File size: 103,736 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 | {
"paper_id": "P13-1042",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T09:36:00.437972Z"
},
"title": "Large-scale Semantic Parsing via Schema Matching and Lexicon Extension",
"authors": [
{
"first": "Qingqing",
"middle": [],
"last": "Cai",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Temple University Computer and Information Sciences",
"location": {}
},
"email": "qingqing.cai@temple.edu"
},
{
"first": "Alexander",
"middle": [],
"last": "Yates",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Temple University",
"location": {}
},
"email": "yates@temple.edu"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Supervised training procedures for semantic parsers produce high-quality semantic parsers, but they have difficulty scaling to large databases because of the sheer number of logical constants for which they must see labeled training data. We present a technique for developing semantic parsers for large databases based on a reduction to standard supervised training algorithms, schema matching, and pattern learning. Leveraging techniques from each of these areas, we develop a semantic parser for Freebase that is capable of parsing questions with an F1 that improves by 0.42 over a purely-supervised learning algorithm.",
"pdf_parse": {
"paper_id": "P13-1042",
"_pdf_hash": "",
"abstract": [
{
"text": "Supervised training procedures for semantic parsers produce high-quality semantic parsers, but they have difficulty scaling to large databases because of the sheer number of logical constants for which they must see labeled training data. We present a technique for developing semantic parsers for large databases based on a reduction to standard supervised training algorithms, schema matching, and pattern learning. Leveraging techniques from each of these areas, we develop a semantic parser for Freebase that is capable of parsing questions with an F1 that improves by 0.42 over a purely-supervised learning algorithm.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Semantic parsing is the task of translating natural language utterances to a formal meaning representation language (Chen et al., 2010; Liang et al., 2009; Clarke et al., 2010; Liang et al., 2011; Artzi and Zettlemoyer, 2011) . There has been recent interest in producing such semantic parsers for large, heterogeneous databases like Freebase (Krishnamurthy and Mitchell, 2012; Cai and Yates, 2013) and Yago2 (Yahya et al., 2012) , which has driven the development of semi-supervised and distantlysupervised training methods for semantic parsing. Previous purely-supervised approaches have been limited to smaller domains and databases, such as the GeoQuery database, in part because of the cost of labeling enough samples to cover all of the logical constants involved in a domain.",
"cite_spans": [
{
"start": 116,
"end": 135,
"text": "(Chen et al., 2010;",
"ref_id": "BIBREF7"
},
{
"start": 136,
"end": 155,
"text": "Liang et al., 2009;",
"ref_id": "BIBREF20"
},
{
"start": 156,
"end": 176,
"text": "Clarke et al., 2010;",
"ref_id": "BIBREF9"
},
{
"start": 177,
"end": 196,
"text": "Liang et al., 2011;",
"ref_id": "BIBREF21"
},
{
"start": 197,
"end": 225,
"text": "Artzi and Zettlemoyer, 2011)",
"ref_id": "BIBREF0"
},
{
"start": 362,
"end": 377,
"text": "Mitchell, 2012;",
"ref_id": null
},
{
"start": 378,
"end": 398,
"text": "Cai and Yates, 2013)",
"ref_id": "BIBREF4"
},
{
"start": 409,
"end": 429,
"text": "(Yahya et al., 2012)",
"ref_id": "BIBREF29"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "This paper investigates a reduction of the problem of building a semantic parser to three standard problems in semantics and machine learning: supervised training of a semantic parser, schema matching, and pattern learning. Figure 1 provides a visualization of our system architecture. We apply an existing supervised training algorithm for semantic parsing to a labeled data set. We ( Figure 1: We reduce the task of learning a largescale semantic parser to a combination of 1) a standard supervised algorithm for learning semantic parsers; 2) our MATCHER algorithm for finding correspondences between words and database symbols; and 3) our LEXTENDER algorithm for integrating (word, database symbol) matches into a semantic parsing lexicon.",
"cite_spans": [
{
"start": 384,
"end": 385,
"text": "(",
"ref_id": null
}
],
"ref_spans": [
{
"start": 224,
"end": 232,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "apply schema matching techniques to the problem of finding correspondences between English words w and ontological symbols s. And we apply pattern learning techniques to incorporate new (w, s) pairs into the lexicon of the trained semantic parser. This reduction allows us to apply standard techniques from each problem area, which in combination provide a large improvement over the purely-supervised approaches. On a dataset of 917 questions taken from 81 domains of the Freebase database, a standard learning algorithm for semantic parsing yields a parser with an F1 of 0.21, in large part because of the number of logical symbols that appear during testing but never appear during training. Our techniques can extend this parser to new logical symbols through schema matching, and yield a semantic parser with an F1 of 0.63 on the same task. On a more challenging task where training and test data are divided so that all logical constants in test are never observed dur-ing training, our approach yields a semantic parser with an F1 of 0.6, whereas the purely supervised approach cannot parse a single test question correctly. These results indicate that it is possible to automatically extend semantic parsers to symbols for which little or no training data has been observed.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The rest of this paper is organized as follows. The next section discusses related work. Section 3 describes our MATCHER algorithm for performing schema matching between a knowledge base and text. Section 4 explains how we use MATCHER's schema matching to extend a standard semantic parser to logical symbols for which it has seen no labeled training data. Section 5 analyzes the performance of MATCHER and our semantic parser. Section 6 concludes.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Two existing systems translate between natural language questions and database queries over large-scale databases. Yahya et al. (2012) report on a system for translating natural language queries to SPARQL queries over the Yago2 (Hoffart et al., 2013) database. Yago2 consists of information extracted from Wikipedia, WordNet, and other resources using manually-defined extraction patterns. The manual extraction patterns pre-define a link between natural language terms and Yago2 relations. Our techniques automate the process of identifying matches between textual phrases and database relation symbols, in order to scale up to databases with more relations, like Freebase. A more minor difference between Yahya et al.'s work and ours is that their system handles SPARQL queries, which do not handle aggregation queries like argmax and count. We rely on an existing semantic parsing technology to learn the language that will translate into such aggregation queries. On the other hand, their test questions involve more conjunctions and complex semantics than ours. Developing a dataset with more complicated semantics in the queries is part of our ongoing efforts. Krishnamurthy and Mitchell (2012) also create a semantic parser for Freebase covering 77 of Freebase's over 2000 relations. Like our work, their technique uses distant supervision to drive training over a collection of sentences gathered from the Web, and they do not require any manually-labeled training data. However, their technique does require manual specification of rules that construct CCG lexical entries from dependency parses. In comparison, we fully automate the process of constructing CCG lexical entries for the semantic parser by making it a prediction task. We also leverage synonym-matching techniques for comparing relations extracted from text with