File size: 80,045 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 | {
"paper_id": "C08-1013",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T12:26:18.896674Z"
},
"title": "ParaMetric: An Automatic Evaluation Metric for Paraphrasing",
"authors": [
{
"first": "Chris",
"middle": [],
"last": "Callison-Burch",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Johns Hopkins University",
"location": {
"addrLine": "3400 N. Charles St",
"postCode": "21218",
"settlement": "Baltimore",
"region": "MD"
}
},
"email": ""
},
{
"first": "Trevor",
"middle": [],
"last": "Cohn",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Edinburgh",
"location": {
"addrLine": "2 Buccleuch Place Edinburgh",
"postCode": "EH8 9LW"
}
},
"email": ""
},
{
"first": "Mirella",
"middle": [],
"last": "Lapata",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Edinburgh",
"location": {
"addrLine": "2 Buccleuch Place Edinburgh",
"postCode": "EH8 9LW"
}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "We present ParaMetric, an automatic evaluation metric for data-driven approaches to paraphrasing. ParaMetric provides an objective measure of quality using a collection of multiple translations whose paraphrases have been manually annotated. ParaMetric calculates precision and recall scores by comparing the paraphrases discovered by automatic paraphrasing techniques against gold standard alignments of words and phrases within equivalent sentences. We report scores for several established paraphrasing techniques.",
"pdf_parse": {
"paper_id": "C08-1013",
"_pdf_hash": "",
"abstract": [
{
"text": "We present ParaMetric, an automatic evaluation metric for data-driven approaches to paraphrasing. ParaMetric provides an objective measure of quality using a collection of multiple translations whose paraphrases have been manually annotated. ParaMetric calculates precision and recall scores by comparing the paraphrases discovered by automatic paraphrasing techniques against gold standard alignments of words and phrases within equivalent sentences. We report scores for several established paraphrasing techniques.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Paraphrasing is useful in a variety of natural language processing applications including natural language generation, question answering, multidocument summarization and machine translation evaluation. These applications require paraphrases for a wide variety of domains and language usage. Therefore building hand-crafted lexical resources such as WordNet (Miller, 1990) would be far too laborious. As such, a number of data-driven approaches to paraphrasing have been developed (Lin and Pantel, 2001 ; Barzilay and McKeown, 2001 ; Barzilay and Lee, 2003; Pang et al., 2003; Quirk et al., 2004; Bannard and Callison-Burch, 2005) . Despite this spate of research, no objective evaluation metric has been proposed.",
"cite_spans": [
{
"start": 358,
"end": 372,
"text": "(Miller, 1990)",
"ref_id": "BIBREF12"
},
{
"start": 481,
"end": 502,
"text": "(Lin and Pantel, 2001",
"ref_id": "BIBREF11"
},
{
"start": 505,
"end": 531,
"text": "Barzilay and McKeown, 2001",
"ref_id": "BIBREF2"
},
{
"start": 534,
"end": 557,
"text": "Barzilay and Lee, 2003;",
"ref_id": "BIBREF1"
},
{
"start": 558,
"end": 576,
"text": "Pang et al., 2003;",
"ref_id": "BIBREF17"
},
{
"start": 577,
"end": 596,
"text": "Quirk et al., 2004;",
"ref_id": "BIBREF19"
},
{
"start": 597,
"end": 630,
"text": "Bannard and Callison-Burch, 2005)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In absence of a repeatable automatic evaluation, the quality of these paraphrasing techniques was gauged using subjective manual evaluations. Section 2 gives a survey of the various evaluation methodologies used in previous research. It has not been possible to directly compare paraphrasing techniques, because each one was evaluated using its own idiosyncratic experimental design. Moreover, because these evaluations were performed manually, they are difficult to replicate.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We introduce an automatic evaluation metric, called ParaMetric, which uses paraphrasing techniques to be compared and enables an evaluation to be easily repeated in subsequent research. Para-Metric utilizes data sets which have been annotated with paraphrases. ParaMetric compares automatic paraphrases against reference paraphrases.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper we:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Present a novel automatic evaluation metric for data-driven paraphrasing methods;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Describe how manual alignments are created by annotating correspondences between words in multiple translations;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Show how phrase extraction heuristics from statistical machine translation can be used to enumerate paraphrases from the alignments;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Report ParaMetric scores for a number of existing paraphrasing methods.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "No consensus has been reached with respect to the proper methodology to use when evaluating paraphrase quality. This section reviews past methods for paraphrase evaluation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Researchers usually present the quality of their automatic paraphrasing technique in terms of a subjective manual evaluation. These have used a variety of criteria. For example, Barzilay and McKeown (2001) evaluated their paraphrases by asking judges whether paraphrases were \"approximately conceptually equivalent. \" Ibrahim et al. (2003) asked judges whether their paraphrases were \"roughly interchangeable given the genre.\" Bannard and Callison-Burch (2005) replaced phrases with paraphrases in a number of sentences and asked judges whether the substitutions \"preserved meaning and remained grammatical.\" These subjective evaluations are rather vaguely defined and not easy to reproduce.",
"cite_spans": [
{
"start": 178,
"end": 205,
"text": "Barzilay and McKeown (2001)",
"ref_id": "BIBREF2"
},
{
"start": 316,
"end": 339,
"text": "\" Ibrahim et al. (2003)",
