File size: 54,966 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 |
{
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T02:09:44.202811Z"
},
"title": "XMU's Simultaneous Translation System at NAACL 2021",
"authors": [
{
"first": "Shuangtao",
"middle": [],
"last": "Li",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Xiamen University",
"location": {
"settlement": "Xiamen",
"country": "China"
}
},
"email": "lishuangtao@stu.xmu.edu.com"
},
{
"first": "Jinming",
"middle": [],
"last": "Hu",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Xiamen University",
"location": {
"settlement": "Xiamen",
"country": "China"
}
},
"email": ""
},
{
"first": "Boli",
"middle": [],
"last": "Wang",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Xiamen University",
"location": {
"settlement": "Xiamen",
"country": "China"
}
},
"email": ""
},
{
"first": "Xiaodong",
"middle": [],
"last": "Shi",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Xiamen University",
"location": {
"settlement": "Xiamen",
"country": "China"
}
},
"email": ""
},
{
"first": "Yidong",
"middle": [],
"last": "Chen",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Xiamen University",
"location": {
"settlement": "Xiamen",
"country": "China"
}
},
"email": "ydchen@xmu.edu.cn"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "This paper describes XMU's two systems submitted to the simultaneous translation evaluation at the 2nd automatic simultaneous translation workshop, which are for Zh->En textto-text translation and Zh->En speech-to-text translation. In both systems, our translation model is based on Transformer. To translate streaming text, we use an adaptive policy to split the text into appropriate segments and translate them monotonically. Our speech-totext system is a pipeline system, in which the MT component is exactly the same as our textto-text system.",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "This paper describes XMU's two systems submitted to the simultaneous translation evaluation at the 2nd automatic simultaneous translation workshop, which are for Zh->En textto-text translation and Zh->En speech-to-text translation. In both systems, our translation model is based on Transformer. To translate streaming text, we use an adaptive policy to split the text into appropriate segments and translate them monotonically. Our speech-totext system is a pipeline system, in which the MT component is exactly the same as our textto-text system.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Simultaneous translation refers to translating the message from the speaker to the audience in realtime without interrupting the speaker. It is widely used in many scenarios such as international conferences and business negotiations. Simultaneous machine translation is a challenging task and has become an increasingly popular research field in recent years.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "There have been some researches on simultaneous translation of speech input (Niehues et al., 2018; Ma et al., 2020b,c; Ren et al., 2020) , and some researches focused on text translation (Arivazhagan et al., 2019; Ma et al., 2020a) .",
"cite_spans": [
{
"start": 76,
"end": 98,
"text": "(Niehues et al., 2018;",
"ref_id": "BIBREF8"
},
{
"start": 99,
"end": 118,
"text": "Ma et al., 2020b,c;",
"ref_id": null
},
{
"start": 119,
"end": 136,
"text": "Ren et al., 2020)",
"ref_id": "BIBREF12"
},
{
"start": 187,
"end": 213,
"text": "(Arivazhagan et al., 2019;",
"ref_id": "BIBREF0"
},
{
"start": 214,
"end": 231,
"text": "Ma et al., 2020a)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper, we describe our two systems submitted to the simultaneous translation evaluation at the 2nd automatic simultaneous translation workshop, which are for Zh->En text-to-text translation and Zh->En speech-to-text translation. We build our systems with the state-of-the-art method , and verify the effectiveness of this method.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this section, we describe our system submitted to Zh->En text-to-text simultaneous translation track.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Text-to-text Track",
"sec_num": "2"
},
{
"text": "The main idea of this system is how human interpreters work. While listening to speakers, human interpreters constantly translate text segments that are appropriate to translate without waiting for more words, and meanwhile making the translation grammatically tolerable. Text segments considered appropriate to translate usually have clear and definite meaning, because the translation of such a segment is not likely to be changed by subsequent text. The authors of refered to such segments Meaningful Units (MU) and gived MUs a precise difinition. See Table 1 for an illustration.",
"cite_spans": [],
"ref_spans": [
{
"start": 555,
