File size: 92,000 Bytes
6fa4bc9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 |
{
"paper_id": "I11-1003",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:32:30.388309Z"
},
"title": "Learning Logical Structures of Paragraphs in Legal Articles",
"authors": [
{
"first": "Xuan",
"middle": [],
"last": "Ngo",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Japan Advanced Institute of Science and Technology",
"location": {
"addrLine": "1-1 Asahidai",
"postCode": "923-1292",
"settlement": "Nomi, Ishikawa",
"country": "Japan"
}
},
"email": ""
},
{
"first": "Nguyen",
"middle": [],
"last": "Bach",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Japan Advanced Institute of Science and Technology",
"location": {
"addrLine": "1-1 Asahidai",
"postCode": "923-1292",
"settlement": "Nomi, Ishikawa",
"country": "Japan"
}
},
"email": "bachnx@jaist.ac.jp"
},
{
"first": "Tran",
"middle": [
"Thi"
],
"last": "Le Minh",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Japan Advanced Institute of Science and Technology",
"location": {
"addrLine": "1-1 Asahidai",
"postCode": "923-1292",
"settlement": "Nomi, Ishikawa",
"country": "Japan"
}
},
"email": ""
},
{
"first": "Akira",
"middle": [],
"last": "Oanh",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Japan Advanced Institute of Science and Technology",
"location": {
"addrLine": "1-1 Asahidai",
"postCode": "923-1292",
"settlement": "Nomi, Ishikawa",
"country": "Japan"
}
},
"email": "oanhtt@jaist.ac.jp"
},
{
"first": "",
"middle": [],
"last": "Shimazu",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Japan Advanced Institute of Science and Technology",
"location": {
"addrLine": "1-1 Asahidai",
"postCode": "923-1292",
"settlement": "Nomi, Ishikawa",
"country": "Japan"
}
},
"email": "shimazu@jaist.ac.jp"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "This paper presents a new task, learning logical structures of paragraphs in legal articles, which is studied in research on Legal Engineering (Katayama, 2007). The goals of this task are recognizing logical parts of law sentences in a paragraph, and then grouping related logical parts into some logical structures of formulas, which describe logical relations between logical parts. We present a two-phase framework to learn logical structures of paragraphs in legal articles. In the first phase, we model the problem of recognizing logical parts in law sentences as a multi-layer sequence learning problem, and present a CRF-based model to recognize them. In the second phase, we propose a graph-based method to group logical parts into logical structures. We consider the problem of finding a subset of complete sub-graphs in a weighted-edge complete graph, where each node corresponds to a logical part, and a complete sub-graph corresponds to a logical structure. We also present an integer linear programming formulation for this optimization problem. Our models achieve 74.37% in recognizing logical parts, 79.59% in recognizing logical structures, and 55.73% in the whole task on the Japanese National Pension Law corpus.",
"pdf_parse": {
"paper_id": "I11-1003",
"_pdf_hash": "",
"abstract": [
{
"text": "This paper presents a new task, learning logical structures of paragraphs in legal articles, which is studied in research on Legal Engineering (Katayama, 2007). The goals of this task are recognizing logical parts of law sentences in a paragraph, and then grouping related logical parts into some logical structures of formulas, which describe logical relations between logical parts. We present a two-phase framework to learn logical structures of paragraphs in legal articles. In the first phase, we model the problem of recognizing logical parts in law sentences as a multi-layer sequence learning problem, and present a CRF-based model to recognize them. In the second phase, we propose a graph-based method to group logical parts into logical structures. We consider the problem of finding a subset of complete sub-graphs in a weighted-edge complete graph, where each node corresponds to a logical part, and a complete sub-graph corresponds to a logical structure. We also present an integer linear programming formulation for this optimization problem. Our models achieve 74.37% in recognizing logical parts, 79.59% in recognizing logical structures, and 55.73% in the whole task on the Japanese National Pension Law corpus.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Legal Engineering (Katayama, 2007) is a new research field which aims to achieve a trustworthy electronic society. Legal Engineering regards that laws are a kind of software for our society. Specifically, laws such as pension law are specifications for information systems such as pension systems.",
"cite_spans": [
{
"start": 18,
"end": 34,
"text": "(Katayama, 2007)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "To achieve a trustworthy society, laws need to be verified about their consistency and contradiction.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Legal texts have some specific characteristics that make them different from other kinds of documents. One of the most important characteristics is that legal texts usually have some specific structures at both sentence and paragraph levels. At the sentence level, a law sentence can roughly be divided into two logical parts: requisite part and effectuation part (Bach, 2011a; Bach et al., 2011b; Tanaka eta al., 1993) . At the paragraph level, a paragraph usually contains a main sentence 1 and one or more subordinate sentences (Takano et al., 2010) .",
"cite_spans": [
{
"start": 364,
"end": 377,
"text": "(Bach, 2011a;",
"ref_id": "BIBREF0"
},
{
"start": 378,
"end": 397,
"text": "Bach et al., 2011b;",
"ref_id": "BIBREF1"
},
{
"start": 398,
"end": 419,
"text": "Tanaka eta al., 1993)",
"ref_id": null
},
{
"start": 531,
"end": 552,
"text": "(Takano et al., 2010)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Analyzing logical structures of legal texts is an important task in Legal Engineering. The outputs of this task will be beneficial to people in understanding legal texts. They can easily understand 1) what does a law sentence say? 2) what cases in which the law sentence can be applied? and 3) what subjects are related to the provision described in the law sentence? This task is the preliminary step, which supports other tasks in legal text processing (translating legal articles into logical and formal representations, legal text summarization, legal text translation, question answering in legal domains, etc) and serves legal text verification, an important goal of Legal Engineering.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "There have been some studies analyzing logical structures of legal texts. (Bach et al., 2011b) presents the RRE task 2 , which recognizes the logical structure of law sentences. (Bach et al., 2010) describes an investigation on contributions of words to the RRE task. (Kimura et al., 2009) focuses on dealing with legal sentences including itemized and referential expressions. These works, however, only analyze logical structures of legal texts at the sentence level. At the paragraph level, (Takano et al., 2010) classifies a legal paragraph into one of six predefined categories: A, B, C, D, E, and F . Among six types, Type A, B, and C correspond to cases in which the main sentence is the first sentence, and subordinate sentences are other sentences. In paragraphs of Type D, E, and F , the main sentence is the first or the second sentence, and a subordinate sentence is an embedded sentence in parentheses within the main sentence.",
