File size: 97,152 Bytes
6fa4bc9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 | {
"paper_id": "P13-1040",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T09:33:26.005796Z"
},
"title": "Extracting bilingual terminologies from comparable corpora",
"authors": [
{
"first": "Ahmet",
"middle": [],
"last": "Aker",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Sheffield",
"location": {}
},
"email": "ahmet.aker@sheffield.ac.uk"
},
{
"first": "Monica",
"middle": [],
"last": "Paramita",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Sheffield",
"location": {}
},
"email": "m.paramita@sheffield.ac.uk"
},
{
"first": "Robert",
"middle": [],
"last": "Gaizauskas",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Sheffield",
"location": {}
},
"email": "r.gaizauskas@sheffield.ac.uk"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "In this paper we present a method for extracting bilingual terminologies from comparable corpora. In our approach we treat bilingual term extraction as a classification problem. For classification we use an SVM binary classifier and training data taken from the EUROVOC thesaurus. We test our approach on a held-out test set from EUROVOC and perform precision, recall and f-measure evaluations for 20 European language pairs. The performance of our classifier reaches the 100% precision level for many language pairs. We also perform manual evaluation on bilingual terms extracted from English-German term-tagged comparable corpora. The results of this manual evaluation showed 60-83% of the term pairs generated are exact translations and over 90% exact or partial translations.",
"pdf_parse": {
"paper_id": "P13-1040",
"_pdf_hash": "",
"abstract": [
{
"text": "In this paper we present a method for extracting bilingual terminologies from comparable corpora. In our approach we treat bilingual term extraction as a classification problem. For classification we use an SVM binary classifier and training data taken from the EUROVOC thesaurus. We test our approach on a held-out test set from EUROVOC and perform precision, recall and f-measure evaluations for 20 European language pairs. The performance of our classifier reaches the 100% precision level for many language pairs. We also perform manual evaluation on bilingual terms extracted from English-German term-tagged comparable corpora. The results of this manual evaluation showed 60-83% of the term pairs generated are exact translations and over 90% exact or partial translations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Bilingual terminologies are important for various applications of human language technologies, including cross-language information search and retrieval, statistical machine translation (SMT) in narrow domains and computer-aided assistance to human translators. Automatic construction of bilingual terminology mappings has been investigated in many earlier studies and various methods have been applied to this task. These methods may be distinguished by whether they work on parallel or comparable corpora, by whether they assume monolingual term recognition in source and target languages (what Moore (2003) calls symmetrical approaches) or only in the source (asymmetric approaches), and by the extent to which they rely on linguistic knowledge as opposed to simply statistical techniques.",
"cite_spans": [
{
"start": 597,
"end": 609,
"text": "Moore (2003)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We focus on techniques for bilingual term extraction from comparable corpora -collections of source-target language document pairs that are not direct translations but are topically related. We choose to focus on comparable corpora because for many less widely spoken languages and for technical domains where new terminology is constantly being introduced, parallel corpora are simply not available. Techniques that can exploit such corpora to deliver bilingual terminologies are of significant practical interest in these cases.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The rest of the paper is structured as follows. In Section 2 we outline our method. In Section 3 we review related work on bilingual term extraction. Section 4 describes feature extraction for term pair classification. In Section 5 we present the data used in our evaluations and discuss our results. Section 6 concludes the paper.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The method we present below for bilingual term extraction is a symmetric approach, i.e. it assumes a method exists for monolingual term extraction in both source and target languages. We do not prescribe what a term must be. In particular we do not place any particular syntactic restrictions on what constitutes an allowable term, beyond the requirement that terms must be contiguous sequences of words in both source and target languages.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "2"
},
{