"ref_id": "BIBREF9"
},
{
"start": 427,
"end": 460,
"text": "Bannard and Callison-Burch (2005)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Others evaluate paraphrases in terms of whether they improve performance on particular tasks. Callison-Burch et al. (2006b) measure improvements in translation quality in terms of Bleu score (Papineni et al., 2002) and in terms of subjective human evaluation when paraphrases are integrated into a statistical machine translation system. Lin and Pantel (2001) manually judge whether a paraphrase might be used to answer questions from the TREC question-answering track. To date, no one has used task-based evaluation to compare different paraphrasing methods. Even if such an evaluation were performed, it is unclear whether the results would hold for a different task. Because of this, we strive for a general evaluation rather than a task-specific one. Dolan et al. (2004) create a set of manual word alignments between pairs of English sentences. We create a similar type of data, as described in Section 4. Dolan et al. use heuristics to draw pairs of English sentences from a comparable corpus of newswire articles, and treat these as potential paraphrases. In some cases these sentence pairs are good examples of paraphrases, and in some cases they are not. Our data differs because it is drawn from multiple translations of the same foreign sentences. Barzilay (2003) suggested that multiple translations of the same foreign source text were a perfect source for \"naturally occurring paraphrases\" because they are samples of text which convey the same meaning but are produced by different writers. That being said, it may be possible to use Dolan et al's data toward a similar end. Cohn et al. (to appear) compares the use of the multiple translation corpus with the MSR corpus for this task.",
"cite_spans": [
{
"start": 94,
"end": 123,
"text": "Callison-Burch et al. (2006b)",
"ref_id": "BIBREF6"
},
{
"start": 191,
"end": 214,
"text": "(Papineni et al., 2002)",
"ref_id": "BIBREF18"
},
{
"start": 338,
"end": 359,
"text": "Lin and Pantel (2001)",
"ref_id": "BIBREF11"
},
{
"start": 755,
"end": 774,
"text": "Dolan et al. (2004)",
"ref_id": "BIBREF8"
},
{
"start": 1259,
"end": 1274,
"text": "Barzilay (2003)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "The work described here is similar to work in summarization evaluation. For example, in the Pyramid Method (Nenkova et al., 2007) content units that are similar across human-generated summaries are hand-aligned. These can have alternative wordings, and are manually grouped. The idea of capturing these and building a resource for evaluating summaries is in the same spirit as our methodology.",
"cite_spans": [
{
"start": 107,
"end": 129,
"text": "(Nenkova et al., 2007)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "There are several problems inherent to automatically evaluating paraphrases. First and foremost, developing an exhaustive list of paraphrases for any given phrase is difficult. Lin and Pantel (2001) illustrate the difficulties that people have generating a complete list of paraphrases, reporting that they missed many examples generated by a system that were subsequently judged to be correct. If a list of reference paraphrases is incomplete, then using it to calculate precision will give inaccurate numbers. Precision will be falsely low if the system produces correct paraphrases which are not in the reference list. Additionally, recall is indeterminable because there is no way of knowing how many correct paraphrases exist. There are further impediments to automatically evaluating paraphrases. Even if we were able to come up with a reasonably exhaustive list of paraphrases for a phrase, the acceptability of each paraphrase would vary depending on the context of the original phrase (Szpektor et al., 2007) . While lexical and phrasal paraphrases can be evaluated by comparing them against a list of known paraphrases (perhaps customized for particular contexts), this cannot be naturally done for structural paraphrases which may transform whole sentences.",
"cite_spans": [
{
"start": 177,
"end": 198,
"text": "Lin and Pantel (2001)",
"ref_id": "BIBREF11"
},
{
"start": 994,
"end": 1017,
"text": "(Szpektor et al., 2007)",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Challenges for Evaluating Paraphrases Automatically",
"sec_num": "3"
},
{
"text": "We attempt to resolve these problems by having annotators indicate correspondences in pairs of equivalent sentences. Rather than having people enumerate paraphrases, we asked that they perform the simper task of aligning paraphrases. After developing these manual \"gold standard alignments\" we can gauge how well different automatic paraphrases are at aligning paraphrases within equivalent sentences. By evaluating the performance of paraphrasing techniques at alignment, rather than at matching a list of reference paraphrases, we obviate the need to have a complete set of paraphrases.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Challenges for Evaluating Paraphrases Automatically",
"sec_num": "3"
},
{
"text": "We describe how sets of reference paraphrases can be extracted from the gold standard alignments. While these sets will obviously be fragmentary, we attempt to make them more complete by aligning groups of equivalent sentences rather than only pairs. The paraphrase sets that we extract are appropriate for the particular contexts. Moreover they may potentially be used to study structural paraphrases, although we do not examine that ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Challenges for Evaluating Paraphrases Automatically",
"sec_num": "3"
},
{
"text": "We asked monolingual English speakers to align corresponding words and phrases across pairs of equivalent English sentences. The English sentences were equivalent because they were translations of the same foreign language text created by different human translators. Our annotators were instructed to align parts of the sentences which had the same meaning. Annotators were asked to prefer smaller one-to-one word alignments, but were allowed to create one-to-many and many-tomany alignments where appropriate. They were given a set of annotation guidelines covering special cases such as repetition, pronouns, genitives, phrasal verbs and omissions (Callison-Burch et al., 2006a) . The manual correspondences are treated as gold standard alignments.",