"end": 562,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Text-to-text Track",
"sec_num": "2"
},
{
"text": "Our system works like a human interpreter described above and is composed of an MU classification model and a machine translation model. Once a segment is classified to be an MU by the MU classifier, the MT model uses forced decoding with previous translation as the prefix to translate the segment, as shown in Table 1 .",
"cite_spans": [],
"ref_spans": [
{
"start": 312,
"end": 319,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Text-to-text Track",
"sec_num": "2"
},
{
"text": "Our machine translation model is implemented with FAIRSEQ 1 (Ott et al., 2019) .",
"cite_spans": [
{
"start": 60,
"end": 78,
"text": "(Ott et al., 2019)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Machine translation",
"sec_num": "2.1"
},
{
"text": "Data and preprocessing. The data we use are CWMT19 2 (9.1M parallel sentences pairs) and the simultaneous translation corpus (39K parallel sentences pairs) provided by the organizer of the workshop.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Machine translation",
"sec_num": "2.1"
},
{
"text": "We do the following steps to preprocess the data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Machine translation",
"sec_num": "2.1"
},
{
"text": "\u2022 Filtering. The sentence pairs whose English sentence is longer than 120 words are filtered out. Table 1 : An illustration of how a human interpreter work. The source sentence is splited to three MUs (separated by \"||\"), and an interpreter translates the MUs in order and makes them form a grammatically correct sentence.",
"cite_spans": [],
"ref_spans": [
{
"start": 98,
"end": 105,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Machine translation",
"sec_num": "2.1"
},
{
"text": "\u2022 There are a few punctuation marks, numbers and letters in the data which are in full width. They are converted to half width characters.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Machine translation",
"sec_num": "2.1"
},
{
"text": "\u2022 There are a few Chinese characters in the data which are traditional characters. They are converted to simplified ones.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Machine translation",
"sec_num": "2.1"
},
{
"text": "\u2022 Chinese segmentation. All Chinese sentences are segmented with Jieba Chinese Segmentation Tool 3 .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Machine translation",
"sec_num": "2.1"
},
{
"text": "\u2022 English tokenization. All English sentences are tokenized and truecased with Moses 4 .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Machine translation",
"sec_num": "2.1"
},
{
"text": "\u2022 Byte-pair-encoding (BPE) (Sennrich et al., 2016) . Both Chinese and English data are encoded by BPE with Subword-NMT 5 . The number of merge operations for each language is set to 30K.",
"cite_spans": [
{
"start": 27,
"end": 50,
"text": "(Sennrich et al., 2016)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Machine translation",
"sec_num": "2.1"
},
{
"text": "Modeling and training. Our translation model's architecture is base Transformer (Vaswani et al., 2017) . We use Adam optimizer (Kingma and Ba, 2015) to optimize the loss. We use weight decay of 1e \u22124 and dropout with probability of 0.2 for regularization. Label smoothing with of 0.1 is applied to our model. During inference, we set beam size to 20. Our model is first pretrained on CWMT and then finetuned on the the training set of the Baidu Speech Translation Corpus (Zhang et al., 2021) . We set learning rate to 5e \u22124 in the pretraining stage and 3e \u22125 in the fine-tuning stage. The learning rate is linearly increased for the first 4000 training steps, and is decreased following an inverse squareroot schedule.",
"cite_spans": [
{
"start": 80,
"end": 102,
"text": "(Vaswani et al., 2017)",
"ref_id": null
},
{
"start": 471,
"end": 491,
"text": "(Zhang et al., 2021)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Machine translation",
"sec_num": "2.1"
},
{
"text": "Modeling and training. The MU classifier is a binary classifier. Given a source word sequence x = {x 1 , x 2 , ..., x n }, the MU classifier determines whether x ends with an MU, and if so the MT will translate x with forced decoding. The input of the classifier is x and m \"future\" words",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "MU classifier",
"sec_num": "2.2"
},
{
"text": "x x f (m = 2) c \u725b\u987f || \u53d1\u73b0 \u4e86 1 \u725b\u987f || \u53d1\u73b0 \u4e86 \u725b\u987f 0 \u725b\u987f || \u53d1\u73b0 \u4e86 || \u725b\u987f \u8fd0\u52a8 1 \u725b\u987f || \u53d1\u73b0 \u4e86 || \u725b\u987f \u8fd0\u52a8 \u5b9a\u5f8b 0",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "MU classifier",
"sec_num": "2.2"
},
{
"text": "x f = {x n+1 , x n+2 , ..., x n+m },",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "MU classifier",
