File size: 107,532 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 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 | {
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:57:52.285474Z"
},
"title": "Incorporating Temporal Information in Entailment Graph Mining",
"authors": [
{
"first": "Liane",
"middle": [],
"last": "Guillou",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Edinburgh",
"location": {}
},
"email": "liane.guillou@ed.ac.uk"
},
{
"first": "Sander",
"middle": [
"Bijl"
],
"last": "De Vroe",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Edinburgh",
"location": {}
},
"email": ""
},
{
"first": "Mohammad",
"middle": [
"Javad"
],
"last": "Hosseini",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Edinburgh",
"location": {}
},
"email": "javad.hosseini@ed.ac.uk"
},
{
"first": "Mark",
"middle": [],
"last": "Steedman",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Edinburgh",
"location": {}
},
"email": "steedman@inf.ed.ac.uk"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "We present a novel method for injecting temporality into entailment graphs to address the problem of spurious entailments, which may arise from similar but temporally distinct events involving the same pair of entities. We focus on the sports domain in which the same pairs of teams play on different occasions, with different outcomes. We present an unsupervised model that aims to learn entailments such as win/lose \u2192 play, while avoiding the pitfall of learning non-entailments such as win \u2192 lose. We evaluate our model on a manually constructed dataset, showing that incorporating time intervals and applying a temporal window around them, are effective strategies.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "We present a novel method for injecting temporality into entailment graphs to address the problem of spurious entailments, which may arise from similar but temporally distinct events involving the same pair of entities. We focus on the sports domain in which the same pairs of teams play on different occasions, with different outcomes. We present an unsupervised model that aims to learn entailments such as win/lose \u2192 play, while avoiding the pitfall of learning non-entailments such as win \u2192 lose. We evaluate our model on a manually constructed dataset, showing that incorporating time intervals and applying a temporal window around them, are effective strategies.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Recognising textual entailment and paraphrases is core to many downstream NLP applications such as question answering and semantic parsing. In the case of open-domain question answering over unstructured data, the answer to a question may not be explicitly stated in the text, but may be recovered via paraphrases and/or entailment rules.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Entailment graphs (Berant et al., 2011; Berant et al., 2015; Hosseini et al., 2018) , in which nodes represent predicates and edges are entailment relations, have been proposed as a means to answer such questions. They can be mined using unsupervised methods applied over large collections of text, by keeping track of which entity pairs occur with which predicates. One common error made by these graphs, however, is that they assert spurious associations between similar but temporally distinct events that occur with the same entity pairs. For example, both the predicates beat and lost against will apply to sports team entity pairs such as (Arsenal, Man United). This is likely to mislead the current methods into incorrectly assigning an entailment relation between these two predicates.",
"cite_spans": [
{
"start": 18,
"end": 39,
"text": "(Berant et al., 2011;",
"ref_id": "BIBREF0"
},
{
"start": 40,
"end": 60,
"text": "Berant et al., 2015;",
"ref_id": "BIBREF1"
},
{
"start": 61,
"end": 83,
"text": "Hosseini et al., 2018)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper we extend the framework of Hosseini et al. (2018) to incorporate the temporal location of events, with the aim of mitigating these spurious entailments. Temporal information can be used to disentangle these groups of highly correlated predicates, because although they will share entity pairs, they will never occur at the same time. For example, in Figure 1 Arsenal and Man United played each other three times in 2019, with three different outcomes: win (beat), lose (lost against), tie (tied with).",
"cite_spans": [
{
"start": 41,
"end": 63,
"text": "Hosseini et al. (2018)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [
{
"start": 364,
"end": 372,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In previous methods, the context in which the predicates occur appears to be the same, because they only consider entity pairs as context. Therefore they mistakenly take the examples in Figure 1 as evidence of entailments or paraphrases between the three outcome predicates (win, lose, and tie), depending on the distributions found in the data. Our method enriches this context to include time interval information, thereby filtering out combinations that are not temporally near each other. Thus we hope to avoid learning that beat \u2192 lost against, while still learning that beat \u2192 play.",
"cite_spans": [],
"ref_spans": [
{
"start": 186,
"end": 194,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "As an initial test domain, we focus on the sports news genre, using extracted relations that involve two sports teams. We evaluate on a manually constructed dataset of 1,312 entailment pairs based on paraphrases of the predicates in the graph on the right hand side of Figure 1 . Our goal is to recover the structure of this graph in an unsupervised way, separating each of the highly correlated outcome predicates while predicting that they all entail play.",
"cite_spans": [],
"ref_spans": [
{
"start": 269,
"end": 277,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Arsenal-played and lost against-Man United 1-3 (25/01/2019) Arsenal-played and beat-Man United 2-0 (10/03/2018) Arsenal-played and tied with-Man United 1-1 (30/09/2019) The contributions of this work are: 1) a model for incorporating relation-level time intervals into an entailment graph mining procedure, outperforming non-temporal models, and 2) a manually constructed evaluation dataset of sports domain predicates. To our knowledge this is the first attempt to incorporate temporal information for learning entailment graphs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Entailment graphs have been constructed for a range of domains, including newswire (Hosseini et al., 2018) , health (Levy et al., 2014) , and commonsense (Yu et al., 2020) . In order to leverage temporal information, our work focuses on the news domain, in which each article has a known publication date and temporal expressions are commonly used.",
"cite_spans": [
{
"start": 83,
"end": 106,
"text": "(Hosseini et al., 2018)",
"ref_id": "BIBREF8"
},
{
"start": 116,
"end": 135,
"text": "(Levy et al., 2014)",
"ref_id": "BIBREF12"
},
{
"start": 154,
"end": 171,
"text": "(Yu et al., 2020)",
"ref_id": "BIBREF24"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Entailment Graphs",
"sec_num": "2.1"
},
{
"text": "A range of node representations have been explored, including Open-IE propositions (Levy et al., 2014) , typed predicates (Berant et al., 2011; Hosseini et al., 2018) , and eventualities (Yu et al., 2020) . In this work we use typed predicates, leveraging the second level in the FIGER hierarchy (Ling and Weld, 2012) , to enable a close examination of events that take place between two sports teams.",
"cite_spans": [
{
"start": 83,
"end": 102,
"text": "(Levy et al., 2014)",
"ref_id": "BIBREF12"
},
{
"start": 122,
"end": 143,
"text": "(Berant et al., 2011;",
"ref_id": "BIBREF0"
},
{
"start": 144,
"end": 166,
"text": "Hosseini et al., 2018)",
"ref_id": "BIBREF8"
},
{
"start": 187,
"end": 204,
"text": "(Yu et al., 2020)",
"ref_id": "BIBREF24"
},
{
"start": 296,
"end": 317,
"text": "(Ling and Weld, 2012)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Entailment Graphs",
"sec_num": "2.1"
},
{
