File size: 90,863 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 | {
"paper_id": "D09-1029",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T16:40:06.613101Z"
},
"title": "Wikipedia as Frame Information Repository",
"authors": [
{
"first": "Sara",
"middle": [],
"last": "Tonelli",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "FBK-irst I-38100",
"location": {
"settlement": "Trento",
"country": "Italy"
}
},
"email": "satonelli@fbk.eu"
},
{
"first": "Claudio",
"middle": [],
"last": "Giuliano",
"suffix": "",
"affiliation": {},
"email": "giuliano@fbk.eu"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "In this paper, we address the issue of automatic extending lexical resources by exploiting existing knowledge repositories. In particular, we deal with the new task of linking FrameNet and Wikipedia using a word sense disambiguation system that, for a given pair frame-lexical unit (F, l), finds the Wikipage that best expresses the the meaning of l. The mapping can be exploited to straightforwardly acquire new example sentences and new lexical units, both for English and for all languages available in Wikipedia. In this way, it is possible to easily acquire good-quality data as a starting point for the creation of FrameNet in new languages. The evaluation reported both for the monolingual and the multilingual expansion of FrameNet shows that the approach is promising.",
"pdf_parse": {
"paper_id": "D09-1029",
"_pdf_hash": "",
"abstract": [
{
"text": "In this paper, we address the issue of automatic extending lexical resources by exploiting existing knowledge repositories. In particular, we deal with the new task of linking FrameNet and Wikipedia using a word sense disambiguation system that, for a given pair frame-lexical unit (F, l), finds the Wikipage that best expresses the the meaning of l. The mapping can be exploited to straightforwardly acquire new example sentences and new lexical units, both for English and for all languages available in Wikipedia. In this way, it is possible to easily acquire good-quality data as a starting point for the creation of FrameNet in new languages. The evaluation reported both for the monolingual and the multilingual expansion of FrameNet shows that the approach is promising.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Many applications in the context of natural language processing or information retrieval have proved to convey significant improvement by exploiting lexical databases with high-quality annotation such as FrameNet (Fillmore et al., 2003) and WordNet (Fellbaum, 1998) . Nevertheless, the practical use of similar resources is often biased by their limited coverage because manual annotation is time-consuming and requires a relevant financial effort. For this reason, some research activities have focused on the automatic enrichment of such resources with annotated information in (near) manual quality. The main strategy proposed was the mapping between resources in order to reciprocally enrich different lexical databases by linking their information layers. This has proved to be useful in several tasks, from verb classification (Chow and Webster, 2007) to semantic role labeling (Giuglea and Moschitti, 2006) , open text semantic parsing (Shi and Mihalcea, 2004) and textual entailment (Burchardt and Frank, 2006) .",
"cite_spans": [
{
"start": 213,
"end": 236,
"text": "(Fillmore et al., 2003)",
"ref_id": "BIBREF11"
},
{
"start": 241,
"end": 265,
"text": "WordNet (Fellbaum, 1998)",
"ref_id": null
},
{
"start": 833,
"end": 857,
"text": "(Chow and Webster, 2007)",
"ref_id": "BIBREF5"
},
{
"start": 884,
"end": 913,
"text": "(Giuglea and Moschitti, 2006)",
"ref_id": "BIBREF12"
},
{
"start": 943,
"end": 967,
"text": "(Shi and Mihalcea, 2004)",
"ref_id": "BIBREF20"
},
{
"start": 991,
"end": 1018,
"text": "(Burchardt and Frank, 2006)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this work, we focus on the automatic enrichment of the FrameNet database for English and we propose a new framework to extend this procedure to new languages. While similar works in the past have mainly proposed to automatically extend the FrameNet database by mapping frames and Word-Net synsets (Shi and Mihalcea (2005) , Johansson and Nugues (2007) , and Tonelli and Pighin (2009) ), we present an explorative approach that for the first time exploits Wikipedia to this purpose. In particular, given a lexical unit l belonging to a frame F , we devise a strategy to link l to the Wikipedia article that best captures the sense of l in F . This is basically a word disambiguation (WSD) problem (Erk, 2004) and to this purpose we employ a state-of-the-art WSD system (Gliozzo et al., 2005) . The mapping between (F, l) pairs and Wikipedia pages could then be exploited for three further subtasks: (a) automatically extract from Wikipedia all sentences pointing to the Wikipage mapped with (F, l) and assign them to F ; (b) automatically expand the lexical units sets in the English FrameNet by exploiting the redirecting and linking strategy of Wikipedia; and (c) since Wikipedia is available in 260 languages, use the English Wikipedia article linked to (F, l) as a bridge to carry out sentence and lexical unit retrieval in other languages. The set of automatically collected data would represent the starting point for the creation of FrameNet in new languages. In fact, having a repository of sentences extracted from Wikipedia which have already been divided by sense would significantly speed up the annotation process. In this way, the annotators would not need to extract all sentences in a corpus containing l and classify them by sense. Instead, they should simply validate the given sentences and assign the correct frame elements.",
"cite_spans": [
{
"start": 300,
"end": 324,
"text": "(Shi and Mihalcea (2005)",
"ref_id": "BIBREF21"
},
{
"start": 327,
"end": 354,
"text": "Johansson and Nugues (2007)",
"ref_id": "BIBREF14"
},
{
"start": 361,
"end": 386,
"text": "Tonelli and Pighin (2009)",
"ref_id": "BIBREF22"
},
{
"start": 699,
"end": 710,
"text": "(Erk, 2004)",
"ref_id": "BIBREF9"
},
{
"start": 771,
"end": 793,
"text": "(Gliozzo et al., 2005)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In the following, we start by providing a brief overview of FrameNet and Wikipedia and we present their structure and organization. Next, we describe the algorithm for mapping lexical units and Wikipages and the word sense disambiguation algorithm employed by the system. In Section 5 we describe the dataset used in the first experiment and report evaluation results of the mapping between (F, l) pairs and Wikipedia senses. In Section 6 we describe an application of the mapping, i.e. the automatic enrichment of English FrameNet. We describe the data extraction process and evaluate the quality of the data. In Section 7 we describe and evaluate another application of the mapping, i.e. the acquisition of data for the automatic creation of Italian FrameNet using the Italian Wikipedia. Finally, we draw conclusions and present future research directions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "FrameNet (Fillmore et al., 2003) is a lexical resource for English based on corpus evidence, whose conceptual model comprises a set of prototypical situations called frames, the frame-evoking words or expressions called lexical units (LUs) and the roles or participants involved in these situations, called frame elements. All