File size: 85,823 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 |
{
"paper_id": "I11-1011",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:31:39.984724Z"
},
"title": "Single and Multi-objective Optimization for Feature Selection in Anaphora Resolution",
"authors": [
{
"first": "Sriparna",
"middle": [],
"last": "Saha",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "IIT Patna",
"location": {
"country": "India"
}
},
"email": "sriparna@iitp.ac.in"
},
{
"first": "Asif",
"middle": [],
"last": "Ekbal",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "IIT Patna",
"location": {
"country": "India"
}
},
"email": ""
},
{
"first": "Olga",
"middle": [],
"last": "Uryupina",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Trento",
"location": {}
},
"email": "uryupina@unitn.it"
},
{
"first": "Massimo",
"middle": [],
"last": "Poesio",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Trento",
"location": {}
},
"email": "poesio@essex.ac.uk"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "There is no generally accepted metric for measuring the performance of anaphora resolution systems, and the existing metrics-MUC, B 3 , CEAF, Blanc, among others-tend to reward significantly different behaviors. Systems optimized according to one metric tend to perform poorly with respect to other ones, making it very difficult to compare anaphora resolution systems, as clearly shown by the results of the SEMEVAL 2010 Multilingual Coreference task. One solution would be to find a single completely satisfactory metric, but it's not clear whether this is possible and at any rate it is not going to happen any time soon. An alternative is to optimize models according to multiple metrics simultaneously. In this paper, we show, first of all, that this is possible to develop such models using Multi-objective Optimization (MOO) techniques based on Genetic Algorithms. Secondly, we show that optimizing according to multiple metrics simultaneously may result in better results with respect to each individual metric than optimizing according to that metric only.",
"pdf_parse": {
"paper_id": "I11-1011",
"_pdf_hash": "",
"abstract": [
{
"text": "There is no generally accepted metric for measuring the performance of anaphora resolution systems, and the existing metrics-MUC, B 3 , CEAF, Blanc, among others-tend to reward significantly different behaviors. Systems optimized according to one metric tend to perform poorly with respect to other ones, making it very difficult to compare anaphora resolution systems, as clearly shown by the results of the SEMEVAL 2010 Multilingual Coreference task. One solution would be to find a single completely satisfactory metric, but it's not clear whether this is possible and at any rate it is not going to happen any time soon. An alternative is to optimize models according to multiple metrics simultaneously. In this paper, we show, first of all, that this is possible to develop such models using Multi-objective Optimization (MOO) techniques based on Genetic Algorithms. Secondly, we show that optimizing according to multiple metrics simultaneously may result in better results with respect to each individual metric than optimizing according to that metric only.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "In anaphora resolution, 1 as in other HLT tasks, optimization to a metric is essential to achieve good performance (Hoste, 2005; Uryupina, 2010) . However, many evaluation metrics have been proposed for anaphora resolution, each capturing what seems to be a key intuition about the task: from MUC (Vilain et al., 1995) to B 3 (Bagga and Baldwin, 1998) , from the ACE metric (Doddington et al., 2004) to CEAF (Luo, 2005) to BLANC (Recasens and Hovy, 2011) . And unlike in other areas of HLT, none has really taken over. This would not matter so much if those metrics were to reward the same systems; but in fact, as dramatically demonstrated by the results of the Coreference Task at SEMEVAL 2010 (Recasens et al., 2010) , the opposite is true-almost every system could come on top depending on which metric was chosen.",
"cite_spans": [
{
"start": 115,
"end": 128,
"text": "(Hoste, 2005;",
"ref_id": "BIBREF6"
},
{
"start": 129,
"end": 144,
"text": "Uryupina, 2010)",
"ref_id": "BIBREF19"
},
{
"start": 297,
"end": 318,
"text": "(Vilain et al., 1995)",
"ref_id": "BIBREF21"
},
{
"start": 337,
"end": 351,
"text": "Baldwin, 1998)",
"ref_id": "BIBREF0"
},
{
"start": 374,
"end": 399,
"text": "(Doddington et al., 2004)",
"ref_id": "BIBREF4"
},
{
"start": 408,
"end": 419,
"text": "(Luo, 2005)",
"ref_id": "BIBREF9"
},
{
"start": 429,
"end": 454,
"text": "(Recasens and Hovy, 2011)",
"ref_id": "BIBREF14"
},
{
"start": 696,
"end": 719,
"text": "(Recasens et al., 2010)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "It seems unlikely that the field will converge on a single metric any time soon. Given that many of the proposed metrics do capture what would seem to be plausible intuitions, it would seem desirable to develop methods to optimize systems according to more than one metric at once-in particular, according to at least one metric of what we might call the 'link-based' cluster of metrics (e.g., MUC) and at least one of what we will call the 'entity-based' cluster (e.g., CEAF).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "As it happens, techniques for doing just that have been developed in the area of Genetic Algorithms: so-called multi-objective optimization (MOO) (Deb, 2001) techniques. In this paper, we will show how these techniques can be used to optimize anaphora resolution models (we focused for the time being on feature selection) by looking for a solution in the space defined by a multiplicity of metrics (we used MUC and CEAF (in two variants) as the optimization functions). Perhaps the most interesting result of this work is the finding that by working in such a multi-metric space it is possible to find solutions that are better with respect to an individual metric than when trying to optimize for that metric alone-which arguably suggests that indeed both families of metrics capture some fundamental intuition about anaphora, and taking into account both intuitions we avoid local optima.",
"cite_spans": [
{
"start": 146,
"end": 157,
"text": "(Deb, 2001)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The structure of the paper is as follows. We first review the literature on using genetic algorithms for both single function and multi function opti-mization. Next, we discuss the particular method of multi-objective optimization we used in this paper, Non-Dominated Sorting Genetic Algorithm II (Deb et al., 2002) . After that we discuss how the method was used, and present our results. We then compare our work with other approaches to optimization for anaphora found in the literature.",
