File size: 92,397 Bytes
6fa4bc9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 | {
"paper_id": "D09-1001",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T16:40:01.601855Z"
},
"title": "Unsupervised Semantic Parsing",
"authors": [
{
"first": "Hoifung",
"middle": [],
"last": "Poon",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Washington Seattle",
"location": {
"postCode": "98195-2350",
"region": "WA",
"country": "U.S.A"
}
},
"email": "hoifung@cs.washington.edu"
},
{
"first": "Pedro",
"middle": [],
"last": "Domingos",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Washington Seattle",
"location": {
"postCode": "98195-2350",
"region": "WA",
"country": "U.S.A"
}
},
"email": "pedrod@cs.washington.edu"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "We present the first unsupervised approach to the problem of learning a semantic parser, using Markov logic. Our USP system transforms dependency trees into quasi-logical forms, recursively induces lambda forms from these, and clusters them to abstract away syntactic variations of the same meaning. The MAP semantic parse of a sentence is obtained by recursively assigning its parts to lambda-form clusters and composing them. We evaluate our approach by using it to extract a knowledge base from biomedical abstracts and answer questions. USP substantially outperforms TextRunner, DIRT and an informed baseline on both precision and recall on this task.",
"pdf_parse": {
"paper_id": "D09-1001",
"_pdf_hash": "",
"abstract": [
{
"text": "We present the first unsupervised approach to the problem of learning a semantic parser, using Markov logic. Our USP system transforms dependency trees into quasi-logical forms, recursively induces lambda forms from these, and clusters them to abstract away syntactic variations of the same meaning. The MAP semantic parse of a sentence is obtained by recursively assigning its parts to lambda-form clusters and composing them. We evaluate our approach by using it to extract a knowledge base from biomedical abstracts and answer questions. USP substantially outperforms TextRunner, DIRT and an informed baseline on both precision and recall on this task.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Semantic parsing maps text to formal meaning representations. This contrasts with semantic role labeling (Carreras and Marquez, 2004) and other forms of shallow semantic processing, which do not aim to produce complete formal meanings. Traditionally, semantic parsers were constructed manually, but this is too costly and brittle. Recently, a number of machine learning approaches have been proposed (Zettlemoyer and Collins, 2005; Mooney, 2007) . However, they are supervised, and providing the target logical form for each sentence is costly and difficult to do consistently and with high quality. Unsupervised approaches have been applied to shallow semantic tasks (e.g., paraphrasing (Lin and Pantel, 2001) , information extraction (Banko et al., 2007) ), but not to semantic parsing.",
"cite_spans": [
{
"start": 105,
"end": 133,
"text": "(Carreras and Marquez, 2004)",
"ref_id": "BIBREF2"
},
{
"start": 400,
"end": 431,
"text": "(Zettlemoyer and Collins, 2005;",
"ref_id": "BIBREF18"
},
{
"start": 432,
"end": 445,
"text": "Mooney, 2007)",
"ref_id": "BIBREF13"
},
{
"start": 688,
"end": 710,
"text": "(Lin and Pantel, 2001)",
"ref_id": "BIBREF10"
},
{
"start": 736,
"end": 756,
"text": "(Banko et al., 2007)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper we develop the first unsupervised approach to semantic parsing, using Markov logic (Richardson and Domingos, 2006) . Our USP system starts by clustering tokens of the same type, and then recursively clusters expressions whose subexpressions belong to the same clusters. Experiments on a biomedical corpus show that this approach is able to successfully translate syntactic variations into a logical representation of their common meaning (e.g., USP learns to map active and passive voice to the same logical form, etc.). This in turn allows it to correctly answer many more questions than systems based on TextRunner (Banko et al., 2007) and DIRT (Lin and Pantel, 2001) .",
"cite_spans": [
{
"start": 97,
"end": 128,
"text": "(Richardson and Domingos, 2006)",
"ref_id": "BIBREF16"
},
{
"start": 631,
"end": 651,
"text": "(Banko et al., 2007)",
"ref_id": "BIBREF1"
},
{
"start": 661,
"end": 683,
"text": "(Lin and Pantel, 2001)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We begin by reviewing the necessary background on semantic parsing and Markov logic. We then describe our Markov logic network for unsupervised semantic parsing, and the learning and inference algorithms we used. Finally, we present our experiments and results.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The standard language for formal meaning representation is first-order logic. A term is any expression representing an object in the domain. An atomic formula or atom is a predicate symbol applied to a tuple of terms. Formulas are recursively constructed from atomic formulas using logical connectives and quantifiers. A lexical entry defines the logical form for a lexical item (e.g., a word). The semantic parse of a sentence is derived by starting with logical forms in the lexical entries and recursively composing the meaning of larger fragments from their parts. In traditional approaches, the lexical entries and meaning-composition rules are both manually constructed. Below are sample rules in a definite clause grammar (DCG) for parsing the sentence: \"Utah borders Idaho\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing",
"sec_num": "2.1"
},
{
"text": "V erb [\u03bby\u03bbx.borders(x, y) ",
"cite_spans": [
{
"start": 6,
"end": 25,
"text": "[\u03bby\u03bbx.borders(x, y)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing",
"sec_num": "2.1"
},
{
"text": "] \u2192 borders N P [Utah] \u2192 U tah N P [Idaho] \u2192 Idaho V P [rel(obj)] \u2192 V erb[rel] N P [obj] S[rel(obj)] \u2192 N P [obj] V P [rel]",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing",
"sec_num": "2.1"
},
{
"text": "The first three lines are lexical entries. They are fired upon seeing the individual words. For example, the first rule applies to the word \"borders\" and generates syntactic category Verb with the meaning \u03bby\u03bbx.borders(x, y) that represents the nextto relation. Here, we use the standard lambdacalculus notation, where \u03bby\u03bbx.borders(x, y) represents a function that is true for any (x, y)pair such that borders(x, y) holds. The last two rules compose the meanings of sub-parts into that of the larger part. For example, after the first and third rules are fired, the fourth rule fires and generates V P [\u03bby\u03bbx.borders(x, y)(Idaho)]; this meaning simplifies to \u03bbx.borders(x, Idaho) by the \u03bb-reduction rule, which substitutes the argument for a variable in a functional application.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing",
"sec_num": "2.1"
},
{
"text": "A major challenge to semantic parsing is syntactic variations of the same meaning, which abound in natural languages. For example, the aforementioned sentence can be rephrased as \"Utah is next to Idaho,\"\"Utah shares a border with Idaho,\" etc. Manually encoding all these variations into the grammar is tedious and error-prone. Supervised semantic parsing addresses this issue by learning to construct the grammar automatically from sample meaning annotations (Mooney, 2007) . Existing approaches differ in the meaning representation languages they use and the amount of annotation required. In the approach of Zettlemoyer and Collins (2005) , the training data consists of sentences paired with their meanings in lambda form. A probabilistic combinatory categorial grammar (PCCG) is learned using a loglinear model, where the probability of the final logical form L and meaning-derivation tree T conditioned on the sentence S is P (L,",