lexical units belonging to the same frame have similar semantics but, differently from WordNet synsets, they can belong to different categories and present different parts of speech. For example, the KILLING frame is described in the FrameNet database 1 as \"A Killer or Cause causes the death of the Victim\". The elements in capitals are the semantic roles (frame elements) typically involved in the KILLING situation. The frame definition comes also with the list of frame-evoking lexical units, namely annihilate.v, annihilation.n, butchery.n, carnage.n, crucify.v, deadly.a, etc. Since FrameNet is a corpus-based resource, every lexical unit should be instantiated by a set of example sentences, where the frame elements are annotated as well. Instead, FrameNet is still an ongoing project and in the latest release (v. 1.3) there are about 3,380 lexical units out of 10,195 that come with no example sentences. In this work we focus on these lexical units and propose how to automatically collect the missing sentences. Anyhow, the algorithm we propose is suitable also for expanding sentence sets already present in FrameNet.",
"cite_spans": [
{
"start": 9,
"end": 32,
"text": "(Fillmore et al., 2003)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "FrameNet and Wikipedia",
"sec_num": "2"
},
{
"text": "1 http://framenet.icsi.berkeley.edu Wikipedia 2 is one of the largest online repositories of encyclopedic knowledge, with millions of articles available for a large number of languages (>2,800,000 for English). The article (or page) is the basic entry in Wikipedia. Every article has an unique reference, i.e., one or more words that identify the page and are present in its URL. For example, Ball (dance) identifies the page that describes several types of ball intended as formal dance, while Dance (musical form) describes the dance as musical genre. Every Wikipedia article is linked to others, and in the body of every page there are plenty of links to connect the most relevant terms to other pages. Another important attribute is the presence of about 3,000,000 redirection pages, that given an identifier that is not present in Wikipedia, automatically display the page with the most semantically similar identifier (for example Killing is redirected to the Murder page). Wikipedia contains also more than 100,000 disambiguation pages listing all senses (pages) for an ambiguous entity. For example, Book has 9 senses, which correspond to 9 different articles. Wikipedia structure and quality make this resource particularly suitable for information extraction and word sense disambiguation tasks (Csomai and Mihalcea (2008) and Milne and Witten (2008) ). In fact, page references can be seen as senses and Wikipedia as a large sense inventory. From this point of view, also linking a lexical unit to the correct Wikipedia page is a word sense disambiguation issue because it implies recognizing what meaning the lexical unit has in the given frame. For example, dance.n in the SOCIAL EVENT frame should be linked to Ball (dance) and not to Dance (musical form).",
"cite_spans": [
{
"start": 1305,
"end": 1332,
"text": "(Csomai and Mihalcea (2008)",
"ref_id": "BIBREF7"
},
{
"start": 1337,
"end": 1360,
"text": "Milne and Witten (2008)",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "FrameNet and Wikipedia",
"sec_num": "2"
},
{
"text": "In this section, we describe how to map a frame -lexical unit pair (F, l) into the Wikipedia article that best captures the sense of l as defined in F . The mapping problem is casted as a supervised WSD problem, in which l must be disambiguated using F to provide the context and Wikipedia to provide the sense inventory and the training data. Even if the idea of using Wikipedia links for disambiguation is not novel (Cucerzan, 2007) , it is applied for the first time to FrameNet lexical units, considering a frame as a sense definition. The proposed algorithm is summarized as follows:",
"cite_spans": [
{
"start": 418,
"end": 434,
"text": "(Cucerzan, 2007)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The Mapping Algorithm",
"sec_num": "3"
},
{
"text": "Step 1 For each lexical unit l, we collect from the English Wikipedia dump 3 all contexts 4 where l is the anchor of an internal link (wiki link). The set of targets represents the senses of l in Wikipedia and the contexts are used as labelled training examples. For example, the lexical unit building.n in the frame Buildings is an anchor in 708 different contexts that point to 42 different Wikipedia pages (senses).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Mapping Algorithm",
"sec_num": "3"
},
{
"text": "Step 2 The set of contexts with their corresponding senses is then used to train the WSD system described in Section 4. For example, the context \"The building, which date from the mid-to-late 19th century, were built in a variety of High Victorian architectural styles.\" is a training example for the sense defined by the Wikipedia page Building.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Mapping Algorithm",
"sec_num": "3"
},
{
"text": "Step 3 Finally, the disambiguation model learned in the previous step is used to map a pair (F, l) to a Wikipedia article. (F, l) is represented as a fictitious-context derived by aggregating the frame definition and all lexical units associated to F . We used the term \"fictitious-context\" to remark the slight difference in structure compared with the training contexts (i.e., the Wikipedia paragraphs). For example, \". . . structures forming an enclosure and providing protection from the elements . . . acropolis arena auditorium bar building . . . \" is the fictitious-context built for the pair (Buildings, building.n). The sense, i.e., the Wikipedia article, assigned to the fictitiouscontext by the disambiguation algorithm uniquely defines the mapping. The previous example is assigned to the Wikipedia page Building.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Mapping Algorithm",
"sec_num": "3"
},
{
"text": "4 The WSD Algorithm Gliozzo et al. (2005) proposed an elegant approach to WSD based on kernel methods. The algorithm proved effective at Senseval-3 (Mihalcea and Edmonds, 2004) and, nowadays, it still represents the state-of-the-art in WSD (Pradhan et al., 2007) . Specifically, they addressed these issues: (i) independently modeling domain and syntagmatic aspects of sense distinction to improve feature representativeness; and (ii) exploiting external knowledge acquired from unlabeled data, with the purpose of drastically reducing the amount of labeled training data. The first direction is based on the linguistic assumption that syntagmatic and domain (associative) relations are crucial for representing sense distictions, but they are originated by different phenomena. Regarding the second direction, it is possible to obtain a more accurate prediction by taking into account unlabeled data relevant for the learning problem (Chapelle et al., 2006) .",
"cite_spans": [
{
"start": 20,
"end": 41,
"text": "Gliozzo et al. (2005)",
"ref_id": "BIBREF13"
},
{
"start": 148,
"end": 176,
"text": "(Mihalcea and Edmonds, 2004)",
"ref_id": "BIBREF16"
},
{
"start": 240,
"end": 262,
"text": "(Pradhan et al., 2007)",
"ref_id": "BIBREF18"
},
{
"start": 935,
"end": 958,
"text": "(Chapelle et al., 2006)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "The Mapping Algorithm",
"sec_num": "3"
},
{