"text": "Our method works by first pairing each term extracted from a source language document S with each term extracted from a target language document T aligned with S in the comparable corpus. We then treat term alignment as a binary classification task, i.e. we extract features for each source-target language potential term pair and decide whether to classify the pair as a term equivalent or not. For classification purposes we use an SVM binary classifier. The training data for the classifier is derived from EUROVOC (Steinberger et al., 2002) , a term thesaurus covering the activities of the EU and the European Parliament. We have run our approach on the 21 official EU languages covered by EUROVOC, constructing 20 language pairs with English as the source language. Considering all these languages allows us to directly compare our method's performance on resource-rich (e.g. German, French, Spanish) and under-resourced languages (e.g. Latvian, Bulgarian, Estonian). We perform two different tests. First, we evaluate the performance of the classifier on a held-out term-pair list from EUROVOC using the standard measures of recall, precision and F-measure. We run this evaluation on all 20 language pairs. Secondly, we test the system's performance on obtaining bilingual terms from comparable corpora. This second test simulates the situation of using the term alignment system in a real world scenario. For this evaluation we collected English-German comparable corpora from Wikipedia, performed monolingual term tagging and ran our tool over the term tagged corpora to extract bilingual terms.",
"cite_spans": [
{
"start": 518,
"end": 544,
"text": "(Steinberger et al., 2002)",
"ref_id": "BIBREF23"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "2"
},
{
"text": "Previous studies have investigated the extraction of bilingual terms from parallel and comparable corpora. For instance, Kupiec (1993) uses statistical techniques and extracts bilingual noun phrases from parallel corpora tagged with terms. Daille et al. (1994) , Fan et al. (2009) and Okita et al. (2010) also apply statistical methods to extract terms/phrases from parallel corpora. In addition to statistical methods Daille et al. use word translation information between two words within the extracted terms as a further indicator of the correct alignment. More recently, Bouamor et al. (2012) use vector space models to align terms. The entries in the vectors are co-occurrence statistics between the terms computed over the entire corpus.",
"cite_spans": [
{
"start": 121,
"end": 134,
"text": "Kupiec (1993)",
"ref_id": "BIBREF13"
},
{
"start": 240,
"end": 260,
"text": "Daille et al. (1994)",
"ref_id": "BIBREF6"
},
{
"start": 263,
"end": 280,
"text": "Fan et al. (2009)",
"ref_id": "BIBREF7"
},
{
"start": 285,
"end": 304,
"text": "Okita et al. (2010)",
"ref_id": "BIBREF18"
},
{
"start": 575,
"end": 596,
"text": "Bouamor et al. (2012)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "3"
},
{
"text": "Bilingual term alignment methods that work on comparable corpora use essentially three sorts of information: (1) cognate information, typically estimated using some sort of transliteration similarity measure (2) context congruence, a measure of the extent to which the words that the source term co-occurs with have the same sort of distribution and co-occur with words with the same sort distribution as do those words that co-occur with the candidate term and (3) translation of component words in the term and/or in context words, where some limited dictionary exists. For example, in Rapp (1995) , Fung and McKeown (1997), Morin et. al. (2007) , Cao and Li (2002) and Ismail and Manandhar (2010) the context of text units is used to identify term mappings. Transliteration and cognate-based information is exploited in Al-Onaizan and Knight (2002) , Graehl (1998), Udupa et. al. (2008) and Aswani and Gaizauskas (2010) .",
"cite_spans": [
{
"start": 588,
"end": 599,
"text": "Rapp (1995)",
"ref_id": "BIBREF20"
},
{
"start": 602,
"end": 610,
"text": "Fung and",
"ref_id": "BIBREF8"
},
{
"start": 611,
"end": 647,
"text": "McKeown (1997), Morin et. al. (2007)",
"ref_id": null
},
{
"start": 650,
"end": 667,
"text": "Cao and Li (2002)",
"ref_id": "BIBREF4"
},
{
"start": 672,
"end": 699,
"text": "Ismail and Manandhar (2010)",
"ref_id": "BIBREF9"
},
{
"start": 838,
"end": 851,
"text": "Knight (2002)",
"ref_id": "BIBREF1"
},
{
"start": 854,
"end": 889,
"text": "Graehl (1998), Udupa et. al. (2008)",
"ref_id": null
},
{
"start": 894,
"end": 922,
"text": "Aswani and Gaizauskas (2010)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "3"
},
{
"text": "Very few approaches have treated term alignment as a classification problem suitable for machine learning (ML) techniques. So far as we are aware, only Cao and Li (2002) , who treat only base noun phrase (NP) mapping, consider the problem this way. However, it naturally lends itself to being viewed as a classification task, assuming a symmetric approach, since the different information sources mentioned above can be treated as features and each source-target language potential term pairing can be treated as an instance to be fed to a binary classifier which decides whether to align them or not. Our work differs from that of Cao and Li (2002) in several ways. First they consider only terms consisting of nounnoun pairs. Secondly for a given source language term N 1 , N 2 , target language candidate terms are proposed by composing all translations (given by a bilingual dictionary) of N 1 into the target language with all translations of N 2 . We remove both these restrictions. By considering all terms proposed by monolingual term extractors we consider terms that are syntactically much richer than nounnoun pairs. In addition, the term pairs we align are not constrained by an assumption that their component words must be translations of each other as found in a particular dictionary resource.",
"cite_spans": [
{
"start": 152,
"end": 169,
"text": "Cao and Li (2002)",
"ref_id": "BIBREF4"
},
{
"start": 632,
"end": 649,
"text": "Cao and Li (2002)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "3"
},
{
"text": "To align or map source and target terms we use an SVM binary classifier (Joachims, 2002) with a linear kernel and the trade-off between training error and margin parameter c = 10. Within the classifier we use language dependent and independent features described in the following sections.",
"cite_spans": [
{