"text": "Whether predicates in the graph entail each other may be determined using a variety of similarity measures. These are inspired by the Distributional Inclusion Hypothesis, which states that a predicate p entails another predicate q if for any context in which p can be used, q may be used in its place (Dagan et al., 1999; Geffet and Dagan, 2005) . They include the symmetric Lin's similarity measure (Lin, 1998) , the directional Weeds' precision and recall measures (Weeds and Weir, 2003) , and the Balanced Inclusion score (BInc) (Szpektor and Dagan, 2008) . BInc, the geometric mean of Lin's similarity and Weed's precision, combines the desirable behaviors of symmetric and directional measures. We adapt and examine each of these similarity measures using our evaluation dataset.",
"cite_spans": [
{
"start": 301,
"end": 321,
"text": "(Dagan et al., 1999;",
"ref_id": "BIBREF6"
},
{
"start": 322,
"end": 345,
"text": "Geffet and Dagan, 2005)",
"ref_id": "BIBREF7"
},
{
"start": 400,
"end": 411,
"text": "(Lin, 1998)",
"ref_id": "BIBREF14"
},
{
"start": 467,
"end": 489,
"text": "(Weeds and Weir, 2003)",
"ref_id": "BIBREF23"
},
{
"start": 532,
"end": 558,
"text": "(Szpektor and Dagan, 2008)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Entailment Graphs",
"sec_num": "2.1"
},
{
"text": "Alternatively, Hosseini et al. (2019) performed link prediction on the set of relation triples extracted from the text, and showed improvements over BInc by augmenting the data with additional predicted triples. We consider this link prediction model to be beyond the scope of this work.",
"cite_spans": [
{
"start": 15,
"end": 37,
"text": "Hosseini et al. (2019)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Entailment Graphs",
"sec_num": "2.1"
},
{
"text": "The construction of entailment datasets has been framed as a number of manual annotation tasks including image captioning (Bowman et al., 2015) and question answering (Levy and Dagan, 2016) .",
"cite_spans": [
{
"start": 122,
"end": 143,
"text": "(Bowman et al., 2015)",
"ref_id": "BIBREF3"
},
{
"start": 167,
"end": 189,
"text": "(Levy and Dagan, 2016)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluating Entailment Graphs",
"sec_num": "2.2"
},
{
"text": "The dataset creation method used by Levy and Dagan (2016) aims to address the bias towards real world knowledge. They ask human annotators to mark possible answers to questions as True/False (entailment/non-entailment), with entities in the answer replaced using tokens representing their type (e.g. London becomes city). The method also aims to address the bias of other datasets such as Zeichner's dataset (Zeichner et al., 2012) and the SherLIic dataset (Schmitt and Sch\u00fctze, 2019) , in which candidate entailments were automatically pre-selected for manual annotation according to a similarity measure. Entailments that exist, but are not captured by these similarity measures will therefore be excluded.",
"cite_spans": [
{
"start": 36,
"end": 57,
"text": "Levy and Dagan (2016)",
"ref_id": "BIBREF11"
},
{
"start": 408,
"end": 431,
"text": "(Zeichner et al., 2012)",
"ref_id": "BIBREF25"
},
{
"start": 457,
"end": 484,
"text": "(Schmitt and Sch\u00fctze, 2019)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluating Entailment Graphs",
"sec_num": "2.2"
},
{
"text": "There has been very little work on the specific problem of evaluating entailment of a temporal nature. The FraCas test suite (Cooper et al., 1996) contains a small section of which only a few examples are entailments between predicates. The TEA dataset (Kober et al., 2019) consists of pairs of sentences in which temporally ordered predicates have varying tense and aspect, such as is visiting \u2192 has arrived, but does not include non-entailments that can be learned through the temporal separation of events (such as the outcome predicates win and lose that we are interested in). Since there is no dataset to evaluate this phenomenon, we construct our own (Section 4.1). Our method for dataset construction is similar to that of Berant et al. (2011) . They manually annotated all edges in 10 typed entailment graphs, resulting in 3,427 edges (entailments) and 35,585 non-edges (non-entailments).",
"cite_spans": [
{
"start": 125,
"end": 146,
"text": "(Cooper et al., 1996)",
"ref_id": "BIBREF5"
},
{
"start": 253,
"end": 273,
"text": "(Kober et al., 2019)",
"ref_id": "BIBREF10"
},
{
"start": 731,
"end": 751,
"text": "Berant et al. (2011)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluating Entailment Graphs",
"sec_num": "2.2"
},
{
"text": "We use a pipeline based on a Combinatory Categorial Grammar (CCG, (Steedman, 2000) ) parser to extract binary relations with time intervals. These relations are used to construct typed entailment graphs using the unsupervised method of Hosseini et al. (2018) , adapted to compare only pairs of relations that are temporally near each other. We extract binary relations of the form arg1-predicate-arg2 (e.g. Arsenal-beat-Man United), following the example of Lewis and Steedman (2013) and Berant et al. (2015) . We use a pipeline approach similar to that described by Hosseini et al. (2018) , which allows us to extract open-domain relations. Relations are extracted from the NewsSpike corpus (Zhang and Weld, 2013) of news articles collected from multiple sources over a period of approximately six weeks.",
"cite_spans": [
{
"start": 66,
"end": 82,
"text": "(Steedman, 2000)",
"ref_id": "BIBREF20"
},
{
"start": 236,
"end": 258,
"text": "Hosseini et al. (2018)",
"ref_id": "BIBREF8"
},
{
"start": 458,
"end": 483,
"text": "Lewis and Steedman (2013)",
"ref_id": "BIBREF13"
},
{
"start": 488,
"end": 508,
"text": "Berant et al. (2015)",
"ref_id": "BIBREF1"
},
{
"start": 567,
"end": 589,
"text": "Hosseini et al. (2018)",
"ref_id": "BIBREF8"
},
{
"start": 692,
"end": 714,
"text": "(Zhang and Weld, 2013)",
"ref_id": "BIBREF26"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Relation Extraction",
"sec_num": "3.1"
},
{
"text": "We traverse dependency graphs generated over the output of the Rotating CCG parser (Stanojevi\u0107 and Steedman, 2019) , starting from verb and preposition nodes, until we reach an argument leaf node. The traversed nodes are used to form (lemmatised) predicate strings, and arguments are classified as either a Named Entity (extracted by the CoreNLP Named Entity recogniser), or a general entity (all other nouns and noun phrases). Predicate strings may include (non-auxiliary) verbs, verb particles, adjectives, and prepositions. Negation nodes are detected via string match (\"not\", \"n't\", and \"never\"), and are included in the predicate if there is a path between the negation node and a node in the predicate. We map passive predicates to active ones. Modifiers such as \"managed to\" as in the example \"Arsenal managed to beat Man United\" are also extracted and included in the predicate. As the modifiers may be rather sparse, we extract the relation both with and without the modifier.",
"cite_spans": [
{
"start": 83,
"end": 114,
"text": "(Stanojevi\u0107 and Steedman, 2019)",
"ref_id": "BIBREF19"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Relation Extraction",
"sec_num": "3.1"
},
{
"text": "We extract and resolve time expressions in the document text, using SUTime (Chang and Manning, 2012), available via CoreNLP. If there is a path in the CCG dependency graph between the time expression and a node in the predicate, the relation is assigned a time interval. Entities are mapped to types by linking to their Freebase (Bollacker et al., 2008) IDs using AIDA-Light (Nguyen et al., 2014) , and subsequently mapping these IDs to their fine-grained FIGER types (Ling and Weld, 2012) .",
"cite_spans": [
{
"start": 329,
"end": 353,
"text": "(Bollacker et al., 2008)",
"ref_id": "BIBREF2"
},
{
"start": 364,
"end": 396,
"text": "AIDA-Light (Nguyen et al., 2014)",
"ref_id": null
},
{
"start": 468,
"end": 489,
"text": "(Ling and Weld, 2012)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Relation Extraction",
"sec_num": "3.1"
},
{