"cite_spans": [
{
"start": 297,
"end": 315,
"text": "(Deb et al., 2002)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "A great number of statistical approaches to anaphora resolution have been proposed in the past ten years. These approaches differ with respect to their underlying models (e.g., mention pair model (Soon et al., 2001 ) vs. tournament model (Iida et al., 2003; Yang et al., 2005) , vs. entity-model (Luo et al., 2004) ), machine learners (e.g., decision trees vs. maximum entropy vs. SVMs vs. TiMBL) and their parameters, and with respect to feature sets used. There have been, however, only few attempts at explicit optimization of these aspects, and in those few cases, optimization tends to be done by hand. An early step in this direction was the work by Ng and Cardie (2002) , who developed a rich feature set including 53 features, but reported no significant improvement over their baseline when all these features were used with the MUC6 and MUC7 corpora. They then proceeded to manually select a subset of features that did yield better results for the MUC-6/7 datasets. A much larger scale and very systematic effort of manual feature selection over the same dataset was carried out by Uryupina (2007) , who evaluated over 600 features.",
"cite_spans": [
{
"start": 196,
"end": 214,
"text": "(Soon et al., 2001",
"ref_id": null
},
{
"start": 238,
"end": 257,
"text": "(Iida et al., 2003;",
"ref_id": "BIBREF7"
},
{
"start": 258,
"end": 276,
"text": "Yang et al., 2005)",
"ref_id": "BIBREF22"
},
{
"start": 296,
"end": 314,
"text": "(Luo et al., 2004)",
"ref_id": "BIBREF8"
},
{
"start": 656,
"end": 676,
"text": "Ng and Cardie (2002)",
"ref_id": "BIBREF11"
},
{
"start": 1093,
"end": 1108,
"text": "Uryupina (2007)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background: Optimizing for Anaphora Resolution",
"sec_num": "2"
},
{
"text": "The first systematic attempt at automatic optimization of anaphora resolution we are aware of was carried out by Hoste (2005) , who investigated the possibility of using genetic algorithms for automatic optimization of both feature selection and of learning parameters, also considering two different machine learners, TiMBL and Ripper. Her results suggest that such techniques yield improvements on the MUC-6/7 datasets. Recasens and Hovy (2009) carried out an investigation of feature selection for Spanish using the ANCORA corpus.",
"cite_spans": [
{
"start": 113,
"end": 125,
"text": "Hoste (2005)",
"ref_id": "BIBREF6"
},
{
"start": 422,
"end": 446,
"text": "Recasens and Hovy (2009)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background: Optimizing for Anaphora Resolution",
"sec_num": "2"
},
{
"text": "These approaches focused on a single metric only; the one proposal simultaneously to consider multiple metrics, Zhao and Ng (2010) still optimized for each metric individually.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background: Optimizing for Anaphora Resolution",
"sec_num": "2"
},
{
"text": "The effect of optimization on anaphora resolution was dramatically demonstrated by Uryupina's contribution to SEMEVAL 2010 Multilingual Coreference Task (Uryupina, 2010) . Uryupina directly optimizes two parameters of her system: the choice of a model (mention-pair vs. ILP with various constraints) and the definition of mention types for training separate classifiers. The optimization is done on the development data in a brute-force fashion, in order to maximize the performance according to a pre-defined metric (MUC, CEAF or BLANC). The results on the SEMEVAL-10 dataset clearly show that existing metrics of coreference rely on different intuitions and therefore a system, optimized for a particular metric, might show inferior results for the other ones. For example, the reported BLANC difference between the runs optimized for BLANC and CEAF is around 10 percentage points.",
"cite_spans": [
{
"start": 153,
"end": 169,
"text": "(Uryupina, 2010)",
"ref_id": "BIBREF19"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background: Optimizing for Anaphora Resolution",
"sec_num": "2"
},
{
"text": "This highlights the importance of the multiobjective optimization (MOO) for coreference, that suggests a family of systems, showing reliable performance according to all the desired metrics. A form of MOO was applied to coreference by Munson et al. (2005) . Their general conclusion was negative, stating that \"ensemble selection seems too unreliable for use in NLP\", but they did see some improvements for coreference.",
"cite_spans": [
{
"start": 235,
"end": 255,
"text": "Munson et al. (2005)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background: Optimizing for Anaphora Resolution",
"sec_num": "2"
},
{
"text": "In this section, we review optimization techniques using genetic algorithms (GAs) (Goldberg, 1989) . We first discuss single objective optimization, that can optimize according to a single objective function, and then multi-objective optimization (MOO), that can optimize more than one objective function, in particular, a popular MOO technique named Non-dominated Sorting Genetic Algorithm (NSGA)-II (Deb et al., 2002) .",
"cite_spans": [
{
"start": 82,
"end": 98,
"text": "(Goldberg, 1989)",
"ref_id": "BIBREF5"
},
{
"start": 401,
"end": 419,
"text": "(Deb et al., 2002)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Optimization with Genetic Algorithms",
"sec_num": "3"
},
{
"text": "Genetic algorithms (GAs) (Goldberg, 1989) are randomized search and optimization techniques guided by the principles of evolution and natural genetics. In GAs the parameters of the search space are encoded in the form of strings (called chromosomes). A collection of such strings is called a population. Initially, a random population is created, which represents different points in the search space. An objective or fitness function is associated with each string that represents the degree of goodness of the string. Based on the principle of survival of the fittest, a few of the strings are selected and each is assigned a number of copies that go into the mating pool. Biologically inspired op-erators like crossover and mutation are applied on these strings to yield a new generation of strings. The processes of selection, crossover and mutation continues for a fixed number of generations or till a termination condition is satisfied.",
"cite_spans": [
{
"start": 25,
"end": 41,
"text": "(Goldberg, 1989)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Genetic Algorithms",
"sec_num": "3.1"
},
{
"text": "Multi-objective optimization (MOO) can be formally stated as follows (Deb, 2001 ). Find the vectors",
"cite_spans": [
{
"start": 69,
"end": 79,
"text": "(Deb, 2001",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-objective Optimization",
"sec_num": "3.2"
},