"cite_spans": [
{
"start": 74,
"end": 94,
"text": "(Bach et al., 2011b)",
"ref_id": "BIBREF1"
},
{
"start": 178,
"end": 197,
"text": "(Bach et al., 2010)",
"ref_id": "BIBREF2"
},
{
"start": 268,
"end": 289,
"text": "(Kimura et al., 2009)",
"ref_id": null
},
{
"start": 494,
"end": 515,
"text": "(Takano et al., 2010)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper, we present a task of learning logical structures of legal articles at the paragraph level. We propose a two-phase framework to complete the task. We also describe experimental results on real legal data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Our main contributions can be summarized in the following points:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Introducing a new task to legal text processing, learning logical structures of paragraphs in legal articles.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Presenting an annotated corpus for the task, the Japanese National Pension Law corpus.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Proposing a two-phase framework and providing solutions to solve the task.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "\u2022 Evaluating our framework on the real annotated corpus.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The rest of this paper is organized as follows. Section 2 describes our task and its two sub-tasks: recognition of logical parts and recognition of logical structures. In Section 3, we present our framework and proposed solutions. Experimental results on real legal articles are described in Section 4. Finally, Section 5 gives some conclusions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Learning logical structures of paragraphs in legal articles is the task of recognition of logical structures between logical parts in law sentences. A logical structure is usually formed from a pair of a requisite part and an effectuation part. These two parts are built from other kinds of logical parts such as topic parts, antecedent parts, consequent parts, and so on (Bach, 2011a; Bach et al., 2011b) 3 . Usually, consequent parts describes a law provision, antecedent parts describes cases in which the law provision can be applied, and topic parts describe subjects which are related to the law provision. In this paper, a logical structure can be defined as a set of some related logical parts. Figure 1 shows two cases of the inputs and outputs of the task. In the first case, the input is a paragraph of two sentences, and the outputs are four logical parts, which are grouped into two logical structures. In the second case, the input is a paragraph consisting of four sentences, and the outputs are four logical parts, which are grouped into three logical structures. An example in natural language 4 is presented in Figure 2 .",
"cite_spans": [
{
"start": 372,
"end": 385,
"text": "(Bach, 2011a;",
"ref_id": "BIBREF0"
},
{
"start": 386,
"end": 405,
"text": "Bach et al., 2011b)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [
{
"start": 703,
"end": 711,
"text": "Figure 1",
"ref_id": "FIGREF0"
},
{
"start": 1129,
"end": 1137,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Formulation",
"sec_num": "2"
},
{
"text": "Let s be a law sentence in the law sentence space S, then s can be represented by a sequence of words s = [w 1 w 2 . . . w n ]. A legal paragraph x in the legal paragraph space X is a sequence of law sentences",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "x = [s 1 s 2 . . . s l ], where s i \u2208 S, \u2200i = 1, 2, . . . , l.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "For each paragraph x, we denote a log-ical part p by a quad-tuple p = (b, e, k, c) where b, e, and k are three integers which indicate position of the beginning word, position of the end word, and sentence position of p, and c is a logical part category in the set of predefined categories C. Formally, the set P of all possible logical parts defined in a paragraph x can be described as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "P = {(b, e, k, c)|1 \u2264 k \u2264 l, 1 \u2264 b \u2264 e \u2264 len(k), c \u2208 C}.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "In the above definition, l is the number of sentences in the paragraph x, and len(k) is the length of the k th sentence.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "In this sub-task, we want to recognize some non-overlapping (but possibly embedded) logical parts in an input paragraph. A solution for this task is a subset y \u2286 P which does not violate the overlapping relationship. We say that two logical parts p 1 and p 2 are overlapping if and only if they are in the same sentence (k 1 = k 2 ) and b 1 < b 2 \u2264 e 1 < e 2 or b 2 < b 1 \u2264 e 2 < e 1 . We denote the overlapping relationship by \u223c. We also say that p 1 is embedded in p 2 if and only if they are in the same sentence (k 1 = k 2 ) and b 2 \u2264 b 1 \u2264 e 1 \u2264 e 2 , and denote the embedded relationship by \u227a. Formally, the solution space can be described as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "Y = {y \u2286 P |\u2200u, v \u2208 y, u \u223c v}.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "The learning problem in this sub-task is to learn a function R : X \u2192 Y from a set of m training samples",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "{(x i , y i )|x i \u2208 X, y i \u2208 Y, \u2200i = 1, 2, . . . , m}.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "In our task, we consider the following types of logical parts:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "1. An antecedent part is denoted by A 2. A consequent part is denoted by C 3. A topic part which depends on the antecedent part is denoted by T 1 4. A topic part which depends on the consequent part is denoted by T 2 5. A topic part which depends on both the antecedent part and the consequent part is denoted by T 3 6. The left part of an equivalent statement is denoted by EL 7. The right part of an equivalent statement is denoted by ER 8. An object part, whose meaning is defined differently in different cases, is denoted by Ob 9. An original replacement part, which will be replaced by other replacement parts (denoted by RepR) in specific cases, is denoted by RepO.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "Compared with previous works (Bach et al., 2011b) , we introduce three new kinds of logical parts: Ob, RepO, and RepR.",