"text": "To restrict the data to the sports domain we filter the set of output relations, accepting only those involving two entities of the fine-grained FIGER type organization/sports team. This results in a set of 78,439 binary relations extracted from 24,147 articles, of which 14,664 (approximately 19%) have time intervals derived from SUTime. The sports domain has the advantage that events are similar and should be easily separable in time, and it provides the straightforward win-lose-tie outcome set. Sports data is common in NewsSpike, and sports teams have reliable Named Entity linking, making it suitable for an initial investigation. In the future we will apply this method to other entity type pairs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Relation Extraction",
"sec_num": "3.1"
},
{
"text": "The input to the graph construction step is the set of typed binary relations paired with their time intervals. As we focus on events that involve two sports teams, the output is a single organization-organization graph, rather than the typical set of graphs (one for every pair of types). Note that these graphs contain only locally learned entailments, and that global inference across graphs is not performed. This is sufficient to demonstrate the benefit of incorporating time intervals.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Graph Construction",
"sec_num": "3.2"
},
{
"text": "In the original method for computing local entailment scores, Hosseini et al. (2018) extract a feature vector for each typed predicate (e.g. play with type pair organization-organization). The entity pairs from the binary relations (e.g. Arsenal, Man United) are used as the feature types, and the pointwise mutual information (PMI) between the predicate and the entity pair is the value. These feature vectors are then used to compute local similarity scores.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Graph Construction",
"sec_num": "3.2"
},
{
"text": "We extend this method to consider the time intervals for each of the binary relations, with the goal of comparing only those events that are temporally near each other. To achieve this, we filter the counts of predicate q according to whether each event's time interval overlaps with any of p's. In other words, an event in q is retained if it is close enough to any event in p. We consider new local similarity scores based on both the filtered counts, and scaled PMI scores.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Graph Construction",
"sec_num": "3.2"
},
{
"text": "Algorithm 1 describes the process of filtering counts using time intervals. The process uses a set of edges E between predicate nodes to store filtered count information. We loop through each entity pair ep and get the list of predicates that occur with that entity pair (line 4). Then, for each pair of predicates, we instantiate edgeObjects (line 7) between predicates p and q (in both directions), to store the filtered count information. We also retrieve p and q's timeObjects, containing a list of the time intervals at which the predicate and entity pair co-occurred (lines 8-9). For each pair of time intervals we compute whether there is an overlap (lines 12-19). The filtered count is the total number of events in predicate p that temporally overlap with any event in predicate q.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Graph Construction",
"sec_num": "3.2"
},
{
"text": "The count is stored in the edgeObject edge p,q . Once all counts have been collected, they are used to compute the similarity measures. The computation of temporal measures is identical to that of the non-temporal counterparts, but they use the filtered counts as input instead of the regular counts. Each edgeObject populates a cell in W, the sparse matrix of all similarity scores between predicates p and q, as presented by Hosseini et al. (2018) . The filtered counts are also used to scale the PMI scores (see section 4.2).",
"cite_spans": [
{
"start": 427,
"end": 449,
"text": "Hosseini et al. (2018)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Graph Construction",
"sec_num": "3.2"
},
{
"text": "Consider the following minimal worked example: Two matches between Arsenal and Man United, one where Arsenal wins (on 10/03/2018), and one where they lose (on 25/01/2019). This initially results in the following extracted predicates and counts: play (2), win (1), lose (1). After filtering, we add a count of 1 to both the win \u2192 play and lose \u2192 play edges, and a count of 0 to the win \u2192 lose edge (and its reverse).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Graph Construction",
"sec_num": "3.2"
},
{
"text": "We consider three possible sources of time intervals: 1) the resolved time expressions extracted from raw text using SUTime, 2) the document creation date (provided as metadata in the NewsSpike corpus), and 3) a combination of the two -using resolved time expressions where these are available, and backing off to the document creation date where they are not. The intuition behind using time expressions extracted from the article text is that these ought to more accurately pinpoint the time interval of the events. However, as such expressions may be sparse, we also investigate the use of the document creation date, under the assumption that sports news is likely to be reported very close the day of the event.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Graph Construction",
"sec_num": "3.2"
},
{
"text": "We also consider a temporal window to extend the time intervals by N days either side. This would mitigate the problem of sports events being reported several days after the event, especially when we fall back to the document creation date. For sports matches we would expect to see a benefit in using a small window of a few days, and a detrimental effect as that window grows increasingly larger. Specifically, we expect that larger windows would render temporal information useless, preventing our model from being able to distinguish between two different matches involving the same pair of teams. Time interval source and window size are event-specific parameters that we experiment with in Section 5.1.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Graph Construction",
"sec_num": "3.2"
},
{
"text": "We propose a semi-automatic method to construct a small evaluation dataset based on manually constructed paraphrase clusters. We start with a small set of predicates for which we know the entailment pattern, in our case {win, play, lose and tie}. We restrict the dataset to include only those binary relations that involve two sports teams, by filtering on the fine-grained FIGER (Ling and Weld, 2012) type organization/sports team. We then order the predicates by their frequency in the corpus, and manually select paraphrases of our small set with a count of at least 20 (the 235 most frequent predicates). This results in four clusters of paraphrases, with sizes of 26, 8, 3 and 5 respectively for win, lose, tie and play. We then automatically generate entailment pairs (1,312 in total), labelling them according to the pattern Algorithm 1 Temporal filtering in local graph computation 1: procedure TEMPORALFILTER(entity pairs, predicates) 2:",
"cite_spans": [
{
"start": 380,
"end": 401,
"text": "(Ling and Weld, 2012)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "E \u2190 initialiseAllEdgeObjects(predicates) Initialise set of edges 3:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "for ep in entity pairs do 4:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "predicatesep \u2190 getP redicatesF orEntityP air(predicates, ep) 5:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "for p \u2190 0 to length(predicatesep) do 6:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "for q \u2190 p + 1 to length(predicatesep) do 7:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "edgep,q, edgeq,p \u2190 getEdgeObjects(E, p, q) 8:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "time objectsep,p \u2190 getT imeObjects(ep, p) 9:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "time objectsep,q \u2190 getT imeObjects(ep, q) 10:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "overlapp \u2190 initialiseV ectorOf Zeros(length(time objectsep,p)) 11:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "overlapq \u2190 initialiseV ectorOf Zeros(length(time objectsep,q)) 12:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "for i \u2190 0 to length(time objectsep,p) do 13:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "for j \u2190 0 to length(time objectsep,q) do 14:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "if return E 27: end procedure in Table 1 , with premises in the rows and hypotheses in the columns.",