"start": 72,
"end": 88,
"text": "(Joachims, 2002)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Feature extraction",
"sec_num": "4"
},
{
"text": "The dictionary based features are language dependent and are computed using bilingual dictionaries which are created with GIZA++ (Och and Ney, 2000; Och and Ney, 2003) . The DGT-TM parallel data (Steinberger et al., 2012) was input to GIZA++ to obtain the dictionaries. Dictionary entries have the form s, t i , p i , where s is a source word, t i is the i-th translation of s in the dictionary and p i is the probability that s is translated by t i , the p i 's summing to 1 for each s in the dictionary. From the dictionaries we removed all entries with p i < 0.05. In addition we also removed every entry from the dictionary where the source word was less than four characters and the target word more than five characters in length and vice versa. This step is performed to try to eliminate translation pairs where a stop word is translated into a non-stop word. After performing these filtering steps we use the dictionaries to extract the following language dependent features:",
"cite_spans": [
{
"start": 129,
"end": 148,
"text": "(Och and Ney, 2000;",
"ref_id": "BIBREF16"
},
{
"start": 149,
"end": 167,
"text": "Och and Ney, 2003)",
"ref_id": "BIBREF17"
},
{
"start": 195,
"end": 221,
"text": "(Steinberger et al., 2012)",
"ref_id": "BIBREF24"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Dictionary based features",
"sec_num": "4.1"
},
{
"text": "\u2022 isFirstWordTranslated is a binary feature indicating whether the first word in the source term is a translation of the first word in the target term. To address the issue of compounding, e.g. for languages like German where what is a multi-word term in English may be expressed as a single compound word, we check whether the compound source term has an initial prefix that matches the translation of the first target word, provided that translation is at least 5 character in length. \u2022 isLastWordTranslated is a binary feature indicating whether the last word in the source term is a translation of the last word in the target term. As with the previous feature in case of compound terms we check whether the source term ends with the translation of the target last word. \u2022 percentageOfTranslatedWords returns the percentage of words in the source term which have their translations in the target term. To address compound terms we check for each source word translation whether it appears anywhere within the target term. \u2022 percentageOfNotTranslatedWords returns the percentage of words of the source term which have no translations in the target term.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dictionary based features",
"sec_num": "4.1"
},
{
"text": "the ratio of the number of words within the longest contiguous sequence of source words which has a translation in the target term to the length of the source term, expressed as a percentage. For compound terms we proceed as with percentageOfTranslatedWords. \u2022 longestNotTranslatedUnitInPercentage returns the percentage of the number of words within the longest sequence of source words which have no translations in the target term.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "\u2022 longestTranslatedUnitInPercentage returns",
"sec_num": null
},
{
"text": "These six features are direction-dependent and are computed in both directions, reversing which language is taken as the source and which as the target. We also compute another feature av-eragePercentageOfTranslatedWords which builds the average between the feature values of percent-ageOfTranslatedWords from source to target and target to source. Thus in total we have 13 dictionary based features. Note for non-compound terms if we compare two words for equality we do not perform string match but rather use the Levenshtein Distance (see Section 4.2) between the two words and treat them as equal if the Levenshtein Distance returns >= 0.95. This is performed to capture words with morphological differences. We set 0.95 experimentally.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "\u2022 longestTranslatedUnitInPercentage returns",
"sec_num": null
},
{
"text": "Dictionaries mostly fail to return translation entries for named entities (NEs) or specialized terminology. Because of this we also use cognate based methods to perform the mapping between source and target words or vice versa. Aker et al. 2012have applied (1) Longest Common Subsequence Ratio, (2) Longest Common Substring Ratio, (3) Dice Similarity, (4) Needleman-Wunsch Distance and (5) Levenshtein Distance in order to extract parallel phrases from comparable corpora. We adopt these measures within our classifier. Each of them returns a score between 0 and 1.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "\u2022 Longest Common Subsequence Ratio (LCSR): The longest common subsequence (LCS) measure measures the longest common non-consecutive sequence of characters between two strings. For instance, the words \"dollars\" and \"dolari\" share a sequence of 5 non-consecutive characters in the same ordering. We make use of dynamic programming (Cormen et al., 2001) to implement LCS, so that its computation is efficient and can be applied to a large number of possible term pairs quickly. We normalize relative to the length of the longest term:",
"cite_spans": [
{
"start": 329,
"end": 350,