"cite_spans": [
{
"start": 29,
"end": 49,
"text": "(Bach et al., 2011b)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 1: Recognition of Logical Parts",
"sec_num": "2.1"
},
{
"text": "In the second sub-task, the goal is to recognize a set of logical structures given a set of logical parts. Let G =< V, E > be a complete undirected graph with the vertex set V and the edge set E. A real value function f is defined on E as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 2: Recognition of Logical Structures",
"sec_num": "2.2"
},
{
"text": "f : E \u2192 R, e \u2208 E \u2192 f (e) \u2208 R. In this sub-task, each vertex of the graph corresponds to a logical part, and a complete sub-graph corresponds to a logical structure. The value on an edge connecting two vertices expresses the degree that the two vertices belong to one logical structure. The positive (negative) value means that two vertices are likely (not likely) to belong to one logical structure.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 2: Recognition of Logical Structures",
"sec_num": "2.2"
},
{
"text": "Let G s be a complete sub-graph of G, then v(G s ) and e(G s ) are the set of vertices and the set of edges of G s , respectively. We define the total value of a sub-graph as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 2: Recognition of Logical Structures",
"sec_num": "2.2"
},
{
"text": "f (G s ) = f (e(G s )) = e\u2208e(Gs) f (e).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 2: Recognition of Logical Structures",
"sec_num": "2.2"
},
{
"text": "Let \u2126 be the set of all complete sub-graphs of G. The problem becomes determining a subset \u03a8 \u2286 \u2126 that satisfies the following constraints:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 2: Recognition of Logical Structures",
"sec_num": "2.2"
},
{
"text": "1. \u2200g \u2208 \u03a8, |v(g)| \u2265 2, 2. \u222a g\u2208\u03a8 v(g) = V , 3. \u2200g 1 , g 2 \u2208 \u03a8|v(g 1 ) \u2286 v(g 2 ) \u21d2 v(g 1 ) = v(g 2 ), 4. \u2200g \u2208 \u03a8, \u222a h\u2208\u03a8,h =g v(h) = V , and",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 2: Recognition of Logical Structures",
"sec_num": "2.2"
},
{
"text": "5. g\u2208\u03a8 f (g) \u2192 maximize. Constraint 1), minimal constraint, says that each logical structure must contain at least two logical parts. There is the case that a logical structure contains only a consequent part. Due to the characteristics of Japanese law sentences, however, our corpus does not contain such cases. A logical structure which contains a consequent part will also contain a topic part or an antecedent part or both of them. So a logical structure contains at least two logical parts. Constraint 2), complete constraint, says that each logical part must belong to at least one logical structure. Constraint 3), maximal constraint, says that we cannot have two different logical structures such that the set of logical parts in one logical structure contains the set of logical parts in the other logical structure. Constraint 4), significant constraint, says that if we remove any logical structure from the solution, Constraint 2) will be violated. Although Constraint 3) is guaranteed by Constraint 4), we introduce it because of its importance.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 2: Recognition of Logical Structures",
"sec_num": "2.2"
},
{
"text": "3 Proposed Solutions",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Sub-Task 2: Recognition of Logical Structures",
"sec_num": "2.2"
},
{
"text": "This sub-section presents our model for recognizing logical parts. We consider the recognition problem as a multi-layer sequence learning problem. First, we give some related notions. Let s be a law sentence, and P be the set of logical parts of s, P = {p 1 , p 2 , . . . , p m }. Layer 1 (s) (outer most layer) is defined as a set of logical parts in P , which are not embedded in any other part. Layer i (s) is defined as a set of logical parts in P \\ \u222a i\u22121 k=1 Layer k (s), which are not embedded in any other part in P \\\u222a i\u22121 k=1 Layer k (s). Formally, we have:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-layer Sequence Learning for Logical Part Recognition",
"sec_num": "3.1"
},
{
"text": "Layer 1 (s) = {p|p \u2208 P, p \u227a q, \u2200q \u2208 P, q = p}.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-layer Sequence Learning for Logical Part Recognition",
"sec_num": "3.1"
},
{
"text": "Layer Figure 3 illustrates a law sentence with four logical parts in three layers: Part 1 and Part 2 in Layer 1 , Part 3 in Layer 2 , and Part 4 in Layer 3 . Let K be the number of layers in a law sentence s, our model will recognize logical parts in K steps. In the k th step we recognize logical parts in Layer k . In each layer, we model the recognition problem as a sequence labeling task in which each word is an element. Logical parts in Layer i\u22121 will be used as input sequence in the i th step (in the first step, we use original sentence as input). Figure 4 gives an example of labeling for an input sentence. The sentence consists of three logical parts in two layers. In our model, we use IOE tag setting: the last element of a part is tagged with E, the other elements of a part are tagged with I, and an element not included in any part is tagged with O.",
"cite_spans": [],
"ref_spans": [
{
"start": 6,
"end": 14,
"text": "Figure 3",
"ref_id": "FIGREF2"
},
{
"start": 558,
"end": 566,
"text": "Figure 4",
"ref_id": "FIGREF3"
}
],
"eq_spans": [],
"section": "Multi-layer Sequence Learning for Logical Part Recognition",
"sec_num": "3.1"
},
{
"text": "i (s) = {p|p \u2208 Q i , p \u227a q, \u2200q \u2208 Q i , q = p}, where Q i = P \\ \u222a i\u22121 k=1 Layer k (s)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-layer Sequence Learning for Logical Part Recognition",
"sec_num": "3.1"
},
{
"text": "Let K * be the maximum number of layers in all law sentences in training data. We learn K * models, in which the k th model is learned from logical parts in the Layer k of training data, using Conditional random fields (Lafferty et al., 2001; Kudo, CRF toolkit) . In the testing phase, we first apply the first model to the input law sentence, and then apply the i th model to the predicted logical parts in Layer i\u22121 .",
"cite_spans": [
{
"start": 219,
"end": 242,
"text": "(Lafferty et al., 2001;",
"ref_id": "BIBREF12"
},
{
"start": 243,
"end": 261,
"text": "Kudo, CRF toolkit)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-layer Sequence Learning for Logical Part Recognition",
"sec_num": "3.1"
},
{