Freebase relations. Finally, we test our results on a dataset of 917 questions covering over 600 Freebase relations, a more extensive test than the 50 questions used by Krishnamurthy and Mitchell. Numerous methods exist for comparing two relations based on their sets of tuples. For instance, the DIRT system (Lin and Pantel, 2001) uses the mutual information between the (X, Y ) argument pairs for two binary relations to measure the similarity between them, and clusters relations accordingly. More recent examples of similar techniques include the Resolver system (Yates and Etzioni, 2009) and Poon and Domingos's USP system (Poon and Domingos, 2009) . Our techniques for comparing relations fit into this line of work, but they are novel in their application of these techniques to the task of comparing database relations and relations extracted from text. Schema matching (Rahm and Bernstein, 2001; Ehrig et al., 2004; Giunchiglia et al., 2005 ) is a task from the database and knowledge representation community in which systems attempt to identify a \"common schema\" that covers the relations defined in a set of databases or ontologies, and the mapping between each individual database and the common schema. Owing to the complexity of the general case, researchers have resorted to defining standard similarity metrics between relations and attributes, as well as machine learning algorithms for learning and predicting matches between relations Wick et al., 2008b; Wick et al., 2008a; Nottelmann and Straccia, 2007; Berlin and Motro, 2006) . These techniques consider only matches between relational databases, whereas we apply these ideas to matches between Freebase and extracted relations. Schema matching in the database sense often considers complex matches between relations (Dhamanka et al., 2004) , whereas as our techniques are currently restricted to matches involving one database relation and one relation extracted from text.",
"cite_spans": [
{
"start": 115,
"end": 134,
"text": "Yahya et al. (2012)",
"ref_id": "BIBREF29"
},
{
"start": 228,
"end": 250,
"text": "(Hoffart et al., 2013)",
"ref_id": "BIBREF16"
},
{
"start": 1167,
"end": 1200,
"text": "Krishnamurthy and Mitchell (2012)",
"ref_id": null
},
{
"start": 2006,
"end": 2033,
"text": "Krishnamurthy and Mitchell.",
"ref_id": null
},
{
"start": 2146,
"end": 2168,
"text": "(Lin and Pantel, 2001)",
"ref_id": "BIBREF22"
},
{
"start": 2404,
"end": 2429,
"text": "(Yates and Etzioni, 2009)",
"ref_id": "BIBREF30"
},
{
"start": 2465,
"end": 2490,
"text": "(Poon and Domingos, 2009)",
"ref_id": "BIBREF24"
},
{
"start": 2715,
"end": 2741,
"text": "(Rahm and Bernstein, 2001;",
"ref_id": "BIBREF25"
},
{
"start": 2742,
"end": 2761,
"text": "Ehrig et al., 2004;",
"ref_id": "BIBREF12"
},
{
"start": 2762,
"end": 2786,
"text": "Giunchiglia et al., 2005",
"ref_id": "BIBREF14"
},
{
"start": 3292,
"end": 3311,
"text": "Wick et al., 2008b;",
"ref_id": "BIBREF28"
},
{
"start": 3312,
"end": 3331,
"text": "Wick et al., 2008a;",
"ref_id": "BIBREF27"
},
{
"start": 3332,
"end": 3362,
"text": "Nottelmann and Straccia, 2007;",
"ref_id": "BIBREF23"
},
{
"start": 3363,
"end": 3386,
"text": "Berlin and Motro, 2006)",
"ref_id": "BIBREF2"
},
{
"start": 3628,
"end": 3651,
"text": "(Dhamanka et al., 2004)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Previous Work",
"sec_num": "2"
},
{
"text": "The textual schema matching task is to identify natural language words and phrases that correspond with each relation and entity in a fixed schema for a relational database. To formalize this task, we first introduce some notation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Formulation",
"sec_num": "3.1"
},
{
"text": "A schema S = (E, R, C, I) consists of a set of entities E, a set of relations R, a set of categories C, and a set of instances I. Categories are one-argument predicates (e.g., film(e)), and relations are two-(or more-) argument predicates (e.g., directed by(e 1 , e 2 )).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Formulation",
"sec_num": "3.1"
},
{
"text": "Instances are known tuples of entities that make a relation or category true, such as film(Titanic) or directed by(Titanic, James Cameron). For a given r \u2208 R (or c \u2208 C), I S (r) indicates the set of known instances of r in schema S (and likewise for I S (c)). Examples of such schemas include Freebase (Bollacker et al., 2008) and Yago2 (Hoffart et al., 2013) . We say a schema is a textual schema if it has been extracted from free text, such as the Nell (Carlson et al., 2010) and ReVerb (Fader et al., 2011) extracted databases.",
"cite_spans": [
{
"start": 302,
"end": 326,
"text": "(Bollacker et al., 2008)",
"ref_id": "BIBREF3"
},
{
"start": 337,
"end": 359,
"text": "(Hoffart et al., 2013)",
"ref_id": "BIBREF16"
},
{
"start": 456,
"end": 478,
"text": "(Carlson et al., 2010)",
"ref_id": null
},
{
"start": 490,
"end": 510,
"text": "(Fader et al., 2011)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Formulation",
"sec_num": "3.1"
},
{
"text": "Given a textual schema T and a database schema D, the textual schema matching task is to identify an alignment or matching M \u2282 R T \u00d7 R D such that (r T , r D ) \u2208 M if and only if r T can be used to refer to r D in normal language usage. The problem would be greatly simplified if M were a 1-1 function, but in practice most database relations can be referred to in many ways by natural language users: for instance, film actor can be referenced by the English verbs \"played,\" \"acted,\" and \"starred,\" along with morphological variants of them. In addition, many English verbs can refer to several different relations in Freebase: \"make\" can refer to computer processor manufacturer or distilled spirits producer, among many others. Our MATCHER algorithm for textual schema matching handles this by producing a confidence score for every possible (r T , r D ) pair, which downstream applications can then use to reason about the possible alignments.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Formulation",
"sec_num": "3.1"
},
{
"text": "Even worse than the ambiguities in alignment, some textual relations do not correspond with any database relation exactly, but instead they correspond with a projection of a relation, or a join between multiple relations, or another com-plex view of a database schema. As a simple example, \"actress\" corresponds to a subset of the Freebase film actor relation that intersects with the set {x: gender(x, female)}. MATCHER can only determine that \"actress\" aligns with film actor or not; it cannot produce an alignment between \"actress\" and a join of film actor and gender. These more complex alignments are an important consideration for future work, but as our experiments will show, quite useful alignments can be produced without handling these more complex cases.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Problem Formulation",
"sec_num": "3.1"
},
{
"text": "MATCHER uses a generate-and-test architecture for determining M . It uses a Web search engine to issue queries for a database relation r D consisting of all the entities in a tuple t \u2208 I D (r D ). 1000 tuples for each r D are randomly chosen for issuing queries. The system then retrieves matching snippets from the search engine results. It uses the top 10 results for each search engine query. It then counts the frequency of each word type in the set of retrieved snippets for r D . The top 500 nonstopword word types are chosen as candidates for matches with r D . We denote the candidate set for",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Identifying candidate matches",
"sec_num": "3.2"
},
{
"text": "r D as C(r D ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Identifying candidate matches",
"sec_num": "3.2"
},
{
"text": "MATCHER's threshold of 500 candidates for C(r D ) results in a maximum possible recall of just less than 0.8 for the alignments in our dataset, but even if we double the threshold to 1000, the recall improves only slightly to 0.82. We therefore settled on 500 as a point with an acceptable upper bound on recall, while also producing an acceptable number of candidate terms for further processing.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Identifying candidate matches",