"text": "On the other hand, kernel methods are theoretically well founded in statistical learning theory and shown good empirical results in many applications (Shawe-Taylor and Cristianini, 2004). The strategy adopted by kernel methods consists of splitting the learning problem into two parts. They first embed the input data in a suitable feature space, and then use a linear algorithm (e.g., support vector machines) to discover nonlinear patterns in the input space. The kernel function is the only task-specific component of the learning algorithm. Thus, to develop a WSD system, one only needs to define appropriate kernel functions to represent the domain and syntagmatic aspects of sense distinction and to exploit the properties of kernel functions in order to define a composite kernel that combines and extends individual kernels.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Mapping Algorithm",
"sec_num": "3"
},
{
"text": "The WSD system described in the following consists of a composite kernel (Section 4.3) that combines the domain and syntagmatic kernels. The former (Section 4.1) models the domain aspects of sense distinction, the latter (Section 4.2) represents the syntagmatic aspects of sense distinction.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "The Mapping Algorithm",
"sec_num": "3"
},
{
"text": "It is been shown that domain information is fundamental for WSD (Magnini et al., 2002) . For instance, the (domain) polysemy between the computer science and the medicine senses of the word \"virus\" can be solved by considering the domain of the context in which it appears.",
"cite_spans": [
{
"start": 64,
"end": 86,
"text": "(Magnini et al., 2002)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "In the context of kernel methods, domain information can be exploited by defining a kernel function that estimates the domain similarity between the contexts of the word to be disambiguated. The simplest method to estimate the domain similarity between two texts is to compute the cosine similarity of their vector representations in the vector space model (VSM). The VSM is a k-dimensional space R k , in which the text t j is represented by a vector t j , where the i th component is the term frequency of the term w i in t j . However, such an approach does not deal well with lexical variability and ambiguity. For instance, despite the fact that the sentences \"he is affected by AIDS\" and \"HIV is a virus\" express concepts closely related, their similarity is zero in the VSM because they have no words in common (they are represented by orthogonal vectors). On the other hand, due to the ambiguity of the word \"virus\" , the similarity between the sentences \"the laptop has been infected by a virus\" and \"HIV is a virus\" is greater than zero, even though they convey very different messages.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "To overcome this problem, Gliozzo et al. (2005) introduced the domain model (DM) and show how to define a domain VSM in which texts and terms are represented in a uniform way. A DM is composed of soft clusters of terms. Each cluster represents a semantic domain, that is, a set of terms that often co-occur in texts having similar topics. A DM is represented by a k \u00d7 k rectangular matrix D, containing the degree of association among terms and domains.",
"cite_spans": [
{
"start": 26,
"end": 47,
"text": "Gliozzo et al. (2005)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "The matrix D is used to define a function D : R k \u2192 R k , that maps the vector t j represented in the standard VSM, into the vector t j in the domain VSM. D is defined by",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "D( tj) = tj(I IDF D) = t j ,",
"eq_num": "(1)"
}
],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "where t j is represented as a row vector, I IDF is a k\u00d7k diagonal matrix such that i IDF i,i = IDF (w i ), and IDF (w i ) is the inverse document frequency of w i .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "In the domain space, the similarity is estimated by taking into account second order relations among terms. For example, the similarity of the two sentences \"He is affected by AIDS\" and \"HIV is a virus\" is very high, because the terms AIDS, HIV and virus are strongly associated with the domain medicine.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "Singular valued decomposition (SVD) is used to acquire in a unsupervised way the DM from a corpus represented by its term-by-document matrix T. SVD decomposes the term-by-document matrix T into three matrixes T V\u03a3 k U T , where V and U are orthogonal matrices (i.e., V T V = I and U T U = I) whose columns are the eigenvectors of TT T and T T T respectively, and \u03a3 k is the diagonal k \u00d7 k matrix containing the highest k k eigenvalues of T, and all the remaining elements set to 0. The parameter k is the dimensionality of the domain VSM and can be fixed in advance. Under this setting, the domain matrix D is defined by",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "D = I N V p \u03a3 k (2)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "where I N is a diagonal matrix such that",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "i N i,i = 1 q w i , w i , w i is the i th row of the matrix V \u221a \u03a3 k .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "The domain kernel is explicitly defined by",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "KD(ti, tj) = D(ti), D(tj) ,",
"eq_num": "(3)"
}
],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "where D is the domain mapping defined in Equation 1. Finally, the domain kernel is further extended to include the standard bag-of-word kernel.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Domain Kernel",
"sec_num": "4.1"
},
{
"text": "Kernel functions are not restricted to operate on vectorial objects x \u2208 R k . In principle, kernels can be defined for any kind of object representation, such as strings and trees. As syntagmatic relations hold among words collocated in a particular temporal order, they can be modeled by analyzing sequences of words. Therefore, the string kernel (Shawe-Taylor and Cristianini, 2004) is a valid tool to represent such relations. It counts how many times a (non-contiguous) subsequence of symbols u of length n occurs in the input string s, and penalizes non-contiguous occurrences according to the number of gaps they contain. Formally, let V be the vocabulary, the feature space associated with the string kernel of length n is indexed by a set I of subsequences over V of length n. The (explicit) mapping function is defined by",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Syntagmatic Kernel",
"sec_num": "4.2"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "\u03c6 n u (s) = X i:u=s(i) \u03bb l(i) , u \u2208 V n ,",
"eq_num": "(4)"
}
],
"section": "Syntagmatic Kernel",
"sec_num": "4.2"
},
{
"text": "where u = s(i) is a subsequence of s in the positions given by the tuple i, l(i) is the length spanned by u, and \u03bb \u2208]0, 1] is the decay factor used to penalize non-contiguous subsequences.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Syntagmatic Kernel",
"sec_num": "4.2"
},
{