"text": "Suppose that G is a sub-graph of G such that G contains all the vertices of G and the degree of each vertex in G is greater than zero, then the set of all the maximal complete sub-graphs (or cliques) of G will satisfy all the minimal, complete, maximal, and significant constraints. We also note that, a set of cliques that satisfies all these four constraints will form a sub-graph that has two properties like properties of G .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "Let \u039b be the set of all such sub-graphs G of G, the sub-task now consists of two steps:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "1. Finding G = argmax G \u2208\u039b f (G ), and 2. Finding all cliques of G .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "Each clique found in the second step will correspond to a logical structure.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "Recently, some researches have shown that integer linear programming (ILP) formulations is an effective way to solve many NLP problems such as semantic role labeling (Punyakanok, 2004) , coreference resolution (Denis and Baldridge, 2007) , summarization (Clarke and Lapata, 2008) , dependency parsing (Martins et al., 2009) , and so on. The advantage of ILP formulations is that we can incorporate non-local features or global constraints easily, which are difficult in traditional algorithms. Although solving an ILP is NP-hard in general, some fast algorithms and available tools 5 make it a practical solution for many NLP problems (Martins et al., 2009) .",
"cite_spans": [
{
"start": 166,
"end": 184,
"text": "(Punyakanok, 2004)",
"ref_id": "BIBREF15"
},
{
"start": 210,
"end": 237,
"text": "(Denis and Baldridge, 2007)",
"ref_id": "BIBREF7"
},
{
"start": 254,
"end": 279,
"text": "(Clarke and Lapata, 2008)",
"ref_id": "BIBREF6"
},
{
"start": 301,
"end": 323,
"text": "(Martins et al., 2009)",
"ref_id": "BIBREF13"
},
{
"start": 635,
"end": 657,
"text": "(Martins et al., 2009)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "In this work, we exploit ILP to solve the first step. Let N be the number of vertices of G, we introduce a set of integer variables {x ij } 1\u2264i<j\u2264N . The values of {x ij } are set as follows. If (i, j) \u2208 e(G ) then x ij = 1, otherwise x ij = 0. ILP formulations for the first step can be described as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "//--------Objective function --------// M aximize : 1\u2264i<j\u2264N f (i, j) * x ij (1) //----------Constraints ----------// Integer : {x ij } 1\u2264i<j\u2264N . (2) 0 \u2264 x ij \u2264 1, (1 \u2264 i < j \u2264 N ). (3) j\u22121 i=1 x ij + N k=j+1 x jk \u2265 1, (1 \u2264 j \u2264 N ). (4)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "The last constraint guarantees that there is at least one edge connecting to each vertex in G .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "The second step, finding all cliques of an undirected graph, is a famous problem in graph theory. Many algorithms have been proposed to solve this problem efficiently. In this work, we exploit the Bron-Kerbosch algorithm, a backtracking algorithm. The main idea of the Bron-Kerbosch algorithm is using a branch-and-bound technique to stop searching on branches that cannot lead to a clique (Bron and Kerbosch, 1973) .",
"cite_spans": [
{
"start": 390,
"end": 415,
"text": "(Bron and Kerbosch, 1973)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "The remaining problem is how to define the value function f . Our solution is that, first we learn a binary classifier C using maximum entropy model. This classifier takes a pair of logical parts as the input, and outputs +1 if two logical parts belong to one logical structure, otherwise it will output \u22121. Then, we define the value function f for two logical parts as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "f (p 1 , p 2 ) = P rob(C(p 1 , p 2 ) = +1) \u2212 0.5. Function f will receive a value from \u22120.5 to +0.5, and it equals to zero in the case that the classifier assigns the same probability to +1 and \u22121.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "ILP for Recognizing Logical Structures",
"sec_num": "3.2"
},
{
"text": "We have built a corpus, Japanese National Pension Law (JNPL) corpus, which consists of 83 legal articles 6 of Japanese national pension law. The architecture of JNPL is shown in Figure 5 . The law 6 Because building corpus is an expensive and timeconsuming task, we only annotate a part of JNPL. consists of articles, articles consist of paragraphs, and paragraphs contain sentences. A sentence may belong to items, sub-items, or sub-sub-items of a paragraph. Figure 6 illustrates the relationship between a law sentence and logical parts. A law sentence may contain some logical parts, and a logical part may be embedded in another one. In our corpus, a logical part is annotated with information about its type (kind of part) and formula-id (logical parts with the same id will be- Figure 7 : An annotated sentence in the JNPL corpus. The sentence contains two logical structures with four logical parts. long to one logical structure). An example of annotated sentence in the JNPL corpus is shown in Figure 7 .",
"cite_spans": [
{
"start": 197,
"end": 198,
"text": "6",
"ref_id": null
}
],
"ref_spans": [
{
"start": 178,
"end": 186,
"text": "Figure 5",
"ref_id": "FIGREF4"
},
{
"start": 460,
"end": 468,
"text": "Figure 6",
"ref_id": "FIGREF5"
},
{
"start": 784,
"end": 792,
"text": "Figure 7",
"ref_id": null
},
{
"start": 1003,
"end": 1011,
"text": "Figure 7",
"ref_id": null
}
],
"eq_spans": [],
"section": "Corpus",
"sec_num": "4.1"
},
{
"text": "We employed two people in a data-making company, who analyzed and annotated our corpus. The corpus consists of 83 legal articles, which contain 119 paragraphs with 426 sentences. On average, each paragraph consists of 3.6 sentences. The total number of logical parts is 807, and the number of logical structures is 351. On average, each paragraph consists of 6.8 logical parts and 3 logical structures. Table 1 shows some statistics on the number of logical parts of each type. Main types of parts are A(35.4%), C(30.7%), T 2 (14.1%), ER(7.1%), and EL(6.8%). Five main types of parts make up more than 94% of all types.",
"cite_spans": [],
"ref_spans": [
{
"start": 403,
"end": 410,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "Corpus",
"sec_num": "4.1"
},
{
"text": "We divided the JNLP corpus into 10 sets, and conducted 10-fold cross-validation tests. For the first sub-task, we evaluated the performance of our system by precision, recall, and F 1 scores as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Methods",
"sec_num": "4.2"
},
{
"text": "precision = |correct parts| |predicted parts| , recall = |correct parts| |actual parts| , F1 = 2 * precision * recall precision+recall .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Methods",
"sec_num": "4.2"
},
{
"text": "For the second sub-task, we used MUC precision, recall, and F 1 scores as described in (Vilain et al., 1995) . We summarize them here for clarity.",
"cite_spans": [