{
"text": "x * = [x * 1 , x * 2 , . . . , x * n ]",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-objective Optimization",
"sec_num": "3.2"
},
{
"text": "T of decision variables that simultaneously optimize the M objective values",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-objective Optimization",
"sec_num": "3.2"
},
{
"text": "{f 1 (x), f 2 (x), . . . , f M (x)}",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-objective Optimization",
"sec_num": "3.2"
},
{
"text": "while satisfying the constraints, if any.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-objective Optimization",
"sec_num": "3.2"
},
{
"text": "An important concept in MOO is that of domination. In the context of a maximization problem, a solution",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-objective Optimization",
"sec_num": "3.2"
},
{
"text": "x i is said to dominate x j if \u2200k \u2208 1, 2, . . . , M , f k (x i ) \u2265 f k (x j ) and \u2203k \u2208 1, 2, . . . , M , such that f k (x i ) > f k (x j ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-objective Optimization",
"sec_num": "3.2"
},
{
"text": "Among a set of solutions P , the nondominated set of solutions P \u2032 are those that are not dominated by any member of the set P . The nondominated set of the entire search space S is called the globally Pareto-optimal set. In general, a MOO algorithm usually admits a set of solutions not dominated by any solution encountered by it.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multi-objective Optimization",
"sec_num": "3.2"
},
{
"text": "Genetic algorithms (GAs) are known to be more effective than classical methods such as weighted metrics, goal programming (Deb, 2001) , for solving MOO primarily because of their populationbased nature. A particularly popular GA of this type is NSGA-II (Deb et al., 2002) . In NSGA-II, initially a random parent population P 0 is created and the population is sorted based on the partial order defined by the non-domination relation. This results in a sequence of nondominated fronts. Each solution is assigned a fitness value which is equal to its non-domination level in the partial order. A child population Q 0 of size N is then created from the parent population P 0 by using binary tournament selection, recombination, and mutation operators. In general, in the t th iteration, a combined population R t = P t + Q t is formed. The size of R t is 2N , as the size of both P t and Q t is N . All the solutions of R t are sorted according to non-domination. If the total number of solutions belonging to the best nondominated set F 1 is smaller than N , then F 1 is to- Figure 1 : Chromosome representation for GA based feature selection tally included in P (t+1) . The remaining members of the population P (t+1) are chosen from the subsequent nondominated fronts in the order of their ranking. To choose exactly N solutions, the solutions of the last included front are sorted using the crowded comparison operator (Deb et al., 2002) and the best among them (i.e., those with lower crowding distance) are selected to fill in the available slots in P (t+1) . The new population P (t+1) is then used for selection, crossover and mutation to create a population Q (t+1) of size N .",
"cite_spans": [
{
"start": 122,
"end": 133,
"text": "(Deb, 2001)",
"ref_id": "BIBREF3"
},
{
"start": 253,
"end": 271,
"text": "(Deb et al., 2002)",
"ref_id": "BIBREF2"
},
{
"start": 1420,
"end": 1438,
"text": "(Deb et al., 2002)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [
{
"start": 1073,
"end": 1081,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Nondominated Sorting Genetic Algorithm-II (NSGA-II)",
"sec_num": "3.3"
},
{
"text": "Below we discuss how single and multi-objective optimization techniques can be used feature selection in the anaphora resolution task.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Two Algorithms for Feature Selection in Anaphora Resolution",
"sec_num": "4"
},
{
"text": "If the total number of features is F , then the length of the chromosome is F . As an example, the encoding of a particular chromosome is represented in Figure 1 . Here F = 12 (i.e., total 12 different features are available). The chromosome represents the use of 7 features for constructing a classifier (first, third, fourth, seventh, tenth, eleventh and twelfth features). The entries of each chromosome are randomly initialized to either 0 or 1. Here, if the i th position of a chromosome is 0 then it represents that i th feature does not participate in constructing the classifier. Else if it is 1 then the i th feature participates in constructing the classifier.",
"cite_spans": [],
"ref_spans": [
{
"start": 153,
"end": 161,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Chromosome Representation and Population Initialization",
"sec_num": "4.1"
},
{
"text": "For fitness computation, the following procedure is executed:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Fitness Computation",
"sec_num": "4.2"
},
{
"text": "1. Suppose there are N number of features present in a particular chromosome (i.e., there are total N number of 1's in that chromosome).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Fitness Computation",
"sec_num": "4.2"
},
{
"text": "2. Construct the coreference resolution system (i.e., BART) with only these N features.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Fitness Computation",
"sec_num": "4.2"
},
{
"text": "3. This coreference system is evaluated on the development data. The recall, precision and F-measure values of three metrics are calculated.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Fitness Computation",
"sec_num": "4.2"
},
{
"text": "In case of single objective optimization (SOO), the objective function corresponding to a particular chromosome is the F-measure value of a single metric. This objective function is optimized using the search capability of GA. For MOO, the objective functions corresponding to a particular chromosome are F M U C (for the MUC metric), F \u03c6 3 (for CEAF using the \u03c6 3 entity alignment function (Luo, 2005) ) and F \u03c6 4 (for CEAF using the \u03c6 4 entity alignment function). These three objective functions are simultaneously optimized using the search capability of NSGA-II.",
"cite_spans": [
{
"start": 391,
"end": 402,
"text": "(Luo, 2005)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Fitness Computation",
"sec_num": "4.2"
},
{