"cite_spans": [
{
"start": 459,
"end": 473,
"text": "(Mooney, 2007)",
"ref_id": "BIBREF13"
},
{
"start": 610,
"end": 640,
"text": "Zettlemoyer and Collins (2005)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing",
"sec_num": "2.1"
},
{
"text": "T |S) = 1 Z exp ( i w i f i (L, T, S)).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing",
"sec_num": "2.1"
},
{
"text": "Here Z is the normalization constant and f i are the feature functions with weights w i . Candidate lexical entries are generated by a domain-specific procedure based on the target logical forms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing",
"sec_num": "2.1"
},
{
"text": "The major limitation of supervised approaches is that they require meaning annotations for example sentences. Even in a restricted domain, doing this consistently and with high quality requires nontrivial effort. For unrestricted text, the complexity and subjectivity of annotation render it essentially infeasible; even pre-specifying the target predicates and objects is very difficult. Therefore, to apply semantic parsing beyond limited domains, it is crucial to develop unsupervised methods that do not rely on labeled meanings.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing",
"sec_num": "2.1"
},
{
"text": "In the past, unsupervised approaches have been applied to some semantic tasks, but not to semantic parsing. For example, DIRT (Lin and Pantel, 2001 ) learns paraphrases of binary relations based on distributional similarity of their arguments; TextRunner (Banko et al., 2007) automatically extracts relational triples in open domains using a self-trained extractor; SNE applies relational clustering to generate a semantic network from TextRunner triples (Kok and Domingos, 2008) . While these systems illustrate the promise of unsupervised methods, the semantic content they extract is nonetheless shallow and does not constitute the complete formal meaning that can be obtained by a semantic parser.",
"cite_spans": [
{
"start": 126,
"end": 147,
"text": "(Lin and Pantel, 2001",
"ref_id": "BIBREF10"
},
{
"start": 255,
"end": 275,
"text": "(Banko et al., 2007)",
"ref_id": "BIBREF1"
},
{
"start": 455,
"end": 479,
"text": "(Kok and Domingos, 2008)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing",
"sec_num": "2.1"
},
{
"text": "Another issue is that existing approaches to semantic parsing learn to parse syntax and semantics together. 1 The drawback is that the complexity in syntactic processing is coupled with semantic parsing and makes the latter even harder. For example, when applying their approach to a different domain with somewhat less rigid syntax, Zettlemoyer and Collins (2007) need to introduce new combinators and new forms of candidate lexical entries. Ideally, we should leverage the enormous progress made in syntactic parsing and generate semantic parses directly from syntactic analysis.",
"cite_spans": [
{
"start": 108,
"end": 109,
"text": "1",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing",
"sec_num": "2.1"
},
{
"text": "In many NLP applications, there exist rich relations among objects, and recent work in statistical relational learning (Getoor and Taskar, 2007) and structured prediction (Bakir et al., 2007) has shown that leveraging these can greatly improve accuracy. One of the most powerful representations for this is Markov logic, which is a probabilistic extension of first-order logic (Richardson and Domingos, 2006) . Markov logic makes it possible to compactly specify probability distributions over complex relational domains, and has been successfully applied to unsupervised coreference resolution (Poon and Domingos, 2008) and other tasks. A Markov logic network (MLN) is a set of weighted first-order clauses. Together with a set of constants, it defines a Markov network with one node per ground atom and one feature per ground clause. The weight of a feature is the weight of the first-order clause that originated it. The probability of a state x in such a network is given by the log-linear model",
"cite_spans": [
{
"start": 119,
"end": 144,
"text": "(Getoor and Taskar, 2007)",
"ref_id": null
},
{
"start": 171,
"end": 191,
"text": "(Bakir et al., 2007)",
"ref_id": "BIBREF0"
},
{
"start": 377,
"end": 408,
"text": "(Richardson and Domingos, 2006)",
"ref_id": "BIBREF16"
},
{
"start": 595,
"end": 620,
"text": "(Poon and Domingos, 2008)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Markov Logic",
"sec_num": "2.2"
},
{
"text": "P (x) = 1 Z exp ( i w i n i (x)),",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Markov Logic",
"sec_num": "2.2"
},
{
"text": "where Z is a normalization constant, w i is the weight of the ith formula, and n i is the number of satisfied groundings.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Markov Logic",
"sec_num": "2.2"
},
{
"text": "Unsupervised semantic parsing (USP) rests on three key ideas. First, the target predicate and object constants, which are pre-specified in supervised semantic parsing, can be viewed as clusters of syntactic variations of the same meaning, and can be learned from data. For example, borders represents the next-to relation, and can be viewed as the cluster of different forms for expressing this relation, such as \"borders\", \"is next to\", \"share the border with\"; Utah represents the state of Utah, and can be viewed as the cluster of \"Utah\", \"the beehive state\", etc. Second, the identification and clustering of candidate forms are integrated with the learning for meaning composition, where forms that are used in composition with the same forms are encouraged to cluster together, and so are forms that are composed of the same sub-forms. This amounts to a novel form of relational clustering, where clustering is done not just on fixed elements in relational tuples, but on arbitrary forms that are built up recursively.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Unsupervised Semantic Parsing with Markov Logic",
"sec_num": "3"
},
{
"text": "Third, while most existing approaches (manual or supervised learning) learn to parse both syntax and semantics, unsupervised semantic parsing starts directly from syntactic analyses and focuses solely on translating them to semantic content. This enables us to leverage advanced syntactic parsers and (indirectly) the available rich resources for them. More importantly, it separates the complexity in syntactic analysis from the semantic one, and makes the latter much easier to perform. In particular, meaning composition does not require domain-specific procedures for generating candidate lexicons, as is often needed by supervised methods.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Unsupervised Semantic Parsing with Markov Logic",
"sec_num": "3"
},
{
"text": "The input to our USP system consists of dependency trees of training sentences. Compared to phrase-structure syntax, dependency trees are the more appropriate starting point for semantic processing, as they already exhibit much of the relation-argument structure at the lexical level.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Unsupervised Semantic Parsing with Markov Logic",
"sec_num": "3"
},
{
"text": "USP first uses a deterministic procedure to convert dependency trees into quasi-logical forms (QLFs). The QLFs and their sub-formulas have natural lambda forms, as will be described later. Starting with clusters of lambda forms at the atom level, USP recursively builds up clusters of larger lambda forms. The final output is a probability distribution over lambda-form clusters and their compositions, as well as the MAP semantic parses of training sentences.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Unsupervised Semantic Parsing with Markov Logic",