"sec_num": "2.2"
},
{
"text": "where m is a hyperparameter. The outputs are the probabilities of two classes p(c = 1) and p(c = 0), which mean x ends with an MU or not. x will be classified into class 1 if p(c = 1) > t, where t is a threshold set based on experience. Obviously, we can control the latency of simultaneous translation by modifying m. Later in experiments, it will be shown that we can also control the latency by modifying t. In our system, m is always set to 2.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "MU classifier",
"sec_num": "2.2"
},
{
"text": "The MU classifier is based on a chinese BERT (Devlin et al., 2019) 6 . We use the base model and fine-tune it with a learning rate of 5e \u22124 .",
"cite_spans": [
{
"start": 45,
"end": 68,
"text": "(Devlin et al., 2019) 6",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "MU classifier",
"sec_num": "2.2"
},
{
"text": "Generating MU samples. To build an MU classifier, we need to generate MU samples just like the samples in Table 2 . For each sentence of length N , we generate N \u2212 m examples for it, and every MU sample is a triple < x, x f , c >. When we generate examples, c is set to 1 if x ends with an MU, else it is set to 0. In our system, we generate MU samples for every sentence pairs in CWMT and the simultaneous translation corpus. Our MU samples are a little different from the MU samples in . In their work, the future words of a sample can be less than m, but not in this paper. We do not need training samples whose future words are less than m, because during inference when the future words are less than m, the sentence is already a whole sentence and thus can be fed into MT.",
"cite_spans": [],
"ref_spans": [
{
"start": 106,
"end": 113,
"text": "Table 2",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "MU classifier",
"sec_num": "2.2"
},
{
"text": "We use the basic method proposed in to generate MU samples.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "MU classifier",
"sec_num": "2.2"
},
{
"text": "In this section, we describe our system submitted to Zh->En speech-to-text simultaneous translation track.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Speech-to-text Track",
"sec_num": "3"
},
{
"text": "This system is a pipeline of three stages: (1) speech recognition, (2) punctuation restoration, and (3) streaming machine translation. The third stage is exactly the same as the system described in Section 2. In other words, the system described in Section 2 is a part of our speech-to-text system, and therefore it will not be repeated in this section. This section only describes the stage (1) and stage (2).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Speech-to-text Track",
"sec_num": "3"
},
{
"text": "Instead of building a speech recognition model, we use Baidu's real time speech recognition service 7 . In our system, We call the API of this service to recognize streaming speech. It is important to note that although the ASR does not output punctuation, it separates different sentences, that is, the ASR outputs are many segmented sentences instead of one sentence.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Speech recognition",
"sec_num": "3.1"
},
{
"text": "The recognition results of Baidu's asr service do not have punctuation, but the input of our MT model needs punctuation. As a result we build a model to restore the punctuation for every recognition result. We use a BERT-based (Devlin et al., 2019) sequence labeling model (Chen and Shi, 2020) to do punctuation restoration. This model labels every Chinese character in a sentence, and for the model we only consider four classes: comma, period, question mark and no punctuation.",
"cite_spans": [
{
"start": 227,
"end": 248,
"text": "(Devlin et al., 2019)",
"ref_id": "BIBREF2"
},
{
"start": 273,
"end": 293,
"text": "(Chen and Shi, 2020)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Punctuation restoration",
"sec_num": "3.2"
},
{
"text": "In this section, we evaluate our two systems on the development set of the Baidu Speech Translation Corpus (Zhang et al., 2021) . The two used metrics are case-sensitive detokenized BLEU (Papineni et al., 2002) and Consecutive Wait (CW) (Neubig et al., 2017) , for translation quality and latency respectively. CW considers on how many source words are waited for consecutively between two target words, and thus larger CW means longer latency. We use SacreBLEU (Post, 2018) to compute BLEU scores. ",
"cite_spans": [
{
"start": 107,
"end": 127,
"text": "(Zhang et al., 2021)",
"ref_id": "BIBREF15"
},
{
"start": 187,
"end": 210,
"text": "(Papineni et al., 2002)",
"ref_id": "BIBREF10"
},
{
"start": 237,
"end": 258,
"text": "(Neubig et al., 2017)",
"ref_id": "BIBREF7"
},
{
"start": 462,
"end": 474,
"text": "(Post, 2018)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "4"
},
{