"text": "The associated string kernel is defined by Gliozzo et al. (2005) modified the generic definition of the string kernel in order to take into account (sparse) collocations. Specifically, they defined syntagmatic kernels as a combination of string kernels applied to sequences of words in a fixed-size window centered on the word to be disambiguated. This formulation allows estimating the number of common (sparse) subsequences of words (i.e., collocations) between two examples, in order to capture syntagmatic similarity. The syntagmatic kernel is defined by",
"cite_spans": [
{
"start": 43,
"end": 64,
"text": "Gliozzo et al. (2005)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Syntagmatic Kernel",
"sec_num": "4.2"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "Kn(si, sj) = \u03c6 n (si), \u03c6 n (sj) = X u\u2208V n \u03c6 n (si)\u03c6 n (sj)",
"eq_num": "(5)"
}
],
"section": "Syntagmatic Kernel",
"sec_num": "4.2"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "KS(si, sj) = p X n=1 Kn(si, sj),",
"eq_num": "(6)"
}
],
"section": "Syntagmatic Kernel",
"sec_num": "4.2"
},
{
"text": "where K n is the string kernel defined in Equation 5 and the parameter n represents the length of the subsequences analyzed when estimating the similarity between contexts. Notice that the syntagmatic kernel is only effective for those fictitious contexts in which the lexical units do occur in meaningful sentences, however this is not guaranteed for the lexical units without examples.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Syntagmatic Kernel",
"sec_num": "4.2"
},
{
"text": "Finally, to combine domain and syntagmatic information, the composite kernel is defined by",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Composite Kernel",
"sec_num": "4.3"
},
{
"text": "EQUATION",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [
{
"start": 0,
"end": 8,
"text": "EQUATION",
"ref_id": "EQREF",
"raw_str": "KW SD (ti, tj) =KD(ti, tj) +KS(ti, tj),",
"eq_num": "(7)"
}
],
"section": "Composite Kernel",
"sec_num": "4.3"
},
{
"text": "whereK D andK S are normalized kernels defined in Equation 3 and 6, respectively. 5 It follows directly from the explicit construction of the feature space and from closure properties of kernels that it is a valid kernel.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Composite Kernel",
"sec_num": "4.3"
},
{
"text": "In this section we report the first experiment, namely the mapping between (F, l) pairs and a Wikipedia pages. We describe the experimental setup and then present the corresponding evaluation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Mapping task",
"sec_num": "5"
},
{
"text": "We applied our algorithm to all lexical units that do not have any example sentence in the FrameNet database. In principle, the proposed approach can be applied to every lexical unit, and we expect the algorithm performance to improve if some example sentences are already available because they could be added to the fictitious-context used to represent (F, l) in the system. Nevertheless, in this explorative study we wanted to focus on the harder cases, even if results are likely to be worse than on the whole FrameNet database. In FrameNet, 3,305 (F, l) pairs have no example sentences (536 pairs with adjectival LU, 1313 verbal LU, 1456 nominal LU). Since Wikipedia is basically a resource organized by concepts, which",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "5.1"
},
{
"text": "5K (xi, xj) = K(x i ,x j ) \u221a K(x j ,x j )K(x i ,x i )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "5.1"
},
{
"text": "are generally expressed by nouns, we decided to restrict our experiment to nominal lexical units. Besides, many verbal and adjectival concepts in Wikipedia are redirected to nominal identifiers. So, we randomly selected 900 pairs with nominal lexical units. For the moment, we decided to discard lexical units expressed by multiwords (about 150), which will be taken into account in a future version of our system. The average ambiguity of the 900 LUs considered is 1.24 in FrameNet. Instead, every LU corresponds to about 35 candidate senses in Wikipedia.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "5.1"
},
{
"text": "In order to perform WSD, we built the domain model from the 200,000 most visited Wikipedia articles. After removing terms that occur less than 5 times, the resulting dictionaries contain about 300,000 terms. We used the SVDLIBC package 6 to compute the SVD, truncated to 100 dimensions. The experiments were performed using the SVM package LIBSVM (Chang and Lin, 2001) customized to embed the kernels described in Section 4.",
"cite_spans": [
{
"start": 347,
"end": 368,
"text": "(Chang and Lin, 2001)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "5.1"
},
{
"text": "In this first evaluation step, we focus on the quality of the mapping between (F, l) pairs and Wikipedia articles. In order to evaluate the system output, we created a gold standard where 250 (F, l) pairs randomly extracted from the nominal subset described above have been manually linked to the Wikipedia page (if available) that best corresponds to the meaning of l in F . The pairs have been chosen in order to maximize the frame variability, i.e. every pair corresponds to a different frame. Since our gold standard contains 34% of all frames in the FrameNet database, we believe that, despite its limited size, it is well representative of FrameNet characteristics. Evaluation was carried out comparing the system output against the gold standard. Results are reported in Table 1 . The baseline was computed considering the most frequent sense of every lexical unit in Wikipedia. This element is obtained by taking into account all occurrences in Wikipedia where the lexical unit LU we consider is anchored to a given page. The most frequent sense for LU is the page to which LU is most frequently linked in Wikipedia. Since about 14% of the lexical units in the gold standard are not present in Wikipedia, we also estimated an upper bound accuracy of 0.86. This confirms our in-tuition that FrameNet and Wikipedia are linkable resources to a large extent and that our task is wellfounded. Wrong assignments include also problematic cases that are not directly connected to proper system errors. One of the most relevant issues is the different granularity between FrameNet frames and Wikipages. For example, the NETWORK frame is defined as \"a set of entities of the same or similar types (Nodes) are linked to each other by Connections to form a Network allowing for the flow of information, resources, etc.\". Even if the listed lexical units (network.n and web.n) and some examples refer to the informatics domain, the situation described in the FrameNet database is more general. Wikipedia instead lists several pages that may be seen as subdomains of NETWORK such as Computer network, Social network, Telecommunications network, etc. In the future, it may be worth modifying the system in order to allow multiple assignments of Wikipages for every frame.",
"cite_spans": [],
"ref_spans": [
{
"start": 778,
"end": 785,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "5.2"
},
{
"text": "In other cases, frame definitions seem not to be very consistent and it is very difficult to discriminate between two frames even for a human annotator. For example, ESTIMATED VALUE and ESTIMATING include both estimation.n as lexical unit, but since their frame definitions are almost the same and the other lexical units in the same frame are not discriminative, the system links both (F, l) pairs to the same Wikipedia article.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Accuracy",