"sec_num": "3"
},
{
"text": "In the remainder of the section, we describe the details of USP. We first present the procedure for generating QLFs from dependency trees. We then introduce their lambda forms and clusters, and show how semantic parsing works in this setting. Finally, we present the Markov logic network (MLN) used by USP. In the next sections, we present efficient algorithms for learning and inference with this MLN.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Unsupervised Semantic Parsing with Markov Logic",
"sec_num": "3"
},
{
"text": "A dependency tree is a tree where nodes are words and edges are dependency labels. To derive the QLF, we convert each node to an unary atom with the predicate being the lemma plus POS tag (below, we still use the word for simplicity), and each edge to a binary atom with the predicate being the dependency label. For example, the node for Utah becomes Utah(n 1 ) and the subject dependency becomes nsubj(n1, n2). Here, the n i are Skolem constants indexed by the nodes. The QLF for a sentence is the conjunction of the atoms for the nodes and edges, e.g., the sentence above will become borders",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Derivation of Quasi-Logical Forms",
"sec_num": "3.1"
},
{
"text": "(n 1 ) \u2227 Utah(n 2 ) \u2227 Idaho(n 3 ) \u2227 nsubj(n 1 , n 2 ) \u2227 dobj(n 1 , n 3 ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Derivation of Quasi-Logical Forms",
"sec_num": "3.1"
},
{
"text": "Parsing in USP Given a QLF, a relation or an object is represented by the conjunction of a subset of the atoms. For example, the next-to relation is represented by borders(n 1 ) \u2227 nsubj(n 1 , n 2 ) \u2227 dobj(n 1 , n 3 ), and the states of Utah and Idaho are represented by Utah(n 2 ) and Idaho(n 3 ). The meaning composition of two sub-formulas is simply their conjunction. This allows the maximum flexibility in learning. In particular, lexical entries are no longer limited to be adjacent words as in Zettlemoyer and Collins (2005) , but can be arbitrary fragments in a dependency tree. For every sub-formula F , we define a corresponding lambda form that can be derived by replacing every Skolem constant n i that does not appear in any unary atom in F with a unique lambda variable x i . Intuitively, such constants represent objects introduced somewhere else (by the unary atoms containing them), and correspond to the arguments of the relation represented by F . For example, the lambda form",
"cite_spans": [
{
"start": 500,
"end": 530,
"text": "Zettlemoyer and Collins (2005)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Lambda-Form Clusters and Semantic",
"sec_num": "3.2"
},
{
"text": "for borders(n 1 ) \u2227 nsubj(n 1 , n 2 ) \u2227 dobj(n 1 , n 3 ) is \u03bbx 2 \u03bbx 3 . borders(n 1 ) \u2227 nsubj(n 1 , x 2 ) \u2227 dobj(n 1 , x 3 ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Lambda-Form Clusters and Semantic",
"sec_num": "3.2"
},
{
"text": "Conceptually, a lambda-form cluster is a set of semantically interchangeable lambda forms. For example, to express the meaning that Utah borders Idaho, we can use any form in the cluster representing the next-to relation (e.g., \"borders\", \"shares a border with\"), any form in the cluster representing the state of Utah (e.g., \"the beehive state\"), and any form in the cluster representing the state of Idaho (e.g., \"Idaho\"). Conditioned on the clusters, the choices of individual lambda forms are independent of each other.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Lambda-Form Clusters and Semantic",
"sec_num": "3.2"
},
{
"text": "To handle variable number of arguments, we follow Davidsonian semantics and further decompose a lambda form into the core form, which does not contain any lambda variable (e.g., borders(n 1 )), and the argument forms, which contain a single lambda variable (e.g., \u03bbx 2 .nsubj(n 1 , x 2 ) and \u03bbx 3 .dobj(n 1 , x 3 )). Each lambda-form cluster may contain some number of argument types, which cluster distinct forms of the same argument in a relation. For example, in Stanford dependencies, the object of a verb uses the dependency dobj in the active voice, but nsubjpass in passive.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Lambda-Form Clusters and Semantic",
"sec_num": "3.2"
},
{
"text": "Lambda-form clusters abstract away syntactic variations of the same meaning. Given an instance of cluster T with arguments of argument types",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Lambda-Form Clusters and Semantic",
"sec_num": "3.2"
},
{
"text": "A 1 , \u2022 \u2022 \u2022 , A k , its abstract lambda form is given by \u03bbx 1 \u2022 \u2022 \u2022 \u03bbx k .T(n) \u2227 k i=1 A i (n, x i )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Lambda-Form Clusters and Semantic",
"sec_num": "3.2"
},
{
"text": ". Given a sentence and its QLF, semantic parsing amounts to partitioning the atoms in the QLF, dividing each part into core form and argument forms, and then assigning each form to a cluster or an argument type. The final logical form is derived by composing the abstract lambda forms of the parts using the \u03bb-reduction rule. 2",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Lambda-Form Clusters and Semantic",
"sec_num": "3.2"
},
{
"text": "Formally, for a QLF Q, a semantic parse L partitions Q into parts p 1 , p 2 , \u2022 \u2022 \u2022 , p n ; each part p is assigned to some lambda-form cluster c, and is further partitioned into core form f and argument forms f 1 , \u2022 \u2022 \u2022 , f k ; each argument form is assigned to an argument type a in c. The USP MLN defines a joint probability distribution over Q and L by modeling the distributions over forms and arguments given the cluster or argument type.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The USP MLN",
"sec_num": "3.3"
},
{
"text": "Before presenting the predicates and formulas in our MLN, we should emphasize that they should not be confused with the atoms and formulas in the QLFs, which are represented by reified constants and variables.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The USP MLN",
"sec_num": "3.3"
},
{
"text": "To model distributions over lambda forms, we introduce the predicates Form(p, f!) and ArgForm(p, i, f!), where p is a part, i is the index of an argument, and f is a QLF subformula. Form(p, f) is true iff part p has core form f, and ArgForm(p, i, f) is true iff the ith argument in p has form f. 3 The \"f!\" notation signifies that each part or argument can have only one form.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The USP MLN",
"sec_num": "3.3"
},
{
"text": "To model distributions over arguments, we introduce three more predicates: ArgType(p, i, a!) signifies that the ith argument of p is assigned to argument type a; Arg(p, i, p \u2032 ) signifies that the ith argument of p is p \u2032 ; Number(p, a, n) signifies that there are n arguments of p that are assigned to type a. The truth value of Number(p, a, n) is determined by the ArgType atoms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The USP MLN",
"sec_num": "3.3"
},
{
"text": "Unsupervised semantic parsing can be captured by four formulas:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The USP MLN",
"sec_num": "3.3"
},
{
"text": "p \u2208 +c \u2227 Form(p, +f) ArgType(p, i, +a) \u2227 ArgForm(p, i, +f) Arg(p, i, p \u2032 ) \u2227 ArgType(p, i, +a) \u2227 p \u2032 \u2208 +c \u2032 Number(p, +a, +n)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The USP MLN",
"sec_num": "3.3"
},
{