"text": "We set the threshold t in the MU classifier to various values and get multiple results, as shown in Figure 1 . It is worth noting that the rightmost point in Figure 1 is not a result of simultaneous translation. This result is got by translating every sentence after it is finished, i.e., we get this result by translating whole sentences.",
"cite_spans": [],
"ref_spans": [
{
"start": 100,
"end": 108,
"text": "Figure 1",
"ref_id": "FIGREF0"
},
{
"start": 158,
"end": 166,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Text-to-text track",
"sec_num": "4.1"
},
{
"text": "The experimental results are shown in Figure 2 . Similarly, the rightmost point is not a result of simultaneous translation. Because the speech in the development set is difficult for ASR, the ASR does not perform well, resulting in a character error rate of 35.3%. The errors caused by ASR are brought to MT, and thus the BLEU is much lower than in the text-to-text track.",
"cite_spans": [],
"ref_spans": [
{
"start": 38,
"end": 46,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Speech-to-text track",
"sec_num": "4.2"
},
{
"text": "From Figure 1 and Figure 2 , we can observe that the larger the threshold t is, the longer the latency is. This is because the larger the threshold t is, the longer the detected MUs are, which further leads to longer waiting time between two translations. We can also observe that the larger the threshold t is, the higher the translation quality is. This is because the larger the threshold t is, the more likely a detected MU is a true MU and thus the translation of the detected MU will not be changed by subsequent incoming text. Table 3 is an illustration for this.",
"cite_spans": [],
"ref_spans": [
{
"start": 5,
"end": 13,
"text": "Figure 1",
"ref_id": "FIGREF0"
},
{
"start": 18,
"end": 26,
"text": "Figure 2",
"ref_id": "FIGREF1"
},
{
"start": 534,
"end": 541,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Analysis",
"sec_num": "4.3"
},
{
"text": "Source: \u597d \uff0c \u8ba9 \u6211\u4eec \u6765\u770b \u4e0b\u4e2a \u4f8b\u5b50 \u3002 okay let we look next example Reference:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Analysis",
"sec_num": "4.3"
},
{
"text": "Okay , let 's look at the next example . Simultaneous Translation (t = 0.7): OK , let 's look at the next example . Simultaneous Translation (t = 0.5): Okay , || let 's look at the next example . Simultaneous Translation (t = 0.3): Okay , || let 's do it . || let 's look at the next example . Table 3 : An illustration of text-to-text simultaneous translation with different threshold t. \"||\" is a symbol for separating the translations of detected MUs. ",
"cite_spans": [],
"ref_spans": [
{
"start": 294,
"end": 301,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Analysis",
"sec_num": "4.3"
},
{
"text": "We have built two systems for text-to-text simultaneous translation and speech-to-text simultaneous translation. The key of our systems is the basic adaptive segmentation policy in . With this policy, simultaneous translation can be achieved without any modification to the MT component, and the latency can be controlled. In our future work, we would like to study how to improve the cooperation of ASR and MT and the performance of MU classification.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "https://github.com/fxsjy/jieba 4 https://github.com/moses-smt/mosesdecoder 5 https://github.com/rsennrich/subword-nmt",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://github.com/649453932/Bert-Chinese-Text-Classification-Pytorch",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Monotonic infinite lookback attention for simultaneous machine translation",
"authors": [
{
"first": "N",
"middle": [],
"last": "Arivazhagan",
"suffix": ""
},
{
"first": "Colin",
"middle": [],
"last": "Cherry",
"suffix": ""
},
{
"first": "Wolfgang",
"middle": [],
"last": "Macherey",
"suffix": ""
},
{
"first": "Chung-Cheng",
"middle": [],
"last": "Chiu",
"suffix": ""
},
{
"first": "Semih",
"middle": [],
"last": "Yavuz",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Pang",
"suffix": ""
},
{
"first": "Wei",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Colin",
"middle": [],
"last": "Raffel",
"suffix": ""
}
],
"year": 2019,
"venue": "ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "N. Arivazhagan, Colin Cherry, Wolfgang Macherey, Chung-Cheng Chiu, Semih Yavuz, R. Pang, Wei Li, and Colin Raffel. 2019. Monotonic infinite look- back attention for simultaneous machine translation. In ACL.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Improving machine simultaneous interpretation by punctuation recovery",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "X",
"middle": [],
"last": "Shi",
"suffix": ""
}
],
"year": 2020,
"venue": "Journal of Computer Applications",
"volume": "40",
"issue": "4",
