File size: 101,071 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 | {
"paper_id": "D07-1018",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T16:20:11.353474Z"
},
"title": "Modelling Polysemy in Adjective Classes by Multi-Label Classification",
"authors": [
{
"first": "Gemma",
"middle": [],
"last": "Boleda",
"suffix": "",
"affiliation": {},
"email": "gemma.boleda@upf.edu"
},
{
"first": "Sabine",
"middle": [],
"last": "Schulte",
"suffix": "",
"affiliation": {},
"email": "schulte@ims.uni-stuttgart.de"
},
{
"first": "Toni",
"middle": [],
"last": "Badia",
"suffix": "",
"affiliation": {},
"email": "toni.badia@upf.edu"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "This paper assesses the role of multi-label classification in modelling polysemy for language acquisition tasks. We focus on the acquisition of semantic classes for Catalan adjectives, and show that polysemy acquisition naturally suits architectures used for multilabel classification. Furthermore, we explore the performance of information drawn from different levels of linguistic description, using feature sets based on morphology, syntax, semantics, and n-gram distribution. Finally, we demonstrate that ensemble classifiers are a powerful and adequate way to combine different types of linguistic evidence: a simple, majority voting ensemble classifier improves the accuracy from 62.5% (best single classifier) to 84%.",
"pdf_parse": {
"paper_id": "D07-1018",
"_pdf_hash": "",
"abstract": [
{
"text": "This paper assesses the role of multi-label classification in modelling polysemy for language acquisition tasks. We focus on the acquisition of semantic classes for Catalan adjectives, and show that polysemy acquisition naturally suits architectures used for multilabel classification. Furthermore, we explore the performance of information drawn from different levels of linguistic description, using feature sets based on morphology, syntax, semantics, and n-gram distribution. Finally, we demonstrate that ensemble classifiers are a powerful and adequate way to combine different types of linguistic evidence: a simple, majority voting ensemble classifier improves the accuracy from 62.5% (best single classifier) to 84%.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "This paper reports on a series of experiments to explore the automatic acquisition of semantic classes for Catalan adjectives. The most important challenge of the classification task is to model the assignment of polysemous lexical instances to multiple semantic classes, combining a) a state-of-the-art Machine Learning architecture for Multi-label Classification (Schapire and Singer, 2000; Ghamrawi and McCallum, 2005) and an Ensemble Classifier (Dietterich, 2002) with b) the definition of features at various levels of linguistic description.",
"cite_spans": [
{
"start": 365,
"end": 392,
"text": "(Schapire and Singer, 2000;",
"ref_id": "BIBREF31"
},
{
"start": 393,
"end": 421,
"text": "Ghamrawi and McCallum, 2005)",
"ref_id": "BIBREF13"
},
{
"start": 449,
"end": 467,
"text": "(Dietterich, 2002)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "A proper treatment of polysemy is essential in the area of lexical acquisition, since polysemy repre-sents a pervasive phenomenon in natural language. However, previous approaches to the automatic acquisition of semantic classes have mostly disregarded the problem (cf. Merlo and Stevenson, 2001 and Stevenson and Joanis, 2003 for English semantic verb classes, or Schulte im Walde, 2006 for German semantic verb classes). There are a few exceptions to this tradition, such as Pereira et al. (1993) , Rooth et al. (1999) , Korhonen et al. (2003) , who used soft clustering methods for multiple assignment to verb semantic classes.",
"cite_spans": [
{
"start": 270,
"end": 295,
"text": "Merlo and Stevenson, 2001",
"ref_id": "BIBREF22"
},
{
"start": 300,
"end": 326,
"text": "Stevenson and Joanis, 2003",
"ref_id": "BIBREF32"
},
{
"start": 477,
"end": 498,
"text": "Pereira et al. (1993)",
"ref_id": "BIBREF25"
},
{
"start": 501,
"end": 520,
"text": "Rooth et al. (1999)",
"ref_id": "BIBREF29"
},
{
"start": 523,
"end": 545,
"text": "Korhonen et al. (2003)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Our work addresses the lack of methodology in modelling a polysemous classification. We implement a multi-label classification architecture to handle polysemy. This paper concentrates on the classification of Catalan adjectives, but the general nature of the architecture should allow related tasks to profit from our insights.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "As target classification for the experiments, a set of 210 Catalan adjectives was manually classified by experts into three simple and three polysemous semantic classes. We deliberately decided in favour of a small-scale, broad classification. So far, there is little work on the semantic classification of adjectives, as opposed to verbal semantic classification. The semantic classification we propose is a first step in characterising adjectival meaning, and can be refined and extended in subsequent work.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The experiments also provide a thorough comparison of feature sets based on different levels of linguistic description (morphology, syntax, semantics). A set of features is defined for each level of description, and its performance is assessed within the series of experiments. An ensemble classifier comple-ments the classification architecture, by optimising the combination of these different types of linguistic evidence.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Our task is motivated by the fact that adjectives play an important role in sentential semantics: they are crucial in determining the reference of NPs, and in defining properties of entities. Even using only three different classes, the information acquired could be applied to, e.g., identify referents in a given context in Dialog or Question Answering systems, and to induce properties of objects within Information Extraction tasks. Furthermore, with the semantic classes corresponding to broad sense representations, they can be exploited for Word Sense Disambiguation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The remainder of this paper is organised as follows. Section 2 provides background on Catalan adjectives, and Section 3 presents the Gold Standard classification. Section 4 introduces the methodology of the multi-label classification experiments, Section 5 discusses the results, and the improved ensemble classifier is presented in Section 6.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The definition and characterisation of our target semantic classification follows the proposal by Raskin and Nirenburg (1998) within the framework of Ontological Semantics (Nirenburg and Raskin, 2004) . In Ontological Semantics, an ontology of concepts modelling the world is explicitly defined, and the semantics of words are mapped onto elements of the ontology. The classification pursued in this paper is drawn up based on the ontological sort of adjectival denotation: all adjectives denote properties, but these properties can be instantiated as simple attributes (basic adjectives), relationships to objects (object-related adjectives), or relationships to events (event-related adjectives).",
"cite_spans": [
{
"start": 98,
"end": 125,
"text": "Raskin and Nirenburg (1998)",
"ref_id": "BIBREF27"
},
{
"start": 172,
"end": 200,
"text": "(Nirenburg and Raskin, 2004)",
"ref_id": "BIBREF24"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Catalan adjective classes",
"sec_num": "2"
},
{