"sec_num": "3.2"
},
{
"text": "The candidate pool C(r D ) of 500 word types is significantly smaller than the set of all textual relations, but it is also extremely noisy. The candidates may include non-relation words, or other frequent but unrelated words. They may also include words that are highly related to r D , but not actually corresponding textual relations. For instance, the candidate set for film director in Freebase includes words like \"directed,\" but also words like \"film,\" \"movie,\" \"written,\" \"produced,\" and \"starring.\" We use a series of filters based on synonym-detection techniques to help select the true matching candidates from C(r D ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pattern-based match selection",
"sec_num": "3.3"
},
{
"text": "1. \"r T in E\" r T ends with \"-ed\" and E has type datetime or location \"founded in 1989\" 2. \"r T by E\" r T ends with \"-ed\" \"invented by Edison\" 3. \"r T such as E\" r T ends with \"-s\" \"directors such as Tarantino\" 4. \"E is a(n) r T \" all cases \"Paul Rudd is an actor\" Table 1 : Patterns used by MATCHER as evidence of a match between r D and r T . E represents an entity randomly selected from the tuples in",
"cite_spans": [],
"ref_spans": [
{
"start": 265,
"end": 272,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Condition Example",
"sec_num": null
},
{
"text": "I D (r D ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Condition Example",
"sec_num": null
},
{
"text": "The first type of evidence we consider for identifying true matches from C(r D ) consists of pattern-matching. Relation words that express r D will often be found in complex grammatical constructions, and often they will be separated from their entity arguments by long-distance dependencies. However, over a large corpus, one would expect that in at least some cases, the relation word will appear in a simple, relatively-unambiguous grammatical construction that connects r T with entities from r D . For instance, entities e from the relationship automotive designer appear in the pattern \"designed by e\" more than 100 times as often as the next most-common patterns, \"considered by e\" and \"worked by e.\"",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Condition Example",
"sec_num": null
},
{
"text": "MATCHER use searches over the Web to count the number of instances where a candidate r T appears in simple patterns that involve entities from r D . Greater counts for these patterns yield greater evidence of a correct match between r D and r T . Table 1 provides a list of patterns that we consider. For each r D and each r T \u2208 C(r D ), MATCHER randomly selects 10 entities from r D 's tuples to include in its pattern queries. Two of the patterns are targeted at past-tense verbs, and the other two patterns at nominal relation words.",
"cite_spans": [],
"ref_spans": [
{
"start": 247,
"end": 254,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Condition Example",
"sec_num": null
},
{
"text": "MATCHER computes statistics similar to pointwise mutual information (PMI) (Turney, 2001) to measure how related r D and r T are, for each pattern p. Let c(p, r D , r T ) indicate the sum of all the counts for a particular pattern p, database relation, and textual relation:",
"cite_spans": [
{
"start": 74,
"end": 88,
"text": "(Turney, 2001)",
"ref_id": "BIBREF26"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Condition Example",
"sec_num": null
},
{
"text": "f p (r T , r D ) = c(p, r D , r T ) r D c(p, r D , r T ) * r T c(p, r D , r T )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Condition Example",
"sec_num": null
},
{
"text": "For the sum over all r D , we use all r D in Freebase for which r T was extracted as a candidate. One downside of the pattern-matching evidence is the sheer number of queries it requires. Freebase currently has over 2,000 relations. For each r D , we have up to 500 candidate r T , up to 4 patterns, and up to 10 entities per pattern. To cover all of Freebase, MATCHER needs 2, 000\u00d7500\u00d74\u00d710 = 40 million queries, or just over 1.25 years if it issues 1 query per second (we covered approximately one-quarter of Freebase's relations in our experiments). Using more patterns and more entities per pattern are desirable for accumulating more evidence about candidate matches, but there is a trade-off with the time required to issue the necessary queries.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Condition Example",
"sec_num": null
},
{
"text": "Open Information Extraction (Open IE) systems (Banko et al., 2007) can often provide a large set of extracted tuples for a given r T , which MATCHER can then use to make much more comprehensive comparisons with the full tuple set for r D than the pattern-matching technique allows. MATCHER employs a form of PMI to compute the degree of relatedness between r D and r T . In its simplest form, MATCHER computes:",
"cite_spans": [
{
"start": 46,
"end": 66,
"text": "(Banko et al., 2007)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Comparing database relations with extracted relations",
"sec_num": "3.4"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "P M I(r T , r D ) = |I D (r D ) \u2229 I T (r T )| |I D (r D )| \u2022 |I T (r T )|",
"eq_num": "(1)"
}
],
"section": "Comparing database relations with extracted relations",
"sec_num": "3.4"
},
{
"text": "While this PMI statistic is already quite useful, we have found that in practice there are many cases where an exact match between tuples in I D (r D ) and tuples in I T (r T ) is too strict of a criterion. MATCHER uses a variety of approximate matches to compute variations of this statistic. Considered as predictors for the true matches in M , these variations of the PMI statistic have a lower precision, in that they are more likely to have high values for incorrect matches. However, they also have a higher recall: that is, they will have a high value for correct candidates in C(r D ) when the strict version of PMI does not. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparing database relations with extracted relations",
"sec_num": "3.4"
},
{
"text": "s \u03ba (r T , r D ) = t D \u2208I D (r D ) t T \u2208I T (r T ) \u03ba(t D , t T ) |I D (r D )|\u2022|I T (r T )| s \u03ba (r T , r D ) = s\u03ba(r T ,r D ) r D s \u03ba (r D , r T ) s (r T , r D ) = |I T (r T )| |I D (r D )|",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparing database relations with extracted relations",
"sec_num": "3.4"
},
{
"text": "\u03ba(t 1 , t 2 ) for comparing tuples t 1 , t 2 strict match: 1, if \u03a6(t 1 ) = \u03a6 (t 2 ) 0, otherwise.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparing database relations with extracted relations",
"sec_num": "3.4"
},
{
"text": "type match: Table 3 : MATCHER's \u03ba functions for computing whether two tuples are similar. cat maps an entity to a category (or type) in the schema. MATCHER has a different \u03ba function for each possible combination of \u03a6 and \u03a6 functions, which are given in Table 4. MATCHER uses an API for the ReVerb Open IE system 1 (Fader et al., 2011) to collect I(r T ), for each r T . The API for ReVerb allows for relational queries in which some subset of the entity strings, entity categories, and relation string are specified. The API returns all matching triples; types must match exactly, but relation or argument strings in the query will match any relation or argument that contains the query string as a substring. MATCHER queries ReVerb with three different types of queries for each r T , specifying the types for both arguments, or just the type of the first argument, or just the second argument. Types for arguments are taken from the types of arguments for a potentially matching r D in Freebase. To avoid overwhelming the ReVerb servers, for our experiments we limited MATCHER to queries 1 http://openie.cs.washington.edu/ \u03a6(t) for tuple t = (e 1 , . . . , e n ) \u2200 i e i (projection to one dimension) (e 1 , . . . , e n ) (full tuple) \u2200 \u03c3(\u2022) (e \u03c3(1) , . . . , e \u03c3(n) ) (permutation) for the top 80 r T \u2208 C(r D ), when they are ranked according to frequency during the candidate identification process.",