"text": "All free variables are implicitly universally quantified. The \"+\" notation signifies that the MLN contains an instance of the formula, with a separate weight, for each value combination of the variables with a plus sign. The first formula models the mixture of core forms given the cluster, and the others model the mixtures of argument forms, argument types, and argument numbers, respectively, given the argument type.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The USP MLN",
"sec_num": "3.3"
},
{
"text": "To encourage clustering and avoid overfitting, we impose an exponential prior with weight \u03b1 on the number of parameters. 4 The MLN above has one problem: it often clusters expressions that are semantically opposite. For example, it clusters antonyms like \"elderly/young\", \"mature/immature\". This issue also occurs in other semantic-processing systems (e.g., DIRT). In general, this is a difficult open problem that only recently has started to receive some attention (Mohammad et al., 2008) . Resolving this is not the focus of this paper, but we describe a general heuristic for fixing this problem. We observe that the problem stems from the lack of negative features for discovering meanings in contrast. In natural languages, parallel structures like conjunctions are one such feature. 5 We thus introduce an exponential prior with weight \u03b2 on the number of conjunctions where the two conjunctive parts are assigned to the same cluster. To detect conjunction, we simply used the Stanford dependencies that begin with \"conj\". This proves very effective, fixing the majority of the errors in our experiments.",
"cite_spans": [
{
"start": 121,
"end": 122,
"text": "4",
"ref_id": null
},
{
"start": 467,
"end": 490,
"text": "(Mohammad et al., 2008)",
"ref_id": "BIBREF12"
},
{
"start": 790,
"end": 791,
"text": "5",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The USP MLN",
"sec_num": "3.3"
},
{
"text": "Given a sentence and the quasi-logical form Q derived from its dependency tree, the conditional probability for a semantic parse L is given by P r",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "4"
},
{
"text": "(L|Q) \u221d exp ( i w i n i (L, Q)). The MAP se- mantic parse is simply arg max L i w i n i (L, Q).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "4"
},
{
"text": "Enumerating all L's is intractable. It is also unnecessary, since most partitions will result in parts whose lambda forms have no cluster they can be assigned to. Instead, USP uses a greedy algorithm to search for the MAP parse. First we introduce some definitions: a partition is called \u03bb-reducible from p if it can be obtained from the current partition by recursively \u03bb-reducing the part containing p with one of its arguments; such a partition is Algorithm 1 USP-Parse(MLN, QLF)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "4"
},
{
"text": "Form parts for individual atoms in QLF and assign each to its most probable cluster repeat for all parts p in the current partition do for all partitions that are \u03bb-reducible from p and feasible do Find the most probable cluster and argument type assignments for the new part and its arguments end for end for Change to the new partition and assignments with the highest gain in probability until none of these improve the probability return current partition and assignments called feasible if the core form of the new part is contained in some cluster. For example, consider the QLF of \"Utah borders Idaho\" and assume that the current partition is",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "4"
},
{
"text": "\u03bbx 2 x 3 .borders(n 1 ) \u2227 nsubj(n 1 , x 2 ) \u2227 dobj(n 1 , x 3 ), Utah(n 2 ), Idaho(n 3 ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "4"
},
{
"text": "Then the following partition is \u03bb-reducible from the first part in the above partition:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "4"
},
{
"text": "\u03bbx 3 .borders(n 1 ) \u2227 nsubj(n 1 , n 2 ) \u2227 Utah(n 2 ) \u2227 dobj(n 1 , x 3 ), Idaho(n 3 ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "4"
},
{
"text": "Whether this new partition is feasible depends on whether the core form of the new part \u03bbx 3 .borders(n 1 ) \u2227 nsubj(n 1 , n 2 ) \u2227 Utah(n 2 ) \u2227 dobj(n 1 , x 3 ) (i.e. borders(n 1 ) \u2227 nsubj(n 1 , n 2 ) \u2227 Utah(n 2 )) is contained in some lambda-form cluster.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "4"
},
{
"text": "Algorithm 1 gives pseudo-code for our algorithm. Given part p, finding partitions that are \u03bbreducible from p and feasible can be done in time O(ST ), where S is the size of the clustering in the number of core forms and T is the maximum number of atoms in a core form. We omit the proof here but point out that it is related to the unordered subtree matching problem which can be solved in linear time (Kilpelainen, 1992) . Inverted indexes (e.g., from p to eligible core forms) are used to further improve the efficiency. For a new part p and a cluster that contains p's core form, there are k m ways of assigning p's m arguments to the k argument types of the cluster. For larger k and m, this is very expensive. We therefore approximate it by assigning each argument to the best type, independent of other arguments. This algorithm is very efficient, and is used repeatedly in learning.",
"cite_spans": [
{
"start": 402,
"end": 421,
"text": "(Kilpelainen, 1992)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Inference",
"sec_num": "4"
},
{
"text": "The learning problem in USP is to maximize the log-likelihood of observing the QLFs obtained from the dependency trees, denoted by Q, summing out the unobserved semantic parses:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "L \u03b8 (Q) = log P \u03b8 (Q) = log L P \u03b8 (Q, L)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "Here, L are the semantic parses, \u03b8 are the MLN parameters, and P \u03b8 (Q, L) are the completion likelihoods. A serious challenge in unsupervised learning is the identifiability problem (i.e., the optimal parameters are not unique) (Liang and Klein, 2008) . This problem is particularly severe for log-linear models with hard constraints, which are common in MLNs. For example, in our USP MLN, conditioned on the fact that p \u2208 c, there is exactly one value of f that can satisfy the formula p \u2208 c \u2227 Form(p, f), and if we add some constant number to the weights of p \u2208 c \u2227 Form(p, f) for all f, the probability distribution stays the same. 6 The learner can be easily confused by the infinitely many optima, especially in the early stages. To address this problem, we impose local normalization constraints on specific groups of formulas that are mutually exclusive and exhaustive, i.e., in each group, we require that k i=1 e w i = 1, where w i are the weights of formulas in the group. Grouping is done in such a way as to encourage the intended mixture behaviors. Specifically, for the rule p \u2208 +c \u2227 Form(p, +f), all instances given a fixed c form a group; for each of the remaining three rules, all instances given a fixed a form a group. Notice that with these constraints the completion likelihood P (Q, L) can be computed in closed form for any L. In particular, each formula group contributes a term equal to the weight of the currently satisfied formula. In addition, the optimal weights that maximize the completion likelihood P (Q, L) can be derived in closed form using empirical relative frequencies. E.g., the optimal weight of p \u2208 c \u2227 Form(p, f) is log(n c,f /n c ), where n c,f is the number of parts p that satisfy both p \u2208 c and Form(p, f), and n c is the number of parts p that satisfy p \u2208 c. 7 We leverage this fact for efficient learning in USP. Another major challenge in USP learning is the summation in the likelihood, which is over all possible semantic parses for a given dependency tree. Even an efficient sampler like MC-SAT (Poon and Domingos, 2006) , as used in Poon & Domingos (2008) , would have a hard time generating accurate estimates within a reasonable amount of time. On the other hand, as already noted in the previous section, the lambda-form distribution is generally sparse. Large lambda-forms are rare, as they correspond to complex expressions that are often decomposable into smaller ones. Moreover, while ambiguities are present at the lexical level, they quickly diminish when more words are present. Therefore, a lambda form can usually only belong to a small number of clusters, if not a unique one. We thus simplify the problem by approximating the sum with the mode, and search instead for the L and \u03b8 that maximize log P \u03b8 (Q, L). Since the optimal weights and log-likelihood can be derived in closed form given the semantic parses L, we simply search over semantic parses, evaluating them using log-likelihood.",