"cite_spans": [
{
"start": 651,
"end": 681,
"text": "(Callison-Burch et al., 2006a)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": "We use a corpus that contains eleven English translations of Chinese newswire documents, which were commissioned from different translation agencies by the Linguistics Data Consortium 1 . The data was created for the Bleu machine translation evaluation metric (Papineni et al., 2002) , which uses multiple translations as a way of capturing allowable variation in translation. Whereas the Bleu metric requires no further information, our method requires a one-off annotation to explicitly show which parts of the multiple translations constitute paraphrases.",
"cite_spans": [
{
"start": 260,
"end": 283,
"text": "(Papineni et al., 2002)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": "The rationale behind using a corpus with eleven translations was that a greater number of translations would likely result in a greater number of paraphrases for each phrase. Figure 1 shows the alignments that were created between one sentence and three of its ten corresponding translations. Table 1 gives a list of non-identical words and phrases that can be paired by way of the word alignments. These are the basic paraphrases contained within the three sentence pairs. Each phrase has up to three paraphrases. The maximum number of paraphrases for a given span in each sentence is bounded by the number of equivalent sentences that it is paired with.",
"cite_spans": [],
"ref_spans": [
{
"start": 175,
"end": 183,
"text": "Figure 1",
"ref_id": "FIGREF0"
},
{
"start": 293,
"end": 300,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": "In addition to these basic paraphrases, longer paraphrases can also be obtained using the heuristic presented in Och and Ney (2004) for extracting phrase pairs (PP) from word alignments A, between a foreign sentence f J 1 and an English sen-some some people, there are those who want propose, are proposing to impeach an indictment against, impeaching and while others some, those who expect want step down resign, leave office voluntarily, tender his resignation Table 1 : Non-identical words and phrases which are identified as being in correspondence by the alignments in Figure 1 . tence e I 1 :",
"cite_spans": [
{
"start": 113,
"end": 131,
"text": "Och and Ney (2004)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [
{
"start": 464,
"end": 471,
"text": "Table 1",
"ref_id": null
},
{
"start": 575,
"end": 583,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": "P P (f J 1 , e I 1 , A) = {(f j+m j , e i+n i )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": ":",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": "\u2200(i , j ) \u2208 A : j \u2264 j \u2264 j + m \u2194 i \u2264 i \u2264 i + n \u2227\u2203(i , j ) \u2208 A : j \u2264 j \u2264 j + m\u2227 \u2194 i \u2264 i \u2264 i + n}",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": "When we apply the phrase extraction heuristic to aligned English sentences, we add the constraint f j+m j = e i+n i to exclude phrases that are identical. This heuristic would allow \"some people propose to impeach him,\" \"some are proposing an indictment against him,\" and \"there are those who propose impeaching him\" to be extracted as paraphrases of \"some want to impeach him.\" The heuristic extracts a total of 142 non-identical phrase pairs from the three sentences given in Figure 1.",
"cite_spans": [],
"ref_spans": [
{
"start": 478,
"end": 484,
"text": "Figure",
"ref_id": null
}
],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": "For the results reported in this paper, annotators aligned 50 groups of 10 pairs of equivalent sentences, for a total of 500 sentence pairs. These were assembled by pairing the first of the LDC translations with the other ten (i.e. 1-2, 1-3, 1-4, ..., 1-11). The choice of pairing one sentence with the others instead of doing all pairwise combinations was made simply because the latter would not seem to add much information. However, the choice of using the first translator as the key was arbitrary.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": "Annotators corrected a set of automatic word alignments that were created using Giza++ (Och and Ney, 2003) , which was trained on a total of 109,230 sentence pairs created from all pairwise combinations of the eleven translations of 993 Chinese sentences.",
"cite_spans": [
{
"start": 87,
"end": 106,
"text": "(Och and Ney, 2003)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": "The average amount of time spent on each of the sentence pairs was 77 seconds, with just over eleven hours spent to annotate all 500 sentence pairs. Although each sentence pair in our data set was annotated by a single annotator, Cohn et al. (to appear) analyzed the inter-annotator agreement for randomly selected phrase pairs from the same corpus, and found inter-annotator agreement of\u0108 = 0.85 over the aligned words and\u0108 = 0.63 over the alignments between basic phrase pairs, where\u0108 is measure of inter-rater agreement in the style of Kupper and Hafner (1989) .",
"cite_spans": [
{
"start": 539,
"end": 563,
"text": "Kupper and Hafner (1989)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Manually Aligning Paraphrases",
"sec_num": "4"
},
{
"text": "We can exploit the manually aligned data to compute scores in two different fashions. First, we can calculate how well an automatic paraphrasing technique is able to align the paraphrases in a sentence pair. Second, we can calculate the lowerbound on precision for a paraphrasing technique and its relative recall by enumerating the paraphrases from each of the sentence groups. The first of these score types does not require groups of sentences, only pairs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ParaMetric Scores",
"sec_num": "5"
},
{