"text": "(Cormen et al., 2001)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "LCSR(X, Y ) = len[LCS(X, Y )] max[len(X), len(Y )]",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "where LCS is the longest common subsequence between two strings and characters in this subsequence need not be contiguous. The shorthand len stands for length. \u2022 Longest Common Substring Ratio (LC-STR): The longest common substring (LCST) measure is similar to the LCS measure, but measures the longest common consecutive string of characters that two strings have in common. I.e. given two terms we need to find the longest character n-gram the terms share. The formula we use for the LCSTR measure is a ratio as in the previous measure:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "LCST R(X, Y ) = len[LCST (X, Y )] max[len(X), len(Y )]",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "\u2022 Dice Similarity:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "dice = 2 * LCST len(X) + len(Y )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "\u2022 Needlemann Wunsch Distance (NWD):",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "N W D = LCST min[len(X) + len(Y )]",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "\u2022 Levenshtein Distance (LD): This method computes the minimum number of operations necessary to transform one string into another. The allowable operations are insertion, deletion, and substitution. Compared to the previous methods, which all return scores between 0 and 1, this method returns a score s that lies between 0 and n. The number n represents the maximum number of operations to convert an arbitrarily dissimilar string to a given string. To have a uniform score across all cognate methods we normalize s so that it lies between 0 and 1, subtracting from 1 to convert it from a distance measure to a similarty measure:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "LD normalized = 1 \u2212 LD max[len(X), len(Y )]",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features",
"sec_num": "4.2"
},
{
"text": "The cognate methods assume that the source and target language strings being compared are drawn from the same character set and fail to capture the corresponding terms if this is not the case. For instance, the cognate methods are not directly applicable to the English-Bulgarian and English-Greek language pairs, as both the Bulgarian and Greek alphabets, which are Cyrillic-based, differ from the English Latin-based alphabet. However, the use of distinct alphabets is not the only problem when comparing source and target terms. Although most EU languages use the Latin alphabet, the occurrence of special characters and diacritics, as well spelling and phonetic variations, are further challenges which are faced by term or entity mapping methods, especially in determining the variants of the same mention of the entity (Snae, 2007; Karimi et al., 2011) . 1 We address this problem by mapping a source term to the target language writing system or vice versa. For mapping we use simple character mappings between the writing systems, such as \u03b1 \u2192 a, \u03c6 \u2192 ph, etc., from Greek to English. The rules allow one character on the lefthand side (source language) to map onto one or more characters on the righthand side (target language). We created our rules manually based on sound similarity between source and target language characters. We created mapping rules for 20 EU language pairs using primarily Wikipedia as a resource for describing phonetic mappings to English. After mapping a term from source to target language we apply the cognate metrics described in 4.2 to the resulting mapped term and the original term in the other language. Since we perform both target to source and source to target mapping, the number of cognate feature scores on the mapped terms is 10 -5 due to source to target mapping and 5 due to target to source mapping.",
"cite_spans": [
{
"start": 825,
"end": 837,
"text": "(Snae, 2007;",
"ref_id": "BIBREF22"
},
{
"start": 838,
"end": 858,
"text": "Karimi et al., 2011)",
"ref_id": "BIBREF11"
},
{
"start": 861,
"end": 862,
"text": "1",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Cognate based features with term matching",
"sec_num": "4.3"
},
{
"text": "We also combined dictionary and cognate based features. The combined features are as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Combined features",
"sec_num": "4.4"
},
{
"text": "\u2022 isFirstWordCovered is a binary feature indicating whether the first word in the source term has a translation (i.e. has a translation entry in the dictionary regardless of the score) or transliteration (i.e. if one of the cognate metric scores is above 0.7 2 ) in the target term. The threshold 0.7 for transliteration similarity is set experimentally using the training data. To do this we iteratively ran feature extraction, trained the classifier and recorded precision on the training data using a threshold value chosen from the interval [0, 1] in steps of 0.1. We selected as final threshold value, the lowest value for which the precision score was the same as when the threshold value was set to 1. \u2022 isLastWordCovered is similar to the previous feature one but indicates whether the last word in the source term has a translation or transliteration in the target term. If this is the case, 1 is returned otherwise 0. \u2022 percentageOfCoverage returns the percentage of source term words which have a translation or transliteration in the target term. \u2022 percentageOfNonCoverage returns the percentage of source term words which have neither a translation nor transliteration in the target term.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Combined features",
"sec_num": "4.4"
},
{
"text": "returns the difference between the last two features.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "\u2022 difBetweenCoverageAndNonCoverage",
"sec_num": null
},
{