"cite_spans": [],
"ref_spans": [
{
"start": 33,
"end": 40,
"text": "Table 1",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "We include the paraphrase category for completeness, although we are more interested in the effect of separating temporally disjoint sports match outcomes. The paraphrase category contains predicates of varying gradation, such as crush suggesting a strong victory or eliminate indicating that a team is knocked out of a tournament. We wish to avoid specific predicates such as eliminate entailing nonspecific predicates like beat. To avoid this issue we manually annotated the predicates for specificity, and for the paraphrase entailments subset we only generate pairs for non-specific predicates. More generally, a set of paraphrase clusters with a total of n predicates yields n 2 \u2212 n pairs (not taking into account the paraphrase subsets reduction). 1 The dataset of 1,312 entailment pairs is available here. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset Construction",
"sec_num": "4.1"
},
{
"text": "We compute both symmetric and directional similarity measures to learn entailments, making use of the temporally filtered counts and PMI scores described in Section 3.2. Specifically, we adapted Lin's similarity measure (Lin, 1998 ), Weeds' precision and recall measures (Weeds and Weir, 2003) , and BInc (Szpektor and Dagan, 2008) . The adaptations of these measures are: Temporal count-based measures using the temporally filtered counts: Weeds' precision, recall, and similarity (harmonic average of precision and recall); Lin's similarity; BInc using Weed's precision and count-based Lin's similarity.",
"cite_spans": [
{
"start": 220,
"end": 230,
"text": "(Lin, 1998",
"ref_id": "BIBREF14"
},
{
"start": 271,
"end": 293,
"text": "(Weeds and Weir, 2003)",
"ref_id": "BIBREF23"
},
{
"start": 305,
"end": 331,
"text": "(Szpektor and Dagan, 2008)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Similarity Measures",
"sec_num": "4.2"
},
{
"text": "Temporal PMI-based measures: As a proxy to computing Conditional PMI between an entity pair, predicate p, and predicate q, which would be computationally expensive (if not infeasible) given the existing graph construction framework, we scale the original PMI scores. We apply two strategies: 1) Ratio: scale the original PMI scores according to the ratio of filtered counts to regular counts, 2) Binary: use the original PMI score if any of the events in predicate p overlap with any event in predicate q, otherwise set the score to zero. The following measures use the Ratio and Binary PMI strategies: Weeds' precision, recall, and similarity; Lin's similarity; BInc using Weed's PMI precision and Lin's similarity.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Similarity Measures",
"sec_num": "4.2"
},
{
"text": "Temporal hybrid BInc measures: Computed using count-based Weeds' precision and PMI-based Lin's similarity, using the temporally filtered counts. We do this for both Ratio and Binary PMI.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Similarity Measures",
"sec_num": "4.2"
},
{
"text": "We also ensure that for every temporal measure, its non-temporal counterpart is also included, and we include cosine similarity as a symmetric baseline. In total, we experiment with 29 similarity measures.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Similarity Measures",
"sec_num": "4.2"
},
{
"text": "5 Experiments, Results and Analysis",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Similarity Measures",
"sec_num": "4.2"
},
{
"text": "As described in Section 3.1 we extract all possible relations from the NewsSpike corpus and map their entities to types using the FIGER hierarchy. We construct a typed entailment graph for the organization/organization type pair using the subset of these relations where both entities are sports teams. We compute entailment scores using the set of 29 similarity measures described in Section 4.2. Due to space constraints we highlight results for eight of these measures: BInc and cosine similarity baselines, and the best performing temporal measures and their non-temporal counterparts.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Settings",
"sec_num": "5.1"
},
{
"text": "We experimented with different values for the event-specific time information source and temporal window described in Section 3.2. We constructed typed entailment graphs using only time expressions (timexOnly), only the document creation date (docDateOnly), and using time expressions where available, otherwise backing off to the document creation date (timexAndDocDate). For each of the time interval sources, we also applied windows of 1, 2, 3, 4, 5, 6, 7, 30 and 3,650 days, as well as no window.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Settings",
"sec_num": "5.1"
},
{
"text": "We used the evaluation dataset described in Section 4.1 and Table 1 to evaluate entailments captured under each of these experimental settings. We evaluate on three different configurations of the dataset: Base (entailment 1 and outcome 0), Directional (entailment 1 and directional 0), and All (entailment 1, outcome 0, directional 0, paraphrase 1). When we introduce parameter tuning in future work, we plan to assign a development/test split to the dataset.",
"cite_spans": [],
"ref_spans": [
{
"start": 60,
"end": 67,
"text": "Table 1",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Experimental Settings",
"sec_num": "5.1"
},
{
"text": "See Table 2 for area under the curve (AUC) scores for each of the three temporal information sources, for a range of temporal (T) and non-temporal similarity measures. To evaluate the similarity measures fairly we calculate AUC under a recall threshold (a recall of 0.75 is reached by all non-timexOnly measures). The timexOnly source has the weakest performance, since it has access to a sparser set of time intervals (with only 19% of relations linked to a time expression in the text). When we focus on the low recall range (< 0.15), however, we find that the temporal measures outperform the non-temporal ones. This is especially promising as it highlights the benefit of using the more accurate time intervals resolved from time expressions in the text; these give temporal measures a larger increase than the document date. When using only the document creation date (docDateOnly), we find that temporal and non temporal measures perform similarly. Results for the timexAndDocDate source show that leveraging both time expression and document time information together leads to the most effective model. Table 2 : Temporal information source: AUC scores for the base evaluation dataset, and a temporal window size of 4 days",
"cite_spans": [],
"ref_spans": [
{
"start": 4,
"end": 11,
"text": "Table 2",
"ref_id": null
},
{
"start": 1110,
"end": 1117,
"text": "Table 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Temporal Information Source",
"sec_num": "5.2"
},
{
"text": "In Figure 2 we can see that there is a sharp improvement in AUC score for all of the temporally-informed similarity measures when a window of one day is applied 2 . This is likely due to data sparsity and because sports articles report on the same event on different days. The horizontal lines represent the nontemporally-informed similarity measures. We can also see that the choice of window size depends on the similarity measure. For the majority of the temporally-informed similarity measures, a window size between one and four days works well. For this class of predicates a window size of 4 seems suitable, as it avoids conflating games that happen on consecutive weekends, while giving some leeway. We discuss the possibility of using a dynamic window in Section 6. ",