"text": "Basic adjectives are the prototypical adjectives which denote attributes or properties and cannot be decomposed further (such as bonic 'beautiful', gran 'big'). In Ontological Semantics, these adjectives are mapped to concepts of type attribute. For instance, the semantics of the adjective gran specifies a mapping to the size-attribute element in the ontology. As for event-related adjectives, they have an event component in their meaning and are therefore mapped onto event concepts in the ontology. For instance, the semantics of tangible ('tangible') includes a pointer to the event element touch in the ontology. Similarly, object-related adjectives are mapped onto object concepts in the ontology: deformaci\u00f3 nasal ('nasal deformity') can be paraphrased as deformity that affects the nose, so nasal evokes the object nose.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Catalan adjective classes",
"sec_num": "2"
},
{
"text": "The semantic distinctions are mirrored at several levels of linguistic description, such as morphology, syntax, and semantics. For instance, there is a clear relationship between morphological type and semantic class: basic adjectives are typically non-derived, object adjectives tend to be denominal, and event adjectives are usually deverbal. This is the default mapping that one expects from the morphology-semantics interface. As an example for syntactic evidence, basic adjectives in Catalan can be used non-restrictively (in a pre-nominal position) and also predicatively, while object adjectives typically cannot.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Catalan adjective classes",
"sec_num": "2"
},
{
"text": "However, the correspondences between the linguistic properties and the semantic classes are not one-to-one mappings. Taking the morphological level as an example, some denominal adjectives are basic (such as vergony\u00f3s 'shy', from vergonya 'shyness'). Conversely, some object adjectives are not synchronically denominal (such as bot\u00e0nic 'botanical'), and some deverbal adjectives are not eventrelated, such as amable (lit. 'suitable to be loved'; has evolved to 'kind, friendly'). In such cases, the semantic class can be better traced in the distributional properties, not the morphological properties of the adjective.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Catalan adjective classes",
"sec_num": "2"
},
{
"text": "The proposed classification accounts for some cases of adjectival polysemy. For instance, familiar has an object reading (related to the Catalan noun for 'family'), and a basic reading (corresponding to the English adjective 'familiar'): Similarly, the participial adjective sabut ('known') has an event-related sense, corresponding to the verb saber ('know'), and a basic sense equivalent to 'wise': The polysemy between our proposed classes, as exemplified in (1) and 2, is the kind of polysemy we aim to model in the acquisition experiments reported in this paper.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Catalan adjective classes",
"sec_num": "2"
},
{
"text": "As a Gold Standard for the experiments to follow, 210 Catalan adjectives were classified by three experts. The adjectives were randomly sampled from an adjective database (Sanrom\u00e0, 2003) , balancing three factors of variability: frequency, morphological type, and suffix. An equal number of adjectives was chosen from three frequency bands (low, medium, high), from four derivational types (denominal, deverbal, non-derived, participle), and from a series of suffixes within each type. The derivational type and suffix of each adjective were available in the adjective database, and had been manually encoded.",
"cite_spans": [
{
"start": 171,
"end": 186,
"text": "(Sanrom\u00e0, 2003)",
"ref_id": "BIBREF30"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Gold Standard classes",
"sec_num": "3"
},
{
"text": "Three experts assigned the 210 lemmata to one out of six classes: each adjective was tagged as basic (B), event (E), object (O), or as polysemous between basic and event (BE), between basic and object (BO), or between event and object (EO). The decisions were reached by consensus. The distribution of the Gold Standard material across classes is shown in the last column of Table 6 (Section 5.2).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Gold Standard classes",
"sec_num": "3"
},
{
"text": "In the acquisition experiments, our aim is to automatically assign a class to each adjective that can be simple (B, E, O) or complex (BE, BO, EO), in case of polysemy.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Gold Standard classes",
"sec_num": "3"
},
{
"text": "Adjective classification was performed within a twolevel architecture for multi-label classification: first, make a binary decision on each of the classes, and then combine the classifications to achieve a final, multi-label classification. We therefore decomposed the global decision on the (possibly polysemous) class of an adjective into three binary decisions: Is it basic or not? Is it event-related or not? Is it objectrelated or not? The individual decisions were then combined into an overall classification that included polysemy. For example, if a lemma was classified both as basic and as object in each of the binary decisions, it was deemed polysemous (BO). The motivation behind this approach was that polysemous adjectives should exhibit properties of all the classes involved. As a result, positive decisions on each binary classification can be made by the algorithm, which can be viewed as implicit polysemous assignments.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Classification method",
"sec_num": "4"
},
{
"text": "This classification architecture is very popular in Machine Learning for multi-label problems, cf. (Schapire and Singer, 2000; Ghamrawi and Mc-Callum, 2005) , and has also been applied to NLP problems such as entity extraction and noun-phrase chunking (McDonald et al., 2005) . The remainder of this section describes other methodological aspects of our experiments.",
"cite_spans": [
{
"start": 99,
"end": 126,
"text": "(Schapire and Singer, 2000;",
"ref_id": "BIBREF31"
},
{
"start": 127,
"end": 156,
"text": "Ghamrawi and Mc-Callum, 2005)",
"ref_id": null
},
{
"start": 252,
"end": 275,
"text": "(McDonald et al., 2005)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Classification method",
"sec_num": "4"
},
{
"text": "As classifier for the binary decisions we chose Decision Trees, one of the most widely used Machine Learning techniques for supervised experiments (Witten and Frank, 2005) . Decision Trees provide a transparent representation of the decisions made by the algorithm, and thus facilitate the inspection of results and the error analysis. The experiments were carried out with the freely available Weka software package. The particular algorithm chosen, Weka's J48, is the latest open source version of C4.5 (Quinlan, 1993) . For an explanation of decision tree induction and C4.5, see Quinlan (1993) and Witten and Frank (2005, Sections 4 .3 and 6.1).",
"cite_spans": [
{
"start": 147,
"end": 171,
"text": "(Witten and Frank, 2005)",
"ref_id": "BIBREF34"
},
{
"start": 505,
"end": 520,
"text": "(Quinlan, 1993)",
"ref_id": "BIBREF26"
},
{
"start": 583,
"end": 597,
"text": "Quinlan (1993)",
"ref_id": "BIBREF26"
},
{
"start": 602,
"end": 636,
"text": "Witten and Frank (2005, Sections 4",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Classifier: Decision Trees",
"sec_num": "4.1"
},
{
"text": "Five levels of linguistic description, formalised as different feature sets, were chosen for our task. They included evidence from morphology (morph), syntax (func, uni, bi), semantics (sem), plus a combination of the five levels (all). Table 1 lists the linguistic levels, their explanations, and the number of features used on each level. 1 Morphological features (morph) encode the derivational type (denominal, deverbal, participial, non-derived) and the suffix (in case the adjective is derived) of each adjective, and correspond to the manually encoded informa- ",
"cite_spans": [],
"ref_spans": [
{
"start": 237,
"end": 244,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Feature definition",
"sec_num": "4.2"
},
{