"cite_spans": [
{
"start": 315,
"end": 335,
"text": "(Fader et al., 2011)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [
{
"start": 12,
"end": 19,
"text": "Table 3",
"ref_id": null
},
{
"start": 254,
"end": 262,
"text": "Table 4.",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Comparing database relations with extracted relations",
"sec_num": "3.4"
},
{
"text": "\uf8f1 \uf8f4 \uf8f2 \uf8f4 \uf8f3 1, if \u2200 k cat(\u03a6(t 1 ) k ) = cat(\u03a6 (t 2 ) k ) 0, otherwise.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparing database relations with extracted relations",
"sec_num": "3.4"
},
{
"text": "Pattern statistics, the ReVerb statistics from Table 2, and the count of r T during the candidate identification step all provide evidence for correct matches between r D and r T . MATCHER uses a regression model to combine these various statistics into a score for (r T , r D ). The regression model is a linear regression with least-squares parameter estimation; we experimented with support vector regression models with non-linear kernels, with no significant improvements in accuracy. Section 5 explains the dataset we use to train this model. Unlike a classifier, MATCHER does not output any single matching M . However, downstream applications can easily convert MATCHER's output into a matching M by, for instance, selecting the top K candidate r T values for each r D , or by selecting all (r T , r D ) pairs with a score over a chosen threshold. Our experiments analyze MATCHER's success by comparing its performance across a range of different values for the number of r T matches for each r D .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Regression models for scoring candidates",
"sec_num": "3.5"
},
{
"text": "An alignment between textual relations and database relations has many possible uses: for example, it might be used to allow queries over a database to be answered using additional information stored in an extracted relation store, or it might be used to deduce clusters of synonymous relation words in English. Here, we describe an application in which we build a questionanswering system for Freebase by extending a standard learning technique for semantic parsing with schema alignment information. As a starting point, we used the UBL system developed by Kwiatkowski et al. (2010) to learn a semantic parser based on probabilistic Combinatory Categorial Grammar (PCCG). Source code for UBL is freely available. Its authors found that it achieves results competitive with the state-of-the-art on a variety of standard semantic parsing data sets, including Geo250 English (0.85 F1). Using a fixed CCG grammar and a procedure based on unification in second-order logic, UBL learns a lexicon \u039b from the training data which includes entries like:",
"cite_spans": [
{
"start": 559,
"end": 584,
"text": "Kwiatkowski et al. (2010)",
"ref_id": "BIBREF19"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Extending a Semantic Parser Using a Schema Alignment",
"sec_num": "4"
},
{
"text": "Example Lexical Entries New York City N P : new york neighborhoods in S\\N P/N P : \u03bbx\u03bby.neighborhoods(x, y)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Extending a Semantic Parser Using a Schema Alignment",
"sec_num": "4"
},
{
"text": "Example CCG Grammar Rules",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Extending a Semantic Parser Using a Schema Alignment",
"sec_num": "4"
},
{
"text": "X/Y : f Y : g \u21d2 X : f (g) Y : g X\\Y : f \u21d2 X : f (g)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Extending a Semantic Parser Using a Schema Alignment",
"sec_num": "4"
},
{
"text": "Using \u039b, UBL selects a logical form z for a sentence S by selecting the z with the most likely parse derivations y: h(S) = arg max z y p(y, z|x; \u03b8, \u039b). The probabilistic model is a log-linear model with features for lexical entries used in the parse, as well as indicator features for relation-argument pairs in the logical form, to capture selectional preferences. Inference (parsing) and parameter estimation are driven by standard dynamic programming algorithms (Clark and Curran, 2007) , while lexicon induction is based on a novel search procedure through the space of possible higher-order logic unification operations that yield the desired logical form for a training sentence.",
"cite_spans": [
{
"start": 465,
"end": 489,
"text": "(Clark and Curran, 2007)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Extending a Semantic Parser Using a Schema Alignment",
"sec_num": "4"
},
{
"text": "Our Freebase data covers 81 of the 86 core domains in Freebase, and 635 of its over 2000 relations, but we wish to develop a semantic parser that can scale to all of Freebase. UBL gets us part of the way there, by inducing a PCCG grammar, as well as lexical entries for function words that must be handled in all domains. It can also learn lexical entries for relations r D that appear in the training data. However, UBL has no way to learn lexical entries for the many valid (r T , r D ) pairs that do not appear during training.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Extending a Semantic Parser Using a Schema Alignment",
"sec_num": "4"
},
{
"text": "We use MATCHER's learned alignment to extend the semantic parser that we get from UBL by automatically adding in lexical entries for Free-base relations. Essentially, for each (r T , r D ) from MATCHER's output, we wish to construct a lexical entry that states that r T 's semantics resembles \u03bbx\u03bby.r D (x, y). However, this simple process is complicated by the fact that the semantic parser requires two additional types of information for each lexical entry: a syntactic category, and a weight. Furthermore, for many cases the appropriate semantics are significantly more complex than this pattern.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Extending a Semantic Parser Using a Schema Alignment",
"sec_num": "4"
},
{
"text": "To extend the learned semantic parser to a semantic parser for all of Freebase, we introduce a prediction task, which we call semantic lexicon extension: given a matching M together with scores for each pair in M , predict the syntactic category Syn, lambda-calculus semantics Sem, and weight W for a full lexical entry for each (r T , r D ) \u2208 M . One advantage of the reduction approach to learning a semantic parser is that we can automatically construct training examples for this prediction task from the other components in the reduction. We use the output lexical entries learned by UBL as (potentially noisy) examples of true lexical entries for (r T , r D ) pairs where r T matches the word in one of UBL's lexical entries, and r D forms part of the semantics in the same lexical entry. For (r T , r D ) pairs in M where r D occurs in UBL's lexical entries, but not paired with r T , we create dummy \"negative\" lexical entries with very low weights, one for each possible syntactic category observed in all lexical entries. Note that in order to train LEXTENDER, we need the output of MATCHER for the relations in UBL's training data, as well as UBL's output lexicon from the training data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Extending a Semantic Parser Using a Schema Alignment",
"sec_num": "4"
},
{