"cite_spans": [
{
"start": 228,
"end": 251,
"text": "(Liang and Klein, 2008)",
"ref_id": "BIBREF9"
},
{
"start": 635,
"end": 636,
"text": "6",
"ref_id": null
},
{
"start": 1807,
"end": 1808,
"text": "7",
"ref_id": null
},
{
"start": 2048,
"end": 2073,
"text": "(Poon and Domingos, 2006)",
"ref_id": "BIBREF14"
},
{
"start": 2087,
"end": 2109,
"text": "Poon & Domingos (2008)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "Algorithm 2 gives pseudo-code for our algorithm. The input consists of an MLN without weights and the QLFs for the training sentences. Two operators are used for updating semantic parses. The first is to merge two clusters, denoted by MERGE(C 1 , C 2 ) for clusters C 1 , C 2 , which does the following:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "and there is the local normalization constraint f e wc,f = 1. The optimal weights wc,f are easily derived by solving this constrained optimization problem.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "1. Create a new cluster C and add all core forms in C 1 , C 2 to C;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "2. Create new argument types for C by merging those in C 1 , C 2 so as to maximize the loglikelihood;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "3. Remove C 1 , C 2 .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "Here, merging two argument types refers to pooling their argument forms to create a new argument type. Enumerating all possible ways of creating new argument types is intractable. USP approximates it by considering one type at a time and either creating a new type for it or merging it to types already considered, whichever maximizes the loglikelihood. The types are considered in decreasing order of their numbers of occurrences so that more information is available for each decision. MERGE clusters syntactically different expressions whose meanings appear to be the same according to the model. The second operator is to create a new cluster by composing two existing ones, denoted by COMPOSE(C R , C A ), which does the following:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "1. Create a new cluster C;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "2. Find all parts r \u2208 C R , a \u2208 C A such that a is an argument of r, compose them to r(a) by \u03bb-reduction and add the new part to C;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "3. Create new argument types for C from the argument forms of r(a) so as to maximize the log-likelihood.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "COMPOSE creates clusters of large lambda-forms if they tend to be composed of the same subforms (e.g., the lambda form for \"is next to\"). These lambda-forms may later be merged with other clusters (e.g., borders). At learning time, USP maintains an agenda that contains operations that have been evaluated and are pending execution. During initialization, USP forms a part and creates a new cluster for each unary atom u(n). It also assigns binary atoms of the form b(n, n \u2032 ) to the part as argument forms and creates a new argument type for each. This forms the initial clustering and semantic parses. USP then merges clusters with the same core form (i.e., the same unary predicate) using MERGE. 8 At each step, USP evaluates the candidate operations and adds them to the agenda if the improvement is above a threshold. 9 The operation with the highest score is executed, and the parameters are updated with the new optimal values. The QLFs which contain an affected part are reparsed, and operations in the agenda whose score might be affected are re-evaluated. These changes are done very efficiently using inverted indexes. We omit the details here due to space limitations. USP terminates when the agenda is empty, and outputs the current MLN parameters and semantic parses.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "USP learning uses the same optimization objective as hard EM, and is also guaranteed to find a local optimum since at each step it improves the log-likelihood. It differs from EM in directly optimizing the likelihood instead of a lower bound.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Learning",
"sec_num": "5"
},
{
"text": "Evaluating unsupervised semantic parsers is difficult, because there is no predefined formal language or gold logical forms for the input sentences. Thus the best way to test them is by using them for the ultimate goal: answering questions based on the input corpus. In this paper, we applied USP to extracting knowledge from biomedical abstracts and evaluated its performance in answering a set of questions that simulate the information needs of biomedical researchers. We used the GENIA dataset (Kim et al., 2003) as the source for knowledge extraction. It contains 1999 PubMed abstracts and marks all mentions of biomedical entities according to the GENIA ontology, such as cell, protein, and DNA. As a first approximation to the questions a biomedical researcher might ask, we generated a set of two thousand questions on relations between entities. Sample questions are: \"What regulates MIP-1alpha?\", \"What does anti-STAT 1 inhibit?\". To simulate the real information need, we sample the relations from the 100 most frequently used verbs (excluding the auxiliary verbs be, have, and do), and sample the entities from those annotated in GENIA, both according to their numbers of occurrences. We evaluated USP by the number of answers it provided and the accuracy as determined by manual labeling. 10",
"cite_spans": [
{
"start": 498,
"end": 516,
"text": "(Kim et al., 2003)",
"ref_id": "BIBREF7"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Task",
"sec_num": "6.1"
},
{