"text": "Irrelevant features typically decrease performance by 5 to 10% when using Decision Trees (Witten and Frank, 2005, p. 288) . We therefore applied a feature selection algorithm. We chose a feature selection method available in Weka (WrapperSubsetEval) that selects a subset of the features according to its performance within the Machine Learning algorithm used for classification. Accuracy for a given subset of features is estimated by cross-validation over the training data. Because the number of subsets increases exponentially with the number of features, this method is computationally very expensive, and we used a best-first search strategy to alleviate this problem. We additionally used the feature selection procedure to select the features for level all: for each class, we used only those features that were selected by the feature selection algorithm in at least 30% of the experiments.",
"cite_spans": [
{
"start": 89,
"end": 121,
"text": "(Witten and Frank, 2005, p. 288)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Feature selection",
"sec_num": "4.3"
},
{
"text": "One of our goals was to test the strengths and weaknesses of each level of linguistic description for the task of adjective classification. This was done by comparing the accuracy results obtained with each of the feature sets in the Machine Learning experiments. Following a standard procedure in Machine Learning, we created several partitions of the data to obtain different estimates of the accuracy of each of the levels, so as to be able to perform a significance test on the differences in accuracy. We performed 10 experiments with 10-fold cross-validation (10x10 cv for short), so that for each class 100 different binary decisions were made for each adjective. For the comparison of accuracies, a standard paired t-test could not be used, because of the inflated Type I er-ror probability when reusing data (Dietterich, 1998) . Instead, we used the corrected resampled t-test as proposed by Nadeau and Bengio (2003) . 2",
"cite_spans": [
{
"start": 817,
"end": 835,
"text": "(Dietterich, 1998)",
"ref_id": "BIBREF9"
},
{
"start": 901,
"end": 925,
"text": "Nadeau and Bengio (2003)",
"ref_id": "BIBREF23"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Differences across linguistic levels",
"sec_num": "4.4"
},
{
"text": "The accuracy results for each of the binary decisions (basic/non-basic, event/non-event, object/nonobject) are depicted in Table 3 . 3 Level bl corresponds to the baseline: the baseline accuracy was determined by assigning all lemmata to the most frequent class. The remaining levels follow the nomenclature in Table 1 above. Each column contains the mean and the standard deviation (marked by \u00b1) of the accuracy for the relevant level of information over the 100 results obtained with 10x10 cv. As one might have expected, the best results were obtained with the all level (bold faced in Table 3) , which is the combination of all feature types. This level achieved a mean improvement of 12.3% over the baseline. The differences in accuracy results between most levels of information were, however, rather small. For the object class, all levels except for func and uni achieved a significant improvement over the baseline. For the basic class, no improve-ment over the baseline was significant according to the corrected resampled t-test. And for the event class, only levels morph and all offered a significant improvement in accuracy; the remaining levels even obtained a slightly lower accuracy score.",
"cite_spans": [
{
"start": 133,
"end": 134,
"text": "3",
"ref_id": null
}
],
"ref_spans": [
{
"start": 123,
"end": 130,
"text": "Table 3",
"ref_id": "TABREF3"
},
{
"start": 311,
"end": 318,
"text": "Table 1",
"ref_id": null
},
{
"start": 589,
"end": 597,
"text": "Table 3)",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Accuracy results",
"sec_num": "5.1"
},
{
"text": "These results concern the three individual binary decisions. However, our goal was not to obtain three separate decisions, but a single classification including polysemy. Table 4 shows the accuracy results for the classification obtained by combining the three individual decisions for each adjective. We report two accuracy measures, full and partial: full accuracy required the class assignments to be identical; partial accuracy only required some overlap in the classification of the Machine Learning algorithm and the Gold Standard for a given class assignment. The motivation for calculating partial overlap was that a class assignment with some overlap with the Gold Standard (even if they were not identical) is generally more useful than a class assignment with no overlap. Table 4 : Accuracy results for combined decisions.",
"cite_spans": [],
"ref_spans": [
{
"start": 171,
"end": 178,
"text": "Table 4",
"ref_id": null
},
{
"start": 783,
"end": 790,
"text": "Table 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Basic",
"sec_num": null
},
{
"text": "Again, the best results were obtained with the all level. The second best results were obtained with level morph. These results could have been expected from the results obtained by the individual decisions (Table 3) ; however, note that the differences between the various levels are much clearer in the combined classification than in the individual binary decisions. Table 5 shows the two-by-two comparisons of the accuracy scores. Each cell contains the difference in accuracy means between two levels of description, as well as the level of significance of the difference. The significance is marked as follows: * for p < 0.05, ** for p < 0.01, *** for p < 0.001. If no asterisk is shown, the difference was not significant.",
"cite_spans": [],
"ref_spans": [
{
"start": 207,
"end": 216,
"text": "(Table 3)",
"ref_id": "TABREF3"
},
{
"start": 370,
"end": 377,
"text": "Table 5",
"ref_id": null
}
],
"eq_spans": [],
"section": "Full",
"sec_num": null
},
{
"text": "Under racy), only levels morph, func, and all significantly improved upon the baseline. Levels morph and all are better than the remaining levels, to a similar extent. In the partial evaluation condition, all levels achieved a highly significant improvement over the baseline (p < 0.001). Therefore, the classifications obtained with any of the feature levels are more useful than the baseline, in the sense that they present more overlap with the Gold Standard.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Full",
"sec_num": null
},
{
"text": "The best result obtained for the full classification of adjectives with our methodology achieved a mean of 62.3% (full accuracy) or 90.7% (partial accuracy), which represents an improvement of 11.3% and 25.5% over the baselines, respectively. Levels including morphological information were clearly superior to levels using only distributional information.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Full",
"sec_num": null
},
{
"text": "These results suggest that morphology is the best single source of evidence for our task. However, recall from Section 3 that the sampling procedure for the Gold Standard explicitly balanced for morphological factors. As a result, denominal and participial adjectives are underrepresented in the Gold Standard, while non-derived and deverbal adjectives are overrepresented. Moreover, previous experiments on different datasets (Boleda et al., 2004; Boleda et al., 2005) provided some evidence that distributional information outperforms morphological information for our task. Therefore, we cannot conclude from the experiments that morphological features are the most important information for the classification of Catalan adjectives in general.",
"cite_spans": [
{
"start": 427,
"end": 448,
"text": "(Boleda et al., 2004;",
"ref_id": "BIBREF2"
},
{
"start": 449,
"end": 469,
"text": "Boleda et al., 2005)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Full",
"sec_num": null
},
{