"sec_num": null
},
{
"text": "In the following part of the experiment, we want to investigate to what extent the FrameNet -Wikipedia mapping can be effectively applied to automatically expand the FrameNet database with new example sentences, and eventually to acquire new lexical units. For every (F, l) pair, we consider the linked Wikipedia sense s and extract all sentences C s in Wikipedia with a reference to s. In this way, we can assume that, if s was linked to (F, l), C s can be included in the example sentences of F . This repository of sentences is already divided by sense and can significantly speed-up manual annotation. On the other hand, the extracted sentences could enrich the training set of machine learning systems for frame annotation to improve the frame identification step. In fact, this task has raised growing interest in the NLP community, with a devoted challenge at the last SemEval campaign (Baker et al., 2007) .",
"cite_spans": [
{
"start": 893,
"end": 913,
"text": "(Baker et al., 2007)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "English FrameNet expansion",
"sec_num": "6"
},
{
"text": "This retrieval process allows also to extract from C s all words W s that have an embedded reference to s in the form <a href=\"/wiki/Wiki Sense\"...>word</a>. In this way, W s are automatically included in F as new lexical units. In this phase, redirecting links are very useful because they automatically connect a word or expression to its nearest sense in case there is no specific page for this word. The information about redirecting allows also to account for orthographic variations of the same lexical unit, for example collectible is redirected to collectable.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "English FrameNet expansion",
"sec_num": "6"
},
{
"text": "We explain the data extraction process in the light of an example from our dataset. Our WSD system assigned to the (F, l) pair (WORD RELATIONS -homonym.n) the Wikipage http://en.wikipedia.org/wiki/Homonym. So, we extracted from the English Wikipedia dump all sentences where the anchor <a href=\"/wiki/Homonym\"... > appears and assumed that the word or multiword expression that is linked to the Homonym site may be a good candidate as lexical unit for the WORD RELATIONS frame. In this case, the example sentences were 186. Apart from homonym, the candidate lexical units are homograph, homophone, homophonous, homonymic, heteronym, same. Among them, only the latter is not appropriate, even if the sentence where it occurs is semantically connected to the WORD RELATIONS frame: \"In Hebrew the word 'thus' has the same triconsonantal root\". Instead, homonymic and heteronym can be acquired as new lexical units for WORD RELATIONS, and homograph, for which no example sentences are provided in FrameNet, can be automatically instantiated by a set of examples.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "English FrameNet expansion",
"sec_num": "6"
},
{
"text": "We considered 893 frame -lexical unit pairs assigned to Wikipedia pages following the algorithm described in Section 3. We discarded 7 pairs for which the system reported an assignment failure, i.e. the best sense delivered is the disambigua-tion page. Then we extracted a set of sentences for every (F, l) pair as described in the previous paragraph. Statistics about the retrieved data is reported in Table 2. English Wikipedia (F, l) pairs 893 N. of extracted sents 964,268 Avg. sents per (F, l) 1,080 ",
"cite_spans": [],
"ref_spans": [
{
"start": 403,
"end": 411,
"text": "Table 2.",
"ref_id": "TABREF1"
},
{
"start": 492,
"end": 498,
"text": "(F, l)",
"ref_id": null
}
],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "6.1"
},
{
"text": "The dimension of the extracted corpus does not allow to carry out a comprehensive evaluation. For this reason, we manually evaluated 1,000 sentences, i.e. we considered 20 (F, l) pairs, and for each of them we evaluated 50 sentences extracted from our large repository. Both (F, l) pairs and the assigned sentences were randomly selected.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "6.2"
},
{
"text": "In particular, the 20 (F, l) pairs do not contain only correctly assigned pairs, in fact three of them are wrong. Anyhow, the 20 pairs seem to be a representative subset of the 893 pairs considered in our experiment because they include both monosemic lexical units (gynaecology.n in MED-ICAL SPECIALTIES) and more ambiguous ones (club.n in the WEAPON frame). Our evaluation shows that 78% of the sentences were correctly linked to (F, l) pairs. This value is higher that the mapping accuracy between (F, l) and Wikipages reported in Section 5.2. In fact, we noticed that even if the Wikipage assigned to (F, l) is not the article that best corresponds to the meaning of l in F , some sentences pointing to it may be appropriate to express l.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "6.2"
},
{
"text": "As we already mentioned in Section 5.2, the different granularity of the information encoded by frames and Wikipages impacts on the output quality. For example, conversion.n in CAUSE CHANGE has a causative meaning, while it implies a personal process in UN-DERGO CHANGE. The mapping, instead, links (CAUSE CHANGE -conversion.n) to the Religious conversion page, and all the sentences collected point to religious conversion, regardless of their causative form or not. Another characteristic of this approach is that we can acquire new lexical units regardless of their part-of-speech, even if we start from nominal lexical units. This proves that we do not need to apply the initial mapping to verbal or adjectival LUs to obtain new data for all parts of speech. For example, we linked (MEDI-CAL SPECIALTIES -gynaecology.n) to the Gynaecology Wikipage. Consequently, we could include the adjective gynaecologic, pointing to the Gynaecology page, into the MEDICAL SPECIALTIES frame for sentences like \"Fellowship training in a gynaecologic subspeciality can range from one to four years\". However, this advantage can also turn into a weakness, because gynaecologist is also redirected to the Gynaecology page, but it belongs to MEDICAL PROFESSIONALS and should not be included into MEDICAL SPECIALTIES.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "6.2"
},
{
"text": "For the 20 (F, l) pairs considered in the given sentences, it was possible also to retrieve 8 lexical units that are not present in FrameNet, for example billy-club for the WEAPON frame. Exploiting redirections and anchoring strategies, our induction method can account for orthographical variations, for example it acquires both memorize and memorise. On the other hand, also misspelled words may be collected, for instance gynaecologial instead of gynaecological.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "6.2"
},
{
"text": "One of the great advantages of Wikipedia is its availability in several languages. The English version is by far the most extended, but a considerable repository of pages is available also for other languages, esp. European ones. In general, articles on the same object in different languages are edited independently and do not have to be translations of one another, but are linked to each other by their authors. In this way, the multilingual versions of Wikipedia can be easily exploited to build comparable corpora, with connected Wikipages in different languages dealing with the same contents.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Multilingual FrameNet expansion",