"text": "Since USP is the first unsupervised semantic parser, conducting a meaningful comparison of it with other systems is not straightforward. Standard question-answering (QA) benchmarks do not provide the most appropriate comparison, because they tend to simultaneously emphasize other aspects not directly related to semantic parsing. Moreover, most state-of-the-art QA systems use supervised learning in their key components and/or require domain-specific engineering efforts. The closest available system to USP in aims and capabilities is TextRunner (Banko et al., 2007) , and we compare with it. TextRunner is the state-of-the-art system for open-domain information extraction; its goal is to extract knowledge from text without using supervised labels. Given that a central challenge to semantic parsing is resolving syntactic variations of the same meaning, we also compare with RESOLVER (Yates and Etzioni, 2009) , a state-of-the-art unsupervised system based on TextRunner for jointly resolving entities and relations, and DIRT (Lin and Pantel, 2001) , which resolves paraphrases of binary relations. Finally, we also compared to an informed baseline based on keyword matching. Keyword: We consider a baseline system based on keyword matching. The question substring containing the verb and the available argument is directly matched with the input text, ignoring case and morphology. We consider two ways to derive the answer given a match. The first one (KW) simply returns the rest of sentence on the other side of the verb. The second one (KW-SYN) is informed by syntax: the answer is extracted from the subject or object of the verb, depending on the question. If the verb does not contain the expected argument, the sentence is ignored. TextRunner: TextRunner inputs text and outputs relational triples in the form (R, A 1 , A 2 ) , where R is the relation string, and A 1 , A 2 the argument strings. Given a triple and a question, we first match their relation strings, and then match the strings for the argument that is present in the question. If both match, we return the other argument string in the triple as an answer. We report results when exact match is used (TR-EXACT), or when the triple string can contain the question one as a substring (TR-SUB). RESOLVER: RESOLVER (Yates and Etzioni, 2009) inputs TextRunner triples and collectively resolves coreferent relation and argument strings. On the GENIA data, using the default parameters, RESOLVER produces only a few trivial relation clusters and no argument clusters. This is not surprising, since RESOLVER assumes high redundancy in the data, and will discard any strings with fewer than 25 extractions. For a fair comparison, we also ran RESOLVER using all extractions, and manually tuned the parameters based on eyeballing of clustering quality. The best result was obtained with 25 rounds of execution and with the entity multiple set to 200 (the default is 30). To answer questions, the only difference from TextRunner is that a question string can match any string in its cluster. As in TextRunner, we report results for both exact match (RS-EXACT) and substring (RS-SUB). DIRT: The DIRT system inputs a path and returns a set of similar paths. To use DIRT in question answering, we queried it to obtain similar paths for the relation of the question, and used these paths while matching sentences. We first used MINIPAR (Lin, 1998) to parse input text using the same dependencies as DIRT. To determine a match, we first check if the sentence contains the question path or one of its DIRT paths. If so, and if the available argument slot in the question is contained in the one in the sentence, it is a match, and we return the other argument slot from the sentence if it is present. Ideally, a fair comparison will require running DIRT on the GENIA text, but we were not able to obtain the source code. We thus resorted to using the latest DIRT database released by the author, which contains paths extracted from a large corpus with more than 1GB of text. This puts DIRT in a very advantageous position compared with other systems. In our experiments, we used the top three similar paths, as including more results in very low precision. USP: We built a system for knowledge extraction and question answering on top of USP. It generated Stanford dependencies (de Marneffe et al., 2006) from the input text using the Stanford parser, and then fed these to USP-Learn 11 , which produced an MLN with learned weights and the MAP semantic parses of the input sentences. These MAP parses formed our knowledge base (KB). To answer questions, the system first parses the questions 12 using USP-Parse with the learned MLN, and then matches the question parse to parses in the KB by testing subsumption (i.e., a question parse matches a KB one iff the former is subsumed by the latter). When a match occurs, our system then looks for arguments of type in accordance with the question. For example, if the question is \"What regulates MIP-1alpha?\", it searches for the argument type of the relation that contains the argument form \"nsubj\" for subject. If such an argument exists for the relation part, it will be returned as the answer. Table 1 shows the results for all systems. USP extracted the highest number of answers, almost doubling that of the second highest (RS-SUB). It obtained the highest accuracy at 88%, and the number of correct answers it extracted is three times that of the second highest system. The informed baseline (KW-SYN) did surprisingly well compared to systems other than USP, in terms of accuracy and number of correct answers. TextRunner achieved good accuracy when exact match is used (TR-EXACT), but only obtained a fraction of the answers compared to USP. With substring match, its recall substantially improved, but precision dropped more than 20 points. RE-SOLVER improved the number of extracted answers by sanctioning more matches based on the clusters it generated. However, most of those additional answers are incorrect due to wrong clustering. DIRT obtained the second highest number of correct answers, but its precision is quite low because the similar paths contain many errors.",
"cite_spans": [
{
"start": 549,
"end": 569,
"text": "(Banko et al., 2007)",
"ref_id": "BIBREF1"
},
{
"start": 890,
"end": 915,
"text": "(Yates and Etzioni, 2009)",
"ref_id": "BIBREF17"
},
{
"start": 1032,
"end": 1054,
"text": "(Lin and Pantel, 2001)",
"ref_id": "BIBREF10"
},
{
"start": 2291,
"end": 2316,
"text": "(Yates and Etzioni, 2009)",
"ref_id": "BIBREF17"
},
{
"start": 3400,
"end": 3411,
"text": "(Lin, 1998)",
"ref_id": "BIBREF11"
},
{
"start": 4340,
"end": 4366,
"text": "(de Marneffe et al., 2006)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [
{
"start": 1825,
"end": 1840,
"text": "(R, A 1 , A 2 )",
"ref_id": null
},
{
"start": 5206,
"end": 5213,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Systems",
"sec_num": "6.2"
},
{
"text": "Manual inspection shows that USP is able to resolve many nontrivial syntactic variations without user supervision. It consistently resolves the syntactic difference between active and passive voices. It successfully identifies many distinct argument forms that mean the same (e.g., \"X stimulates Y\" \u2248 \"Y is stimulated with X\", \"expression of X\" \u2248 \"X expression\"). It also resolves many nouns correctly and forms meaningful groups of relations. Here are some sample clusters in core forms: {investigate, examine, evaluate, analyze, study, assay} {diminish, reduce, decrease, attenuate} {synthesis, production, secretion, release} {dramatically, substantially, significantly} An example question-answer pair, together with the source sentence, is shown below: Q: What does IL-13 enhance? A: The 12-lipoxygenase activity of murine macrophages.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Qualitative Analysis",
"sec_num": "6.4"
},
{
"text": "Sentence: The data presented here indicate that (1) the 12-lipoxygenase activity of murine macrophages is upregulated in vitro and in vivo by IL-4 and/or IL-13, . . .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Qualitative Analysis",
"sec_num": "6.4"
},
{
"text": "This paper introduces the first unsupervised approach to learning semantic parsers. Our USP system is based on Markov logic, and recursively clusters expressions to abstract away syntactic variations of the same meaning. We have successfully applied USP to extracting a knowledge base from biomedical text and answering questions based on it.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "7"
},
{
"text": "Directions for future work include: better handling of antonyms, subsumption relations among expressions, quantifier scoping, more complex lambda forms, etc.; use of context and discourse to aid expression clustering and semantic parsing; more efficient learning and inference; application to larger corpora; etc.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "7"