"cite_spans": [],
"ref_spans": [
{
"start": 3,
"end": 11,
"text": "Figure 2",
"ref_id": "FIGREF2"
}
],
"eq_spans": [],
"section": "Temporal Window Size",
"sec_num": "5.3"
},
{
"text": "We can see from Figure 3 that BInc, a state-of-the-art measure for relation entailment, does not perform well within the temporal setting and is outperformed by a number of the temporal measures. Of those, T. BInc count sims, the version of BInc that uses the temporally filtered counts, produces the best results on the base subset of the evaluation dataset. The two temporally informed measures that use scaled PMI scores (T. Ratio BInc sims and T. Binary BInc sims) also outperform BInc but to a lesser degree. This may be due to sparsity in the set of binary relations making it difficult to estimate accurate scaled PMI scores. We hope to alleviate this problem by moving to a larger corpus in future work 3 . In more general terms, for each similarity measure, the temporal version performs better (with the exception of Weed's probabilistic precision, for which there is no change).",
"cite_spans": [],
"ref_spans": [
{
"start": 16,
"end": 24,
"text": "Figure 3",
"ref_id": "FIGREF3"
}
],
"eq_spans": [],
"section": "Comparing Similarity Measures",
"sec_num": "5.4"
},
{
"text": "Our main interest is in performance on the base dataset, but for completeness and comparison to previous work, which tests directionality more, we also evaluated on the all and directional subsets (see Table 3 ).",
"cite_spans": [],
"ref_spans": [
{
"start": 202,
"end": 209,
"text": "Table 3",
"ref_id": "TABREF5"
}
],
"eq_spans": [],
"section": "Performance on Data Subsets",
"sec_num": "5.5"
},
{
"text": "To investigate the challenge of directionality in entailment we consider the set of entailments and their reverse, e.g. play \u2192 win (\"entailment 1\" and \"directional 0\" in Table 1(b)). We find that in general the temporal similarity measures still perform strongly. T. Weeds' precision, the only purely directional measure in the set, performs comparably to its non-temporal counterpart on the directional subset. The second best measure is BInc which is unsurprising given that it also captures directionality, and that the dataset no longer tests temporality.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Performance on Data Subsets",
"sec_num": "5.5"
},
{
"text": "We also evaluate on the complete dataset (all) which includes paraphrases (\"paraphrase 1\" in Table 1(b)). Here we find that T. Weeds' precision is again the best measure, closely followed by the PMI-based T. BInc scores (Binary and Ratio). We expect that the strong performance of T. Weeds' precision is due to correctly identifying the directional entailments in the dataset, and that the drop in performance between the directional and all subsets is due to the inclusion of paraphrases, which are challenging for all measures, but particularly for the Weed's precision measures which have no symmetric component. BInc also performs reasonably well on this subset, showing that temporally-uninformed measures remain competitive when multiple phenomena are tested. In general we find that at least one temporally-informed similarity measure still performs strongly for each of the subsets in the dataset. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Performance on Data Subsets",
"sec_num": "5.5"
},
{
"text": "To see the benefits of including temporal information, we compare the AUC scores for the temporal and non-temporal similarity measures. We use the timexAndDocDate temporal information source, and a window size of four days (see Figure 3) . The organization-organization graph has 46,899 nodes and between 7,888,372 and 8,256,182 edges depending on the similarity measure. We can see that the temporal similarity measure T. BInc count outperforms BInc, the state-of-the-art non-temporal similarity measure for relation entailment employed by Berant et al. (2011) and Hosseini et al. (2018) . We conclude that incorporating temporal information is beneficial for accurately predicting entailments and non-entailments for highly correlated predicates that are part of distinct events. The NewsSpike corpus covers a period of approximately six weeks so the outcomes of two matches between two teams within this period may be fairly similar (as there will have been few changes to the teams management, players, etc.). In future work we plan to move to a corpus covering a larger time period, for which we would expect to observe a greater effect. Expanding beyond the sports team domain would also allow us to study events with longer duration, such as a president holding office, preceded by their campaign, and election.",
"cite_spans": [
{
"start": 541,
"end": 561,
"text": "Berant et al. (2011)",
"ref_id": "BIBREF0"
},
{
"start": 566,
"end": 588,
"text": "Hosseini et al. (2018)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [
{
"start": 228,
"end": 237,
"text": "Figure 3)",
"ref_id": "FIGREF3"
}
],
"eq_spans": [],
"section": "Effects of Using Temporal Information",
"sec_num": "5.6"
},
{
"text": "Expanding the domain will also give us a collection of local graphs of types beyond just organizationorganization, across which we can then learn globally consistent similarity scores (as in Hosseini et al. (2018) ). We can then also collapse cliques in the graph into paraphrase clusters with a single relation identifier, which we hope will improve performance, especially on sparser predicates. Our dataset is perfectly suited to evaluating benefits from this addition due to its origin in paraphrase clusters.",
"cite_spans": [
{
"start": 191,
"end": 213,
"text": "Hosseini et al. (2018)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Effects of Using Temporal Information",
"sec_num": "5.6"
},
{
"text": "Whilst we can observe a positive effect when using temporal information, the effect is modest. Upon closer inspection we found that this was due to relatively few events being filtered. An analysis of a subset of sentences revealed that relations were being extracted spuriously due to various linguistic phenomena. Issues are caused by conditionals (e.g. \"if Arsenal win\"), modals (\"I still expect Arsenal. . . \"), incorrect future predictions (\"Arsenal will win\") and counterfactuals (\"had Arsenal won,...\"). These types of predictions appear to be especially common in the sports domain. Another issue arose due to an incorrect application of passive to active conversion (Arsenal-lost to-Man United from \"Man United lost to Arsenal\") resulting from incorrect verb feature labels in the CCG parses. Finally, SUTime sometimes provides partial time information which can result in a whole year being used as an interval, creating spurious overlaps 4 . Addressing these issues should lead to a larger effect from using temporal information, because it would reduce overlaps and allow more filtering.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Effects of Using Temporal Information",
"sec_num": "5.6"
},
{
"text": "Our algorithm lends itself naturally to mining entailments with a temporal relation such as visit \u2192 arrived. We plan to achieve this by splitting the window into a before and after frame, producing separate entailment scores for different orderings. We also plan to investigate setting the window dynamically. In the current setup, events stay relevant for a similar amount of time, but different predicates should allow comparison for different granularities of time. For example, the window around a person being president should be larger than a person visiting a location. Initially we might aim to learn a different window size per predicate (for example by taking into account average predicate duration and granularity). Dynamic windowing could become particularly valuable with a broader domain.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Effects of Using Temporal Information",
"sec_num": "5.6"
},
{