"text": "Our system for this prediction task, which we call LEXTENDER (for Lexicon eXtender), factors into three components: P (Sem|r D , r T , score), P (Syn|Sem, r D , r T , score), and P (W |Syn, Sem, r D , r T , score). This factorization is trivial in that it introduces no independence assumptions, but it helps in designing models for the task. We set the event space for random variable Sem to be the set of all lambda calculus expressions observed in UBL's output lexicon, modulo the names of specific Freebase relations. For instance, if the lexicon includes two entries whose semantics are \u03bbx\u03bby . film actor(x, y) and \u03bbx\u03bby . book author(x, y), the event space would include the single expression in which relations film actor and book author were replaced by a new variable: \u03bbp\u03bbx\u03bby.p(x, y). The final semantics for a lexical entry is then constructed by substituting r D for p, or more formally, by a function application Sem(r D ). The event space for Syn consists of all syntactic categories in UBL's output lexicon, and W ranges over R. LEXTENDER's model for Sem and Syn are Na\u00efve Bayes classifiers (NBC), with features for the part-of-speech for r T (taken from a POS tagger), the suffix of r T , the number of arguments of r D , and the argument types of r D . For Syn, we add a feature for the predicted value of Sem. For W , we use a linear regression model whose features are the score from MATCHER, the probabilities from the Syn and Sem NBC models, and the average weight of all lexical entries in UBL with matching syntax and semantics. Using the predictions from these models, LEXTENDER extends UBL's learned lexicon with all possible lexical entries with their predicted weights, although typically only a few lexical entries have high enough weight to make a difference during parsing. Pruning entries with low weights could improve the memory and time requirements for parsing, but these were not an issue in our experiments, so we did not investigate this further.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Extending a Semantic Parser Using a Schema Alignment",
"sec_num": "4"
},
{
"text": "We conducted experiments to test the ability of MATCHER and LEXTENDER to produce a semantic parser for Freebase. We first analyze MATCHER on the task of finding matches between Freebase relations and textual relations. We then compare the performance of the semantic parser learned by UBL with its extension provided by LEXTENDER on a dataset of English questions posed to Freebase.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "5"
},
{
"text": "Freebase (Bollacker et al., 2008 ) is a free, online, user-contributed, relational database (www.freebase.com) covering many different domains of knowledge. The full schema and contents are available for download. The \"Freebase Commons\" subset of Freebase, which is our focus, consists of 86 domains, an average of 25 relations per domain (total of 2134 relations), and 615,000 known instances per domain (53 million instances total). As a reference point, the GeoQuery database -which is a standard benchmark database for semantic parsing -Examples 1. What are the neighborhoods in New York City? \u03bbx . neighborhoods(new york, x) 2. How many countries use the rupee?",
"cite_spans": [
{
"start": 9,
"end": 32,
"text": "(Bollacker et al., 2008",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Setup",
"sec_num": "5.1"
},
{
"text": "count(x) . countries used(rupee, x)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental Setup",
"sec_num": "5.1"
},
{
"text": "3. How many Peabody Award winners are there? count(x) . \u2203y . award honor(y) \u2227 award winner(y, x) \u2227 award(y, peabody award) Figure 2 : Example questions with their logical forms. The logical forms make use of Freebase symbols as logical constants, as well as a few additional symbols such as count and argmin, to allow for aggregation queries.",
"cite_spans": [],
"ref_spans": [
{
"start": 123,
"end": 131,
"text": "Figure 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Experimental Setup",
"sec_num": "5.1"
},
{
"text": "contains a single domain (geography), 8 relations, and 880 total instances. Our dataset contains 917 questions (on average, 6.3 words per question) and a meaning representation for each question written in a variant of lambda calculus 2 . 81 domains are represented in the data set, and the lambda calculus forms contain 635 distinct Freebase relations. The most common domains, film and business, each took up no more than 6% of the overall dataset. Several examples are listed in Fig. 2 . The questions were provided by two native English speakers. No restrictions were placed on the type of questions they should produce, except that they should produce questions for multiple domains. By inspection, a large majority of the questions appear to be answerable from Freebase, although no instructions were given to restrict questions to this sort. We also created a dataset of alignments from these annotated questions by creating an alignment for each Freebase relation mentioned in the logical form for a question, paired with a manually-selected word from the question.",
"cite_spans": [],
"ref_spans": [
{
"start": 482,
"end": 488,
"text": "Fig. 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Experimental Setup",
"sec_num": "5.1"
},
{
"text": "We measured the precision and recall of MATCHER's output against the manually labeled data. Let M be the set of (r T , r D ) matches produced by the system, and G the set of matches in the gold-standard manual data. We define precision and recall as: Figure 3 shows a Precision-Recall (PR) curve for MATCHER and three baselines: a \"Frequency\" model that ranks candidate matches for r D by their frequency during the candidate identification step; a \"Pattern\" model that uses MATCHER's linear regression model for ranking, but is restricted to only the pattern-based features; and an \"Extractions\" model that similarly restricts the ranking model to ReVerb features. We have three folds in our data; the alignments for relation r D in one fold are predicted by models trained on the other two folds. Once all of the alignments in all three folds are scored, we generate points on the PR curve by applying a threshold to the model's ranking, and treating all alignments above the threshold as the set of predicted alignments.",
"cite_spans": [],
"ref_spans": [
{
"start": 251,
"end": 259,
"text": "Figure 3",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Alignment Tests",
"sec_num": "5.2"
},
{
"text": "P = |M \u2229 G| |M | , R = |M \u2229 G| |G|",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Alignment Tests",
"sec_num": "5.2"
},
{
"text": "All regression models for learning alignments outperform the Frequency ranking by a wide margin. The Pattern model outperforms the Extractions model at the high-precision, low-recall end of the curve. At the high-recall points, the Pattern model drops quickly in precision. However, the combination of the two kinds of features in MATCHER yields improved precision at all levels of recall.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Alignment Tests",
"sec_num": "5.2"
},
{
"text": "While our alignment tests can tell us in relative terms how well different models are performing, it is difficult to assess these models in absolute terms, since alignments are not typical applications that people care about in their own right. We now compare our alignments on a semantic parsing task for Freebase.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing Tests",
"sec_num": "5.3"
},
{
"text": "In a first semantic parsing experiment, we train UBL, MATCHER, and LEXTENDER on a random sample of 70% of the questions, and test them on the remaining 30%. In a second test, we focus on the hard case where all questions from the test set contain logical constants that have never been seen before during training. We split the data into 3 folds, making sure that no Freebase domain has symbols appearing in questions in more than one fold. We then perform 3-fold crossvalidation for all of our supervised models. We varied the number of matches that the alignment model (MATCHER, Pattern, Extractions, or Frequency) could make for each Freebase relation, and measured semantic parsing performance as a function of the number of matches. Figure 4 shows the F1 scores for these semantic parsers, judged by exact match between the top-scoring logical form from the parser and the manually-produced logical form. Exact-match tests are overly-strict, in the sense that the system may be judged incorrect even when the logical form that is produced is logically equivalent to the correct logical form. However, by inspection such cases appear to be very rare in our data, and the exact-match criterion is often used in other semantic parsing experimental settings.",