"pages": "972--977",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Y. Chen and X. Shi. 2020. Improving machine simulta- neous interpretation by punctuation recovery. Jour- nal of Computer Applications, 40(4):972-977.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Bert: Pre-training of deep bidirectional transformers for language understanding",
"authors": [
{
"first": "J",
"middle": [],
"last": "Devlin",
"suffix": ""
},
{
"first": "Ming-Wei",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "Kenton",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Kristina",
"middle": [],
"last": "Toutanova",
"suffix": ""
}
],
"year": 2019,
"venue": "NAACL-HLT",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirec- tional transformers for language understanding. In NAACL-HLT.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Adam: A method for stochastic optimization",
"authors": [
{
"first": "P",
"middle": [],
"last": "Diederik",
"suffix": ""
},
{
"first": "Jimmy",
"middle": [],
"last": "Kingma",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Ba",
"suffix": ""
}
],
"year": 2015,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Diederik P. Kingma and Jimmy Ba. 2015. Adam: A method for stochastic optimization. CoRR, abs/1412.6980.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Simulmt to simulst: Adapting simultaneous text translation to end-to-end simultaneous speech translation",
"authors": [
{
"first": "Xutai",
"middle": [],
"last": "Ma",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Pino",
"suffix": ""
},
{
"first": "Philipp",
"middle": [],
"last": "Koehn",
"suffix": ""
}
],
"year": 2020,
"venue": "AACL/IJCNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xutai Ma, J. Pino, and Philipp Koehn. 2020b. Simulmt to simulst: Adapting simultaneous text translation to end-to-end simultaneous speech translation. In AACL/IJCNLP.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Streaming simultaneous speech translation with augmented memory transformer",
"authors": [
{
"first": "Xutai",
"middle": [],
"last": "Ma",
"suffix": ""
},
{
"first": "Yongqiang",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Dousti",
"suffix": ""
},
{
"first": "Philipp",
"middle": [],
"last": "Koehn",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Pino",
"suffix": ""
}
],
"year": 2020,
"venue": "ArXiv",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xutai Ma, Yongqiang Wang, M. Dousti, Philipp Koehn, and J. Pino. 2020c. Streaming simultaneous speech translation with augmented memory transformer. ArXiv, abs/2011.00033.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Learning to translate in real-time with neural machine translation",
"authors": [
{
"first": "Graham",
"middle": [],
"last": "Neubig",
"suffix": ""
},
{
"first": "Kyunghyun",
"middle": [],
"last": "Cho",
"suffix": ""
},
{
"first": "Jiatao",
"middle": [],
"last": "Gu",
"suffix": ""
},
{
"first": "O",
"middle": [
"K"
],
"last": "Victor",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Li",
"suffix": ""
}
],
"year": 2017,
"venue": "EACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Graham Neubig, Kyunghyun Cho, Jiatao Gu, and Vic- tor O. K. Li. 2017. Learning to translate in real-time with neural machine translation. In EACL.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Low-latency neural speech translation",
"authors": [
{
"first": "J",
"middle": [],
"last": "Niehues",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Pham",
"suffix": ""
},
{
"first": "Thanh-Le",
"middle": [],
"last": "Ha",
"suffix": ""
},
{
"first": "Matthias",
"middle": [],
"last": "Sperber",
"suffix": ""
},
{
"first": "Alexander",
"middle": [
"H"
],
"last": "Waibel",
"suffix": ""
}
],
"year": 2018,
"venue": "INTERSPEECH",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Niehues, N. Pham, Thanh-Le Ha, Matthias Sperber, and Alexander H. Waibel. 2018. Low-latency neural speech translation. In INTERSPEECH.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "fairseq: A fast, extensible toolkit for sequence modeling",
"authors": [
{
"first": "Myle",
"middle": [],
"last": "Ott",
"suffix": ""
},
{
"first": "Sergey",
"middle": [],
"last": "Edunov",
"suffix": ""
},
{
"first": "Alexei",
"middle": [],
"last": "Baevski",
"suffix": ""
},
{
"first": "Angela",
"middle": [],
"last": "Fan",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Gross",
"suffix": ""
},
{
"first": "Nathan",
"middle": [],
"last": "Ng",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Grangier",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Auli",
"suffix": ""
}
],
"year": 2019,
"venue": "NAACL-HLT",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, S. Gross, Nathan Ng, David Grangier, and M. Auli. 2019. fairseq: A fast, extensible toolkit for sequence modeling. In NAACL-HLT.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Bleu: a method for automatic evaluation of machine translation",