"text": "In case of SOO, a single point crossover operation is used with a user defined crossover probability, \u00b5 c . A mutation operator is applied to each entry of the chromosome with a mutation probability, \u00b5 m , where the entry is randomly replaced by either 0 or 1. In this approach, the processes of fitness computation, selection, crossover, and mutation are executed for a maximum number of generations. The best string seen up to the last generation provides the solution to the above feature selection problem. Elitism has been implemented at each generation by preserving the best string seen upto that generation in a location outside the population. Thus on termination, this location contains the best feature combination.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Genetic Operators",
"sec_num": "4.3"
},
{
"text": "We use crowded binary tournament selection as in NSGA-II, followed by conventional crossover and mutation for the MOO based feature selection. The most characteristic part of NSGA-II is its elitism operation, where the non-dominated solutions (Deb, 2001 ) among the parent and child populations are propagated to the next generation. The near-Pareto-optimal strings of the last generation provide the different solutions to the feature selection problem.",
"cite_spans": [
{
"start": 243,
"end": 253,
"text": "(Deb, 2001",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Genetic Operators",
"sec_num": "4.3"
},
{
"text": "For our experiments, we use BART (Versley et al., 2008) , a modular toolkit for anaphora reso-lution that supports state-of-the-art statistical approaches to the task and enables efficient feature engineering. BART implements different models of anaphora resolution (mention-pair and entitymention; best-first vs. ranking), has interfaces to different machine learners (MaxEnt, SVM, decision trees) and provides a large set of linguistically motivated features, along with the possibility to design new ones. It is thus ideally suited for experimenting with optimization and feature selection.",
"cite_spans": [
{
"start": 33,
"end": 55,
"text": "(Versley et al., 2008)",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The BART System",
"sec_num": "5.1"
},
{
"text": "In this study, we specifically focus on feature selection. 2 The complete list of features currently implemented in BART is listed in Table 1 ; all were considered in the present experiments. We used a simple mention-pair model without ranking as in (Soon et al., 2001 ). In the mention-pair model, anaphora resolution is recast as a binary classification problem. Each classification instance consists of two mentions, i.e. an anaphor M j and its potential antecedent M i (i < j). Instances are modeled as feature vectors (cf. Table 1 ) and are handed over to a binary classifier that decides, whether the anaphor and its candidate antecedent are mentions of the same entity or not. All the feature values are computed automatically.",
"cite_spans": [
{
"start": 250,
"end": 268,
"text": "(Soon et al., 2001",
"ref_id": null
}
],
"ref_spans": [
{
"start": 134,
"end": 141,
"text": "Table 1",
"ref_id": "TABREF0"
},
{
"start": 528,
"end": 535,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "The BART System",
"sec_num": "5.1"
},
{
"text": "We train a maximum entropy classifier and follow the approach of (Soon et al., 2001) to partition mentions into coreference sets given the classifier's decisions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The BART System",
"sec_num": "5.1"
},
{
"text": "We evaluated our approach on the ACE-02 dataset, which is divided in three subsets: bnews, npaper, and nwire. We provide results for both gold (handannotated) versions of the datasets (gbnews, gnpaper, gnwire) and system mentions extracted with CARAFE 3 (cbnews, cnpaper, cnwire). Table 2 compares the performance level obtained using all the features in Table 1 with that of a loose re-implementation of the system proposed by Soon et al. (2001) , commonly used as baseline and relying only on very shallow information. Our reimplementation of the Soon et al. model uses only a subset of features: those marked with an asterisk in Table 1 . We also provide in Table 2 typical state-of-the-art figures on the ACE-02 dataset, as presented in an overview by Poon and Domin- Soon et al. (2001) 71.6 67.2 69.6 67.8 62.6 67.5 66.7 67.9 69.7 All features (Table 1) 75 Soon et al. (2001) 61.3 56.7 55.9 63.3 57.6 54.0 60.8 58.2 57.0 All features (Table 1) 62.3 57.9 57.5 65.5 55.9 52.7 60.6 56.8 55.6 (Uryupina, 2007; Bengtson and Roth, 2008) , the set of features already included in BART is sufficient to achieve results well above the state of the art on the dataset we used.",
"cite_spans": [
{
"start": 428,
"end": 446,
"text": "Soon et al. (2001)",
"ref_id": null
},
{
"start": 772,
"end": 790,
"text": "Soon et al. (2001)",
"ref_id": null
},
{
"start": 862,
"end": 880,
"text": "Soon et al. (2001)",
"ref_id": null
},
{
"start": 994,
"end": 1010,
"text": "(Uryupina, 2007;",
"ref_id": "BIBREF18"
},
{
"start": 1011,
"end": 1035,
"text": "Bengtson and Roth, 2008)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [
{
"start": 281,
"end": 288,
"text": "Table 2",
"ref_id": "TABREF1"
},
{
"start": 355,
"end": 362,
"text": "Table 1",
"ref_id": "TABREF0"
},
{
"start": 632,
"end": 639,
"text": "Table 1",
"ref_id": "TABREF0"
},
{
"start": 661,
"end": 668,
"text": "Table 2",
"ref_id": "TABREF1"
},
{
"start": 849,
"end": 858,
"text": "(Table 1)",
"ref_id": "TABREF0"
},
{
"start": 939,
"end": 948,
"text": "(Table 1)",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "The Data Sets",
"sec_num": "5.2"
},
{
"text": "F M U C F \u03c6 3 F \u03c6 4 F M U C F \u03c6 3 F \u03c6 4 F M U C F \u03c6 3 F \u03c6 4 following",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Data Sets",
"sec_num": "5.2"
},
{
"text": "F M U C F \u03c6 3 F \u03c6 4 F M U C F \u03c6 3 F \u03c6 4 F M U C F \u03c6 3 F \u03c6 4 following",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Data Sets",
"sec_num": "5.2"
},
{
"text": "The results in Table 2 also confirm the intuition that, contrary to what is suggested by some of the early papers (Soon et al., 2001; Ng and Cardie, 2002) working on smaller datasets, linguistic factors do play a crucial role in anaphora resolution and therefore rich feature sets may lead to performance improvements once larger datasets are considered (a similar result was also obtained by Bengtson and Roth (2008) ). Such improvements, however, come at high costs, as both using larger datasets and larger sets of features learning a model becomes slower and requires much more memory.",
"cite_spans": [
{
"start": 114,
"end": 133,
"text": "(Soon et al., 2001;",
"ref_id": null
},
{
"start": 134,
"end": 154,
"text": "Ng and Cardie, 2002)",
"ref_id": "BIBREF11"
},
{
"start": 393,
"end": 417,
"text": "Bengtson and Roth (2008)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [
{
"start": 15,
"end": 22,
"text": "Table 2",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "The Data Sets",
"sec_num": "5.2"
},
{