"text": "Like the dictionary based features, these five features are direction-dependent and are computed in both directions -source to target and target to source, resulting in 10 combined features.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "\u2022 difBetweenCoverageAndNonCoverage",
"sec_num": null
},
{
"text": "In total we have 38 features -13 features based on dictionary translation as described in Section 4.1, 5 cognate related features as outlined in Section 4.2, 10 cognate related features derived from character mappings over terms as described in Section 4.3 and 10 combined features.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "\u2022 difBetweenCoverageAndNonCoverage",
"sec_num": null
},
{
"text": "In our experiments we use two different data resources: EUROVOC terms and comparable corpora collected from Wikipedia.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data Sources",
"sec_num": "5.1"
},
{
"text": "EUROVOC is a term thesaurus covering the activities of the EU and the European Parliament in particular. It contains 6797 term entries in 24 different languages including 22 EU languages and Croatian and Serbian (Steinberger et al., 2002) .",
"cite_spans": [
{
"start": 212,
"end": 238,
"text": "(Steinberger et al., 2002)",
"ref_id": "BIBREF23"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "EUROVOC terms",
"sec_num": "5.1.1"
},
{
"text": "We also built comparable corpora in the information technology (IT) and automotive domains by gathering documents from Wikipedia for the English-German language pair. First, we manually chose one seed document in English as a starting point for crawling in each domain 3 . We then identified all articles to which the seed document is linked and added them to the crawling queue. This process is performed recursively for each document in the queue. Since our aim is to build a comparable corpus, we only added English 3 http://en.wikipedia.org/wiki/Information technology for IT and http://en.wikipedia.org/wiki/Automotive industry for automotive domain. documents which have an inter-language link in Wikipedia to a German document. We set a maximum depth of 3 in the recursion to limit size of the crawling set, i.e. documents are crawled only if they are within 3 clicks of the seed documents. A score is then calculated to represent the importance of each document d i in this domain:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparable Corpora",
"sec_num": "5.1.2"
},
{
"text": "score d i = n j=1 f req d ij depth d j",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparable Corpora",
"sec_num": "5.1.2"
},
{
"text": "where n is the total number of documents in the queue, f req d ij is 1 if d i is linked to d j , or 0 otherwise, and depth d j is the number of clicks between d j and the seed document. After all documents in the queue were assigned a score, we gathered the top 1000 documents and used inter-language link information to extract the corresponding article in the target language.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Comparable Corpora",
"sec_num": "5.1.2"
},
{
"text": "We pre-processed each Wikipedia article by performing monolingual term tagging using TWSC (Pinnis et al., 2012) . TWSC is a term extraction tool which identifies terms ranging from one to four tokens in length. First, it POS-tags each document. For German POS-tagging we use TreeTagger (Schmid, 1995) . Next, it uses term grammar rules, in the form of sequences of POS tags or non-stop words, to identify candidate terms. Finally, it filters the candidate terms using various statistical measures, such as pointwise mutual information and TF*IDF.",
"cite_spans": [
{
"start": 90,
"end": 111,
"text": "(Pinnis et al., 2012)",
"ref_id": "BIBREF19"
},
{
"start": 286,
"end": 300,
"text": "(Schmid, 1995)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Comparable Corpora",
"sec_num": "5.1.2"
},
{
"text": "To test the classifier's performance we evaluated it against a list of positive and negative examples of bilingual term pairs using the measures of precision, recall and F -measure. We used 21 EU official languages, including English, and paired each non-English language with English, leading to 20 language pairs. 4 In the evaluation we used 600 positive term pairs taken randomly from the EU-ROVOC term list. We also created around 1.3M negative term pairs by pairing a source term with 200 randomly chosen distinct target terms. We select such a large number to simulate the real application scenario where the classifier will be confronted with a huge number of negative cases and a relatively small number of positive pairs. The 600 positive examples contain 200 single term pairs (i.e. single word on both sides), 200 term pairs with a single word on only one side (either source or target) and 200 term pairs with more than one word on each side. For training we took the remaining 6200 positive term pairs from EU-ROVOC and constructed another 6200 term pairs as negative examples, leading to total of 12400 term pairs. To construct the 6200 negative examples we used the 6200 terms on the source side and paired each source term with an incorrect target term. Note that we ensure that in both training and testing the set of negative and positive examples do not overlap. Furthermore, we performed data selection for each language pair separately. This means that the same pairs found in, e.g., English-German are not necessarily the same as in English-Italian. The reason for this is that the translation lengths, in number of words, vary between language pairs. For instance adult education is translated into Erwachsenenbildung in German and contains just a single word (although compound). The same term is translated into istruzione degli adulti in Italian and contains three words. For this reason we carry out the data preparation process separately for each language pair in order to obtain the three term pair sets consisting of term pairs with only a single word on each side, term pairs with a single word on just one side and term pairs with multiple words on both sides.",