"text": "The error analysis focuses on the two best feature sets, morph and all. Table 6 compares the errors made by the experiment classifications (based on the two sets of features) against the Gold Standard classification. To obtain a unique experiment classification for each feature level in this comparison, we applied majority voting across the 10 different classifications obtained in the 10 experiment runs for each of the linguistic levels. The table rows correspond to the Gold Standard classification and the columns correspond to the experiment classifications with the feature levels all and morph, respectively. The matches (the diagonal elements) are in italics, and off-diagonal cells representing the largest numbers of mismatches are boldfaced. The overall number of mistakes made by both levels with majority voting is almost the same: 86 (morph) vs. 89 (all). However, the mismatches are qualitatively quite different.",
"cite_spans": [],
"ref_spans": [
{
"start": 72,
"end": 79,
"text": "Table 6",
"ref_id": "TABREF6"
}
],
"eq_spans": [],
"section": "Error analysis",
"sec_num": "5.2"
},
{
"text": "Level morph uniformly mapped denominal adjectives to both basic and object (BO). Because of this overgeneration of BOs, 31 lemmata that were tagged as either basic or object in the Gold Standard were assigned to BO. In contrast, level all was overly discriminative: most of the BO cases (16 out of 23), as well as 16 object adjectives, were assigned to basic. This type of confusion could be explained by the fact that some non-prototypical basic adjectives were as- signed to the basic class in the Gold Standard, because they did not fit the narrower definitions of the event and object classes, but these adjectives do not behave like typical basic adjectives.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error analysis",
"sec_num": "5.2"
},
{
"text": "As for event adjectives, the morph level assigned almost all deverbal adjectives to the event class, which worked well in most cases (26). However, this mapping cannot distinguish deverbal adjectives with a basic meaning (11 basic and 6 BE adjectives in the Gold Standard). Level all, including morphological and distributional cues, also shows difficulties with the event class, but of a different nature. Feature examination showed that the distributional differences between basic and event adjectives are not robust. For instance, according to t-tests performed on the Gold Standard (\u03b1 = 0.05), only three of the 18 semantic features exhibit significant mean differences for classes basic and event. In contrast, ANOVA across the 6 classes (\u03b1 = 0.05) yields significant differences for 16 out of the 18 features, which indicates that most features serve to distinguish object adjectives from basic and event adjectives. As a result of the lack of robust distributional differences between basic and event adjectives, 35 basic or event adjectives were classified as BE when using the all level as feature set.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error analysis",
"sec_num": "5.2"
},
{
"text": "Further 23 event adjectives were incorrectly classified as BE by the all level, but correctly classified by the morph level, because they are deverbal adjectives. These cases involved adjectives derived from stative verbs, such as abundant ('abundant') or preferible ('preferable'). Feature analysis revealed that deverbal adjectives derived from stative verbs are more similar to basic adjectives than those derived from process-denoting verbs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error analysis",
"sec_num": "5.2"
},
{
"text": "To sum up, the default morphological mapping mentioned in Section 2 works well in most cases but has a clear ceiling, as it cannot account for deviations from the expected mapping. Distributional cues are more sensitive to these deviations, but fail mostly in the distinction between basic and event, because the differences in syntactic distribution between these classes are not robust.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Error analysis",
"sec_num": "5.2"
},
{
"text": "The error analysis in the previous section has shown that, although the number of mistakes made with level morph and all is comparable, the kinds of mistakes are qualitatively very different. This suggests that mixing features for the construction of a single Decision Tree, as is done in level all, is not the optimal way to combine the strengths of each level of description. An alternative combination can be achieved with an ensemble classifier, a type of classifier that has received much attention in the Machine Learning community in the last decade (Dietterich, 2002) . When building an ensemble classifier, several class proposals for each item are obtained, and one of them is chosen on the basis of majority voting, weighted voting, or more sophisticated decision methods. It has been shown that in most cases, the accuracy of the ensemble classifier is higher than the best individual classifier (Freund and Schapire, 1996; Dietterich, 2000; Breiman, 2001) . Within NLP, ensemble classifiers have been applied, for instance, to genus term disambiguation in machinereadable dictionaries (Rigau et al., 1997) , using a majority voting scheme upon several heuristics, and to part of speech tagging, by combining the class predictions of different algorithms (van Halteren et al., 1998) . The main reason for the general success of ensemble classifiers is that they gloss over the biases introduced by the individual systems. We implemented an ensemble classifier by using the different levels of description as different subsets of features, and applying majority voting across the class proposals from each level. Intuitively, this architecture is analogous to having a team of linguists and NLP engineers, each contributing their knowledge on morphology, n-gram distribution, syntactic properties, etc., and have them reach a consensus classification. We thus established a different classification for each of the 10 cross-validation runs by assigning each adjective to the class that received most votes. To enable a majority vote, at least three levels have to be combined. Table 7 contains a representative selection of the combinations, together with their accuracies. Also, the accuracies obtained with the baseline (bl) and the best single level (all) are included for comparison.",
"cite_spans": [
{
"start": 557,
"end": 575,
"text": "(Dietterich, 2002)",
"ref_id": "BIBREF11"
},
{
"start": 908,
"end": 935,
"text": "(Freund and Schapire, 1996;",
"ref_id": "BIBREF12"
},
{
"start": 936,
"end": 953,
"text": "Dietterich, 2000;",
"ref_id": "BIBREF10"
},
{
"start": 954,
"end": 968,
"text": "Breiman, 2001)",
"ref_id": "BIBREF7"
},
{
"start": 1098,
"end": 1118,
"text": "(Rigau et al., 1997)",
"ref_id": "BIBREF28"
},
{
"start": 1284,
"end": 1294,
"text": "al., 1998)",
"ref_id": null
}
],
"ref_spans": [
{
"start": 2088,
"end": 2095,
"text": "Table 7",
"ref_id": "TABREF8"
}
],
"eq_spans": [],
"section": "An improved classifier",
"sec_num": "6"
},
{
"text": "In any of the combinations tested, accuracy improved over 10% with respect to the all level. The best result, a mean of 84% (full accuracy), was obtained by combining all levels of description. These results represent a raw improvement over the baseline of 33%, and 21.7% over the best single classifier. Also note that with this procedure 95.7% of the classifications obtained with the ensemble classifier present partial overlap with the class assignments in the Gold Standard.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "An improved classifier",
"sec_num": "6"
},
{
"text": "These results show that the combination of different sources of linguistic evidence is more important than the type of information used. As an example, consider the second ensemble classifier in Table 7 : this classifier excludes the two levels that contain morphological information (morph and all), which represents the most successful individual source of information for our dataset. Nevertheless, the combination achieved 19.2/20.9% more accuracy than levels all and morph, respectively.",