"text": "This suggests that automatic feature selection may be essential not just to improve performance but also to be able to train a model-i.e., that an efficient coreference resolution system should combine rich linguistic feature sets with automatic feature selection mechanisms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Data Sets",
"sec_num": "5.2"
},
{
"text": "We set the following parameter values for both single (i.e., GA) and MOO (i.e., NSGA-II): population size=20, number of generations=30, probability of mutation \u00b5 m = 0.2 and probability of crossover \u00b5 c = 0.9. Both approaches are executed on devel-opment data to determine the optimal feature vector(s). Final results are reported on the test data. It is to be noted that GA is a stochastic approach and outputs different results for trials with different seeds and initial populations. Initial seeds and population are chosen randomly. Thus for each data set we executed the proposed single and multi objective based approaches 3 times. Finally, we report the maximum values of these 3 runs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Genetic Algorithm Parameter Setting",
"sec_num": "5.3"
},
{
"text": "Single objective GA based feature selection was executed on the six data sets to determine the appropriate set of features. For each data set three sets of experiments were carried out by optimizing the F-measure values of the three different evaluation metrics. The binary-valued feature vectors identified by the single objective GA based feature selection technique for the six data sets and the corresponding F-measure values are shown in Table 3. The order of the features in the vector corresponds to their order in Table 1 ; the values of 0's and 1's represent the absence and presence of the corresponding features. Significant improvements over the classifier based on all the features are indicated with \u2020 (sign test, p < 0.05).",
"cite_spans": [],
"ref_spans": [
{
"start": 522,
"end": 529,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "Single Objective Optimization",
"sec_num": "6.1"
},
{
"text": "These results show that for all the datasets, the proposed single objective GA-based feature selection technique performs better than the baseline approach of using all features. Moreover, the results show that the technique based on SOO (i.e., conventional GA-based method) with different objective functions provides different evaluation figures. Thus, it is meaningful to optimize each objective function separately.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Single Objective Optimization",
"sec_num": "6.1"
},
{
"text": "It is also evident from Table 3 that the optimal feature set obtained by optimizing a single objective function may not be optimal with respect to another objective function. Thus, it is not possible to come up with common patterns in the set of optimal features. For example, in case of gbnews, the F-measure value of the first metric, i.e. of M U C corresponding to the optimal feature vector optimizing second metric, i.e. \u03c6 3 is 76.7. This is obviously less than the evaluation figure obtained by optimizing the first metric.",
"cite_spans": [],
"ref_spans": [
{
"start": 24,
"end": 31,
"text": "Table 3",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Single Objective Optimization",
"sec_num": "6.1"
},
{
"text": "Thereafter we apply our proposed MOO based feature selection technique on the six data sets. The MOO approach provides a set of non-dominated solutions on the final Pareto optimal front. All the solutions are equally important from the algorithmic point of view. In Table 4 , we show the final solutions obtained by the MOO based approach for all the data sets. Significant improvements over the classifier based on all the features are indicated with \u2020 (sign test, p < 0.05).",
"cite_spans": [],
"ref_spans": [
{
"start": 266,
"end": 273,
"text": "Table 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Multi-objective Optimization",
"sec_num": "6.2"
},
{
"text": "The results in Table 4 indicate that the MOO based technique achieves higher performance than the single objective GA based approach. For the gbnews data set, MOO achieves 0.6, 0.3 and 0.8 Fmeasure points increments for three metrics over the single objective GA based technique. For the gnpaper data set, there are increments of 2.5 F-measure points on second metric and 1.0 Fmeasure point on third metric over the corresponding single objective GA based technique. Similarly, for all other datasets the MOO based approach attains superior performance over the SOObased approach.",
"cite_spans": [],
"ref_spans": [
{
"start": 15,
"end": 22,
"text": "Table 4",
"ref_id": null
}
],
"eq_spans": [],
"section": "Multi-objective Optimization",
"sec_num": "6.2"
},
{
"text": "As discussed in Section 2 most work on optimization in anaphora resolution relies on manual optimization; the one significant exception is the work of Hoste (2005) .",
"cite_spans": [
{
"start": 151,
"end": 163,
"text": "Hoste (2005)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison with Related Work",
"sec_num": "7"
},
{
"text": "There are two major differences between the approach of Hoste (2005) and that followed in our study. First, the scope of (Hoste, 2005 ) is restricted to single-objective optimization. As we saw above, this might provide unstable solutions, that are too tailored to a particular scoring metric. Second, the feature set of Hoste (2005) is relatively small and therefore does not provide an efficient test-bed for a feature selection approach. Not surprising, parameter optimization shows a more consistent effect on the overall performance than feature selection in (Hoste, 2005) 's experiments.",
"cite_spans": [
{
"start": 56,
"end": 68,
"text": "Hoste (2005)",
"ref_id": "BIBREF6"
},
{
"start": 121,
"end": 133,
"text": "(Hoste, 2005",
"ref_id": "BIBREF6"
},
{
"start": 321,
"end": 333,
"text": "Hoste (2005)",
"ref_id": "BIBREF6"
},
{
"start": 564,
"end": 577,
"text": "(Hoste, 2005)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Comparison with Related Work",
"sec_num": "7"
},
{
"text": "In this paper we showed that it may not be necessary to choose one among the existing metrics for anaphora resolution-in fact, that developing systems attempting to optimize according to a combination of them may lead to better results.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion and Conclusions",
"sec_num": "8"
},
{
"text": "In subsequent work, we plan to expand the optimization technique to consider also learning parameters optimization, classifier selection, and learning model selection.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion and Conclusions",
"sec_num": "8"
},
{