"text": "We calculate alignment accuracy by comparing the manual alignments for the sentence pairs in the test corpus with the alignments that the automatic paraphrasing techniques produce for the same sentence pairs. We enumerate all non-identical phrase pairs within the manually word-aligned sentence pairs and within the automatically word aligned sentence pairs using P P . We calculate the precision and recall of the alignments by taking the intersection of the paraphrases extracted from the manual alignments M , and the paraphrases extracted from a system's alignments S:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ParaMetric Scores",
"sec_num": "5"
},
{
"text": "AlignP rec = <e 1 ,e 2 >\u2208C |P P (e1, e2, S) \u2229 P P (e1, e2, M )| <e 1 ,e 2 >\u2208C |P P (e1, e2, S)| Align Recall = <e 1 ,e 2 >\u2208C |P P (e1, e2, S) \u2229 P P (e1, e2, M )| <e 1 ,e 2 >\u2208C |P P (e1, e2, M )|",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ParaMetric Scores",
"sec_num": "5"
},
{
"text": "Where e 1 , e 2 are pairs of English sentence from the test corpus.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ParaMetric Scores",
"sec_num": "5"
},
{
"text": "Measuring a paraphrasing method's performance on the task of aligning the paraphrases is somewhat different than what most paraphrasing methods do. Most methods produce a list of paraphrases for a given input phrase, drawing from a large set of rules or a corpus larger than our small test set. We therefore also attempt to measure precision and recall by comparing the set of paraphrases that method M produces for phrase p that occurs in sentence s. We denote this set as para M (p, s), where s is an optional argument for methods that constrain their paraphrases based on context.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ParaMetric Scores",
"sec_num": "5"
},
{
"text": "Our reference sets of paraphrases are generated in a per group fashion. We enumerate the reference paraphrases for phrase p in sentence s in group G as",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ParaMetric Scores",
"sec_num": "5"
},
{
"text": "paraREF (p1, s1, G) = {p2 : \u2200(p1, p2) \u2208 <s 1 ,s 2 ,A>\u2208G P P (s1, s2, A)}",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ParaMetric Scores",
"sec_num": "5"
},
{
"text": "The maximum size of this set is the number of sentence pairs in G. Because this set of reference paraphrases is incomplete, we can only calculate a lower bound on the precision of a paraphrasing method and its recall relative to the reference paraphrases. We call these LB-P recision and Rel-Recall and calculate them as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ParaMetric Scores",
"sec_num": "5"
},
{
"text": "LB-P recision = <s,G>\u2208C p\u2208s |paraM (p, s) \u2229 paraREF (p1, s, G)| |paraM (p, s)| Rel-Recall = <s,G>\u2208C p\u2208s |paraM (p, s) \u2229 paraREF (p1, s, G)| |paraREF (p1, s, G)|",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ParaMetric Scores",
"sec_num": "5"
},
{
"text": "For these metrics we require the test corpus C to be a held-out set and restrict the automatic paraphrasing techniques from drawing paraphrases from it. The idea is instead to see how well these techniques are able to draw paraphrases from the other sources of data which they would normally use.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ParaMetric Scores",
"sec_num": "5"
},
{
"text": "There are a number of established methods for extracting paraphrases from data. We describe the following methods in this section and evaluate them in the next:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Paraphrasing Techniques",
"sec_num": "6"
},
{
"text": "\u2022 Pang et al. (2003) used syntactic alignment to merge parse trees of multiple translations,",
"cite_spans": [
{
"start": 2,
"end": 20,
"text": "Pang et al. (2003)",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Paraphrasing Techniques",
"sec_num": "6"
},
{
"text": "\u2022 Quirk et al. (2004) Figure 2 : Pang et al. (2003) created word graphs by merging parse trees. Paths with the same start and end nodes are treated as paraphrases. Pang et al. (2003) use multiple translations to learn paraphrases using a syntax-based alignment algorithm, illustrated in Figure 2 . Parse trees were merged into forests by grouping constituents of the same type (for example, the two NPs and two VPs are grouped). Parse forests were mapped onto finite state word graphs by creating alternative paths for every group of merged nodes. Different paths within the resulting word lattice are treated as paraphrases of each other. For example, in the word lattice in Figure 2 , people were killed, persons died, persons were killed, and people died are all possible paraphrases of each other. Quirk et al. (2004) treated paraphrasing as \"monolingual statistical machine translation.\" They created a \"parallel corpus\" containing pairs of English sentences by drawing sentences with a low edit distance from news articles that were written about the same topic on the same date, but published by different newspapers. They formulated the problem of paraphrasing in probabilistic terms in the same way it had been defined in the statistical machine translation literature: Where p(e 1 |e 2 ) is estimated by training word alignment models over the \"parallel corpus\" as in the IBM Models (Brown et al., 1993) , and phrase translations are extracted from word alignments as in the Alignment Template Model (Och, 2002) . Bannard and Callison-Burch (2005) also used techniques from statistical machine translation to identify paraphrases. Rather than drawing pairs of English sentences from a comparable corpus, Bannard and Callison-Burch (2005) used bilingual parallel corpora. They identified English paraphrases by pivoting through phrases in another language. They located foreign language translations of an English phrase, and treated the other English translations of those foreign phrases as potential paraphrases. Figure 3 illustrates how a Spanish phrase can be used as a point of identification for English paraphrases in this way. Bannard and Callison-Burch (2005) defined a paraphrase probability p(e 2 |e 1 ) in terms of the translation model probabilities p(f |e 1 ) and p(e 2 |f ). Since e 1 can translate as multiple foreign language phrases, they sum over f , and since multiple parallel corpora can be used they summed over each parallel corpus C: Table 2 : Summary results for scoring the different paraphrasing techniques using our proposed automatic evaluations.",