"cite_spans": [
{
"start": 316,
"end": 317,
"text": "4",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Performance test of the classifier",
"sec_num": "5.2"
},
{
"text": "For this evaluation we used the Wikipedia comparable corpora collected for the English-German (EN-DE) language pair. For each pair of Wikipedia articles we used the terms tagged by TWSC and aligned each source term with every target term. This means if both source and target articles contain 100 terms then this leads to 10K term pairs. We extracted features for each pair of terms and ran the classifier to decide whether the pair is positive or negative. Table 1 shows the number of term pairs processed and the count of pairs classified as positive. Table 2 shows five positive term pairs extracted from the English-German comparable corpora for each of the IT and automotive domains. We manually assessed a subset of the positive examples. We asked human assessors to categorize each term pair into one of the following categories:",
"cite_spans": [],
"ref_spans": [
{
"start": 458,
"end": 465,
"text": "Table 1",
"ref_id": "TABREF0"
},
{
"start": 554,
"end": 561,
"text": "Table 2",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.3"
},
{
"text": "1. Equivalence: The terms are exact translations/transliterations of each other.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.3"
},
{
"text": "Not an exact translation/transliteration, but an exact translation/transliteration of one term is entirely contained within the term in the other language, e.g: \"F1 car racing\" vs \"Autorennen (car racing)\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inclusion:",
"sec_num": "2."
},
{
"text": "3. Overlap: Not category 1 or 2, but the terms share at least one translated/transliterated word, e.g: \"hybrid electric vehicles\" vs \"hybride bauteile (hybrid components)\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Inclusion:",
"sec_num": "2."
},
{
"text": "No word in either term is a translation/transliteration of a word in the other.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Unrelated:",
"sec_num": "4."
},
{
"text": "In the evaluation we randomly selected 300 pairs for each domain and showed them to two German native speakers who were fluent in English. We asked the assessors to place each of the term pair into one of the categories 1 to 4.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Unrelated:",
"sec_num": "4."
},
{
"text": "The results of the classifier evaluation are shown in Table 3 . The results show that the overall performance of the classifier is very good. In many cases the precision scores reach 100%. The lowest precision score is obtained for Lithuanian (LT) with 67%. For this language we performed an error analysis. In total there are 221 negative examples classified as positive. All these terms are multi-term, i.e. each term pair contains at least two words on each side. For the majority of the misclassified terms -209 in total -50% or more of the words on one side are either translations or cognates of words on the other side. Of these, 187 contained 50% or more translation due to cognate words -examples of such cases are capital increase -kapitalo eksportas or Arab organisation -Arabu lyga with the cognates capital -kapitalo and Arab -Arabu respectively. For the remainder, 50% or more of the words on one side are dictionary translations of words on the other side. In order to understand the reason why the classifier treats such cases as positive we examined the ",
"cite_spans": [],
"ref_spans": [
{
"start": 54,
"end": 61,
"text": "Table 3",
"ref_id": "TABREF2"
}
],
"eq_spans": [],
"section": "Performance test of the classifier",
"sec_num": "5.4.1"
},
{
"text": "Automotive chromatographic technique -chromatographie methode distribution infrastructure -versorgungsinfrastruktur electrolytic capacitor -elektrolytkondensatoren ambient temperature -au\u00dfenlufttemperatur natural user interfaces -nat\u00fcrliche benutzerschnittstellen higher cetane number -erh\u00f6hter cetanzahl anode voltage -anodenspannung fuel tank -kraftstoffpumpe digital subscriber loop -digitaler teilnehmeranschluss hydrogen powered vehicle -wasserstoff fahrzeug .67 .72 .82 .69 .81 .77 .78 .65 .82 .66 .66 .7 .77 .84 .72 .78 .69 .8 .78 .79 F .80 .83 .89 .81 .89 .86 .87 .78 .75 .79 .79 .82 .71 .91 .83 .87 .81 .88 .87 .88 training data and found 467 positive pairs which had the same characteristics as the negative examples in the testing set classified. We removed these 467 entries from the training set and re-trained the classifier. The results with the new classifier are 99% precision, 68% recall and 80% F score. In addition to Lithuanian, two further languages, Portuguese (PT) and Slovak (SK), also had substantially lower precision scores. For these languages we also removed positive entries falling into the same problem categories as the LT ones and trained new classifiers with the filtered training data. The precision results increased substantially for both PT and SK -95% precision, 76% recall, 84% F score for PT and 94% precision, 72% recall, 81% F score for SK. The recall scores are lower than the precision scores, ranging from 65% to 84%. We have investigated the recall problem for FI, which has the lowest recall score at 65%. We observed that all the missing term pairs were not cognates. Thus, the only way these terms could be recognized as positive is if they are found in the GIZA++ dictionaries. However, due to data sparsity in these dictionaries this did not happen in these cases. For these term pairs either the source or target terms were not found in the dictionaries. For instance, for the term pair offshoringuudelleensijoittautuminen the GIZA++ dictionary contains the entry offshoring but according to the dictionary it is not translated into uudelleensijoittautuminen, which is the matching term in EU-ROVOC.",