"sec_num": "7"
},
{
"text": "In this research step, we focus on this aspect of Wikipedia and propose a methodology that, using the English Wikipages as a bridge, automatically acquires new lexical units and example sentences also for other languages. This would represent the starting point towards the creation of FrameNet for new languages. Indeed, FrameNet structure comprises a language-independent level of information, namely frame and frame element definitions, and a language-dependent one, i.e. the lexical units and the example sentences. This makes the resource particularly suitable to corpus-based (semi) automatic creation of FrameNet for new languages, because the descriptive part can be pre-served and the language-dependent layer can be populated with new instances in other languages (Crespo and Buitelaar, 2008) .",
"cite_spans": [
{
"start": 774,
"end": 802,
"text": "(Crespo and Buitelaar, 2008)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Multilingual FrameNet expansion",
"sec_num": "7"
},
{
"text": "We apply our extraction algorithm to the Italian Wikipedia. Since several approaches have been experimented to (semi) automatically build Italian FrameNet using WordNet (De Cao et al. (2008) and Tonelli and Pighin (2009) ), we believe that our new proposal to exploit Wikipedia may be of interest in the research community. Anyhow, the approach can be exploited in principle for every language available in Wikipedia.",
"cite_spans": [
{
"start": 169,
"end": 190,
"text": "(De Cao et al. (2008)",
"ref_id": "BIBREF2"
},
{
"start": 195,
"end": 220,
"text": "Tonelli and Pighin (2009)",
"ref_id": "BIBREF22"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Multilingual FrameNet expansion",
"sec_num": "7"
},
{
"text": "Similarly to the data extraction process described in Section 6, we consider for every (F, l) pair in English the linked Wikipedia sense s, in English as well. Then, we retrieve the Italian Wikipedia sense s i linked to s and extract all sentences C i in the Italian Wikipedia dump 7 with a reference to s i . In this way, we can assume that C i are example sentences of F and that the words or expressions W i in C i containing an embedded reference to s i are good candidate lexical units of F in the Italian FrameNet. For example, if we link http: //en.wikipedia.org/wiki/Court to the JUDI-CIAL BODY frame, we first retrieve the Italian version of the site http://it.wikipedia.org/ wiki/Tribunale. Then, with a top-down strategy, we further extract all Italian sentences pointing to the Tribunale page and acquire as lexical units all words with an embedded reference to this concept, for example tribunale and corte. In this way, we can include the extracted lexical units and the sentences where they occur in the JUDI-CIAL BODY frame for Italian.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "7.1"
},
{
"text": "Given the 893 (F, l) pairs in English and the linked Wikipedia senses described in 6.2, we first extracted the Italian Wikipages that are linked to the English ones. Then for every linked Wikipage in Italian, we retrieved all sentences with a reference pointing to that page in the Italian Wikipedia dump. Statistics about the extracted data are reported in Table 3 .",
"cite_spans": [],
"ref_spans": [
{
"start": 358,
"end": 365,
"text": "Table 3",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "7.1"
},
{
"text": "Since the Italian Wikipedia is about one fifth of the English one, it was not possible to map every English Wikipage with an Italian article. In fact, only 371 senses out of 893 in English were linked to an Italian page. Also the average num- . 1,080) . Anyhow, this does not represent a problem because in the English FrameNet, the lexical units whose annotation is considered to be complete are usually instantiated by set of 20 annotated sentences on average. So, according to the FrameNet standard, 60 sentences are more than enough to represent the meaning of a lexical unit in a frame.",
"cite_spans": [],
"ref_spans": [
{
"start": 243,
"end": 251,
"text": ". 1,080)",
"ref_id": null
}
],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "7.1"
},
{
"text": "In this evaluation part, we took into account 1,000 sentences, in order to have a comparable dataset w.r.t. the evaluation for English. However, the sets of Italian sentences extracted for every (F, l), i.e. for every Wikipedia article, were much smaller, so we increased the number of randomly chosen (F, l) pairs to 80. Our evaluation is focused on the quality of the sentences and aims at assessing if the given sentences are correctly assigned to the (F, l) pairs. We report 69% accuracy, which is 9% lower than for English. Apart from the same errors and issues reported for English, a decrease in performance can be explained by the fact that, since less articles are present w.r.t. the English version, redirections and internal links tend to be less precise and fine-grained. For example, the word \"diritti\" in the sense of \"(human) rights\" redirects to the article about Diritto, corresponding to Law as a system of rules. On the contrary, Law and Rights have two different pages in English. Besides, the different quality of the two resources can also depend on the smaller number of users that edit and check the Italian articles. From the 1,000 sentences evaluated we extracted 145 new lexical units: since Italian FrameNet does not exist yet, every lexical unit in a sentence that is correct can be straightforwardly included in the first version of the resource.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "7.2"
},
{
"text": "In this work, we have proposed to apply a word sense disambiguation system to a new task, namely the linking between FrameNet and Wikipedia. Results are promising and show that the task is adequately substantiated. The proposed approach can help enriching FrameNet with new example sentences and lexical units and provide a starting point for the creation of FrameNet-like resources in all Wikipedia languages. On the one hand, the retrieved data could speed up human annotation, requiring only a manual validation. On the other hand, the extracted sentences could provide enough training data to machine learning systems for frame assignment, since insufficient frame attestations in the FrameNet database are a major problem for such systems.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Future work",
"sec_num": "8"
},
{
"text": "In the next research step, we plan to carry out an extended evaluation process in order to compute inter-annotator agreement and eventually point out validation problems. Then, we want to extend the mapping and the data extraction process to all (F, l) pairs in FrameNet (about 10,000). The retrieved sentences will be made available as training or annotation material. Besides, we want to create an online resource where the links between (F, l) pairs and Wikipages are made explicit and where users can browse the retrieved sentences. The resource can be produced and made available with a reduced effort for every language in Wikipedia. Anyway, the English version has proved to be more precise, while the resource for new languages would require a more accurate revision.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Future work",
"sec_num": "8"
},
{