"cite_spans": [
{
"start": 2,
"end": 21,
"text": "Quirk et al. (2004)",
"ref_id": "BIBREF19"
},
{
"start": 33,
"end": 51,
"text": "Pang et al. (2003)",
"ref_id": "BIBREF17"
},
{
"start": 164,
"end": 182,
"text": "Pang et al. (2003)",
"ref_id": "BIBREF17"
},
{
"start": 802,
"end": 821,
"text": "Quirk et al. (2004)",
"ref_id": "BIBREF19"
},
{
"start": 1393,
"end": 1413,
"text": "(Brown et al., 1993)",
"ref_id": "BIBREF4"
},
{
"start": 1510,
"end": 1521,
"text": "(Och, 2002)",
"ref_id": "BIBREF16"
},
{
"start": 1524,
"end": 1557,
"text": "Bannard and Callison-Burch (2005)",
"ref_id": "BIBREF0"
},
{
"start": 1714,
"end": 1747,
"text": "Bannard and Callison-Burch (2005)",
"ref_id": "BIBREF0"
},
{
"start": 2145,
"end": 2178,
"text": "Bannard and Callison-Burch (2005)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [
{
"start": 22,
"end": 30,
"text": "Figure 2",
"ref_id": null
},
{
"start": 287,
"end": 295,
"text": "Figure 2",
"ref_id": null
},
{
"start": 676,
"end": 684,
"text": "Figure 2",
"ref_id": null
},
{
"start": 2025,
"end": 2033,
"text": "Figure 3",
"ref_id": "FIGREF1"
},
{
"start": 2469,
"end": 2476,
"text": "Table 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Paraphrasing Techniques",
"sec_num": "6"
},
{
"text": "e 2 =",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Paraphrasing Techniques",
"sec_num": "6"
},
{
"text": "e 2 = arg max e 2 =e 1 p(e 2 |e 1 ) \u2248 arg max e 2 =e 1 C f in C p(f |e 1 )p(e 2 |f )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Paraphrasing Techniques",
"sec_num": "6"
},
{
"text": "Bo Pang provided syntactic alignments for the 500 sentence pairs. The word lattices combine the groups of sentences. When measuring alignment quality, we took pains to try to limit the extracted phrase pairs to those which occurred in each sentence pair, but we acknowledge that our methodology may be flawed.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Paraphrasing Techniques",
"sec_num": "6"
},
{
"text": "We created training data for the monolingual statistical machine translation method using all pairwise combination of eleven English translations in LDC2002T01. All combinations of the eleven translations of the 993 sentences in that corpus resulted in 109,230 sentence pairs with 3,266,769 words on each side. We used this data to train an alignment model, and applied it to the 500 sentence pairs in our test set.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Paraphrasing Techniques",
"sec_num": "6"
},
{
"text": "We used the parallel corpus method to align each pair of English sentences by creating intermediate alignments through their Chinese source sentences. The bilingual word alignment model was trained on a Chinese-English parallel corpus from the NIST MT Evaluation consisting of 40 million words. This was used to align the 550 Chinese-English sentence pairs constructed from the test set.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Paraphrasing Techniques",
"sec_num": "6"
},
{
"text": "Each of the paraphrasing methods generated paraphrases for LB-P recision and Rel-Recall using larger training sets of data than for the alignments. For the syntax-based alignment method, we excluded the 50 word lattices corresponding to the test set. We used the remaining 849 lattices for the LDC multiple translation corpus. For the monolingual statistical machine translation method, we downloaded the Microsoft Research Paraphrase Phrase Table, which contained paraphrases for nearly 9 million phrases, gener-ated from the method described in Quirk et al. (2004) . For the parallel corpus method, we derived paraphrases from the entire Europarl corpus, which contains parallel corpora between English and 10 other languages, with approximately 30 million words per language. We limited both the Quirk et al. (2004) and the Bannard and Callison-Burch (2005) paraphrases to those with a probability greater than or equal to 1%. Table 2 gives a summary of how each of the paraphrasing techniques scored using the four different automatic metrics. The precision of their alignments was in the same ballpark, with each paraphrasing method reaching above 60%. The monolingual SMT method vastly outstripped the others in terms of recall and therefore seems to be the best on the simplified task of aligning paraphrases within pairs of equivalent sentences.",
"cite_spans": [
{
"start": 547,
"end": 566,
"text": "Quirk et al. (2004)",
"ref_id": "BIBREF19"
},
{
"start": 799,
"end": 818,
"text": "Quirk et al. (2004)",
"ref_id": "BIBREF19"
},
{
"start": 827,
"end": 860,
"text": "Bannard and Callison-Burch (2005)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [
{
"start": 930,
"end": 937,
"text": "Table 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Training data for precision and recall",
"sec_num": "7.2"
},
{
"text": "For the task of generating paraphrases from unrestricted resources, the monolingual SMT method again had the highest precision, although time time its recall was quite low. The 500 manually aligned sentence pairs contained 14,078 unique paraphrases for phrases of 5 words or less. The monolingual SMT method only posited 230 paraphrases with 156 of them being correct. By contrast, the syntactic alignment method posited 1,213 with 399 correct, and the parallel corpus method posited 6,914 with 998 correct. Since the reference lists are incomplete by their very nature, the LB-P recision score gives a lower-bound on the precision, and the Rel-Recall gives recall only with respect to the partial list of paraphrases. Table 3 gives the performance of the different paraphrasing techniques for different phrase lengths.",
"cite_spans": [],
"ref_spans": [
{
"start": 719,
"end": 726,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Results",
"sec_num": "7.3"
},
{