"text": "We use the term 'anaphora resolution' to refer to the task perhaps most commonly referred to as 'coreference resolution,' which many including us find a misnomer. For the purposes of the present paper the two terms could be seen as interchangeable.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "The choice of the best model and the best machine learner, along with its parameters, is the main direction of our future work.3 http://sourceforge.net/projects/ carafe",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Algorithms for scoring coreference chains",
"authors": [
{
"first": "Amit",
"middle": [],
"last": "References",
"suffix": ""
},
{
"first": "Breck",
"middle": [],
"last": "Bagga",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Baldwin",
"suffix": ""
}
],
"year": 1998,
"venue": "Proc. of the LREC workshop on Linguistic Coreference",
"volume": "",
"issue": "",
"pages": "563--566",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "References Amit Bagga and Breck Baldwin. 1998. Algorithms for scoring coreference chains. In Proc. of the LREC workshop on Linguistic Coreference, pages 563- 566, Granada.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Understanding the value of features for coreference resolution",
"authors": [
{
"first": "Eric",
"middle": [],
"last": "Bengtson",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Roth",
"suffix": ""
}
],
"year": 2008,
"venue": "Proc. of EMNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Eric Bengtson and Dan Roth. 2008. Understanding the value of features for coreference resolution. In Proc. of EMNLP.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "A fast and elitist multiobjective genetic algorithm: NSGA-II",
"authors": [
{
"first": "Kalyanmoy",
"middle": [],
"last": "Deb",
"suffix": ""
},
{
"first": "Amrit",
"middle": [],
"last": "Pratap",
"suffix": ""
},
{
"first": "Sameer",
"middle": [],
"last": "Agarwal",
"suffix": ""
},
{
"first": "T",
"middle": [],
"last": "Meyarivan",
"suffix": ""
}
],
"year": 2002,
"venue": "IEEE Transactions on Evolutionary Computation",
"volume": "6",
"issue": "2",
"pages": "181--197",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kalyanmoy Deb, Amrit Pratap, Sameer Agarwal, and T. Meyarivan. 2002. A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE Transactions on Evolutionary Computation, 6(2):181-197.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Multi-objective Optimization Using Evolutionary Algorithms",
"authors": [
{
"first": "Kalyanmoy",
"middle": [],
"last": "Deb",
"suffix": ""
}
],
"year": 2001,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kalyanmoy Deb. 2001. Multi-objective Optimization Using Evolutionary Algorithms. John Wiley and Sons, Ltd, England.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "The automatic content extraction (ACE) program-tasks, data, and evaluation",
"authors": [
{
"first": "George",
"middle": [],
"last": "Doddington",
"suffix": ""
},
{
"first": "Alexis",
"middle": [],
"last": "Mitchell",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Przybocki",
"suffix": ""
},
{
"first": "Lance",
"middle": [],
"last": "Ramshaw",
"suffix": ""
},
{
"first": "Stephanie",
"middle": [],
"last": "Strassell",
"suffix": ""
},
{
"first": "Ralph",
"middle": [],
"last": "Weischedel",
"suffix": ""
}
],
"year": 2004,
"venue": "Proc. of LREC",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "George Doddington, Alexis Mitchell, Mark Przybocki, Lance Ramshaw, Stephanie Strassell, and Ralph Weischedel. 2004. The automatic content extrac- tion (ACE) program-tasks, data, and evaluation. In Proc. of LREC.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Genetic Algorithms in Search, Optimization and Machine Learning",
"authors": [
{
"first": "David",
"middle": [
"E"
],
"last": "Goldberg",
"suffix": ""
}
],
"year": 1989,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "David E. Goldberg. 1989. Genetic Algorithms in Search, Optimization and Machine Learning. Addison-Wesley, New York.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Optimization Issues in Machine Learning of Coreference Resolution",
"authors": [
{
"first": "Veronique",
"middle": [],
"last": "Hoste",
"suffix": ""
}
],
"year": 2005,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Veronique Hoste. 2005. Optimization Issues in Ma- chine Learning of Coreference Resolution. Ph.D. thesis, Antwerp University.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Incorporating contextual cues in trainable models for coreference resolution",
"authors": [
{
"first": "Ryu",
"middle": [],
"last": "Iida",
"suffix": ""
},
{
"first": "Kentaro",
"middle": [],
"last": "Inui",
"suffix": ""
},
{
"first": "Hiroya",
"middle": [],
"last": "Takamura",
"suffix": ""
},
{
"first": "Yuji",
"middle": [],
"last": "Matsumoto",
"suffix": ""
}
],
"year": 2003,
"venue": "Proc. EACL Workshop on the Computational Treatment of Anaphora",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ryu Iida, Kentaro Inui, Hiroya Takamura, and Yuji Matsumoto. 2003. Incorporating contextual cues in trainable models for coreference resolution. In Proc. EACL Workshop on the Computational Treatment of Anaphora.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "A mentionsynchronous coreference resolution algorithm based on the Bell Tree",
"authors": [
{
"first": "Xiaoqiang",
"middle": [],
"last": "Luo",
"suffix": ""
},
{
"first": "Abe",
"middle": [],
"last": "Ittycheriah",
"suffix": ""
},
{
"first": "Hongyan",
"middle": [],
"last": "Jing",
"suffix": ""
},
{
"first": "Nanda",
"middle": [],
"last": "Kambhatla",
"suffix": ""
},
{
"first": "Salim",
"middle": [],
"last": "Roukos",
"suffix": ""
}
],
"year": 2004,
"venue": "Proc. of ACL",
"volume": "",
"issue": "",
"pages": "136--143",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xiaoqiang Luo, Abe Ittycheriah, Hongyan Jing, Nanda Kambhatla, and Salim Roukos. 2004. A mention- synchronous coreference resolution algorithm based on the Bell Tree. In Proc. of ACL 2004, pages 136- 143.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "On coreference resolution performance metrics",
"authors": [
{
"first": "Xiaoqiang",
"middle": [],
"last": "Luo",
"suffix": ""
}
],
"year": 2005,
"venue": "Proc. NAACL / EMNLP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xiaoqiang Luo. 2005. On coreference resolution per- formance metrics. In Proc. NAACL / EMNLP, Van- couver.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Optimizing to arbitrary NLP metrics using ensemble selection",
"authors": [
{
"first": "Art",
"middle": [],
"last": "Munson",
"suffix": ""
},
{
"first": "Claire",
"middle": [],
"last": "Cardie",
"suffix": ""
},
{
"first": "Rich",
"middle": [],
"last": "Caruana",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of HLT/EMNLP",
"volume": "",
"issue": "",