"cite_spans": [],
"ref_spans": [
{
"start": 195,
"end": 202,
"text": "Table 7",
"ref_id": "TABREF8"
}
],
"eq_spans": [],
"section": "An improved classifier",
"sec_num": "6"
},
{
"text": "Adjectives have received less attention than verbs and nouns within Lexical Acquisition. Work by Hatzivassiloglou and colleagues (Hatzivassiloglou and McKeown, 1993; Hatzivassiloglou and McKeown, 1997; Hatzivassiloglou and Wiebe, 2000) used clustering methods to automatically identify adjectival scales from corpora.",
"cite_spans": [
{
"start": 129,
"end": 165,
"text": "(Hatzivassiloglou and McKeown, 1993;",
"ref_id": "BIBREF14"
},
{
"start": 166,
"end": 201,
"text": "Hatzivassiloglou and McKeown, 1997;",
"ref_id": "BIBREF15"
},
{
"start": 202,
"end": 235,
"text": "Hatzivassiloglou and Wiebe, 2000)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related work",
"sec_num": "7"
},
{
"text": "Coordination information was used in Bohnet et al. (2002) for a classification task similar to the task we pursue, using a bootstrapping approach. The authors, however, pursued a classification that is not purely semantic, between quantitative adjectives (similar to determiners, like viele 'many'), referential adjectives (heutige, 'of today'), qualitative adjectives (equivalent to basic adjectives), classificatory adjectives (equivalent to object adjectives), and adjectives of origin (Stuttgarter, 'from Stuttgart').",
"cite_spans": [
{
"start": 37,
"end": 57,
"text": "Bohnet et al. (2002)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related work",
"sec_num": "7"
},
{
"text": "In a recent paper, Yallop et al. (2005) reported experiments on the acquisition of syntactic subcategorisation patterns for English adjectives.",
"cite_spans": [
{
"start": 19,
"end": 39,
"text": "Yallop et al. (2005)",
"ref_id": "BIBREF35"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related work",
"sec_num": "7"
},
{
"text": "Apart from the above research with a classificatory flavour, other lines of research exploited lexical relations among adjectives for Word Sense Disambiguation (Justeson and Katz, 1995; Chao and Dyer, 2000) . Work by Lapata (2001) , contrary to the studies mentioned so far, focused on the meaning of adjective-noun combinations, not on that of adjectives alone.",
"cite_spans": [
{
"start": 160,
"end": 185,
"text": "(Justeson and Katz, 1995;",
"ref_id": "BIBREF17"
},
{
"start": 186,
"end": 206,
"text": "Chao and Dyer, 2000)",
"ref_id": "BIBREF8"
},
{
"start": 217,
"end": 230,
"text": "Lapata (2001)",
"ref_id": "BIBREF19"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related work",
"sec_num": "7"
},
{
"text": "This paper has presented an architecture for the semantic classification of Catalan adjectives that explicitly includes polysemous classes. The focus of the architecture was on two issues: (i) finding an appropriate set of linguistic features, and (ii) defining an adequate architecture for the task. The investigation and comparison of features at various linguistic levels has shown that morphology plays a major role for the target classification, despite the caveats raised in the discussion. Morphological features related to derivational processes are among the simplest types of features to extract, so that the approach can be straightforwardly extended to languages similar to Catalan with no extensive need of resources.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "8"
},
{
"text": "Furthermore, we have argued that polysemy acquisition naturally suits multi-label classification architectures. We have implemented a standard architecture for this class of problems, and demonstrated its applicability and success. The general nature of the architecture should be useful for related tasks that involve polysemy within the area of automatic lexical acquisition.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "8"
},
{
"text": "Our work has focused on a broad classification of the adjectives, similarly to Merlo and Stevenson (2001) , who classified transitive English verbs into three semantic classes. The small number of classes might be considered as an over-simplification of adjective semantics, but the simplified setup facilitates a detailed qualitative evaluation. In addition, as there has been virtually no work on the acquisition of semantic classes for adjectives, it seems sensible to start with a small number of classes and incrementally build upon that. Previous work has demonstrated that multi-label classification is applicable also to a large number of classes as used in, e.g., document categorisation (Schapire and Singer, 2000) . This potential can be exploited in future work, addressing a finer-grained adjective classification.",
"cite_spans": [
{
"start": 79,
"end": 105,
"text": "Merlo and Stevenson (2001)",
"ref_id": "BIBREF22"
},
{
"start": 697,
"end": 724,
"text": "(Schapire and Singer, 2000)",
"ref_id": "BIBREF31"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "8"
},
{
"text": "Finally, we have demonstrated that the combination of different types of linguistic evidence boosts the performance of the system beyond the best single type of information: ensemble classifiers are a more adequate way to combine the linguistic levels of description than simply merging all features for tree construction. Using a simple, majority voting ensemble classifier, the accuracy jumped from 62.5% (best single classifier) to 84%. This result is impressive by itself, and also in comparison to similar work such as (Rigau et al., 1997) , who achieved a 9% improvement on a similar task. Our insights are therefore useful in related work which involves the selection of linguistic features in Machine Learning experiments.",
"cite_spans": [
{
"start": 524,
"end": 544,
"text": "(Rigau et al., 1997)",
"ref_id": "BIBREF28"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "8"
},
{
"text": "Future work involves three main lines of research. First, the refinement of the classification itself, based on the results of the experiments presented. Second, the use of additional linguistic evidence that contributes to the semantic class distinctions (e.g., selectional restrictions). Third, the application of the acquired information to broader NLP tasks. For example, given that each semantic class exhibits a particular syntactic behaviour, infor-mation on the semantic class should improve POStagging for adjective-noun and adjective-participle ambiguities, probably the most difficult distinctions both for humans and computers (Marcus et al., 1993; Brants, 2000) . Also, semantic classes might be useful in terminology extraction, where, presumably, object adjectives participate in terms more often than basic adjectives. 4 ",
"cite_spans": [
{
"start": 639,
"end": 660,
"text": "(Marcus et al., 1993;",
"ref_id": "BIBREF20"
},
{
"start": 661,
"end": 674,
"text": "Brants, 2000)",
"ref_id": "BIBREF5"
},
{
"start": 835,
"end": 836,
"text": "4",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "8"
},
{
"text": "In level all, different features were used for each of the three classes.Table 1reports the mean number of features across the three classes.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Note that the corrected resampled t-test can only compare accuracies obtained under two conditions (algorithms or, as is our case, feature sets); ANOVA would be more adequate. In the field of Machine Learning, there is no established correction for ANOVA for the purposes of testing differences in accuracy(Bouckaert, 2004). Therefore, we used multiple t-tests instead, which increases the overall error probability of the results for the significance tests.3 The accuracy for each decision was computed independently. For instance, a BE adjective was judged correct within the basic class iff the decision was basic; correct within the event class iff the decision was event; and correct within the object class iff the decision was non-object.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "We thank Roser Sanrom\u00e0 for providing us with the manually annotated database of adjectives, and for being part of the Gold Standard annotation committee. Special thanks also to the Institut d'Estudis Catalans for proding us with the research corpus. The comments of three anonymous reviewers helped improve the paper. Finally, the financial support of the Universitat Pompeu Fabra and its Translation and Philology Department and the Fundaci\u00f3n Caja Madrid is gratefully acknowledged.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "CATCG: a general purpose parsing tool applied",