"text": "http://en.wikipedia.org",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "http://download.wikimedia.org/enwiki/ 20090306 4 A context corresponds to a line of text in the Wikipedia dump and it is represented as a paragraph in a Wikipedia article.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "http://tedlab.mit.edu/\u02dcdr/svdlibc/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "http://download.wikimedia.org/itwiki/ 20090203",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "Claudio Giuliano is supported by the ITCH project (http://itch.fbk.eu), sponsored by the Italian Ministry of University and Research and by the Autonomous Province of Trento and the X-Media project (http://www. x-media-project.org), sponsored by the European Commission as part of the Information Society Technologies (IST) programme under EC grant number IST-FP6-026978.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgments",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "SemEval-2007 Task 10: Frame Semantic Structure Extraction",
"authors": [
{
"first": "Collin",
"middle": [
"F"
],
"last": "Baker",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Ellsworth",
"suffix": ""
},
{
"first": "Katrin",
"middle": [],
"last": "Erk",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the Fourth International Workshop on Semantic Evaluations (SemEval-2007)",
"volume": "",
"issue": "",
"pages": "99--104",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Collin F. Baker, Michael Ellsworth, and Katrin Erk. 2007. SemEval-2007 Task 10: Frame Semantic Structure Extraction. In Proceedings of the Fourth International Workshop on Semantic Evaluations (SemEval-2007), pages 99-104, Prague, CZ, June.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Approximating Textual Entailment with LFG and FrameNet Frames",
"authors": [
{
"first": "Aljoscha",
"middle": [],
"last": "Burchardt",
"suffix": ""
},
{
"first": "Anette",
"middle": [],
"last": "Frank",
"suffix": ""
}
],
"year": 2006,
"venue": "Proceedings of the 2nd PASCAL RTE Workshop",
"volume": "",
"issue": "",
"pages": "92--97",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Aljoscha Burchardt and Anette Frank. 2006. Approxi- mating Textual Entailment with LFG and FrameNet Frames. In Proceedings of the 2nd PASCAL RTE Workshop, pages 92-97, Venice, Italy.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Combining Word Sense and Usage for modeling Frame Semantics",
"authors": [
{
"first": "Diego",
"middle": [],
"last": "De Cao",
"suffix": ""
},
{
"first": "Danilo",
"middle": [],
"last": "Croce",
"suffix": ""
},
{
"first": "Marco",
"middle": [],
"last": "Pennacchiotti",
"suffix": ""
},
{
"first": "Roberto",
"middle": [],
"last": "Basili",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of STEP",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Diego De Cao, Danilo Croce, Marco Pennacchiotti, and Roberto Basili. 2008. Combining Word Sense and Usage for modeling Frame Semantics. In Pro- ceedings of STEP 2008, Venice, Italy.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "LIB-SVM: a library for support vector machines",
"authors": [
{
"first": "Chih-Chung",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "Chih-Jen",
"middle": [],
"last": "Lin",
"suffix": ""
}
],
"year": 2001,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Chih-Chung Chang and Chih-Jen Lin, 2001. LIB- SVM: a library for support vector machines. Software available at http://www.csie.ntu. edu.tw/\u02dccjlin/libsvm.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Semi-Supervised Learning",
"authors": [
{
"first": "Olivier",
"middle": [],
"last": "Chapelle",
"suffix": ""
},
{
"first": "Bernhard",
"middle": [],
"last": "Sch\u00f6lkopf",
"suffix": ""
},
{
"first": "Alexander",
"middle": [],
"last": "Zien",
"suffix": ""
}
],
"year": 2006,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Olivier Chapelle, Bernhard Sch\u00f6lkopf, and Alexander Zien. 2006. Semi-Supervised Learning. MIT Press, Cambridge, MA.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Integration of Linguistic Resources for Verb Classification: FrameNet Frame, WordNet Verb and Suggested Upper Merged Ontology",
"authors": [
{
"first": "Ian",
"middle": [],
"last": "Chow",
"suffix": ""
},
{
"first": "Jonathan",
"middle": [],
"last": "Webster",
"suffix": ""
}
],
"year": 2007,
"venue": "Computational Linguistics and Intelligent Text Processing",
"volume": "",
"issue": "",
"pages": "1--11",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ian Chow and Jonathan Webster. 2007. Integra- tion of Linguistic Resources for Verb Classification: FrameNet Frame, WordNet Verb and Suggested Up- per Merged Ontology. Computational Linguistics and Intelligent Text Processing, pages 1-11.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Domainspecific English-to-Spanish Translation of FrameNet",
"authors": [
{
"first": "Mario",
"middle": [],
"last": "Crespo",
"suffix": ""
},
{
"first": "Paul",
"middle": [],
"last": "Buitelaar",
"suffix": ""
}
],
"year": 2008,
"venue": "Proc. of LREC",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mario Crespo and Paul Buitelaar. 2008. Domain- specific English-to-Spanish Translation of FrameNet. In Proc. of LREC 2008, Marrakech.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Linking Documents to Encyclopedic Knowledge",
"authors": [
{
"first": "Andras",
"middle": [],
"last": "Csomai",
"suffix": ""
},
{
"first": "Rada",
"middle": [],
"last": "Mihalcea",
"suffix": ""
}
],
"year": 2008,
"venue": "Natural Language Processing for the Web",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Andras Csomai and Rada Mihalcea. 2008. Linking Documents to Encyclopedic Knowledge. IEEE In- telligent Systems, special issue on \"Natural Lan- guage Processing for the Web\".",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Large-scale named entity disambiguation based on Wikipedia data",
"authors": [
{
"first": "",
"middle": [],
"last": "Silviu Cucerzan",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP-CoNLL)",
"volume": "",
"issue": "",
"pages": "708--716",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Silviu Cucerzan. 2007. Large-scale named entity disambiguation based on Wikipedia data. In Pro- ceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Com- putational Natural Language Learning (EMNLP- CoNLL), pages 708-716, Prague, Czech Republic, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Frame assignment as Word Sense Disambiguation",
"authors": [
{
"first": "Katrin",
"middle": [],
"last": "Erk",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of IWCS-6",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Katrin Erk. 2004. Frame assignment as Word Sense Disambiguation. In Proceedings of IWCS-6, Tilburg, NL.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "WordNet: An Electronic Lexical Database",
"authors": [],
"year": 1998,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Christiane Fellbaum, editor. 1998. WordNet: An Elec- tronic Lexical Database. MIT Press.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Background to FrameNet",
"authors": [
{
"first": "C",
"middle": [
"J"
],
"last": "Fillmore",
"suffix": ""
},
{
"first": "C",
"middle": [
"R"
],
"last": "Johnson",
"suffix": ""
},
{
"first": "M",
"middle": [
"R L"
],
"last": "Petruck",
"suffix": ""
}
],
"year": 2003,
"venue": "International Journal of Lexicography",
"volume": "16",
"issue": "",