"text": "In this paper we defined a number of automatic scores for data-driven approaches to paraphrasing, which we collectively dub \"ParaMetric\". We discussed the inherent difficulties in automatically assessing paraphrase quality. These are due primarily to the fact that it is exceedingly difficult to create an exhaustive list of paraphrases. To address this problem, we introduce an artificial task of aligning paraphrases within pairs of equivalent English sentences, which guarantees accurate precision and recall numbers. In order to measure alignment quality, we create a set of gold standard alignments. While the creation of this data does require some effort, it seems to be a manageable amount, and the inter-annotator agreement seems reasonable.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions",
"sec_num": "8"
},
{
"text": "Since alignment is not perfectly matched with what we would like automatic paraphrasing techniques to do, we also use the gold standard alignment data to measure a lower bound on the precision of a method's paraphrases, as well as its recall relative to the limited set of paraphrases. Future studies should examine how well these scores rank different paraphrasing methods when compared to human judgments. Follow up work should investigate the number of equivalent English sentences that are required for reasonably complete lists of paraphrases. In this work we aligned sets of eleven different English sentences, but we acknowledge that such a data set is rare and might make it difficult to port this method to other domains or languages.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions",
"sec_num": "8"
},
{
"text": "The goal of this work is to develop a set of scores that both allows different paraphrasing techniques to be compared objectively and provides an easily repeatable method for automatically evaluating paraphrases. This has hitherto not been possible. The availability of an objective, automatic evaluation metric for paraphrasing has the potential to impact research in the area in a number of ways. It not only allows for the comparison of different approaches to paraphrasing, as shown in this paper, but also provides a way to tune the parameters of a single system in order to optimize its quality. Table 3 : Results for paraphrases of continuous subphrases of various lengths.",
"cite_spans": [],
"ref_spans": [
{
"start": 602,
"end": 609,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Conclusions",
"sec_num": "8"
},
{
"text": "See LDC catalog number 2002T01.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "The authors are grateful to Bo Pang for providing the word lattices from her method, to Stefan Riezler for his comments on an early draft of this paper, and to Michelle Bland for proofreading. This work was supported by the National Science Foundation under Grant No. 0713448. The views and findings are the authors' alone.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgments",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Paraphrasing with bilingual parallel corpora",
"authors": [
{
"first": "Colin",
"middle": [],
"last": "Bannard",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Callison-Burch",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL-2005)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bannard, Colin and Chris Callison-Burch. 2005. Para- phrasing with bilingual parallel corpora. In Proceed- ings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL-2005), Ann Ar- bor, Michigan.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Learning to paraphrase: An unsupervised approach using multiple-sequence alignment",
"authors": [
{
"first": "Regina",
"middle": [],
"last": "Barzilay",
"suffix": ""
},
{
"first": "Lillian",
"middle": [],
"last": "Lee",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of HLT/NAACL-2003",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Barzilay, Regina and Lillian Lee. 2003. Learn- ing to paraphrase: An unsupervised approach us- ing multiple-sequence alignment. In Proceedings of HLT/NAACL-2003, Edmonton, Alberta.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Extracting paraphrases from a parallel corpus",
"authors": [
{
"first": "Regina",
"middle": [],
"last": "Barzilay",
"suffix": ""
},
{
"first": "Kathleen",
"middle": [],
"last": "Mckeown",
"suffix": ""
}
],
"year": 2001,
"venue": "Proceedings of the 39th Annual Meeting of the Association for Computational Linguistics (ACL-2001)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Barzilay, Regina and Kathleen McKeown. 2001. Ex- tracting paraphrases from a parallel corpus. In Pro- ceedings of the 39th Annual Meeting of the Associa- tion for Computational Linguistics (ACL-2001).",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Information Fusion for Mutlidocument Summarization: Paraphrasing and Generation",
"authors": [
{
"first": "Regina",
"middle": [],
"last": "Barzilay",
"suffix": ""
}
],
"year": 2003,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Barzilay, Regina. 2003. Information Fusion for Mutli- document Summarization: Paraphrasing and Gener- ation. Ph.D. thesis, Columbia University, New York.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "The mathematics of machine translation: Parameter estimation",
"authors": [
{
"first": "Peter",
"middle": [],
"last": "Brown",
"suffix": ""
},
{
"first": "Stephen",
"middle": [
"Della"
],
"last": "Pietra",
"suffix": ""
},
{
"first": "Vincent",
"middle": [
"Della"
],
"last": "Pietra",
"suffix": ""
},
{
"first": "Robert",
"middle": [],
"last": "Mercer",
"suffix": ""
}
],
"year": 1993,
"venue": "Computational Linguistics",
"volume": "19",
"issue": "2",
"pages": "263--311",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Brown, Peter, Stephen Della Pietra, Vincent Della Pietra, and Robert Mercer. 1993. The mathematics of machine translation: Parameter estimation. Com- putational Linguistics, 19(2):263-311, June.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Annotation guidelines for paraphrase alignment",
"authors": [
{
"first": "Chris",
"middle": [],
"last": "Callison-Burch",
"suffix": ""
},
{
"first": "Trevor",
"middle": [],
"last": "Cohn",
"suffix": ""
},
{
"first": "Mirella",
"middle": [],
"last": "Lapata",
"suffix": ""
}