"authors": [
{
"first": "\u00c0",
"middle": [],
"last": "Alsina",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Badia",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Boleda",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Bott",
"suffix": ""
},
{
"first": "\u00c0",
"middle": [],
"last": "Gil",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Quixal",
"suffix": ""
},
{
"first": "O",
"middle": [],
"last": "Valent\u00edn",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of Third International Conference on Language Resources and Evaluation (LREC-02)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "\u00c0. Alsina, T. Badia, G. Boleda, S. Bott, \u00c0. Gil, M. Quixal, and O. Valent\u00edn. 2002. CATCG: a general purpose parsing tool applied. In Proceedings of Third International Conference on Language Resources and Evaluation (LREC-02), Las Palmas, Spain.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "An approach to automatic annotation of functional information to adjectives with an application to German",
"authors": [
{
"first": "B",
"middle": [],
"last": "Bohnet",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Klatt",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Wanner",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the LREC Workshop on Linguistic Knowledge Acquisition and Representation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "B. Bohnet, S. Klatt, and L. Wanner. 2002. An approach to automatic annotation of functional information to adjectives with an application to German. In Proceed- ings of the LREC Workshop on Linguistic Knowledge Acquisition and Representation, Las Palmas, Spain.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Acquisition of semantic classes for adjectives from distributional evidence",
"authors": [
{
"first": "G",
"middle": [],
"last": "Boleda",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Badia",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Batlle",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of the 20th International Conference on Computational Linguistics (COLING 2004)",
"volume": "",
"issue": "",
"pages": "1119--1125",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "G. Boleda, T. Badia, and E. Batlle. 2004. Acquisition of semantic classes for adjectives from distributional evidence. In Proceedings of the 20th International Conference on Computational Linguistics (COLING 2004), pages 1119-1125, Geneva, Switzerland.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Morphology vs. syntax in adjective class acquisition",
"authors": [
{
"first": "G",
"middle": [],
"last": "Boleda",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Badia",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Schulte Im Walde",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of the ACL-SIGLEX 2005 Workshop on Deep Lexical Acquisition",
"volume": "",
"issue": "",
"pages": "1119--1125",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "G. Boleda, T. Badia, and S. Schulte im Walde. 2005. Morphology vs. syntax in adjective class acquisition. In Proceedings of the ACL-SIGLEX 2005 Workshop on Deep Lexical Acquisition, pages 1119-1125, Ann Arbor, USA.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Estimating replicability of classifier learning experiments",
"authors": [
{
"first": "R",
"middle": [],
"last": "Bouckaert",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of ICML",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Bouckaert. 2004. Estimating replicability of classifier learning experiments. In Proceedings of ICML.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Inter-annotator agreement for a german newspaper corpus",
"authors": [
{
"first": "T",
"middle": [],
"last": "Brants",
"suffix": ""
}
],
"year": 2000,
"venue": "Second International Conference on Language Resources and Evaluation (LREC-2000)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T. Brants. 2000. Inter-annotator agreement for a german newspaper corpus. In Second International Confer- ence on Language Resources and Evaluation (LREC- 2000), Athens, Greece.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Random forests",
"authors": [
{
"first": "L",
"middle": [],
"last": "Breiman",
"suffix": ""
}
],
"year": 2001,
"venue": "Mach. Learn",
"volume": "45",
"issue": "",
"pages": "5--23",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "L. Breiman. 2001. Random forests. Mach. Learn., 45:5- 23.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Word sense disambiguation of adjectives using probabilistic networks",
"authors": [
{
"first": "G",
"middle": [],
"last": "Chao",
"suffix": ""
},
{
"first": "M",
"middle": [
"G"
],
"last": "Dyer",
"suffix": ""
}
],
"year": 2000,
"venue": "Proceedings of COLING",
"volume": "",
"issue": "",
"pages": "152--158",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "G. Chao and M. G. Dyer. 2000. Word sense disambigua- tion of adjectives using probabilistic networks. In Pro- ceedings of COLING, pages 152-158.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Approximate statistical tests for comparing supervised classification learning algorithms",
"authors": [
{
"first": "T",
"middle": [
"G"
],
"last": "Dietterich",
"suffix": ""
}
],
"year": 1998,
"venue": "Neural Computation",
"volume": "10",
"issue": "7",
"pages": "1895--1924",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T.G. Dietterich. 1998. Approximate statistical tests for comparing supervised classification learning algo- rithms. Neural Computation, 10(7):1895-1924.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "An experimental comparison of three methods for constructing ensembles of decision trees: Bagging, boosting, and randomization",
"authors": [
{
"first": "T",
"middle": [
"G"
],
"last": "Dietterich",
"suffix": ""
}
],
"year": 2000,
"venue": "Mach. Learn",
"volume": "40",
"issue": "",
"pages": "5--23",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T.G. Dietterich. 2000. An experimental comparison of three methods for constructing ensembles of decision trees: Bagging, boosting, and randomization. Mach. Learn., 40:5-23.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "The Handbook of Brain Theory and Neural Networks",
"authors": [
{
"first": "T",
"middle": [
"G"
],
"last": "Dietterich",
"suffix": ""
}
],
"year": 2002,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T.G. Dietterich. 2002. Ensemble learning. In M. A. Ar- bib, editor, The Handbook of Brain Theory and Neural Networks. The MIT Press.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Experiments with a new boosting algorithm",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Freund",
"suffix": ""
},
{
"first": "R",
"middle": [
"E"
],
"last": "Schapire",
"suffix": ""
}
],
"year": 1996,
"venue": "Proceedings of ICML",
"volume": "",
"issue": "",
"pages": "148--156",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Y. Freund and R.E. Schapire. 1996. Experiments with a new boosting algorithm. In Proceedings of ICML, pages 148-156.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Collective multilabel classification",
"authors": [
{
"first": "N",
"middle": [],