"cite_spans": [],
"ref_spans": [
{
"start": 738,
"end": 746,
"text": "Figure 4",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Semantic Parsing Tests",
"sec_num": "5.3"
},
{
"text": "The semantic parsers produced by MATCHER+LEXTENDER and the other alignment techniques significantly outperform the baseline semantic parser learned by UBL, which achieves an overall F1 of 0.21 on these questions in the 70/30 split of the data, and an F1 of 0 in the cross-domain experiment.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing Tests",
"sec_num": "5.3"
},
{
"text": "Purely-supervised approaches to this data are severely limited, since they have almost no chance of correctly parsing questions that refer to logical symbols that never appeared during training. However, MATCHER and LEXTENDER combine with UBL to produce an effective semantic parser. The best semantic parser we tested, which was produced by UBL, MATCHER, and LEXTENDER with 9 matches per Freebase relation, had a precision of 0.67 and a recall of 0.59 on the 70/30 split experiment.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing Tests",
"sec_num": "5.3"
},
{
"text": "The difference in alignment performance between MATCHER, Pattern, and Extractions carries over to semantic parsing. MATCHER drops in F1 with more matches as additional matches tend to be low-quality and low-probability, whereas Pat- tern and Extractions keep improving as more lowprobability alignments are added. Interestingly, the Extractions model begins to overtake the Pattern model in F1 at higher numbers of matches, and all three models trend toward convergence in F1 with increasing numbers of matches. Nevertheless, MATCHER clearly improves over both, and reaches a higher F1 than either Pattern or Extractions using a small number of matches, which corresponds to a smaller lexicon and a leaner model. To place these results in context, many different semantic parsers for databases like GeoQuery and ATIS (including parsers produced by UBL) have achieved F1 scores of 0.85 and higher. However, in all such tests, the test questions refer to logical constants that also appeared during training, allowing supervised techniques for learning semantic parsers to achieve strong accuracy. As we have argued, Freebase is large enough that is difficult to produce enough labeled training data to cover all of its logical constants. An unsupervised semantic parser for GeoQuery has achieved an F1 score of 0.66 (Goldwasser et al., 2011) , impressive in its own right and slightly better than our F1 score. However, this parser was given questions which it knew a priori to contain words that refer to the logical constants in the database. Our MATCHER and LEXTENDER systems address a different challenge: how to learn a semantic parser for Freebase given the Web and a set of initial labeled questions.",
"cite_spans": [
{
"start": 1315,
"end": 1340,
"text": "(Goldwasser et al., 2011)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Semantic Parsing Tests",
"sec_num": "5.3"
},
{
"text": "Scaling semantic parsing to large databases requires an engineering effort to handle large datasets, but also novel algorithms to extend se-mantic parsing models to testing examples that look significantly different from labeled training data. The MATCHER and LEXTENDER algorithms represent an initial investigation into such techniques, with early results indicating that semantic parsers can handle Freebase questions on a large variety of domains with an F1 of 0.63.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "We hope that our techniques and datasets will spur further research into this area. In particular, more research is needed to handle more complex matches between database and textual relations, and to handle more complex natural language queries. As mentioned in section 3.1, words like \"actress\" cannot be addressed by the current methodology, since MATCHER assumes that a word maps to a single Freebase relation, but the closest Freebase equivalent to the meaning of \"actress\" involves the two relations film actor and gender. Another limitation is that our current methodology focuses on finding matches for nouns and verbs. Other important limitations of the current methodology include:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "\u2022 the assumption that function words have no domain-specific meaning, which prepositions in particular can violate;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "\u2022 low accuracy when there are few relevant results among the set of extracted relations;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "\u2022 and the restriction to a single database (Freebase) for finding answers.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "While significant challenges remain, the reduction of large-scale semantic parsing to a combination of schema matching and supervised learning offers a new path toward building high-coverage semantic parsers.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "The data is available from the second author's website.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Bootstrapping Semantic Parsers from Conversations",
"authors": [
{
"first": "Yoav",
"middle": [],
"last": "Artzi",
"suffix": ""
},
{
"first": "Luke",
"middle": [],
"last": "Zettlemoyer",
"suffix": ""
}
],
"year": 2011,
"venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yoav Artzi and Luke Zettlemoyer. 2011. Bootstrap- ping Semantic Parsers from Conversations. In Pro- ceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP).",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Open information extraction from the web",
"authors": [
{
"first": "M",
"middle": [],
"last": "Banko",
"suffix": ""
},
{
"first": "M",
"middle": [
"J"
],
"last": "Cafarella",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Soderland",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Broadhead",
"suffix": ""
},
{
"first": "O",
"middle": [],
"last": "Etzioni",
"suffix": ""
}
],
"year": 2007,
"venue": "IJCAI",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M. Banko, M. J. Cafarella, S. Soderland, M. Broad- head, and O. Etzioni. 2007. Open information ex- traction from the web. In IJCAI.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Database schema matching using machine learning with feature selection",
"authors": [
{
"first": "Jacob",
"middle": [],
"last": "Berlin",
"suffix": ""
},
{
"first": "Amihai",
"middle": [],
"last": "Motro",
"suffix": ""
}
],
"year": 2006,
"venue": "Advanced Information Systems Engineering",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jacob Berlin and Amihai Motro. 2006. Database schema matching using machine learning with fea- ture selection. In Advanced Information Systems Engineering. Springer.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"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 International Conference on Management of Data (SIG-MOD)",
"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 col- laboratively created graph database for structuring human knowledge. In Proceedings of the Interna- tional Conference on Management of Data (SIG- MOD), pages 1247-1250.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Semantic Parsing Freebase: Towards Open-Domain Semantic Parsing",
"authors": [
{
"first": "Qingqing",
"middle": [],
"last": "Cai",
"suffix": ""
},
{
"first": "Alexander",
"middle": [],
"last": "Yates",
"suffix": ""
}
],
"year": 2013,
"venue": "Second Joint Conference on Lexical and Computational Semantics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Qingqing Cai and Alexander Yates. 2013. Semantic Parsing Freebase: Towards Open-Domain Semantic Parsing. In Second Joint Conference on Lexical and Computational Semantics.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Toward an Architecture for Never-Ending Language Learning",
"authors": [
{
"first": "",
"middle": [],
"last": "Mitchell",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the Twenty-Fourth Conference on Artificial Intelligence (AAAI 2010)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mitchell. 2010. Toward an Architecture for Never- Ending Language Learning. In Proceedings of the Twenty-Fourth Conference on Artificial Intelligence (AAAI 2010).",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Training a Multilingual Sportscaster: Using Perceptual Context to Learn Language",