},
{
"text": "The only exception that we are aware of isGe and Mooney (2009).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Currently, we do not handle quantifier scoping or semantics for specific closed-class words such as determiners. These will be pursued in future work.3 There are hard constraints to guarantee that these assignments form a legal partition. We omit them for simplicity.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Excluding weights of \u221e or \u2212\u221e, which signify hard constraints.5 For example, in the sentence \"IL-2 inhibits X in A and induces Y in B\", the conjunction between \"inhibits\" and \"induces\" suggests that they are different. If \"inhibits\" and \"induces\" are indeed synonyms, such a sentence will sound awkward and would probably be rephrased as \"IL-2 inhibits X in A and Y in B\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Regularizations, e.g., Gaussian priors on weights, alleviate this problem by penalizing large weights, but it remains true that weights within a short range are roughly equivalent.7 To see this, notice that for a given c, the total contribution to the completion likelihood from all groundings in its formula group is f wc,fnc,f. In addition,f nc,f = nc",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Word-sense disambiguation can be handled by including a new kind of operator that splits a cluster into subclusters. We leave this to future work.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "We currently set it to 10 to favor precision and guard against errors due to inexact estimates.10 The labels and questions are available at http://alchemy.cs.washington.edu/papers/poon09.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "\u03b1 and \u03b2 are set to \u22125 and \u221210.12 The question slot is replaced by a dummy word.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "We thank the anonymous reviewers for their comments. This ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": "8"
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Predicting Structured Data",
"authors": [
{
"first": "G",
"middle": [],
"last": "Bakir",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Hofmann",
"suffix": ""
},
{
"first": "B",
"middle": [
"B"
],
"last": "Sch\u00f6lkopf",
"suffix": ""
}
],
"year": 2007,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "G. Bakir, T. Hofmann, B. B. Sch\u00f6lkopf, A. Smola, B. Taskar, S. Vishwanathan, and (eds.). 2007. Pre- dicting Structured Data. MIT Press, Cambridge, MA.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Open information extraction from the web",
"authors": [
{
"first": "Michele",
"middle": [],
"last": "Banko",
"suffix": ""
},
{
"first": "Michael",
"middle": [
"J"
],
"last": "Cafarella",
"suffix": ""
},
{
"first": "Stephen",
"middle": [],
"last": "Soderland",
"suffix": ""
},
{
"first": "Matt",
"middle": [],
"last": "Broadhead",
"suffix": ""
},
{
"first": "Oren",
"middle": [],
"last": "Etzioni",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the Twentieth International Joint Conference on Artificial Intelligence",
"volume": "",
"issue": "",
"pages": "2670--2676",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Michele Banko, Michael J. Cafarella, Stephen Soder- land, Matt Broadhead, and Oren Etzioni. 2007. Open information extraction from the web. In Pro- ceedings of the Twentieth International Joint Con- ference on Artificial Intelligence, pages 2670-2676, Hyderabad, India. AAAI Press.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Introduction to the CoNLL-2004 shared task: Semantic role labeling",
"authors": [
{
"first": "Xavier",
"middle": [],
"last": "Carreras",
"suffix": ""
},
{
"first": "Luis",
"middle": [],
"last": "Marquez",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of the Eighth Conference on Computational Natural Language Learning",
"volume": "",
"issue": "",
"pages": "89--97",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xavier Carreras and Luis Marquez. 2004. Introduction to the CoNLL-2004 shared task: Semantic role la- beling. In Proceedings of the Eighth Conference on Computational Natural Language Learning, pages 89-97, Boston, MA. ACL.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Generating typed dependency parses from phrase structure parses",
"authors": [
{
"first": "Marie-Catherine",
"middle": [],
"last": "De Marneffe",
"suffix": ""
},
{
"first": "Bill",
"middle": [],
"last": "Maccartney",
"suffix": ""
},
{
"first": "Christopher",
"middle": [
"D"
],
"last": "Manning",
"suffix": ""
}
],
"year": 2006,
"venue": "Proceedings of the Fifth International Conference on Language Resources and Evaluation",
"volume": "",
"issue": "",
"pages": "449--454",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marie-Catherine de Marneffe, Bill MacCartney, and Christopher D. Manning. 2006. Generating typed dependency parses from phrase structure parses. In Proceedings of the Fifth International Conference on Language Resources and Evaluation, pages 449- 454, Genoa, Italy. ELRA.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Learning a compositional semantic parser using an existing syntactic parser",
"authors": [
{
"first": "Ruifang",
"middle": [],
"last": "Ge",
"suffix": ""
},
{
"first": "Raymond",
"middle": [
"J"
],
"last": "Mooney",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the Forty Seventh Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ruifang Ge and Raymond J. Mooney. 2009. Learning a compositional semantic parser using an existing syntactic parser. In Proceedings of the Forty Sev- enth Annual Meeting of the Association for Compu- tational Linguistics, Singapore. ACL.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Introduction to Statistical Relational Learning",
"authors": [],
"year": 2007,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lise Getoor and Ben Taskar, editors. 2007. Introduc- tion to Statistical Relational Learning. MIT Press, Cambridge, MA.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Tree Matching Problems with Applications to Structured Text databases",
"authors": [
{
"first": "Pekka",
"middle": [],
"last": "Kilpelainen",
"suffix": ""
}
],
"year": 1992,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Pekka Kilpelainen. 1992. Tree Matching Prob- lems with Applications to Structured Text databases. Ph.D. Thesis, Department of Computer Science, University of Helsinki.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "GENIA corpus -a semantically annotated corpus for bio-textmining",
"authors": [
{
"first": "Jin-Dong",
"middle": [],
"last": "Kim",
"suffix": ""
},
{
"first": "Tomoko",
"middle": [],
"last": "Ohta",
"suffix": ""
},
{
"first": "Yuka",
"middle": [],
"last": "Tateisi",
"suffix": ""
},
{
"first": "Jun'ichi",
"middle": [],
"last": "Tsujii",
"suffix": ""
}
],
"year": 2003,
"venue": "Bioinformatics",
"volume": "19",
"issue": "",
"pages": "180--82",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jin-Dong Kim, Tomoko Ohta, Yuka Tateisi, and Jun'ichi Tsujii. 2003. GENIA corpus -a seman- tically annotated corpus for bio-textmining. Bioin- formatics, 19:180-82.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Extracting semantic networks from text via relational clustering",
"authors": [
{
"first": "Stanley",
"middle": [],
"last": "Kok",
"suffix": ""
},
{
"first": "Pedro",
"middle": [],
"last": "Domingos",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the Nineteenth European Conference on Machine Learning",
"volume": "",
"issue": "",