{
"start": 87,
"end": 108,
"text": "(Vilain et al., 1995)",
"ref_id": "BIBREF19"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Methods",
"sec_num": "4.2"
},
{
"text": "Let P 1 , P 2 , . . . , P n be n predicted logical structures, and G 1 , G 2 , . . . , G m be the correct answers or gold logical structures. To calculate recall, for each gold logical structure G i (i = 1, 2, . . . , m), let k(G i ) be the smallest number such that there exist k(G i ) predicted structures",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Methods",
"sec_num": "4.2"
},
{
"text": "P i 1 , P i 2 , . . . , P i k(G i ) which satisfy G i \u2286 \u222a k(G i ) j=1 P i j : recall = m i=1 (|G i |\u2212k(G i )) m i=1 (|G i |\u22121)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Methods",
"sec_num": "4.2"
},
{
"text": ". To calculate precision, we switch the roles of predicted structures and gold structures. Finally, F 1 score is computed in a similar manner as in the first sub-task. We chose the Filter-Ranking (FR) Perceptron algorithm proposed by (Carreras and Marquez, 2005; Carreras et al., 2002) as our baseline model because of its effectiveness on phrase recognition problems, especially on problems that accept the embedded relationship 7 . We use FR-perceptron algorithm to recognize logical parts in law sentences one by one in an input paragraph. For beginning/end predictors, we got features of words, POS tags, and Bunsetsu 8 tags in a window size 2. Moreover, with beginning predictor, we used a feature for checking whether this position is the beginning of the sentence or not. Similarly, with end predictor, we use a feature for checking whether this position is the end of the sentence or not.",
"cite_spans": [
{
"start": 234,
"end": 262,
"text": "(Carreras and Marquez, 2005;",
"ref_id": "BIBREF4"
},
{
"start": 263,
"end": 285,
"text": "Carreras et al., 2002)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Methods",
"sec_num": "4.2"
},
{
"text": "With each logical part candidate, we extract following kinds of features:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments on",
"sec_num": "4.3"
},
{
"text": "1. Length of the part 2. Internal structure: this feature is the concatenation of the top logical parts, punctuation marks, parenthesis, and quotes inside the candidate. An example about internal structure may be (A+, +C + .) (plus is used to concatenate items)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments on",
"sec_num": "4.3"
},
{
"text": "3. Word (POS) uni-gram, word (POS) bi-gram, and word (POS) tri-gram.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments on",
"sec_num": "4.3"
},
{
"text": "In our experiments, we focus on paragraphs in Type A, B, and C defined in (Takano et al., 2010) . In these types, the first sentence is the main sentence, which usually contains more logical parts than other sentences. The other sentences often have a few logical parts, and in most cases these logical parts only appear in one layer. The first sentences usually contain logical parts in two layers.",
"cite_spans": [
{
"start": 74,
"end": 95,
"text": "(Takano et al., 2010)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.3.2"
},
{
"text": "We divided sentences into two groups. The first group consists of the first sentences in paragraphs, and the second group consists of other sentences. We set the number of layers k to 2 for sentences in the first group, and to 1 for sentences in the second group. To learn sequence labeling models, we used CRFs (Lafferty et al., 2001; Kudo, CRF toolkit) .",
"cite_spans": [
{
"start": 312,
"end": 335,
"text": "(Lafferty et al., 2001;",
"ref_id": "BIBREF12"
},
{
"start": 336,
"end": 354,
"text": "Kudo, CRF toolkit)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.3.2"
},
{
"text": "Experimental results on the JNPL corpus are described in Table 2 . We conducted experiments with four feature sets: words; words and POS tags; words and Bunsetsu tags; and words, POS tags, and Bunsetsu tags. To extract features from source sentences, we used the CaboCha tool (Kudo, Cabocha) , a Japanese morphological and syntactic analyzer. The best model (word and Bunsetsu tag features) achieved 74.37% in F 1 score. It improves 11.04% in F 1 score (30.11% in error rate) compared with the baseline model. Table 3 shows experimental results of our best model in more detail. Our model got good results on most main parts: C(78.98%), A(80.42%), and T 2 (82.14%). The model got low results on the other types of parts. It is understandable because three types of logical parts C, A, and T 2 make up more than 80%, while six other types only make up 20% of all types.",
"cite_spans": [
{
"start": 276,
"end": 291,
"text": "(Kudo, Cabocha)",
"ref_id": null
}
],
"ref_spans": [
{
"start": 57,
"end": 64,
"text": "Table 2",
"ref_id": "TABREF1"
},
{
"start": 510,
"end": 517,
"text": "Table 3",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.3.2"
},
{
"text": "Our baseline is a heuristic algorithm to solve this sub-task on graphs. This is an approximate algorithm which satisfies minimal, complete, maximal, and significant constraints. The main idea of our algorithm is picking up as many positive edges as possible, and as few negative edges as possible. We consider two cases: 1) There is no positive value edge on the input graph; and 2) There are some positive value edges on the input graph.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Baseline: a Heuristic Algorithm",
"sec_num": "4.4.1"
},
{
"text": "In the first case, because all the edges have negative values, we build logical structures with as few logical parts as possible. In this case, each logical structure contains exactly two logical parts. So we gradually choose two nodes in the graph with the maximum value on the edge connecting them. An example of the first case is illustrated in Figure 8 . The maximum value on an edge is \u22120.1, so the first logical structure will contain node 1 and node 3. The second logical structure contains node 2 and node 4 9 . In the second case, we first consider the subgraph which only contains non-negative value edges. In this sub-graph, we repeatedly build logical structures with as many logical parts as possi- ble. After building successfully a logical structure, we remove all the nodes and the edges according to it on the graph. When have no positive edge, we will build logical structures with exactly two logical parts.",
"cite_spans": [],
"ref_spans": [
{
"start": 348,
"end": 356,
"text": "Figure 8",
"ref_id": "FIGREF7"
}
],
"eq_spans": [],
"section": "Baseline: a Heuristic Algorithm",
"sec_num": "4.4.1"
},
{