"cite_spans": [
{
"start": 464,
"end": 623,
"text": ".67 .72 .82 .69 .81 .77 .78 .65 .82 .66 .66 .7 .77 .84 .72 .78 .69 .8 .78 .79 F .80 .83 .89 .81 .89 .86 .87 .78 .75 .79 .79 .82 .71 .91 .83 .87 .81 .88 .87 .88",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "IT",
"sec_num": null
},
{
"text": "The results of the manual evaluation are shown in Table 4 . From the results we can see that both assessors judge above 80% of the IT domain terms as category 1 -the category containing equivalent term pairs. Only a small proportion of the term pairs are judged as belonging to category 4 (3-7%) -the category containing unrelated term pairs. For the automotive domain the proportion of equivalent term pairs varies between 60 and 66%. For unrelated term pairs this is below 10% for both assessors.",
"cite_spans": [],
"ref_spans": [
{
"start": 50,
"end": 57,
"text": "Table 4",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "We investigated the inter-annotator agreement. Across the four classes the percentage agreement was 83% for the automotive domain term pairs and 86% for the IT domain term pairs. The kappa statistic, \u03ba, was .69 for the automotive domain pairs and .52 for the IT domain. We also considered two class agreement where we treated term pairs within categories 2 and 3 as belonging to category 4 (i.e. as \"incorrect\" translations). In this case, for the automotive domain the percentage agreement was 90% and \u03ba = 0.72 and for the IT domain percentage agreement was 89% with \u03ba = 0.55. The agreement in the automotive domain is higher than in the IT one although both judges were computer scientists. We analyzed the differences and found that they differ in cases where the German and the English term are both in English. One of the annotators treated such cases as correct translation, whereas the other did not.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "We also checked to ensure our technique was not simply rediscovering our dictionaries. Since the GIZA++ dictionaries contain only single word-single word mappings, we examined the newly aligned term pairs that consisted of one word on both source and target sides. Taking both the IT and automotive domains together, our algorithm proposed 5021 term pairs of which 2751 (55%) were word-word term pairs. 462 of these (i.e. 17% of the word-word term pairs or 9% of the overall set of aligned term pairs) were already in either the EN-DE or DE-EN GIZA++ dictionaries. Thus, of our newly extracted term pairs a relatively small proportion are rediscovered dictionary entries. We also checked our evaluation data to see what proportion of the assessed term pairs were already to be found in the GIZA++ dictionaries. A total of 600 term pairs were put in front of the judges of which 198 (33%) were word-word term pairs. Of these 15 (less than 8% of the word-word pairs and less then 3% of the overall assessed set of assessed term pairs) were word-word pairs already in the dictionaries. We conclude that our evaluation results are not unduly affected by assessing term pairs which were given to the algorithm.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "Error analysis For both domains we performed an error analysis for the unrelated, i.e. category 4 term pairs. We found that in both domains the main source of errors is due to terms with different meanings but similar spellings such as the following example (1).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "(1) accelerator -decelerator For this example the cognate methods, e.g. the Levenshtein similarity measure, returns a score of 0.81. This problem could be addressed in different ways. First, it could be resolved by applying a very high threshold for the cognate methods. Any cognate score below that threshold could be regarded as zero -as we did for the combined features (cf. Section 4.4). However, setting a similarity threshold higher than 0.9 -to filter out cases as in (1) -will cause real cognates with greater variation in the spellings to be missed. This will, in particular, affect languages with a lot of inflection, such as Latvian. Another approach to address this problem would be to take the contextual or distributional properties of the terms into consideration. To achieve this, training data consisting of term pairs along with contextual information is required. However, such training data does not currently exist (i.e. resources like EUROVOC do not contain contextual information) and it would need to be collected as a first step towards applying this approach to the problem.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "Partial Translation The assessors assigned 6 -7% of the term pairs in the IT domain and 12 -16% in the automotive domain to categories 2 and 3. In both categories the term pairs share translations or cognates.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "Clearly, if humans such as professional translators are the end users of these terms, then it could be helpful for them to find some translation units within the terms. In category 2 this will be the entire translation of one term in the other such as the following examples. 5 (2) visible graphical interface -grafische benutzerschnittstelle",