"authors": [
{
"first": "David",
"middle": [
"L"
],
"last": "Chen",
"suffix": ""
},
{
"first": "Joohyun",
"middle": [],
"last": "Kim",
"suffix": ""
},
{
"first": "Raymond",
"middle": [
"J"
],
"last": "Mooney",
"suffix": ""
}
],
"year": 2010,
"venue": "Journal of Artificial Intelligence Research",
"volume": "37",
"issue": "",
"pages": "397--435",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "David L. Chen, Joohyun Kim, and Raymond J. Mooney. 2010. Training a Multilingual Sportscaster: Using Perceptual Context to Learn Language. Journal of Artificial Intelligence Re- search, 37:397-435.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Widecoverage efficient statistical parsing with ccg and log-linear models",
"authors": [
{
"first": "Stephen",
"middle": [],
"last": "Clark",
"suffix": ""
},
{
"first": "James",
"middle": [
"R"
],
"last": "Curran",
"suffix": ""
}
],
"year": 2007,
"venue": "Computational Linguistics",
"volume": "33",
"issue": "4",
"pages": "493--552",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Stephen Clark and James R. Curran. 2007. Wide- coverage efficient statistical parsing with ccg and log-linear models. Computational Linguistics, 33(4):493-552.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Driving semantic parsing from the world's response",
"authors": [
{
"first": "J",
"middle": [],
"last": "Clarke",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Goldwasser",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Roth",
"suffix": ""
}
],
"year": 2010,
"venue": "Computational Natural Language Learning (CoNLL)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Clarke, D. Goldwasser, M. Chang, and D. Roth. 2010. Driving semantic parsing from the world's re- sponse. In Computational Natural Language Learn- ing (CoNLL).",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "iMAP: Discovering Complex Semantic Matches between Database Schemas",
"authors": [
{
"first": "R",
"middle": [],
"last": "Dhamanka",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Doan",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Halevy",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Domingos",
"suffix": ""
}
],
"year": 2004,
"venue": "SIGMOD",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Dhamanka, Y. Lee, A. Doan, A. Halevy, and P. Domingos. 2004. iMAP: Discovering Complex Semantic Matches between Database Schemas. In SIGMOD.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Ontology Matching: A Machine Learning Approach",
"authors": [
{
"first": "A",
"middle": [],
"last": "Doan",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Madhavan",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Domingos",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Halevy",
"suffix": ""
}
],
"year": 2004,
"venue": "Handbook on Ontologies in Information Systems",
"volume": "",
"issue": "",
"pages": "397--416",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A. Doan, J. Madhavan, P. Domingos, and A. Halevy. 2004. Ontology Matching: A Machine Learning Approach. In S. Staab and R. Studer, editors, Hand- book on Ontologies in Information Systems, pages 397-416. Springer-Verlag.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Similarity for ontologies-a comprehensive framework",
"authors": [
{
"first": "M",
"middle": [],
"last": "Ehrig",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Haase",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Stojanovic",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Hefke",
"suffix": ""
}
],
"year": 2004,
"venue": "Workshop Enterprise Modelling and Ontology: Ingredients for Interoperability",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M. Ehrig, P. Haase, N. Stojanovic, and M. Hefke. 2004. Similarity for ontologies-a comprehensive framework. In Workshop Enterprise Modelling and Ontology: Ingredients for Interoperability, PAKM.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Identifying Relations for Open Information Extraction",
"authors": [
{
"first": "Anthony",
"middle": [],
"last": "Fader",
"suffix": ""
},
{
"first": "Stephen",
"middle": [],
"last": "Soderland",
"suffix": ""
},
{
"first": "Oren",
"middle": [],
"last": "Etzioni",
"suffix": ""
}
],
"year": 2011,
"venue": "Conference on Empirical Methods in Natural Language Processing (EMNLP)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Anthony Fader, Stephen Soderland, and Oren Etzioni. 2011. Identifying Relations for Open Information Extraction. In Conference on Empirical Methods in Natural Language Processing (EMNLP).",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Semantic schema matching",
"authors": [
{
"first": "Fausto",
"middle": [],
"last": "Giunchiglia",
"suffix": ""
},
{
"first": "Pavel",
"middle": [],
"last": "Shvaiko",
"suffix": ""
},
{
"first": "Mikalai",
"middle": [],
"last": "Yatskevich",
"suffix": ""
}
],
"year": 2005,
"venue": "On the Move to Meaningful Internet Systems 2005: CoopIS, DOA, and ODBASE",
"volume": "",
"issue": "",
"pages": "347--365",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Fausto Giunchiglia, Pavel Shvaiko, and Mikalai Yatskevich. 2005. Semantic schema matching. On the Move to Meaningful Internet Systems 2005: CoopIS, DOA, and ODBASE, pages 347-365.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Confidence driven unsupervised semantic parsing",
"authors": [
{
"first": "D",
"middle": [],
"last": "Goldwasser",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Reichart",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Clarke",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Roth",
"suffix": ""
}
],
"year": 2011,
"venue": "Association for Computational Linguistics (ACL)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "D. Goldwasser, R. Reichart, J. Clarke, and D. Roth. 2011. Confidence driven unsupervised semantic parsing. In Association for Computational Linguis- tics (ACL).",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "YAGO2: A Spatially and Temporally Enhanced Knowledge Base from Wikipedia",
"authors": [
{
"first": "Johannes",
"middle": [],
"last": "Hoffart",
"suffix": ""
},
{
"first": "Fabian",
"middle": [
"M"
],
"last": "Suchanek",
"suffix": ""
},
{
"first": "Klaus",
"middle": [],
"last": "Berberich",
"suffix": ""
},
{
"first": "Gerhard",
"middle": [],
"last": "Weikum",
"suffix": ""
}
],
"year": 2013,
"venue": "Artificial Intelligence",
"volume": "194",
"issue": "",
"pages": "28--61",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Johannes Hoffart, Fabian M. Suchanek, Klaus Berberich, and Gerhard Weikum. 2013. YAGO2: A Spatially and Temporally Enhanced Knowl- edge Base from Wikipedia. Artificial Intelligence, 194:28-61, January.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Weakly Supervised Training of Semantic Parsers",
"authors": [],
"year": null,
"venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Weakly Supervised Training of Semantic Parsers. In Proceedings of the Conference on Empirical Meth- ods in Natural Language Processing (EMNLP).",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Inducing Probabilistic CCG Grammars from Logical Form with Higherorder Unification",
"authors": [
{
"first": "Tom",
"middle": [],
"last": "Kwiatkowski",
"suffix": ""
},
{
"first": "Luke",
"middle": [],
"last": "Zettlemoyer",
"suffix": ""
},
{
"first": "Sharon",
"middle": [],
"last": "Goldwater",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Steedman",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tom Kwiatkowski, Luke Zettlemoyer, Sharon Goldwa- ter, and Mark Steedman. 2010. Inducing Probabilis- tic CCG Grammars from Logical Form with Higher- order Unification. In Proceedings of the Conference on Empirical Methods in Natural Language Pro- cessing (EMNLP).",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Learning semantic correspondences with less supervision",
"authors": [
{
"first": "P",
"middle": [],
"last": "Liang",
"suffix": ""
},
{
"first": "M",
"middle": [