"text": "More generally, our method could incorporate in its filtering any function of the contextualised events to determine whether their co-occurrence should contribute to an entailment score. Currently a binary decision is made based on time interval overlap, but one might use features such as (lexical) aspect, tense, the presence of other entities, etc. Previous work was limited to using the presence of two entities as a proxy for entailment relevance; with our refinements we could expand to involving not only time but also other features of the contextualised events.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Effects of Using Temporal Information",
"sec_num": "5.6"
},
{
"text": "We will also explore the use of other temporal resolution systems and aim to develop more sophisticated ways of linking times to events, which currently only occurs through CCG dependencies. More time intervals might be propagated using the TempEval (UzZaman et al., 2013) ordering approach or through other means, for instance by reasoning about tense, Reichenbachian reference time (Reichenbach, 1947) , or event coreference (within, or across documents).",
"cite_spans": [
{
"start": 384,
"end": 403,
"text": "(Reichenbach, 1947)",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Effects of Using Temporal Information",
"sec_num": "5.6"
},
{
"text": "We injected temporal information into the local entailment graph construction method of Hosseini et al. (2018) , with the goal of comparing only those events that are temporally near each other. This is achieved by filtering the counts of predicate p according to whether its events' time intervals overlap with the those of predicate q. We considered a range of new local similarity scores based on both temporally filtered counts and scaled PMI scores, which we evaluate on a semi-automatically constructed dataset, based on manually constructed paraphrase clusters.",
"cite_spans": [
{
"start": 88,
"end": 110,
"text": "Hosseini et al. (2018)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions",
"sec_num": "7"
},
{
"text": "Our temporal similarity measures outperform their non-temporal counterparts, including BInc, the state-of-the-art measure for relation entailment. We also show that using a combination of time expressions recovered from the text and the document creation date performed better than using only one of these sources, and that adding a temporal window around the time intervals of the events is essential. The performance of the temporal similarity measures over the non-temporal measures is particularly strong at the low recall range when only time expressions from the text are used. This is especially promising as it suggests that there is much room for improvement in using more sophisticated temporal resolution systems and methods for linking times to events.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions",
"sec_num": "7"
},
{
"text": "The subtracted term comes from duplicate pairs like defeat-defeat",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "With no window, the temporally informed similarity measures perform poorly (between 0.24 and 0.26)3 We have already compiled a dataset of 10 years worth of news data from multiple sources",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Focusing only on short time intervals failed to offset the other sources of spurious overlaps",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "We thank Milo\u0161 Stanojevi\u0107 for assistance with the Rotating CCG parser, Nick McKenna and Ian Wood for helpful discussions, and the reviewers for their valuable feedback. This work was funded by the ERC H2020 Advanced Fellowship GA 742137 SEMANTAX, a grant from The University of Edinburgh and Huawei Technologies, and The Alan Turing Institute under EPSRC grant EP/N510129/1.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Global learning of typed entailment rules",
"authors": [
{
"first": "Jonathan",
"middle": [],
"last": "Berant",
"suffix": ""
},
{
"first": "Ido",
"middle": [],
"last": "Dagan",
"suffix": ""
},
{
"first": "Jacob",
"middle": [],
"last": "Goldberger",
"suffix": ""
}
],
"year": 2011,
"venue": "Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies",
"volume": "",
"issue": "",
"pages": "610--619",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jonathan Berant, Ido Dagan, and Jacob Goldberger. 2011. Global learning of typed entailment rules. In Proceed- ings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technolo- gies, pages 610-619, Portland, Oregon, USA, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Efficient global learning of entailment graphs",
"authors": [
{
"first": "Jonathan",
"middle": [],
"last": "Berant",
"suffix": ""
},
{
"first": "Noga",
"middle": [],
"last": "Alon",
"suffix": ""
},
{
"first": "Ido",
"middle": [],
"last": "Dagan",
"suffix": ""
},
{
"first": "Jacob",
"middle": [],
"last": "Goldberger",
"suffix": ""
}
],
"year": 2015,
"venue": "Computational Linguistics",
"volume": "41",
"issue": "2",
"pages": "221--263",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jonathan Berant, Noga Alon, Ido Dagan, and Jacob Goldberger. 2015. Efficient global learning of entailment graphs. Computational Linguistics, 41(2):221-263, June.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Freebase: A collaboratively created graph database for structuring human knowledge",
"authors": [
{
"first": "Kurt",
"middle": [],
"last": "Bollacker",
"suffix": ""
},
{
"first": "Colin",
"middle": [],
"last": "Evans",
"suffix": ""
},
{
"first": "Praveen",
"middle": [],
"last": "Paritosh",
"suffix": ""
},
{
"first": "Tim",
"middle": [],
"last": "Sturge",
"suffix": ""
},
{
"first": "Jamie",
"middle": [],
"last": "Taylor",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the 2008 ACM SIGMOD International Conference on Management of Data, SIGMOD '08",
"volume": "",
"issue": "",
"pages": "1247--1250",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kurt Bollacker, Colin Evans, Praveen Paritosh, Tim Sturge, and Jamie Taylor. 2008. Freebase: A collaboratively created graph database for structuring human knowledge. In Proceedings of the 2008 ACM SIGMOD Interna- tional Conference on Management of Data, SIGMOD '08, page 1247-1250, New York, NY, USA. Association for Computing Machinery.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "A large annotated corpus for learning natural language inference",
"authors": [
{
"first": "R",
"middle": [],
"last": "Samuel",
"suffix": ""
},
{
"first": "Gabor",
"middle": [],
"last": "Bowman",
"suffix": ""
},
{
"first": "Christopher",
"middle": [],
"last": "Angeli",
"suffix": ""
},
{
"first": "Christopher",
"middle": [
"D"
],
"last": "Potts",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Manning",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "632--642",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. 2015. A large annotated corpus for learning natural language inference. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 632-642, Lisbon, Portugal, September. Association for Computational Linguistics.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "SUTime: A library for recognizing and normalizing time expressions",
"authors": [
{
"first": "X",
"middle": [],
"last": "Angel",
"suffix": ""
},
{
"first": "Christopher",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Manning",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12)",
"volume": "",
"issue": "",
"pages": "3735--3740",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Angel X. Chang and Christopher Manning. 2012. SUTime: A library for recognizing and normalizing time expressions. In Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12), pages 3735-3740, Istanbul, Turkey, May. European Language Resources Association (ELRA).",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Using the framework",
"authors": [
{
"first": "Robin",
"middle": [],
"last": "Cooper",
"suffix": ""
},
{
"first": "Dick",
"middle": [],
"last": "Crouch",
"suffix": ""
},
{
"first": "Jan",
"middle": [],
"last": "Van Eijck",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Fox",
"suffix": ""
},
{
"first": "Josef",
"middle": [],
"last": "Van Genabith",
"suffix": ""
},
{
"first": "Jan",
"middle": [],
"last": "Jaspars",