"authors": [
{
"first": ",",
"middle": [
"S"
],
"last": "Kishore Papineni",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Roukos",
"suffix": ""
},
{
"first": "Wei-Jing",
"middle": [],
"last": "Ward",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Zhu",
"suffix": ""
}
],
"year": 2002,
"venue": "ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kishore Papineni, S. Roukos, T. Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In ACL.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "A call for clarity in reporting BLEU scores",
"authors": [
{
"first": "Matt",
"middle": [],
"last": "Post",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the Third Conference on Machine Translation: Research Papers",
"volume": "",
"issue": "",
"pages": "186--191",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Matt Post. 2018. A call for clarity in reporting BLEU scores. In Proceedings of the Third Conference on Machine Translation: Research Papers, pages 186- 191, Belgium, Brussels. Association for Computa- tional Linguistics.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Simulspeech: End-to-end simultaneous speech to text translation",
"authors": [
{
"first": "J",
"middle": [],
"last": "Yi Ren",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Xu Tan",
"suffix": ""
},
{
"first": "Tao",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Zhou",
"middle": [],
"last": "Qin",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Zhao",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Liu",
"suffix": ""
}
],
"year": 2020,
"venue": "ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yi Ren, J. Liu, Xu Tan, C. Zhang, Tao Qin, Zhou Zhao, and T. Liu. 2020. Simulspeech: End-to-end simulta- neous speech to text translation. In ACL.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Neural machine translation of rare words with subword units",
"authors": [
{
"first": "Rico",
"middle": [],
"last": "Sennrich",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Haddow",
"suffix": ""
},
{
"first": "Alexandra",
"middle": [],
"last": "Birch",
"suffix": ""
}
],
"year": 2016,
"venue": "ArXiv",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rico Sennrich, B. Haddow, and Alexandra Birch. 2016. Neural machine translation of rare words with sub- word units. ArXiv, abs/1508.07909.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Bstc: A large-scale chinese-english speech translation dataset",
"authors": [
{
"first": "Ruiqing",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Xiyang",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Chuanqiang",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Zhongjun",
"middle": [],
"last": "He",
"suffix": ""
},
{
"first": "Hua",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Zhi",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Haifeng",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Ying",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Qinfei",
"middle": [],
"last": "Li",
"suffix": ""
}
],
"year": 2021,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:2104.03575"
]
},
"num": null,
"urls": [],
"raw_text": "Ruiqing Zhang, Xiyang Wang, Chuanqiang Zhang, Zhongjun He, Hua Wu, Zhi Li, Haifeng Wang, Ying Chen, and Qinfei Li. 2021. Bstc: A large-scale chinese-english speech translation dataset. arXiv preprint arXiv:2104.03575.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Learning adaptive segmentation policy for simultaneous translation",
"authors": [
{
"first": "Ruiqing",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Chuanqiang",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Zhongjun",
"middle": [],
"last": "He",
"suffix": ""
},
{
"first": "Hua",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Haifeng",
"middle": [],
"last": "Wang",
"suffix": ""
}
],
"year": 2020,
"venue": "EMNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "RuiQing Zhang, Chuanqiang Zhang, Zhongjun He, Hua Wu, and Haifeng Wang. 2020. Learning adap- tive segmentation policy for simultaneous transla- tion. In EMNLP.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"text": "7 https://cloud.baidu.com/doc/SPEECH/s/2k5dllqxj Translation quality against latency of different thresholds t. The rightmost point is not a result of simultaneous translation, but a result got by translating complete sentences.",
"uris": null,
"num": null,
"type_str": "figure"
},
"FIGREF1": {
"text": "Translation quality against latency of different thresholds t. The rightmost point is not a result of simultaneous translation, but a result got by translating complete sentences.",
"uris": null,
"num": null,
"type_str": "figure"
},
"TABREF1": {
"text": "MU samples for the MU classifier. \"||\" is a symbol to separate MUs. c = 1 means that x ends with an MU, otherwise not.",
"html": null,
"num": null,
"type_str": "table",
"content": "<table/>"
}
}
}
} |