"text": "An example of the second case is illustrated in Figure 9 . First, we consider the subgraph with positive edges. This sub-graph consists of five nodes {1, 2, 3, 4, 5} and four edges {(1, 2), (1, 3), (2, 3), (2, 4)}. First, we have a logical structure with three nodes {1, 2, 3}. We remove these nodes and the positive edges connecting to these nodes. We have two nodes {4, 5} with no positive edges. Now we build logical structures with exactly two nodes. We consider node 4. Among edges connecting to node 4, the edge (2, 4) has maximal value. So we have the second logical structure with two nodes {2, 4}. Next, we consider node 5, and we have the third logical structure with two nodes {1, 5}.",
"cite_spans": [],
"ref_spans": [
{
"start": 48,
"end": 56,
"text": "Figure 9",
"ref_id": "FIGREF8"
}
],
"eq_spans": [],
"section": "Baseline: a Heuristic Algorithm",
"sec_num": "4.4.1"
},
{
"text": "In our experiments, to learn a maximum entropy binary classification we used the implementation of Tsuruoka (Tsuruoka, MEM) . With a pair of logical parts, we extracted the following features (and combinations of them):",
"cite_spans": [
{
"start": 108,
"end": 123,
"text": "(Tsuruoka, MEM)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.4.2"
},
{
"text": "\u2022 Categories of two parts.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.4.2"
},
{
"text": "\u2022 Layers of two parts.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.4.2"
},
{
"text": "\u2022 The positions of the sentences that contain two parts (the first sentence or not).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.4.2"
},
{
"text": "\u2022 Categories of other parts in the input paragraph.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.4.2"
},
{
"text": "We conducted experiments on this sub-task in two settings. In the first setting, we used annotated logical parts (gold inputs) as the inputs to the system. The purpose of this experiment is to evaluate the performance of the graph-based method on Sub-task 2. In the second setting, predicted logical parts (end-to-end) outputted by the Sub-task 1 were used as the inputs to the system. The purpose of this experiment is to evaluate the performance of our framework on the whole task.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.4.2"
},
{
"text": "In the second setting, end-to-end setting, because input logical parts may differ from the correct logical parts, we need to modify the MUC scores. Let P 1 , P 2 , . . . , P n be n predicted logical structures, and G 1 , G 2 , . . . , G m be the gold logical structures. For each gold logical structure G i (i = 1, 2, . . . , m), let D i be the set of logical parts in G i which are not included in the set of input logical parts. D i = {p \u2208 G i |p / \u2208 \u222a n j=1 P j }. Let k(G i ) be the smallest number such that there exist k(G i ) predicted structures P i 1 , P i 2 , . . . , P i k(G i ) which satisfy",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.4.2"
},
{
"text": "G i \u2286 (\u222a k(G i ) j=1 P i j ) \u222a D i . recall = m i=1 (|G i |\u2212|D i |\u2212k(G i )) m i=1 (|G i |\u22121)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.4.2"
},
{
"text": ". To calculate the precision, we switch the roles of predicted structures and gold structures. Table 4 shows experimental results on the second sub-task. The ILP model outperformed the baseline model in both settings. It improved 3.70% in the F 1 score (15.35% in error rate) in the gold-input setting, and 4.61% in the F 1 score (9.43% in error rate) in the end-to-end setting compared with the baseline model (heuristic algorithm).",
"cite_spans": [],
"ref_spans": [
{
"start": 95,
"end": 102,
"text": "Table 4",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.4.2"
},
{
"text": "We have introduced the task of learning logical structures of paragraphs in legal articles, a new task which has been studied in research on Legal Engineering. We presented the Japanese National Pension Law corpus, an annotated corpus of real legal articles for the task. We also described a two-phase framework with multi-layer sequence learning model and ILP formulation to complete the task. Our results provide a baseline for further researches on this interesting task.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "In the future, we will continue to improve this task. On the other hand, we also investigate the task of translating legal articles into logical and formal representations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "Usually, the first sentence is the main sentence.2 The task of Recognition of Requisite part and Effectuation part in law sentences.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "We only recognize logical structures (a set of related logical parts). The task of translating legal articles into logical and formal representations is not covered in this paper.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Because law sentences are very long and complicated, we use toy sentences to illustrate the task.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "We used lp-solve from http://lpsolve.sourceforge.net/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "We re-implement the FR-perceptron algorithm by ourself.8 In Japanese, a Bunsetsu is an unit of sentence which is similar to a chunk in English.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "If the number of nodes is odd, the final logical structure will consist of the final node and another node, so that the edge connecting them has the maximal value.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "This work was partly supported by the 21st Century COE program 'Verifiable and Evolvable e-Society', Grant-in-Aid for Scientific Research, Education and Research Center for Trustworthy e-Society, and JAIST Overseas Training Program for 3D Program Students.We would like to give special thanks to Kenji Takano and Yoshiko Oyama, who analyzed law sentences and built the corpus, and the reviewers, who gave us valuable comments.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgments",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "A Study on Recognition of Requisite Part and Effectuation Part in Law Sentences",
"authors": [
{
"first": "N",
"middle": [
"X"
],
"last": "Bach",
"suffix": ""
}
],
"year": 2011,
"venue": "School of Information Science",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "N.X. Bach. 2011a. A Study on Recognition of Requisite Part and Effectuation Part in Law Sen- tences. Master Thesis, School of Information Sci- ence, Japan Advanced Institute of Science and Tech- nology.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "RRE Task: The Task of Recognition of Requisite Part and Effectuation Part in Law Sentences",
"authors": [
{
"first": "N",
"middle": [
"X"
],
"last": "Bach",
"suffix": ""
},
{
"first": "N",
"middle": [
"L"
],
"last": "Minh",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Shimazu",
"suffix": ""
}
],
"year": 2011,
"venue": "In International Journal of Computer Processing Of Languages (IJCPOL)",
"volume": "23",