],
"year": 2006,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Callison-Burch, Chris, Trevor Cohn, and Mirella Lap- ata. 2006a. Annotation guidelines for paraphrase alignment. Tech report, University of Edinburgh.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Improved statistical machine translation using paraphrases",
"authors": [
{
"first": "Chris",
"middle": [],
"last": "Callison-Burch",
"suffix": ""
},
{
"first": "Philipp",
"middle": [],
"last": "Koehn",
"suffix": ""
},
{
"first": "Miles",
"middle": [],
"last": "Osborne",
"suffix": ""
}
],
"year": 2006,
"venue": "Proceedings of HLT/NAACL-2006",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Callison-Burch, Chris, Philipp Koehn, and Miles Os- borne. 2006b. Improved statistical machine translation using paraphrases. In Proceedings of HLT/NAACL-2006, New York, New York.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Constructing corpora for the development and evaluation of paraphrase systems",
"authors": [
{
"first": "Trevor",
"middle": [],
"last": "Cohn",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Callison-Burch",
"suffix": ""
},
{
"first": "Mirella",
"middle": [],
"last": "Lapata",
"suffix": ""
}
],
"year": null,
"venue": "Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Cohn, Trevor, Chris Callison-Burch, and Mirella Lap- ata. to appear. Constructing corpora for the develop- ment and evaluation of paraphrase systems. Compu- tational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Unsupervised construction of large paraphrase corpora: Exploiting massively parallel news sources",
"authors": [
{
"first": "Bill",
"middle": [],
"last": "Dolan",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Quirk",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Brockett",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of the 20th International Conference on Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dolan, Bill, Chris Quirk, and Chris Brockett. 2004. Unsupervised construction of large paraphrase cor- pora: Exploiting massively parallel news sources. In Proceedings of the 20th International Conference on Computational Linguistics.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Extracting structural paraphrases from aligned monolingual corpora",
"authors": [
{
"first": "Ali",
"middle": [],
"last": "Ibrahim",
"suffix": ""
},
{
"first": "Boris",
"middle": [],
"last": "Katz",
"suffix": ""
},
{
"first": "Jimmy",
"middle": [],
"last": "Lin",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of the Second International Workshop on Paraphrasing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ibrahim, Ali, Boris Katz, and Jimmy Lin. 2003. Ex- tracting structural paraphrases from aligned mono- lingual corpora. In Proceedings of the Second Inter- national Workshop on Paraphrasing (ACL 2003).",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "On assessing interrater agreement for multiple attribute responses",
"authors": [
{
"first": "Lawrence",
"middle": [
"L"
],
"last": "Kupper",
"suffix": ""
},
{
"first": "Kerry",
"middle": [
"B"
],
"last": "Hafner",
"suffix": ""
}
],
"year": 1989,
"venue": "Biometrics",
"volume": "45",
"issue": "3",
"pages": "957--967",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kupper, Lawrence L. and Kerry B. Hafner. 1989. On assessing interrater agreement for multiple attribute responses. Biometrics, 45(3):957-967.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Discovery of inference rules from text",
"authors": [
{
"first": "Dekang",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "Patrick",
"middle": [],
"last": "Pantel",
"suffix": ""
}
],
"year": 2001,
"venue": "Natural Language Engineering",
"volume": "7",
"issue": "3",
"pages": "343--360",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lin, Dekang and Patrick Pantel. 2001. Discovery of inference rules from text. Natural Language Engi- neering, 7(3):343-360.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Wordnet: An on-line lexical database. Special Issue of the",
"authors": [
{
"first": "George",
"middle": [
"A"
],
"last": "Miller",
"suffix": ""
}
],
"year": 1990,
"venue": "International Journal of Lexicography",
"volume": "3",
"issue": "4",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Miller, George A. 1990. Wordnet: An on-line lexical database. Special Issue of the International Journal of Lexicography, 3(4).",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "The pyramid method: incorporating human content selection variation in summarization evaluation",
"authors": [
{
"first": "Ani",
"middle": [],
"last": "Nenkova",
"suffix": ""
},
{
"first": "Rebecca",
"middle": [],
"last": "Passonneau",
"suffix": ""
},
{
"first": "Kathleen",
"middle": [],
"last": "Mckeown",
"suffix": ""
}
],
"year": 2007,
"venue": "ACM Transactions on Speech and Language Processing",
"volume": "4",
"issue": "2",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nenkova, Ani, Rebecca Passonneau, and Kathleen McKeown. 2007. The pyramid method: incorporat- ing human content selection variation in summariza- tion evaluation. ACM Transactions on Speech and Language Processing, 4(2).",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "A systematic comparison of various statistical alignment models",
"authors": [
{
"first": "Franz",
"middle": [
"Josef"
],
"last": "Och",
"suffix": ""
},
{
"first": "Hermann",
"middle": [],
"last": "Ney",
"suffix": ""
}
],
"year": 2003,
"venue": "Computational Linguistics",
"volume": "29",
"issue": "1",
"pages": "19--51",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Och, Franz Josef and Hermann Ney. 2003. A system- atic comparison of various statistical alignment mod- els. Computational Linguistics, 29(1):19-51.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "The alignment template approach to statistical machine translation",
"authors": [
{
"first": "Franz",
"middle": [
"Josef"
],
"last": "Och",
"suffix": ""
},
{
"first": "Hermann",
"middle": [],
"last": "Ney",
"suffix": ""
}
],
"year": 2004,
"venue": "Computational Linguistics",
"volume": "30",
"issue": "4",