"suffix": ""
},
{
"first": "Hans",
"middle": [],
"last": "Kamp",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Milward",
"suffix": ""
},
{
"first": "Manfred",
"middle": [],
"last": "Pinkal",
"suffix": ""
},
{
"first": "Massimo",
"middle": [],
"last": "Poesio",
"suffix": ""
},
{
"first": "Steve",
"middle": [],
"last": "Pulman",
"suffix": ""
},
{
"first": "Ted",
"middle": [],
"last": "Briscoe",
"suffix": ""
},
{
"first": "Holger",
"middle": [],
"last": "Maier",
"suffix": ""
},
{
"first": "Karsten",
"middle": [],
"last": "Konrad",
"suffix": ""
}
],
"year": 1996,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Robin Cooper, Dick Crouch, Jan Van Eijck, Chris Fox, Josef Van Genabith, Jan Jaspars, Hans Kamp, David Milward, Manfred Pinkal, Massimo Poesio, Steve Pulman, Ted Briscoe, Holger Maier, and Karsten Konrad. 1996. Using the framework.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Similarity-based models of word cooccurrence probabilities",
"authors": [
{
"first": "Lillian",
"middle": [],
"last": "Ido Dagan",
"suffix": ""
},
{
"first": "Fernando",
"middle": [
"C N"
],
"last": "Lee",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Pereira",
"suffix": ""
}
],
"year": 1999,
"venue": "Machine Learning",
"volume": "34",
"issue": "",
"pages": "43--69",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ido Dagan, Lillian Lee, and Fernando C. N. Pereira. 1999. Similarity-based models of word cooccurrence proba- bilities. Machine Learning, 34:43-69.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "The distributional inclusion hypotheses and lexical entailment",
"authors": [
{
"first": "Maayan",
"middle": [],
"last": "Geffet",
"suffix": ""
},
{
"first": "Ido",
"middle": [],
"last": "Dagan",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL'05)",
"volume": "",
"issue": "",
"pages": "107--114",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Maayan Geffet and Ido Dagan. 2005. The distributional inclusion hypotheses and lexical entailment. In Proceed- ings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL'05), pages 107-114, Ann Arbor, Michigan, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Learning typed entailment graphs with global soft constraints",
"authors": [
{
"first": "Mohammad Javad",
"middle": [],
"last": "Hosseini",
"suffix": ""
},
{
"first": "Nathanael",
"middle": [],
"last": "Chambers",
"suffix": ""
},
{
"first": "Siva",
"middle": [],
"last": "Reddy",
"suffix": ""
},
{
"first": "Xavier",
"middle": [
"R"
],
"last": "Holt",
"suffix": ""
},
{
"first": "Shay",
"middle": [
"B"
],
"last": "Cohen",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Johnson",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Steedman",
"suffix": ""
}
],
"year": 2018,
"venue": "Transactions of the Association for Computational Linguistics",
"volume": "6",
"issue": "",
"pages": "703--717",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mohammad Javad Hosseini, Nathanael Chambers, Siva Reddy, Xavier R. Holt, Shay B. Cohen, Mark Johnson, and Mark Steedman. 2018. Learning typed entailment graphs with global soft constraints. Transactions of the Association for Computational Linguistics, 6:703-717.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Duality of link prediction and entailment graph induction",
"authors": [
{
"first": "Mohammad Javad",
"middle": [],
"last": "Hosseini",
"suffix": ""
},
{
"first": "Shay",
"middle": [
"B"
],
"last": "Cohen",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Johnson",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Steedman",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "4736--4746",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mohammad Javad Hosseini, Shay B. Cohen, Mark Johnson, and Mark Steedman. 2019. Duality of link prediction and entailment graph induction. In Proceedings of the 57th Annual Meeting of the Association for Computa- tional Linguistics, pages 4736-4746, Florence, Italy, July. Association for Computational Linguistics.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Temporal and aspectual entailment",
"authors": [
{
"first": "Thomas",
"middle": [],
"last": "Kober",
"suffix": ""
},
{
"first": "Sander",
"middle": [],
"last": "Bijl De Vroe",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Steedman",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 13th International Conference on Computational Semantics -Long Papers",
"volume": "",
"issue": "",
"pages": "103--119",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Thomas Kober, Sander Bijl de Vroe, and Mark Steedman. 2019. Temporal and aspectual entailment. In Pro- ceedings of the 13th International Conference on Computational Semantics -Long Papers, pages 103-119, Gothenburg, Sweden, May. Association for Computational Linguistics.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Annotating relation inference in context via question answering",
"authors": [
{
"first": "Omer",
"middle": [],
"last": "Levy",
"suffix": ""
},
{
"first": "Ido",
"middle": [],
"last": "Dagan",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics",
"volume": "2",
"issue": "",
"pages": "249--255",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Omer Levy and Ido Dagan. 2016. Annotating relation inference in context via question answering. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 249-255, Berlin, Germany, August. Association for Computational Linguistics.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Focused entailment graphs for open IE propositions",
"authors": [
{
"first": "Omer",
"middle": [],
"last": "Levy",
"suffix": ""
},
{
"first": "Ido",
"middle": [],
"last": "Dagan",
"suffix": ""
},
{
"first": "Jacob",
"middle": [],
"last": "Goldberger",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the Eighteenth Conference on Computational Natural Language Learning",
"volume": "",
"issue": "",
"pages": "87--97",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Omer Levy, Ido Dagan, and Jacob Goldberger. 2014. Focused entailment graphs for open IE propositions. In Proceedings of the Eighteenth Conference on Computational Natural Language Learning, pages 87-97, Ann Arbor, Michigan, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Combined distributional and logical semantics",
"authors": [
{
"first": "Mike",
"middle": [],
"last": "Lewis",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Steedman",
"suffix": ""
}
],
"year": 2013,
"venue": "Transactions of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "179--192",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mike Lewis and Mark Steedman. 2013. Combined distributional and logical semantics. Transactions of the Association for Computational Linguistics, 1:179-192.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "An information-theoretic definition of similarity",
"authors": [
{
"first": "Dekang",
"middle": [],
"last": "Lin",
"suffix": ""
}
],
"year": 1998,
"venue": "Proceedings of the Fifteenth International Conference on Machine Learning, ICML '98",
"volume": "",
"issue": "",
"pages": "296--304",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dekang Lin. 1998. An information-theoretic definition of similarity. In Proceedings of the Fifteenth International Conference on Machine Learning, ICML '98, page 296-304, San Francisco, CA, USA. Morgan Kaufmann Publishers Inc.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Fine-grained entity recognition",
"authors": [
{
"first": "Xiao",
"middle": [],
"last": "Ling",
"suffix": ""
},
{
"first": "Daniel",
"middle": [
"S"
],
"last": "Weld",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the Twenty-Sixth AAAI Conference on Artificial Intelligence, AAAI'12",
"volume": "",
"issue": "",
"pages": "94--100",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xiao Ling and Daniel S. Weld. 2012. Fine-grained entity recognition. In Proceedings of the Twenty-Sixth AAAI Conference on Artificial Intelligence, AAAI'12, page 94-100. AAAI Press.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Aida-light: High-throughput named-entity disambiguation",
"authors": [
{
"first": "Johannes",
"middle": [],
"last": "Dat Ba Nguyen",
"suffix": ""
},
{
"first": "Martin",