"issue": "2",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "N.X. Bach, N.L. Minh, A. Shimazu. 2011b. RRE Task: The Task of Recognition of Requisite Part and Effectuation Part in Law Sentences. In Inter- national Journal of Computer Processing Of Lan- guages (IJCPOL), Volume 23, Number 2.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Exploring Contributions of Words to Recognition of Requisite Part and Effectuation Part in Law Sentences",
"authors": [
{
"first": "N",
"middle": [
"X"
],
"last": "Bach",
"suffix": ""
},
{
"first": "N",
"middle": [
"L"
],
"last": "Minh",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Shimazu",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of JURISIN",
"volume": "",
"issue": "",
"pages": "121--132",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "N.X. Bach, N.L. Minh, A. Shimazu. 2010. Exploring Contributions of Words to Recognition of Requisite Part and Effectuation Part in Law Sentences. In Pro- ceedings of JURISIN, pp. 121-132.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Algorithm 457: Finding All Cliques of an Undirected Graph",
"authors": [
{
"first": "C",
"middle": [],
"last": "Bron",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Kerbosch",
"suffix": ""
}
],
"year": 1973,
"venue": "Communications of the ACM",
"volume": "16",
"issue": "",
"pages": "575--577",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "C. Bron and J. Kerbosch. 1973. Algorithm 457: Find- ing All Cliques of an Undirected Graph. In Commu- nications of the ACM, Volume 16, Issue 9, pp. 575- 577.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Filtering-Ranking Perceptron Learning for Partial Parsing",
"authors": [
{
"first": "X",
"middle": [],
"last": "Carreras",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Marquez",
"suffix": ""
}
],
"year": 2005,
"venue": "Machine Learning",
"volume": "60",
"issue": "",
"pages": "41--71",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "X. Carreras and L. Marquez. 2005. Filtering-Ranking Perceptron Learning for Partial Parsing. In Machine Learning, Volume 60, Issue 1-3, pp. 41-71.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Learning and Inference for Clause Identification",
"authors": [
{
"first": "X",
"middle": [],
"last": "Carreras",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Marquez",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Punyakanok",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Roth",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of ECML",
"volume": "",
"issue": "",
"pages": "35--47",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "X. Carreras, L. Marquez, V. Punyakanok, D. Roth. 2002. Learning and Inference for Clause Identifi- cation. In Proceedings of ECML, pp. 35-47.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Global Inference for Sentence Compression: An Integer Linear Programming Approach",
"authors": [
{
"first": "J",
"middle": [],
"last": "Clarke",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Lapata",
"suffix": ""
}
],
"year": 2008,
"venue": "In Journal of Artificial Intelligence Research",
"volume": "31",
"issue": "",
"pages": "399--429",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Clarke and M. Lapata. 2008. Global Inference for Sentence Compression: An Integer Linear Program- ming Approach. In Journal of Artificial Intelligence Research (JAIR), Volume 31, pp. 399-429.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Joint Determination of Anaphoricity and Coreference Resolution Using Integer Programming",
"authors": [
{
"first": "P",
"middle": [],
"last": "Denis",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Baldridge",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of HLT-NAACL",
"volume": "",
"issue": "",
"pages": "236--243",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "P. Denis and J. Baldridge. 2007. Joint Determination of Anaphoricity and Coreference Resolution Us- ing Integer Programming. In Proceedings of HLT- NAACL, pp. 236-243.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Legal Engineering -An Engineering Approach to Laws in e-Society Age",
"authors": [
{
"first": "T",
"middle": [],
"last": "Katayama",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of JURISIN",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T. Katayama. 2007. Legal Engineering -An Engineer- ing Approach to Laws in e-Society Age. In Proceed- ings of JURISIN.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Treatment of Legal Sentences Including Itemized and Referential Expressions -Towards Translation into Logical Forms",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Kimura",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Nakamura",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Shimazu",
"suffix": ""
}
],
"year": null,
"venue": "New Frontiers in Artificial Intelligence",
"volume": "",
"issue": "",
"pages": "242--253",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Y. Kimura, M. Nakamura, A. Shimazu. Treatment of Legal Sentences Including Itemized and Referen- tial Expressions -Towards Translation into Logical Forms. New Frontiers in Artificial Intelligence, vol- ume 5447 of LNAI, pp.242-253.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Yet Another Japanese Dependency Structure Analyzer",
"authors": [
{
"first": "T",
"middle": [],
"last": "Kudo",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T. Kudo. Yet Another Japanese Depen- dency Structure Analyzer. http://chasen.org/",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "CRF++: Yet Another CRF toolkit",
"authors": [
{
"first": "T",
"middle": [],
"last": "Kudo",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T. Kudo. CRF++: Yet Another CRF toolkit. http://crfpp.sourceforge.net/.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data",
"authors": [
{
"first": "J",
"middle": [],
"last": "Lafferty",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Mccallum",
"suffix": ""
},
{
"first": "F",
"middle": [],
"last": "Pereira",
"suffix": ""
}
],
"year": 2001,
"venue": "Proceedings of ICML",
"volume": "",
"issue": "",
"pages": "282--289",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Lafferty, A. McCallum, F. Pereira. 2001. Condi- tional Random Fields: Probabilistic Models for Seg- menting and Labeling Sequence Data. In Proceed- ings of ICML, pp.282-289.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Concise Integer Linear Programming Formulations for Dependency Parsing",
"authors": [
{
"first": "A",
"middle": [
"F T"
],
"last": "Martins",
"suffix": ""
},
{
"first": "N",
"middle": [
"A"
],
"last": "Smith",
"suffix": ""
},
{
"first": "E",
"middle": [
"P"
],
"last": "Xing",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "342--350",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A.F.T. Martins, N.A. Smith, E.P. Xing. 2009. Concise Integer Linear Programming Formulations for De- pendency Parsing. In Proceedings of ACL, pp.342- 350.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Towards Translation of Legal Sentences into Logical Forms",
"authors": [
{
