003.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Discriminative Training and Maximum Entropy Models for Statistical Machine Translation",
"authors": [
{
"first": "F",
"middle": [
"J"
],
"last": "Och",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Ney",
"suffix": ""
}
],
"year": 2002,
"venue": "ACL 2002: Proc. of the 40th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "295--302",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Och, F. J. and Ney, H., \"Discriminative Training and Maximum Entropy Models for Statistical Machine Translation\". In ACL 2002: Proc. of the 40th Annual Meeting of the Association for Computational Linguistics, pp. 295-302, Philadelphia, PA, July 2002.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Minimum Error Rate Training for Statistical Machine Translation",
"authors": [
{
"first": "F",
"middle": [
"J"
],
"last": "Och",
"suffix": ""
}
],
"year": 2003,
"venue": "ACL 2003: Proc. of the 41st Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Och, F. J., \"Minimum Error Rate Training for Statistical Machine Translation\", In ACL 2003: Proc. of the 41st Annual Meeting of the Association for Computational Linguistics, Japan, Sapporo, July 2003.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Considerations in Minimum Classification Error and Maximum Mutual Information Training for Statistical Machine Translation",
"authors": [
{
"first": "A",
"middle": [],
"last": "Venugopal",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Vogel",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of the Tenth Conference of the European Association for Machine Translation (EAMT-05)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Venugopal, A. and Vogel S., \"Considerations in Minimum Classification Error and Maximum Mutual Information Training for Statistical Machine Translation\", In Proceedings of the Tenth Conference of the European Association for Machine Translation (EAMT-05), Budapest, Hungary, May 2005.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Pharaoh: A Beam Search Decoder for Phrase-Based Statistical Machine Translation Models",
"authors": [
{
"first": "P",
"middle": [],
"last": "Koehn",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of the Association of Machine Translation in the Americas (AMTA-2004)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Koehn, P., \"Pharaoh: A Beam Search Decoder for Phrase-Based Statistical Machine Translation Models\", In Proceedings of the Association of Machine Translation in the Americas (AMTA-2004), Washington, DC, October, 2004.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Generation of Word Graphs in Statistical Machine Translation",
"authors": [
{
"first": "N",
"middle": [],
"last": "Ueffing",
"suffix": ""
},
{
"first": "F",
"middle": [
"J"
],
"last": "Och",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Ney",
"suffix": ""
}
],
"year": 2002,
"venue": "Proc. Conference on Empirical Methods for Natural Language Processing",
"volume": "",
"issue": "",
"pages": "156--163",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ueffing, N., Och, F. J., Ney, H., \"Generation of Word Graphs in Statistical Machine Translation\", In Proc. Conference on Empirical Methods for Natural Language Processing, pp. 156-163, Philadelphia, PA, July 2002.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "The ISI/USC MT System",
"authors": [
{
"first": "I",
"middle": [],
"last": "Thayer",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Ettelaie",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Knight",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Marcu",
"suffix": ""
},
{
"first": "D",
"middle": [
"S"
],
"last": "Munteanu",
"suffix": ""
},
{
"first": "F",
"middle": [
"J"
],
"last": "Och",
"suffix": ""
},
{
"first": "Q",
"middle": [],
"last": "Tipu",
"suffix": ""
}
],
"year": 2004,
"venue": "Proc. of the International Workshop on Spoken Language Translation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Thayer, I., Ettelaie, E., Knight, K., Marcu, D., Munteanu D. S., Och F. J., and Tipu, Q., \"The ISI/USC MT System\", In Proc. of the International Workshop on Spoken Language Translation, Kyoto, Japan, 2004.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "SRILM -An Extensible Language Modeling Toolkit",
"authors": [
{
"first": "A",
"middle": [],
"last": "Stolcke",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the International Conference on Spoken Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Stolcke, A., \"SRILM -An Extensible Language Modeling Toolkit\", In Proceedings of the International Conference on Spoken Language Processing, Denver, CO, 2002.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "An Empirical Study of Smoothing Techniques for Language Modeling",
"authors": [
{
"first": "S",
"middle": [
"F"
],
"last": "Chen",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Goodman",
"suffix": ""
}
],
"year": 1999,
"venue": "Computer Speech and Language",
"volume": "13",
"issue": "",
"pages": "359--394",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Chen, S. F. and Goodman, J., \"An Empirical Study of Smoothing Techniques for Language Modeling\", Computer Speech and Language, 13:359-394, October, 1999.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"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., \"A Bit of Progress in Language Modeling\", Computer Speech and Language, pp. 403-434, 2001.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Minimum Error Rate Training of Log-Linear Translation Models",
"authors": [
{
"first": "M",
"middle": [],
"last": "Cettolo",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Federico",
"suffix": ""
}
],
"year": 2004,
"venue": "Proc. of the International Workshop on Spoken Language Translation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cettolo, M. and Federico, M., \"Minimum Error Rate Training of Log-Linear Translation Models\", In Proc. of the International Workshop on Spoken Language Translation, Kyoto, Japan, 2004.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Alignment Templates: the RWTH SMT System",
"authors": [
{
"first": "O",
"middle": [],
"last": "Bender",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Zens",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Matusov",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Ney",
"suffix": ""
}
],
"year": 2004,
"venue": "Proc. of the International Workshop on Spoken Language Translation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bender, O., Zens, R., Matusov, E. and Ney, H., \"Alignment Templates: the RWTH SMT System\", In Proc. of the International Workshop on Spoken Language Translation, Kyoto, Japan, 2004.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "The Google MT System",
"authors": [
{
"first": "F",
"middle": [],
"last": "Och",
"suffix": ""
}
],
"year": 2005,
"venue": "Presentation at NIST MT Workshop",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Och, F., \"The Google MT System\", Presentation at NIST MT Workshop, Bethesda, MD, 2005.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"text": "Basic Statistical Translation Architecture",
"type_str": "figure",
"uris": null,
"num": null
},
"FIGREF1": {
"text": "the feature function associated with estimates from each individual translation/language model.",
"type_str": "figure",
"uris": null,
"num": null
},
"FIGREF2": {
"text": "Evaluation data from CSTAR 2003 Evaluation data from IWSLT 2004We used these tests to construct 3 other dev sets for more robust parameter optimization:\u2022 Devset 3: first \u00bd of Devsets 1 and Devset 4: second \u00bd of Devsets 1 and Devset 5: sum of Devset 1 and 2",
"type_str": "figure",
"uris": null,
"num": null
},
"TABREF2": {
"html": null,
"type_str": "table",
"num": null,
"text": "",
"content": "<table/>"
},
"TABREF3": {
"html": null,
"type_str": "table",
"num": null,
"text": "shows scores from different development run configurations.",
"content": "<table><tr><td>Dev</td><td>1</td><td>2</td></tr><tr><td>Test</td><td/><td/></tr><tr><td>1</td><td/><td>36.64</td></tr><tr><td>2</td><td>42.00</td><td/></tr></table>"
},
"TABREF4": {
"html": null,
"type_str": "table",
"num": null,
"text": "Devset 1 and 2 Results",
"content": "<table><tr><td>Dev</td><td>3</td><td>4</td></tr><tr><td>Test</td><td/><td/></tr><tr><td>3</td><td/><td>42.44</td></tr><tr><td>4</td><td>33.84</td><td/></tr></table>"
},
"TABREF6": {
"html": null,
"type_str": "table",
"num": null,
"text": "",
"content": "<table/>"
},
"TABREF8": {
"html": null,
"type_str": "table",
"num": null,
"text": "Devset 2 results with various ASR transcripts.",
"content": "<table/>"
}
}
}
} |