"pages": "235--250",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "C.J. Fillmore, C.R. Johnson, and M. R. L. Petruck. 2003. Background to FrameNet. International Journal of Lexicography, 16:235-250, September.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Semantic role labeling via FrameNet, VerbNet and PropBank",
"authors": [
{
"first": "Ana-Maria",
"middle": [],
"last": "Giuglea",
"suffix": ""
},
{
"first": "Alessandro",
"middle": [],
"last": "Moschitti",
"suffix": ""
}
],
"year": 2006,
"venue": "Proceedings of the 21st International Conference on Computational Linguistics and the 44th annual ACL meeting",
"volume": "",
"issue": "",
"pages": "929--936",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ana-Maria Giuglea and Alessandro Moschitti. 2006. Semantic role labeling via FrameNet, VerbNet and PropBank. In Proceedings of the 21st International Conference on Computational Linguistics and the 44th annual ACL meeting, pages 929-936, Morris- town, US.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Domain kernels for word sense disambiguation",
"authors": [
{
"first": "A",
"middle": [],
"last": "Gliozzo",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Giuliano",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Strapparava",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of the 43 rd annual meeting of the Association for Computational Linguistics (ACL-05)",
"volume": "",
"issue": "",
"pages": "403--410",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A. Gliozzo, C. Giuliano, and C. Strapparava. 2005. Domain kernels for word sense disambiguation. In Proceedings of the 43 rd annual meeting of the As- sociation for Computational Linguistics (ACL-05), pages 403-410, Ann Arbor, Michigan, June.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Using Word-Net to extend FrameNet coverage",
"authors": [
{
"first": "R",
"middle": [],
"last": "Johansson",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Nugues",
"suffix": ""
}
],
"year": 2007,
"venue": "Proc. of the Workshop on Building Frame-semantic Resources for Scandinavian and Baltic Languages",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Johansson and P. Nugues. 2007. Using Word- Net to extend FrameNet coverage. In Proc. of the Workshop on Building Frame-semantic Resources for Scandinavian and Baltic Languages, at NODAL- IDA, Tartu.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "The Role of Domain Information in Word Sense Disambiguation",
"authors": [
{
"first": "B",
"middle": [],
"last": "Magnini",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Strapparava",
"suffix": ""
},
{
"first": "G",
"middle": [],
"last": "Pezzulo",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Gliozzo",
"suffix": ""
}
],
"year": 2002,
"venue": "Natural Language Engineering",
"volume": "8",
"issue": "4",
"pages": "359--373",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "B. Magnini, C. Strapparava, G. Pezzulo, and A. Gliozzo. 2002. The Role of Domain Information in Word Sense Disambiguation. Natural Language Engineering, 8(4):359-373.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Proceedings of SENSEVAL-3",
"authors": [
{
"first": "R",
"middle": [],
"last": "Mihalcea",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Edmonds",
"suffix": ""
}
],
"year": 2004,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Mihalcea and P. Edmonds, editors. 2004. Proceed- ings of SENSEVAL-3, Barcelona, Spain, July.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Learning to link with Wikipedia",
"authors": [
{
"first": "David",
"middle": [],
"last": "Milne",
"suffix": ""
},
{
"first": "Ian",
"middle": [
"H"
],
"last": "Witten",
"suffix": ""
}
],
"year": 2008,
"venue": "CIKM '08: Proceedings of the 17th ACM conference on Information and knowledge management",
"volume": "",
"issue": "",
"pages": "509--518",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "David Milne and Ian H. Witten. 2008. Learning to link with Wikipedia. In CIKM '08: Proceedings of the 17th ACM conference on Information and knowl- edge management, pages 509-518, NY, USA. ACM.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Semeval-2007 Task-17: English Lexical Sample, SRL and All Words",
"authors": [
{
"first": "Sameer",
"middle": [],
"last": "Pradhan",
"suffix": ""
},
{
"first": "Edward",
"middle": [],
"last": "Loper",
"suffix": ""
},
{
"first": "Dmitriy",
"middle": [],
"last": "Dligach",
"suffix": ""
},
{
"first": "Martha",
"middle": [],
"last": "Palmer",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the Fourth International Workshop on Semantic Evaluations (SemEval-2007)",
"volume": "",
"issue": "",
"pages": "87--92",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sameer Pradhan, Edward Loper, Dmitriy Dligach, and Martha Palmer. 2007. Semeval-2007 Task-17: En- glish Lexical Sample, SRL and All Words. In Pro- ceedings of the Fourth International Workshop on Semantic Evaluations (SemEval-2007), pages 87- 92, Prague, Czech Republic, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Kernel Methods for Pattern Analysis",
"authors": [
{
"first": "J",
"middle": [],
"last": "Shawe-Taylor",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Cristianini",
"suffix": ""
}
],
"year": 2004,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Shawe-Taylor and N. Cristianini. 2004. Kernel Methods for Pattern Analysis. Cambridge Univer- sity Press.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Open Text Semantic Parsing Using FrameNet and WordNet",
"authors": [
{
"first": "Lei",
"middle": [],
"last": "Shi",
"suffix": ""
},
{
"first": "Rada",
"middle": [],
"last": "Mihalcea",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of HLT-NAACL",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lei Shi and Rada Mihalcea. 2004. Open Text Seman- tic Parsing Using FrameNet and WordNet. In Pro- ceedings of HLT-NAACL 2004.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Putting Pieces Together: Combining FrameNet, VerbNet and Word-Net for Robust Semantic Parsing",
"authors": [
{
"first": "Lei",
"middle": [],
"last": "Shi",
"suffix": ""
},
{
"first": "Rada",
"middle": [],
"last": "Mihalcea",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of CICLing 2005",
"volume": "",
"issue": "",
"pages": "100--111",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lei Shi and Rada Mihalcea. 2005. Putting Pieces To- gether: Combining FrameNet, VerbNet and Word- Net for Robust Semantic Parsing. In Proceedings of CICLing 2005, pages 100-111. Springer.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "New features for FrameNet -WordNet Mapping",
"authors": [
{
"first": "Sara",
"middle": [],
"last": "Tonelli",
"suffix": ""
},
{
"first": "Daniele",
"middle": [],
"last": "Pighin",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the Thirteenth Conference on Computational Natural Language Learning",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sara Tonelli and Daniele Pighin. 2009. New features for FrameNet -WordNet Mapping. In Proceedings of the Thirteenth Conference on Computational Nat- ural Language Learning, Boulder, CO, USA.",
"links": null
}
},
"ref_entries": {
"TABREF1": {
"num": null,
"type_str": "table",
"content": "<table/>",
"html": null,
"text": "Extracted data from English Wikipedia"
},
"TABREF3": {
"num": null,
"type_str": "table",
"content": "<table/>",
"html": null,
"text": "Extracted data from Italian Wikipedia ber of sentences extracted for every sense is much smaller (62 vs"
}
}
}
} |