"pages": "539--546",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Art Munson, Claire Cardie, and Rich Caruana. 2005. Optimizing to arbitrary NLP metrics using ensem- ble selection. In Proceedings of HLT/EMNLP, pages 539-546.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Improving machine learning approaches to coreference resolution",
"authors": [
{
"first": "Vincent",
"middle": [],
"last": "Ng",
"suffix": ""
},
{
"first": "Claire",
"middle": [],
"last": "Cardie",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the 40th Annual Meeting on Association for C omputational Linguistics",
"volume": "",
"issue": "",
"pages": "104--111",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Vincent Ng and Claire Cardie. 2002. Improving ma- chine learning approaches to coreference resolution. In Proceedings of the 40th Annual Meeting on Asso- ciation for C omputational Linguistics, pages 104- 111.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Joint unsupervised coreference resolution with Markov logic",
"authors": [
{
"first": "Hoifung",
"middle": [],
"last": "Poon",
"suffix": ""
},
{
"first": "Pedro",
"middle": [],
"last": "Domingos",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hoifung Poon and Pedro Domingos. 2008. Joint unsupervised coreference resolution with Markov logic. In Proceedings of the Conference on Empiri- cal Methods in Natural Language Processing.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "A deeper look into features for coreference resolution",
"authors": [
{
"first": "Marta",
"middle": [],
"last": "Recasens",
"suffix": ""
},
{
"first": "Eduard",
"middle": [],
"last": "Hovy",
"suffix": ""
}
],
"year": 2009,
"venue": "Anaphora Processing and Applications (DAARC 2009, number 5847 in LNAI",
"volume": "",
"issue": "",
"pages": "29--42",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marta Recasens and Eduard Hovy. 2009. A deeper look into features for coreference resolution. In S. Lalitha Devi, A. Branco, and R. Mitkov, editors, Anaphora Processing and Applications (DAARC 2009, number 5847 in LNAI, pages 29-42, Berlin / Heidelberg. Springer-Verlag.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Blanc: Implementing the rand index for coreference evaluation",
"authors": [
{
"first": "Marta",
"middle": [],
"last": "Recasens",
"suffix": ""
},
{
"first": "Eduard",
"middle": [],
"last": "Hovy",
"suffix": ""
}
],
"year": 2011,
"venue": "Natural Language Engineering",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marta Recasens and Eduard Hovy. 2011. Blanc: Im- plementing the rand index for coreference evalua- tion. Natural Language Engineering.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Semeval-2010 task 1: Coreference resolution in multiple languages",
"authors": [],
"year": null,
"venue": "Proc. SEMEVAL 2010",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Semeval-2010 task 1: Coreference resolution in multiple languages. In Proc. SEMEVAL 2010, Upp- sala.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "A machine learning approach to coreference resolution of noun phrases",
"authors": [],
"year": 2001,
"venue": "Computational Linguistic",
"volume": "27",
"issue": "4",
"pages": "521--544",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Wee Meng Soon, Hwee Tou Ng, and Daniel Chung Yong Lim. 2001. A machine learning ap- proach to coreference resolution of noun phrases. Computational Linguistic, 27(4):521-544.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Knowledge Acquisition for Coreference Resolution",
"authors": [
{
"first": "Olga",
"middle": [],
"last": "Uryupina",
"suffix": ""
}
],
"year": 2007,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Olga Uryupina. 2007. Knowledge Acquisition for Coreference Resolution. Ph.D. thesis, University of the Saarland.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Corry: a system for coreference resolution",
"authors": [
{
"first": "Olga",
"middle": [],
"last": "Uryupina",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the 5th International Workshop on Semantic Eval uation (SemEval'10)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Olga Uryupina. 2010. Corry: a system for coreference resolution. In Proceedings of the 5th International Workshop on Semantic Eval uation (SemEval'10).",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "BART: a modular toolkit for coreference resolution",
"authors": [
{
"first": "Yannick",
"middle": [],
"last": "Versley",
"suffix": ""
},
{
"first": "Simone",
"middle": [
"Paolo"
],
"last": "Ponzetto",
"suffix": ""
},
{
"first": "Massimo",
"middle": [],
"last": "Poesio",
"suffix": ""
},
{
"first": "Vladimir",
"middle": [],
"last": "Eidelman",
"suffix": ""
},
{
"first": "Alan",
"middle": [],
"last": "Jern",
"suffix": ""
},
{
"first": "Jason",
"middle": [],
"last": "Smith",
"suffix": ""
},
{
"first": "Xiaofeng",
"middle": [],
"last": "Yang",
"suffix": ""
},
{
"first": "Alessandro",
"middle": [],
"last": "Moschitti",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the 46th Annual Meeting of the Association for Computational Linguistics on Human Language Technologies",
"volume": "",
"issue": "",
"pages": "9--12",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yannick Versley, Simone Paolo Ponzetto, Massimo Poesio, Vladimir Eidelman, Alan Jern, Jason Smith, Xiaofeng Yang, and Alessandro Moschitti. 2008. BART: a modular toolkit for coreference resolution. In Proceedings of the 46th Annual Meeting of the As- sociation for Computational Linguistics on Human Language Technologies, pages 9-12.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "A modeltheoretic coreference scoring scheme",
"authors": [
{
"first": "Marc",
"middle": [],
"last": "Vilain",
"suffix": ""
},
{
"first": "John",
"middle": [],
"last": "Burger",
"suffix": ""
},
{
"first": "John",
"middle": [],
"last": "Aberdeen",
"suffix": ""
}
],
"year": 1995,
"venue": "Proc. of the Sixth Message Understanding Conference",
"volume": "",
"issue": "",
"pages": "45--52",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marc Vilain, John Burger, John Aberdeen, Dennis Con- nolly, and Lynette Hirschman. 1995. A model- theoretic coreference scoring scheme. In Proc. of the Sixth Message Understanding Conference, pages 45-52.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "A twin-candidate model of coreference resolution with non-anaphor identification capability",
"authors": [
{
"first": "Xiaofeng",
"middle": [],
"last": "Yang",
"suffix": ""
},
{
"first": "Jian",
"middle": [],
"last": "Su",
"suffix": ""
},
{
"first": "Chew Lim",
"middle": [],
"last": "Tan",
"suffix": ""
}
],
"year": 2005,
"venue": "Proc. of IJCNLP",
"volume": "",
"issue": "",
"pages": "719--730",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Xiaofeng Yang, Jian Su, and Chew Lim Tan. 2005. A twin-candidate model of coreference resolution with non-anaphor identification capability. In Proc. of IJCNLP, pages 719-730.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "Maximum metric score training for coreference resolution",