"first": "M",
"middle": [],
"last": "Nakamura",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Nobuoka",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Shimazu",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of JURISIN",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M. Nakamura, S. Nobuoka, A. Shimazu. 2007. To- wards Translation of Legal Sentences into Logical Forms. In Proceedings of JURISIN.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Semantic Role Labeling Via Integer Linear Programming Inference",
"authors": [
{
"first": "V",
"middle": [],
"last": "Punyakanok",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Roth",
"suffix": ""
},
{
"first": "W",
"middle": [],
"last": "Yih",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Zimak",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of COLING",
"volume": "",
"issue": "",
"pages": "1346--1352",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "V. Punyakanok, D. Roth, W. Yih, D. Zimak. 2004. Se- mantic Role Labeling Via Integer Linear Program- ming Inference. In Proceedings of COLING, pp. 1346-1352.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Semantic Analysis of Paragraphs Consisting of Multiple Sentences -Towards Development of a Logical Formulation System",
"authors": [
{
"first": "K",
"middle": [],
"last": "Takano",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Nakamura",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Oyama",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Shimazu",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of JU-RIX",
"volume": "",
"issue": "",
"pages": "117--126",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "K. Takano, M. Nakamura, Y. Oyama, A. Shimazu. 2010. Semantic Analysis of Paragraphs Consisting of Multiple Sentences -Towards Development of a Logical Formulation System. In Proceedings of JU- RIX, pp. 117-126.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Standard Structure of Legal Provisions -for the Legal Knowledge Processing by Natural Language",
"authors": [
{
"first": "K",
"middle": [],
"last": "Tanaka",
"suffix": ""
},
{
"first": "I",
"middle": [],
"last": "Kawazoe",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Narita",
"suffix": ""
}
],
"year": 1993,
"venue": "IPSJ Research Report on Natural Language Processing",
"volume": "",
"issue": "",
"pages": "79--86",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "K. Tanaka, I. Kawazoe, H. Narita. 1993 Stan- dard Structure of Legal Provisions -for the Legal Knowledge Processing by Natural Language -(in Japanese). In IPSJ Research Report on Natural Lan- guage Processing, pp. 79-86.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "A simple C++ library for maximum entropy classification",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Tsuruoka",
"suffix": ""
}
],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Y. Tsuruoka. A simple C++ library for maxi- mum entropy classification. http://www-tsujii.is.s.u- tokyo.ac.jp/ tsuruoka/maxent/.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "A Model-Theoretic Coreference Scoring Scheme",
"authors": [
{
"first": "M",
"middle": [],
"last": "Vilain",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Burger",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Aberdeen",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Connolly",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Hirschman",
"suffix": ""
}
],
"year": 1995,
"venue": "Proceedings of MUC-6",
"volume": "",
"issue": "",
"pages": "45--52",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M. Vilain, J. Burger, J. Aberdeen, D. Connolly, L. Hirschman. 1995. A Model-Theoretic Coreference Scoring Scheme. In Proceedings of MUC-6, pp. 45- 52.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "Two cases of inputs and outputs of the task."
},
"FIGREF1": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "An example in natural language (E means Effectuation part, R means Requisite part, and LS means Logical Structure)."
},
"FIGREF2": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "A law sentence with logical parts in three layers."
},
"FIGREF3": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "An example of labeling in the multilayer model."
},
"FIGREF4": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "The architecture of JNPL."
},
"FIGREF5": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "Relationship between a sentence and logical parts."
},
"FIGREF7": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "An example of the first case."
},
"FIGREF8": {
"type_str": "figure",
"num": null,
"uris": null,
"text": "An example of the second case."
},
"TABREF0": {
"html": null,
"type_str": "table",
"content": "<table><tr><td>Logical Part</td><td>C</td><td>A</td><td colspan=\"3\">T 1 T 2 T 3 EL ER Ob RepO RepR</td></tr><tr><td>Number</td><td colspan=\"3\">248 286 0 114 12 55 57</td><td>9</td><td>12</td><td>14</td></tr></table>",
"text": "Statistics on logical parts of the JNPL corpus",
"num": null
},
"TABREF1": {
"html": null,
"type_str": "table",
"content": "<table><tr><td colspan=\"4\">: Experimental results for Sub-task 1 on the</td></tr><tr><td colspan=\"4\">JNLP corpus(W:Word; P: POS tag; B: Bunsetsu</td></tr><tr><td>tag)</td><td/><td/><td/></tr><tr><td>Model</td><td colspan=\"3\">Prec(%) Recall(%) F 1 (%)</td></tr><tr><td>Baseline</td><td>79.70</td><td>52.54</td><td>63.33</td></tr><tr><td>W</td><td>79.18</td><td>69.27</td><td>73.89</td></tr><tr><td>W+P</td><td>77.62</td><td>68.77</td><td>72.93</td></tr><tr><td>W+B W+P+B</td><td>79.63 77.89</td><td>69.76 69.39</td><td>74.37 73.39</td></tr></table>",
"text": "",
"num": null
},
"TABREF2": {
"html": null,
"type_str": "table",
"content": "<table><tr><td colspan=\"4\">Logical Part Prec(%) Recall(%) F 1 (%)</td></tr><tr><td>C EL</td><td>83.41 76.74</td><td>75.00 60.00</td><td>78.98 67.35</td></tr><tr><td>ER</td><td>41.94</td><td>22.81</td><td>29.55</td></tr><tr><td>Ob</td><td>0.00</td><td>0.00</td><td>0.00</td></tr><tr><td>A RepO</td><td>80.42 100</td><td>80.42 16.67</td><td>80.42 28.57</td></tr><tr><td>RepR</td><td>100</td><td>28.57</td><td>44.44</td></tr><tr><td>T 2 T 3 Overall</td><td>83.64 60.00 79.63</td><td>80.70 25.00 69.76</td><td>82.14 35.29 74.37</td></tr></table>",
"text": "Experimental results in more details",
"num": null
},
"TABREF3": {
"html": null,
"type_str": "table",
"content": "<table><tr><td/><td colspan=\"3\">Experiments on Sub-task 2</td></tr><tr><td/><td colspan=\"3\">Gold Input Setting</td></tr><tr><td>Model Heuristic ILP</td><td colspan=\"3\">Prec(%) Recall(%) F 1 (%) 71.19 75.89 81.24 76.56 82.87 79.59</td></tr><tr><td/><td colspan=\"3\">End-to-End Setting</td></tr><tr><td>Model</td><td colspan=\"3\">Prec(%) Recall(%) F 1 (%)</td></tr><tr><td>Heuristic</td><td>54.88</td><td>47.84</td><td>51.12</td></tr><tr><td>ILP</td><td>57.51</td><td>54.06</td><td>55.73</td></tr></table>",
"text": "",
"num": null
}
}
}
} |