"last": "Ghamrawi",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Mccallum",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of 14th Conf. on Information and Knowledge Management",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "N. Ghamrawi and A. McCallum. 2005. Collective multi- label classification. In Proceedings of 14th Conf. on Information and Knowledge Management.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Towards the automatic identification of adjectival scales: Clustering adjectives according to meaning",
"authors": [
{
"first": "V",
"middle": [],
"last": "Hatzivassiloglou",
"suffix": ""
},
{
"first": "K",
"middle": [
"R"
],
"last": "Mckeown",
"suffix": ""
}
],
"year": 1993,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "172--182",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "V. Hatzivassiloglou and K. R. McKeown. 1993. Towards the automatic identification of adjectival scales: Clus- tering adjectives according to meaning. In Proceed- ings of ACL, pages 172-182.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Predicting the semantic orientation of adjectives",
"authors": [
{
"first": "V",
"middle": [],
"last": "Hatzivassiloglou",
"suffix": ""
},
{
"first": "K",
"middle": [
"R"
],
"last": "Mckeown",
"suffix": ""
}
],
"year": 1997,
"venue": "Proceedings of ACL/EACL",
"volume": "",
"issue": "",
"pages": "174--181",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "V. Hatzivassiloglou and K.R. McKeown. 1997. Predict- ing the semantic orientation of adjectives. In Proceed- ings of ACL/EACL, pages 174-181.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Effects of adjective orientation and gradability on sentence subjectivity",
"authors": [
{
"first": "V",
"middle": [],
"last": "Hatzivassiloglou",
"suffix": ""
},
{
"first": "J",
"middle": [
"M"
],
"last": "Wiebe",
"suffix": ""
}
],
"year": 2000,
"venue": "Proceedings of COLING",
"volume": "",
"issue": "",
"pages": "299--305",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "V. Hatzivassiloglou and J. M. Wiebe. 2000. Effects of adjective orientation and gradability on sentence sub- jectivity. In Proceedings of COLING, pages 299-305, Morristown, NJ, USA. Association for Computational Linguistics.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Principled disambiguation: Discriminating adjective senses with modified nouns",
"authors": [
{
"first": "J",
"middle": [
"S"
],
"last": "Justeson",
"suffix": ""
},
{
"first": "S",
"middle": [
"M"
],
"last": "Katz",
"suffix": ""
}
],
"year": 1995,
"venue": "Computational Linguistics",
"volume": "21",
"issue": "1",
"pages": "1--27",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. S. Justeson and S. M. Katz. 1995. Principled disam- biguation: Discriminating adjective senses with modi- fied nouns. Computational Linguistics, 21(1):1-27.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Clustering polysemic subcategorization frame distributions semantically",
"authors": [
{
"first": "A",
"middle": [],
"last": "Korhonen",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Krymolowski",
"suffix": ""
},
{
"first": "Z",
"middle": [],
"last": "Marx",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "64--71",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A. Korhonen, Y. Krymolowski, and Z. Marx. 2003. Clustering polysemic subcategorization frame distri- butions semantically. In Proceedings of ACL, pages 64-71.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "A corpus-based account of regular polysemy: The case of context-sensitive adjectives",
"authors": [
{
"first": "M",
"middle": [],
"last": "Lapata",
"suffix": ""
}
],
"year": 2001,
"venue": "Proceedings of NAACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M. Lapata. 2001. A corpus-based account of regular polysemy: The case of context-sensitive adjectives. In Proceedings of NAACL.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Building a large annotated corpus of English: The Penn Treebank",
"authors": [
{
"first": "M",
"middle": [],
"last": "Marcus",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Santorini",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Marcinkiewicz",
"suffix": ""
}
],
"year": 1993,
"venue": "Computational Linguistics",
"volume": "19",
"issue": "",
"pages": "313--330",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M. Marcus, B. Santorini, and M. Marcinkiewicz. 1993. Building a large annotated corpus of English: The Penn Treebank. Computational Linguistics, 19:313- 330.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Flexible text segmentation with structured multilabel classification",
"authors": [
{
"first": "R",
"middle": [],
"last": "Mcdonald",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Crammer",
"suffix": ""
},
{
"first": "F",
"middle": [],
"last": "Pereira",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of HLT-EMNLP",
"volume": "",
"issue": "",
"pages": "987--994",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. McDonald, K. Crammer, and F. Pereira. 2005. Flexi- ble text segmentation with structured multilabel classi- fication. In Proceedings of HLT-EMNLP, pages 987- 994.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "Automatic verb classification based on statistical distributions of argument structure",
"authors": [
{
"first": "P",
"middle": [],
"last": "Merlo",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Stevenson",
"suffix": ""
}
],
"year": 2001,
"venue": "Comp. Ling",
"volume": "27",
"issue": "3",
"pages": "373--408",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "P. Merlo and S. Stevenson. 2001. Automatic verb clas- sification based on statistical distributions of argument structure. Comp. Ling., 27(3):373-408.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "Inference for the generalization error",
"authors": [
{
"first": "C",
"middle": [],
"last": "Nadeau",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Bengio",
"suffix": ""
}
],
"year": 2003,
"venue": "Mach. Learn",
"volume": "52",
"issue": "3",
"pages": "239--281",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "C. Nadeau and Y. Bengio. 2003. Inference for the gener- alization error. Mach. Learn., 52(3):239-281.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Ontological Semantics",
"authors": [
{
"first": "S",
"middle": [],
"last": "Nirenburg",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Raskin",
"suffix": ""
}
],
"year": 2004,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "S. Nirenburg and V. Raskin. 2004. Ontological Seman- tics. MIT Press.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Distributional Clustering of English Words",
"authors": [
{
"first": "F",
"middle": [],
"last": "Pereira",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Tishby",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Lee",
"suffix": ""
}
],
"year": 1993,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "183--190",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "F. Pereira, N. Tishby, and L. Lee. 1993. Distributional Clustering of English Words. In Proceedings of ACL, pages 183-190.",
"links": null
},
"BIBREF26": {
"ref_id": "b26",
"title": "C4.5: Programs for Machine Learning",
"authors": [
{
"first": "R",
"middle": [],
"last": "Quinlan",
"suffix": ""
}
],
"year": 1993,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Quinlan. 1993. C4.5: Programs for Machine Learn- ing. Morgan Kaufmann.",
"links": null
},
"BIBREF27": {
"ref_id": "b27",
"title": "An applied ontological semantic microtheory of adjective meaning for natural language processing",
"authors": [
{
"first": "V",
"middle": [],
"last": "Raskin",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Nirenburg",
"suffix": ""
}
],
"year": 1998,
"venue": "Mach. Trans",
"volume": "13",
"issue": "2-3",
"pages": "135--227",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "V. Raskin and S. Nirenburg. 1998. An applied ontologi- cal semantic microtheory of adjective meaning for nat- ural language processing. Mach. Trans., 13(2-3):135- 227.",
"links": null
},
"BIBREF28": {
"ref_id": "b28",