"authors": [
{
"first": "Shanheng",
"middle": [],
"last": "Zhao",
"suffix": ""
},
{
"first": "Hwee Tou",
"middle": [],
"last": "Ng",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceeding of COLING-2010",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shanheng Zhao and Hwee Tou Ng. 2010. Maximum metric score training for coreference resolution. In Proceeding of COLING-2010.",
"links": null
}
},
"ref_entries": {
"TABREF0": {
"type_str": "table",
"text": "Features used by BART: each feature describes a pair of mentions {M i , M j }, i < j, where M i is a candidate antecedent and M j is a candidate anaphor Mi and Mj , as identified in Soon et al. Mi and Mj are in an apposition Appositive2 Mi and Mj are adjacent Coordination Mi is a coordination; same for Mj HeadPartOfSpeech POS of Mi's head; same for Mj and the pair SynPos depth of Mi's node in the parse tree Attributes Mi and Mj have incompatible premodifiers Relations Mi and Mj have incompatible postmodifiers Matching StringMatch* Mi and Mj have the same surface form after stripping off the determiners NonPro StringMatch both Mi and Mj are non-pronominal and Stringmatch(Mi, Mj ) == 1 Pro StringMatch both Mi and Mj are pronominal and Stringmatch(Mi, Mj ) == 1 NE StringMatch both Mi and Mj are NE and Stringmatch(Mi, Mj ) == 1 HeadMatch Mi and Mj have the same head MinSame Mi and Mj have the same minimal span LeftRightMatch Mi, number of preceding mentions with the same head lemma Web Wiki Mi and Mj have the same wikipedia entry Yago Mi and Mj are linked in Yago via means or typeof relation WebPatterns specific contexts for co-reference extracted from the web Proximity DistanceMarkable distance in mentions between Mi and Mj DistanceSentenceInt* distance in sentences between Mi and Mj DistanceSentence log-distance in sentences between Mi and Mj DistanceSentence2 log-distance in sentences between Mi and Mj , different formula DistDiscrete distance in sentences between Mi and Mj discretized into {0,1,>=2} Miscellaneous SpeechMi is in quoted speech; same for Mj and the pair",
"content": "<table><tr><td/><td>Mention types and subtypes</td></tr><tr><td colspan=\"2\">MentionType* relevant types of MentionType Ante Salient</td><td>Mi is demonstrative; Mi is an NE</td></tr><tr><td>MentionType Ante Extra</td><td/><td>Mi is a pronoun</td></tr><tr><td>MentionType Ana</td><td colspan=\"2\">Mj is a definite, demonstrative or indefinite NP, or pronoun of a specific type</td></tr><tr><td>MentionType2</td><td colspan=\"2\">relevant types of Mi and Mj , as identified in Soon et al.</td></tr><tr><td>MentionType Salience</td><td colspan=\"2\">combination of M entionT ype and M entionT ype Ana</td></tr><tr><td>FirstSecondPerson</td><td colspan=\"2\">Mi is a pronoun of the 1st/second person, same for Mj</td></tr><tr><td>PronounLeftRight PronounWordForm</td><td colspan=\"2\">4 possible values for < Mi is a pronoun> * < Mj is a pronoun > lemma for Mi if it's a pronoun; same for Mj</td></tr><tr><td>SemClassValue</td><td colspan=\"2\">semantic class of Mi, and Mj and the pair</td></tr><tr><td>BothLocation</td><td colspan=\"2\">both Mi and Mj are locations or geo-political</td></tr><tr><td/><td>Agreement</td></tr><tr><td>GenderAgree*</td><td/><td>Mi and Mj agree in gender</td></tr><tr><td>NumberAgree*</td><td/><td>Mi and Mj agree in number</td></tr><tr><td>AnimacyAgree*</td><td/><td>Mi and Mj agree in animacy</td></tr><tr><td/><td>Aliasing</td></tr><tr><td>Alias*</td><td/><td>heuristical NE-matching</td></tr><tr><td>BetterNames</td><td colspan=\"2\">heuristical matching for personal names</td></tr><tr><td/><td>Syntax</td></tr><tr><td>Appositive*</td><td/></tr><tr><td/><td colspan=\"2\">Mj is a prefix or suffix substring of Mi or vice versa</td></tr><tr><td>StringMatchExtra</td><td/><td>extra string-macthing for bare plurals</td></tr><tr><td>StringKernel</td><td/><td>approximate matching</td></tr><tr><td/><td>Salience</td></tr><tr><td>First Mention</td><td/><td>Mi is the first mention in its sentence</td></tr><tr><td>CorefChain</td><td colspan=\"2\">Size of the coreference chain suggested for Mi so far (with a threshold)</td></tr><tr><td>NonProSalience</td><td>for non-pronominal</td></tr></table>",
"num": null,
"html": null
},
"TABREF1": {
"type_str": "table",
"text": "Baseline performance on the ACE-02 dataset",
"content": "<table><tr><td/><td>gold mentions</td><td/></tr><tr><td>gbnews</td><td>gnpaper</td><td>gnwire</td></tr></table>",
"num": null,
"html": null
},
"TABREF3": {
"type_str": "table",
"text": "Feature vectors identified via single-objective optimization.",
"content": "<table><tr><td colspan=\"2\">DataSet Metric</td><td>Features Selected</td><td colspan=\"2\">F M U C F \u03c6 3</td><td>F \u03c6 4</td></tr><tr><td/><td>opt.</td><td/><td/></tr><tr><td colspan=\"2\">gbnews MUC</td><td colspan=\"2\">00100110110111100111111000111001001111111010 76.8</td><td>71.5 74.5</td></tr><tr><td/><td>\u03c6 3 ,\u03c6 4</td><td colspan=\"2\">10011000111010110000101101010011011011000001 76.7</td><td>71.8 \u2020 74.9 \u2020</td></tr><tr><td colspan=\"2\">gnpaper MUC</td><td colspan=\"2\">10000001001111110101011101110000101010100111 74.6</td><td>67.1 70.1 \u2020</td></tr><tr><td/><td>\u03c6 3</td><td colspan=\"2\">10101001100100110100100000010100010001101100 72.2</td><td>67.6 69.1</td></tr><tr><td/><td>\u03c6 4</td><td colspan=\"2\">11101001100100110100111000100101110010001100 71.4</td><td>65.2 70.3</td></tr><tr><td>gnwire</td><td>MUC</td><td colspan=\"2\">10111011011111110010101010011010011011001011 74.0 \u2020</td><td>70.3 \u2020 73.1 \u2020</td></tr><tr><td/><td>\u03c6 3</td><td colspan=\"2\">11011011100001000011110110101111011110001101 71.4</td><td>72.3 73.6</td></tr><tr><td/><td>\u03c6 4</td><td colspan=\"2\">11101001100100110100111000100101110010001100 71.7</td><td>72.1 74.4</td></tr><tr><td>cbnews</td><td colspan=\"3\">MUC,\u03c6 3 11111001100101000011011100101101101111001100 64.6</td><td>59.7 58.4</td></tr><tr><td/><td>\u03c6 4</td><td colspan=\"2\">11111001100001000011110100101111101110001101 63.6</td><td>59.6 58.8</td></tr><tr><td colspan=\"4\">cnpaper MUC,\u03c6 3 01000100100101011001000010111100101100001000 66.5 \u03c6 4 10100101101011100011111110010100100010010011 66.2</td><td>59.7 \u2020 54.7 \u2020 59.1 55.6 \u2020</td></tr><tr><td>cnwire</td><td>MUC</td><td colspan=\"2\">00101111101110101001100000010101001011001001 63.8</td><td>60.0 58.1</td></tr><tr><td/><td>\u03c6 3 ,\u03c6 4</td><td colspan=\"2\">00011000101110100010000010011000100110000100 63.4</td><td>61.2 58.4</td></tr><tr><td colspan=\"3\">gos (2008). The results clearly show that although</td><td/></tr><tr><td colspan=\"3\">even larger sets of features have been proposed</td><td/></tr></table>",
"num": null,
"html": null
}
}
}
} |