"I"
],
"last": "Jordan",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Klein",
"suffix": ""
}
],
"year": 2009,
"venue": "Association for Computational Linguistics and International Joint Conference on Natural Language Processing (ACL-IJCNLP)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "P. Liang, M. I. Jordan, and D. Klein. 2009. Learning semantic correspondences with less supervision. In Association for Computational Linguistics and In- ternational Joint Conference on Natural Language Processing (ACL-IJCNLP).",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Learning dependency-based compositional semantics",
"authors": [
{
"first": "P",
"middle": [],
"last": "Liang",
"suffix": ""
},
{
"first": "M",
"middle": [
"I"
],
"last": "Jordan",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Klein",
"suffix": ""
}
],
"year": 2011,
"venue": "Association for Computational Linguistics (ACL)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "P. Liang, M. I. Jordan, and D. Klein. 2011. Learn- ing dependency-based compositional semantics. In Association for Computational Linguistics (ACL).",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "DIRT -Discovery of Inference Rules from Text",
"authors": [
{
"first": "D",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Pantel",
"suffix": ""
}
],
"year": 2001,
"venue": "KDD",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "D. Lin and P. Pantel. 2001. DIRT -Discovery of In- ference Rules from Text. In KDD.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "Information retrieval and machine learning for probabilistic schema matching. Information processing & management",
"authors": [
{
"first": "Henrik",
"middle": [],
"last": "Nottelmann",
"suffix": ""
},
{
"first": "Umberto",
"middle": [],
"last": "Straccia",
"suffix": ""
}
],
"year": 2007,
"venue": "",
"volume": "43",
"issue": "",
"pages": "552--576",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Henrik Nottelmann and Umberto Straccia. 2007. In- formation retrieval and machine learning for proba- bilistic schema matching. Information processing & management, 43(3):552-576.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Unsupervised semantic parsing",
"authors": [
{
"first": "Hoifung",
"middle": [],
"last": "Poon",
"suffix": ""
},
{
"first": "Pedro",
"middle": [],
"last": "Domingos",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, EMNLP '09",
"volume": "",
"issue": "",
"pages": "1--10",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hoifung Poon and Pedro Domingos. 2009. Unsu- pervised semantic parsing. In Proceedings of the 2009 Conference on Empirical Methods in Natu- ral Language Processing, EMNLP '09, pages 1-10, Stroudsburg, PA, USA. Association for Computa- tional Linguistics.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "A survey of approaches to automatic schema matching",
"authors": [
{
"first": "E",
"middle": [],
"last": "Rahm",
"suffix": ""
},
{
"first": "P",
"middle": [
"A"
],
"last": "Bernstein",
"suffix": ""
}
],
"year": 2001,
"venue": "The VLDB Journal",
"volume": "10",
"issue": "",
"pages": "334--350",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "E. Rahm and P.A. Bernstein. 2001. A survey of ap- proaches to automatic schema matching. The VLDB Journal, 10:334-350.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "Mining the Web for Synonyms: PMI-IR versus LSA on TOEFL",
"authors": [
{
"first": "P",
"middle": [
"D"
],
"last": "Turney",
"suffix": ""
}
],
"year": 2001,
"venue": "Procs. of the Twelfth European Conference on Machine Learning (ECML)",
"volume": "",
"issue": "",
"pages": "491--502",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "P. D. Turney. 2001. Mining the Web for Synonyms: PMI-IR versus LSA on TOEFL. In Procs. of the Twelfth European Conference on Machine Learning (ECML), pages 491-502, Freiburg, Germany.",
"links": null
},
"BIBREF27": {
"ref_id": "b27",
"title": "A discriminative approach to ontology mapping",
"authors": [
{
"first": "M",
"middle": [],
"last": "Wick",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Rohanimanesh",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Mccallum",
"suffix": ""
},
{
"first": "A",
"middle": [
"H"
],
"last": "Doan",
"suffix": ""
}
],
"year": 2008,
"venue": "International Workshop on New Trends in Information Integration (NTII) at VLDB WS",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M. Wick, K. Rohanimanesh, A. McCallum, and A.H. Doan. 2008a. A discriminative approach to ontol- ogy mapping. In International Workshop on New Trends in Information Integration (NTII) at VLDB WS.",
"links": null
},
"BIBREF28": {
"ref_id": "b28",
"title": "A unified approach for schema matching, coreference and canonicalization",
"authors": [
{
"first": "M",
"middle": [
"L"
],
"last": "Wick",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Rohanimanesh",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Schultz",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Mc-Callum",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceeding of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M.L. Wick, K. Rohanimanesh, K. Schultz, and A. Mc- Callum. 2008b. A unified approach for schema matching, coreference and canonicalization. In Pro- ceeding of the 14th ACM SIGKDD international conference on Knowledge discovery and data min- ing.",
"links": null
},
"BIBREF29": {
"ref_id": "b29",
"title": "Natural Language Questions for the Web of Data",
"authors": [
{
"first": "Mohamed",
"middle": [],
"last": "Yahya",
"suffix": ""
},
{
"first": "Klaus",
"middle": [],
"last": "Berberich",
"suffix": ""
},
{
"first": "Shady",
"middle": [],
"last": "Elbassuoni",
"suffix": ""
},
{
"first": "Maya",
"middle": [],
"last": "Ramanath",
"suffix": ""
},
{
"first": "Gerhard",
"middle": [],
"last": "Volker Tresp",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Weikum",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mohamed Yahya, Klaus Berberich, Shady Elbas- suoni, Maya Ramanath, Volker Tresp, and Gerhard Weikum. 2012. Natural Language Questions for the Web of Data. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP).",
"links": null
},
"BIBREF30": {
"ref_id": "b30",
"title": "Unsupervised methods for determining object and relation synonyms on the web",
"authors": [
{
"first": "Alexander",
"middle": [],
"last": "Yates",
"suffix": ""
},
{
"first": "Oren",
"middle": [],
"last": "Etzioni",
"suffix": ""
}
],
"year": 2009,
"venue": "Journal of Artificial Intelligence Research (JAIR)",
"volume": "34",
"issue": "",
"pages": "255--296",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Alexander Yates and Oren Etzioni. 2009. Unsuper- vised methods for determining object and relation synonyms on the web. Journal of Artificial Intelli- gence Research (JAIR), 34:255-296, March.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"text": "MATCHER's Pattern features and Extractions features complement one another, so that in combination they outperform either subset on its own, especially at the high-recall end of the curve.",
"type_str": "figure",
"num": null,
"uris": null
},
"FIGREF1": {
"text": "Semantic parsers produced by UBL+MATCHER+LEXTENDER outperform the purelysupervised baseline semantic parser on a random 70/30 split of the data (left) by as much as 0.42 in F1. In the case of this split and in the case of a cross-domain experiment (right), UBL+MATCHER+LEXTENDER outperforms UBL+Pattern+LEXTENDER by as much as 0.06 in F1.",
"type_str": "figure",
"num": null,
"uris": null
},
"TABREF1": {
"content": "<table/>",
"type_str": "table",
"num": null,
"text": "",
"html": null
},
"TABREF2": {
"content": "<table/>",
"type_str": "table",
"num": null,
"text": "MATCHER statistics: for each \u03ba function for comparing two tuples (given inTable 3), MATCHER computes the statistics above to compare r D and r T . The PMI statistic in Equation 1 corresponds to s \u03ba where \u03ba =strict match over \u03a6 =full tuples.",
"html": null
},
"TABREF3": {
"content": "<table/>",
"type_str": "table",
"num": null,
"text": "MATCHER's \u03a6 functions for projecting or permuting a tuple. \u03c3 indicates a permutation of the indices.",
"html": null
}
}
}
} |