"title": "Combining unsupervised lexical knowledge methods for word sense disambiguation",
"authors": [
{
"first": "G",
"middle": [],
"last": "Rigau",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Atserias",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Agirre",
"suffix": ""
}
],
"year": 1997,
"venue": "Proceedings of EACL",
"volume": "",
"issue": "",
"pages": "48--55",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "G. Rigau, J. Atserias, and E. Agirre. 1997. Combin- ing unsupervised lexical knowledge methods for word sense disambiguation. In Proceedings of EACL, pages 48-55.",
"links": null
},
"BIBREF29": {
"ref_id": "b29",
"title": "Inducing a Semantically Annotated Lexicon via EM-Based Clustering",
"authors": [
{
"first": "M",
"middle": [],
"last": "Rooth",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Riezler",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Prescher",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Carroll",
"suffix": ""
},
{
"first": "F",
"middle": [],
"last": "Beil",
"suffix": ""
}
],
"year": 1999,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M. Rooth, S. Riezler, D. Prescher, G. Carroll, and F. Beil. 1999. Inducing a Semantically Annotated Lexicon via EM-Based Clustering. In Proceedings of ACL.",
"links": null
},
"BIBREF30": {
"ref_id": "b30",
"title": "Aspectes morfol\u00f2gics i sint\u00e0ctics dels adjectius en catal\u00e0",
"authors": [
{
"first": "R",
"middle": [],
"last": "Sanrom\u00e0",
"suffix": ""
}
],
"year": 2003,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Sanrom\u00e0. 2003. Aspectes morfol\u00f2gics i sint\u00e0ctics dels adjectius en catal\u00e0. Master's thesis, Universitat Pom- peu Fabra.",
"links": null
},
"BIBREF31": {
"ref_id": "b31",
"title": "Boostexter: A boosting-based system for text categorization",
"authors": [
{
"first": "R",
"middle": [
"E"
],
"last": "Schapire",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Singer",
"suffix": ""
}
],
"year": 2000,
"venue": "Mach. Learn",
"volume": "39",
"issue": "2-3",
"pages": "135--168",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R.E. Schapire and Y. Singer. 2000. Boostexter: A boosting-based system for text categorization. Mach. Learn., 39(2-3):135-168.",
"links": null
},
"BIBREF32": {
"ref_id": "b32",
"title": "Semi-supervised verb class discovery using noisy features",
"authors": [
{
"first": "S",
"middle": [],
"last": "Stevenson",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Joanis",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of CoNLL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "S. Stevenson and E. Joanis. 2003. Semi-supervised verb class discovery using noisy features. In Proceedings of CoNLL.",
"links": null
},
"BIBREF33": {
"ref_id": "b33",
"title": "Improving data driven wordclass tagging by system combination",
"authors": [
{
"first": "H",
"middle": [],
"last": "Van Halteren",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Zavrel",
"suffix": ""
},
{
"first": "W",
"middle": [],
"last": "Daelemans",
"suffix": ""
}
],
"year": 1998,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "491--497",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "H. van Halteren, J. Zavrel, and W. Daelemans. 1998. Im- proving data driven wordclass tagging by system com- bination. In Proceedings of ACL, pages 491-497.",
"links": null
},
"BIBREF34": {
"ref_id": "b34",
"title": "Data Mining: Practical Machine Learning Tools and Techniques with Java Implementations",
"authors": [
{
"first": "I",
"middle": [
"H"
],
"last": "Witten",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Frank",
"suffix": ""
}
],
"year": 2005,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "I.H. Witten and E. Frank. 2005. Data Mining: Practi- cal Machine Learning Tools and Techniques with Java Implementations. Morgan Kaufmann.",
"links": null
},
"BIBREF35": {
"ref_id": "b35",
"title": "Automatic acquisition of adjectival subcategorization from corpora",
"authors": [
{
"first": "J",
"middle": [],
"last": "Yallop",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Korhonen",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Briscoe",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of ACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Yallop, A. Korhonen, and T. Briscoe. 2005. Auto- matic acquisition of adjectival subcategorization from corpora. In Proceedings of ACL, Ann Arbor, Michi- gan.",
"links": null
}
},
"ref_entries": {
"TABREF0": {
"type_str": "table",
"html": null,
"content": "<table><tr><td colspan=\"2\">Level Explanation</td><td/><td># Features</td></tr><tr><td colspan=\"3\">morph morphological (derivational) properties</td><td>2</td></tr><tr><td>func</td><td>syntactic function</td><td/><td>4</td></tr><tr><td>uni</td><td colspan=\"2\">uni-gram distribution</td><td>24</td></tr><tr><td>bi</td><td colspan=\"2\">bi-gram distribution</td><td>50</td></tr><tr><td>sem</td><td colspan=\"2\">distributional cues of semantic properties</td><td>18</td></tr><tr><td>all</td><td colspan=\"2\">combination of the 5 linguistic levels</td><td>10.3</td></tr><tr><td/><td colspan=\"2\">Table 1: Linguistic levels as feature sets.</td></tr><tr><td colspan=\"3\">tion from the adjective database. Syntactic and se-</td></tr><tr><td colspan=\"3\">mantic features encode distributional properties of</td></tr><tr><td colspan=\"3\">adjectives. Syntactic features comprise three sub-</td></tr><tr><td colspan=\"3\">types: (i) the syntactic function (level func) of the</td></tr><tr><td colspan=\"3\">adjective, as assigned by a shallow Constraint Gram-</td></tr><tr><td colspan=\"3\">mar (Alsina et al., 2002), distinguishing the modifier</td></tr><tr><td colspan=\"3\">(pre-nominal or post-nominal) and predicative func-</td></tr><tr><td colspan=\"3\">tions; (ii) a unigram distribution (level uni), inde-</td></tr><tr><td colspan=\"3\">pendently encoding the parts of speech (POS) of the</td></tr><tr><td colspan=\"3\">words preceding and following the adjective, respec-</td></tr><tr><td>tively; and property</td><td/><td>#</td></tr><tr><td>non-restrictivity</td><td/><td>1</td></tr><tr><td>predicativity</td><td/><td>4</td></tr><tr><td>gradability</td><td/><td>4</td></tr><tr><td colspan=\"2\">syntactic function of head noun</td><td>3</td></tr><tr><td>distance to the head noun</td><td/><td>1</td></tr><tr><td colspan=\"3\">binaryhood (adjectives with two arguments) 1</td></tr><tr><td>agreement properties</td><td/><td>2</td></tr></table>",
"num": null,
"text": "(iii) a bigram distribution (level bi), the POS bigram around the target adjective, considering only the 50 most frequent bigrams to avoid sparse features. Semantic features (level sem) expand syntactic features with heterogeneous shallow cues of semantic properties.Table 2lists the semantic properties encoded in the features, as well as the number of heuristic cues defined for each property. As an example, one of the shallow cues used for gradability was the presence of degree adverbs (m\u00e9s 'more', menys 'less') to the left of the target adjectives. The last set of features, all, combines features from all levels of description. However, it does not contain all features, but a selection of the most relevant ones (further details in Section 4.3)."
},
"TABREF1": {
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null,
"text": ""
},
"TABREF3": {
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null,
"text": "Accuracy results for binary decisions."
},
"TABREF6": {
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null,
"text": "Levels all and morph against the Gold Standard."
},
"TABREF7": {
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null,
"text": "84.0 \u00b10.06 95.7 \u00b10.02 func+uni+bi+sem 81.5 \u00b10.04 95.9 \u00b10.01 morph+func+sem+all 72.4 \u00b10.03 89.3 \u00b10.02 bl 51.0 \u00b10.0 65.2 \u00b10.0 all 62.3 \u00b12.3 90.7 \u00b11.6"
},
"TABREF8": {
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null,
"text": "Results for ensemble classifier."
}
}
}
} |