"pages": "624--639",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Stanley Kok and Pedro Domingos. 2008. Extract- ing semantic networks from text via relational clus- tering. In Proceedings of the Nineteenth European Conference on Machine Learning, pages 624-639, Antwerp, Belgium. Springer.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Analyzing the errors of unsupervised learning",
"authors": [
{
"first": "Percy",
"middle": [],
"last": "Liang",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Klein",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the Forty Sixth Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "879--887",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Percy Liang and Dan Klein. 2008. Analyzing the er- rors of unsupervised learning. In Proceedings of the Forty Sixth Annual Meeting of the Association for Computational Linguistics, pages 879-887, Colum- bus, OH. ACL.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "DIRT -discovery of inference rules from text",
"authors": [
{
"first": "Dekang",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "Patrick",
"middle": [],
"last": "Pantel",
"suffix": ""
}
],
"year": 2001,
"venue": "Proceedings of the Seventh ACM SIGKDD International Conference on Knowledge Discovery and Data Mining",
"volume": "",
"issue": "",
"pages": "323--328",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dekang Lin and Patrick Pantel. 2001. DIRT -dis- covery of inference rules from text. In Proceedings of the Seventh ACM SIGKDD International Con- ference on Knowledge Discovery and Data Mining, pages 323-328, San Francisco, CA. ACM Press.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Dependency-based evaluation of MINIPAR",
"authors": [
{
"first": "Dekang",
"middle": [],
"last": "Lin",
"suffix": ""
}
],
"year": 1998,
"venue": "Proceedings of the Workshop on the Evaluation of Parsing Systems",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dekang Lin. 1998. Dependency-based evaluation of MINIPAR. In Proceedings of the Workshop on the Evaluation of Parsing Systems, Granada, Spain. ELRA.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Computing word-pair antonymy",
"authors": [
{
"first": "Saif",
"middle": [],
"last": "Mohammad",
"suffix": ""
},
{
"first": "Bonnie",
"middle": [],
"last": "Dorr",
"suffix": ""
},
{
"first": "Graeme",
"middle": [],
"last": "Hirst",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "982--991",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Saif Mohammad, Bonnie Dorr, and Graeme Hirst. 2008. Computing word-pair antonymy. In Proceed- ings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 982-991, Honolulu, HI. ACL.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Learning for semantic parsing",
"authors": [
{
"first": "J",
"middle": [],
"last": "Raymond",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Mooney",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the Eighth International Conference on Computational Linguistics and Intelligent Text Processing",
"volume": "",
"issue": "",
"pages": "311--324",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Raymond J. Mooney. 2007. Learning for semantic parsing. In Proceedings of the Eighth International Conference on Computational Linguistics and Intel- ligent Text Processing, pages 311-324, Mexico City, Mexico. Springer.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Sound and efficient inference with probabilistic and deterministic dependencies",
"authors": [
{
"first": "Hoifung",
"middle": [],
"last": "Poon",
"suffix": ""
},
{
"first": "Pedro",
"middle": [],
"last": "Domingos",
"suffix": ""
}
],
"year": 2006,
"venue": "Proceedings of the Twenty First National Conference on Artificial Intelligence",
"volume": "",
"issue": "",
"pages": "458--463",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hoifung Poon and Pedro Domingos. 2006. Sound and efficient inference with probabilistic and determin- istic dependencies. In Proceedings of the Twenty First National Conference on Artificial Intelligence, pages 458-463, Boston, MA. AAAI Press.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Joint unsupervised coreference resolution with Markov logic",
"authors": [
{
"first": "Hoifung",
"middle": [],
"last": "Poon",
"suffix": ""
},
{
"first": "Pedro",
"middle": [],
"last": "Domingos",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "649--658",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hoifung Poon and Pedro Domingos. 2008. Joint unsu- pervised coreference resolution with Markov logic. In Proceedings of the 2008 Conference on Empiri- cal Methods in Natural Language Processing, pages 649-658, Honolulu, HI. ACL.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Markov logic networks",
"authors": [
{
"first": "M",
"middle": [],
"last": "Richardson",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Domingos",
"suffix": ""
}
],
"year": 2006,
"venue": "Machine Learning",
"volume": "62",
"issue": "",
"pages": "107--136",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M. Richardson and P. Domingos. 2006. Markov logic networks. Machine Learning, 62:107-136.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Unsupervised methods for determining object and relation synonyms on the web",
"authors": [
{
"first": "Alexander",
"middle": [],
"last": "Yates",
"suffix": ""
},
{
"first": "Oren",
"middle": [],
"last": "Etzioni",
"suffix": ""
}
],
"year": 2009,
"venue": "Journal of Artificial Intelligence Research",
"volume": "34",
"issue": "",
"pages": "255--296",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Alexander Yates and Oren Etzioni. 2009. Unsuper- vised methods for determining object and relation synonyms on the web. Journal of Artificial Intelli- gence Research, 34:255-296.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Learning to map sentences to logical form: Structured classification with probabilistic categorial grammers",
"authors": [
{
"first": "Luke",
"middle": [
"S"
],
"last": "Zettlemoyer",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Collins",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of the Twenty First Conference on Uncertainty in Artificial Intelligence",
"volume": "",
"issue": "",
"pages": "658--666",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Luke S. Zettlemoyer and Michael Collins. 2005. Learning to map sentences to logical form: Struc- tured classification with probabilistic categorial grammers. In Proceedings of the Twenty First Conference on Uncertainty in Artificial Intelligence, pages 658-666, Edinburgh, Scotland. AUAI Press.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Online learning of relaxed CCG grammars for parsing to logical form",
"authors": [
{
"first": "Luke",
"middle": [
"S"
],
"last": "Zettlemoyer",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Collins",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning",
"volume": "",
"issue": "",
"pages": "878--887",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Luke S. Zettlemoyer and Michael Collins. 2007. On- line learning of relaxed CCG grammars for parsing to logical form. In Proceedings of the Joint Con- ference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 878-887, Prague, Czech. ACL.",
"links": null
}
},
"ref_entries": {
"TABREF1": {
"html": null,
"type_str": "table",
"content": "<table><tr><td/><td colspan=\"3\"># Total # Correct Accuracy</td></tr><tr><td>KW</td><td>150</td><td>67</td><td>45%</td></tr><tr><td>KW-SYN</td><td>87</td><td>67</td><td>77%</td></tr><tr><td>TR-EXACT</td><td>29</td><td>23</td><td>79%</td></tr><tr><td>TR-SUB</td><td>152</td><td>81</td><td>53%</td></tr><tr><td>RS-EXACT</td><td>53</td><td>24</td><td>45%</td></tr><tr><td>RS-SUB</td><td>196</td><td>81</td><td>41%</td></tr><tr><td>DIRT</td><td>159</td><td>94</td><td>59%</td></tr><tr><td>USP</td><td>334</td><td>295</td><td>88%</td></tr></table>",
"text": "Comparison of question answering results on the GENIA dataset.",
"num": null
}
}
}
} |