"middle": [],
"last": "Hoffart",
"suffix": ""
},
{
"first": "Gerhard",
"middle": [],
"last": "Theobald",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Weikum",
"suffix": ""
}
],
"year": 2014,
"venue": "Workshop on Linked Data on the Web",
"volume": "1184",
"issue": "",
"pages": "1--10",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Dat Ba Nguyen, Johannes Hoffart, Martin Theobald, and Gerhard Weikum. 2014. Aida-light: High-throughput named-entity disambiguation. Workshop on Linked Data on the Web, 1184:1-10.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "The tenses of verbs. Time: From Concept to Narrative Construct: a Reader",
"authors": [
{
"first": "Hans",
"middle": [],
"last": "Reichenbach",
"suffix": ""
}
],
"year": 1947,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hans Reichenbach. 1947. The tenses of verbs. Time: From Concept to Narrative Construct: a Reader.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "SherLIiC: A typed event-focused lexical inference benchmark for evaluating natural language inference",
"authors": [
{
"first": "Martin",
"middle": [],
"last": "Schmitt",
"suffix": ""
},
{
"first": "Hinrich",
"middle": [],
"last": "Sch\u00fctze",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "902--914",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Martin Schmitt and Hinrich Sch\u00fctze. 2019. SherLIiC: A typed event-focused lexical inference benchmark for evaluating natural language inference. In Proceedings of the 57th Annual Meeting of the Association for Com- putational Linguistics, pages 902-914, Florence, Italy, July. Association for Computational Linguistics.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "CCG parsing algorithm with incremental tree rotation",
"authors": [
{
"first": "Milo\u0161",
"middle": [],
"last": "Stanojevi\u0107",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Steedman",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
"volume": "1",
"issue": "",
"pages": "228--239",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Milo\u0161 Stanojevi\u0107 and Mark Steedman. 2019. CCG parsing algorithm with incremental tree rotation. In Proceed- ings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 228-239, Minneapolis, Minnesota, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "The Syntactic Process",
"authors": [
{
"first": "Mark",
"middle": [],
"last": "Steedman",
"suffix": ""
}
],
"year": 2000,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mark Steedman. 2000. The Syntactic Process. MIT Press, Cambridge, MA, USA.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Learning entailment rules for unary templates",
"authors": [
{
"first": "Idan",
"middle": [],
"last": "Szpektor",
"suffix": ""
},
{
"first": "Ido",
"middle": [],
"last": "Dagan",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the 22nd International Conference on Computational Linguistics",
"volume": "",
"issue": "",
"pages": "849--856",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Idan Szpektor and Ido Dagan. 2008. Learning entailment rules for unary templates. In Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 849-856, Manchester, UK, August. Coling 2008 Organizing Committee.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Semeval-2013 task 1: Tempeval-3: Evaluating time expressions, events, and temporal relations",
"authors": [
{
"first": "Naushad",
"middle": [],
"last": "Uzzaman",
"suffix": ""
},
{
"first": "Hector",
"middle": [],
"last": "Llorens",
"suffix": ""
},
{
"first": "Leon",
"middle": [],
"last": "Derczynski",
"suffix": ""
},
{
"first": "James",
"middle": [],
"last": "Allen",
"suffix": ""
},
{
"first": "Marc",
"middle": [],
"last": "Verhagen",
"suffix": ""
},
{
"first": "James",
"middle": [],
"last": "Pustejovsky",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of the Seventh International Workshop on Semantic Evaluation",
"volume": "2",
"issue": "",
"pages": "1--9",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Naushad UzZaman, Hector Llorens, Leon Derczynski, James Allen, Marc Verhagen, and James Pustejovsky. 2013. Semeval-2013 task 1: Tempeval-3: Evaluating time expressions, events, and temporal relations. In Second Joint Conference on Lexical and Computational Semantics (* SEM), Volume 2: Proceedings of the Seventh International Workshop on Semantic Evaluation (SemEval 2013), volume 2, pages 1-9.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "A general framework for distributional similarity",
"authors": [
{
"first": "Julie",
"middle": [],
"last": "Weeds",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Weir",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "81--88",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Julie Weeds and David Weir. 2003. A general framework for distributional similarity. In Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing, pages 81-88.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Enriching large-scale eventuality knowledge graph with entailment relations",
"authors": [
{
"first": "Changlong",
"middle": [],
"last": "Yu",
"suffix": ""
},
{
"first": "Hongming",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Yangqiu",
"middle": [],
"last": "Song",
"suffix": ""
},
{
"first": "Wilfred",
"middle": [],
"last": "Ng",
"suffix": ""
},
{
"first": "Lifeng",
"middle": [],
"last": "Shang",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Changlong Yu, Hongming Zhang, Yangqiu Song, Wilfred Ng, and Lifeng Shang. 2020. Enriching large-scale eventuality knowledge graph with entailment relations. In Automated Knowledge Base Construction.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Crowdsourcing inference-rule evaluation",
"authors": [
{
"first": "Naomi",
"middle": [],
"last": "Zeichner",
"suffix": ""
},
{
"first": "Jonathan",
"middle": [],
"last": "Berant",
"suffix": ""
},
{
"first": "Ido",
"middle": [],
"last": "Dagan",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics",
"volume": "2",
"issue": "",
"pages": "156--160",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Naomi Zeichner, Jonathan Berant, and Ido Dagan. 2012. Crowdsourcing inference-rule evaluation. In Proceed- ings of the 50th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 156-160, Jeju Island, Korea, July. Association for Computational Linguistics.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "Harvesting parallel news streams to generate paraphrases of event relations",
"authors": [
{
"first": "Congle",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Daniel S Weld",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "1776--1786",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Congle Zhang and Daniel S Weld. 2013. Harvesting parallel news streams to generate paraphrases of event relations. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1776-1786.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"uris": null,
"text": "Example sentences (left) and their resulting (collapsed) entailment/non-entailment graph (right)",
"num": null,
"type_str": "figure"
},
"FIGREF1": {
"uris": null,
"text": "(a) 1 = entailment, 0 = non-entailment. Blue = base (entailments, and non-entailments from temporally disjoint outcomes), orange = directional non-entailment, green = para-Examples from the evaluation dataset",
"num": null,
"type_str": "figure"
},
"FIGREF2": {
"uris": null,
"text": "Effects of window size for the timexAndDocDate temporal information source",
"num": null,
"type_str": "figure"
},
"FIGREF3": {
"uris": null,
"text": "Best results on the base evaluation dataset: timexAndDocDate with a temporal window size of 4 days. For calculating AUC the recall threshold is set to < 0.75 for all similarity measures",
"num": null,
"type_str": "figure"
},
"TABREF2": {
"text": "Entailment pairs evaluation dataset",
"content": "<table/>",
"num": null,
"type_str": "table",
"html": null
},
"TABREF5": {
"text": "AUC scores for different subsets of the evaluation dataset. Temporal information source is timexAndDocDate, temporal window size is 4 days",
"content": "<table><tr><td>6 Discussion and Future Work</td></tr></table>",
"num": null,
"type_str": "table",
"html": null
}
}
}
} |