"pages": "417--449",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Och, Franz Josef and Hermann Ney. 2004. The align- ment template approach to statistical machine trans- lation. Computational Linguistics, 30(4):417-449.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Statistical Machine Translation: From Single-Word Models to Alignment Templates",
"authors": [
{
"first": "Franz",
"middle": [],
"last": "Och",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Josef",
"suffix": ""
}
],
"year": 2002,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Och, Franz Josef. 2002. Statistical Machine Transla- tion: From Single-Word Models to Alignment Tem- plates. Ph.D. thesis, RWTH Aachen Department of Computer Science, Aachen, Germany.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Syntax-based alignment of multiple translations: Extracting paraphrases and generating new sentences",
"authors": [
{
"first": "Bo",
"middle": [],
"last": "Pang",
"suffix": ""
},
{
"first": "Kevin",
"middle": [],
"last": "Knight",
"suffix": ""
},
{
"first": "Daniel",
"middle": [],
"last": "Marcu",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of HLT/NAACL-2003",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Pang, Bo, Kevin Knight, and Daniel Marcu. 2003. Syntax-based alignment of multiple translations: Ex- tracting paraphrases and generating new sentences. In Proceedings of HLT/NAACL-2003, Edmonton, Alberta.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Bleu: A method for automatic evaluation of machine translation",
"authors": [
{
"first": "Kishore",
"middle": [],
"last": "Papineni",
"suffix": ""
},
{
"first": "Salim",
"middle": [],
"last": "Roukos",
"suffix": ""
},
{
"first": "Todd",
"middle": [],
"last": "Ward",
"suffix": ""
},
{
"first": "Wei-Jing",
"middle": [],
"last": "Zhu",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics (ACL-2002)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Papineni, Kishore, Salim Roukos, Todd Ward, and Wei- Jing Zhu. 2002. Bleu: A method for automatic eval- uation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Compu- tational Linguistics (ACL-2002), Philadelphia, Penn- sylvania.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Monolingual machine translation for paraphrase generation",
"authors": [
{
"first": "Chris",
"middle": [],
"last": "Quirk",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Brockett",
"suffix": ""
},
{
"first": "William",
"middle": [],
"last": "Dolan",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Quirk, Chris, Chris Brockett, and William Dolan. 2004. Monolingual machine translation for para- phrase generation. In Proceedings of the 2004 Con- ference on Empirical Methods in Natural Language Processing (EMNLP-2004), Barcelona, Spain.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Instance-based evaluation of entailment rule acquisition",
"authors": [
{
"first": "Idan",
"middle": [],
"last": "Szpektor",
"suffix": ""
},
{
"first": "Eyal",
"middle": [],
"last": "Shnarch",
"suffix": ""
},
{
"first": "Ido",
"middle": [],
"last": "Dagan",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the 45th Annual Meeting of the Association for Computational Linguistics (ACL-2007)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Szpektor, Idan, Eyal Shnarch, and Ido Dagan. 2007. Instance-based evaluation of entailment rule acquisi- tion. In Proceedings of the 45th Annual Meeting of the Association for Computational Linguistics (ACL- 2007), Prague, Czech Republic.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "Pairs of English sentences were aligned by hand. Black squares indicate paraphrase correspondences. in this paper."
},
"FIGREF1": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "Bannard and Callison-Burch (2005) extracted paraphrases by equating English phrases that share a common translation."
},
"TABREF3": {
"type_str": "table",
"num": null,
"html": null,
"text": "Align P rec",
"content": "<table><tr><td/><td/><td/><td/><td/><td/><td/><td/><td colspan=\"4\">Align Recall</td><td/><td/><td colspan=\"4\">LB-P recision</td><td/><td/><td colspan=\"4\">Rel-Recall</td></tr><tr><td/><td>Parallel</td><td>Corpora</td><td>Syntactic</td><td>Alignment</td><td>Monolingual</td><td>SMT</td><td>Parallel</td><td>Corpora</td><td>Syntactic</td><td>Alignment</td><td>Monolingual</td><td>SMT</td><td>Parallel</td><td>Corpora</td><td>Syntactic</td><td>Alignment</td><td>Monolingual</td><td>SMT</td><td>Parallel</td><td>Corpora</td><td>Syntactic</td><td>Alignment</td><td>Monolingual</td><td>SMT</td></tr><tr><td>Length = 1</td><td colspan=\"2\">.54</td><td colspan=\"2\">.48</td><td colspan=\"2\">.64</td><td colspan=\"2\">.24</td><td colspan=\"2\">.18</td><td colspan=\"2\">.56</td><td colspan=\"2\">.15</td><td colspan=\"2\">.25</td><td colspan=\"2\">.59</td><td colspan=\"2\">.20</td><td colspan=\"2\">.16</td><td colspan=\"2\">.02</td></tr><tr><td>Length \u2264 2</td><td colspan=\"2\">.56</td><td colspan=\"2\">.56</td><td colspan=\"2\">.69</td><td colspan=\"2\">.19</td><td colspan=\"2\">.13</td><td colspan=\"2\">.52</td><td colspan=\"2\">.15</td><td colspan=\"2\">.31</td><td colspan=\"2\">.66</td><td colspan=\"2\">.18</td><td colspan=\"2\">.10</td><td colspan=\"2\">.03</td></tr><tr><td>Length \u2264 3</td><td colspan=\"2\">.59</td><td colspan=\"2\">.60</td><td colspan=\"2\">.71</td><td colspan=\"2\">.14</td><td colspan=\"2\">.12</td><td colspan=\"2\">.49</td><td colspan=\"2\">.15</td><td colspan=\"2\">.32</td><td colspan=\"2\">.66</td><td colspan=\"2\">.13</td><td colspan=\"2\">.06</td><td colspan=\"2\">.02</td></tr><tr><td>Length \u2264 4</td><td colspan=\"2\">.60</td><td colspan=\"2\">.63</td><td colspan=\"2\">.72</td><td colspan=\"2\">.12</td><td colspan=\"2\">.11</td><td colspan=\"2\">.48</td><td colspan=\"2\">.14</td><td colspan=\"2\">.33</td><td colspan=\"2\">.68</td><td colspan=\"2\">.09</td><td colspan=\"2\">.04</td><td colspan=\"2\">.01</td></tr><tr><td>Length \u2264 5</td><td colspan=\"2\">.62</td><td colspan=\"2\">.65</td><td colspan=\"2\">.73</td><td colspan=\"2\">.11</td><td colspan=\"2\">.10</td><td colspan=\"2\">.46</td><td colspan=\"2\">.14</td><td colspan=\"2\">.33</td><td colspan=\"2\">.68</td><td colspan=\"2\">.07</td><td colspan=\"2\">.03</td><td colspan=\"2\">.01</td></tr></table>"
}
}
}
} |