"cite_spans": [
{
"start": 276,
"end": 277,
"text": "5",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "(3) modern turbocharger systems -moderne turbolader",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "In example (3) the a translation of the German term is to be found entirely within in the English term but the English term has the additional word visible, a translation of which is not found in the German term. In example (4), again the translation of the German term is entirely found in the English term, but as in the previous example, one of the English words -systems -in this case, has no match within the German term. In category 3 there are only single word translation overlaps between the terms as shown in the following examples.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "(4) national standard language niederl\u00e4ndischen standardsprache (5) thermoplastic material -thermoplastische elastomere",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "In example (5) standard language is translated to standardsprache and in example (6) thermoplastic to thermoplastische. The other words within the terms are not translations of each other.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "Another application of the extracted term pairs is to use them to enhance existing parallel corpora to train SMT systems. In this case, including the partially correct terms may introduce noise. This is especially the case for the terms within category 3. However, the usefulness of terms in both these scenarios requires further investigation, which we aim to do in future work.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Manual evaluation",
"sec_num": "5.4.2"
},
{
"text": "In this paper we presented an approach to align terms identified by a monolingual term extractor in bilingual comparable corpora using a binary classifier. We trained the classifier using data from the EUROVOC thesaurus. Each candidate term pair was pre-processed to extract various features which are cognate-based or dictionary-based. We measured the performance of our classifier using Information Retrieval (IR) metrics and a manual evaluation. In the IR evaluation we tested the performance of the classifier on a held out test set taken from EUROVOC. We used 20 EU language pairs with English being always the source language. The performance of our classifier in this evaluation reached the 100% precision level for many language pairs. In the manual evaluation we had our algorithm extract pairs of terms from Wikipedia articles -articles forming comparable corpora in the IT and automotive domains -and asked native speakers to categorize a selection of the term pairs into categories reflecting the level of translation of the terms. In the manual evaluation we used the English-German language pair and showed that over 80% of the extracted term pairs were exact translations in the IT domain and over 60% in the automotive domain. For both domains over 90% of the extracted term pairs were either exact or partial translations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "We also performed an error analysis and highlighted problem cases, which we plan to address in future work. Exploring ways to add contextual or distributional features to our term representations is also an avenue for future work, though it clearly significantly complicates the approach, one of whose advantages is its simplicitiy. Furthermore, we aim to extend the existing dictionaries and possibly our training data with terms extracted from comparable corpora. Finally, we plan to investigate the usefulness of the terms in different application scenarios, including computer assisted translation and machine translation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "Assuming the terms are correctly spelled, otherwise the misspelling is another problem.2 Note that we use the cognate scores obtained on the character mapped terms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Note that we do not use the Maltese-English language pair, as for this pair we found that 5861 out of 6797 term pairs were identical, i.e. the English and the Maltese terms were the same. Excluding Maltese, the average number of identical terms between a non-English language and English in the EUROVOC data is 37.7 (out of a possible 6797).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "In our data it is always the case that the target term is entirely translated within the English one and the other way round.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "The research reported was funded by the TaaS project, European Union Seventh Framework Programme, grant agreement no. 296312. The authors would like to thank the manual annotators for their helpful contributions. We would also like to thank partners at Tilde SIA and at the University of Zagreb for supplying the TWSC term extraction tool, developed within the EU funded project AC-CURAT.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Automatic bilingual phrase extraction from comparable corpora",
"authors": [
{
"first": "A",
"middle": [],
"last": "Aker",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Feng",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Gaizauskas",
"suffix": ""
}
],
"year": 2012,
"venue": "24th International Conference on Computational Linguistics (COLING 2012), IIT Bombay",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A. Aker, Y. Feng, and R. Gaizauskas. 2012. Auto- matic bilingual phrase extraction from comparable corpora. In 24th International Conference on Com- putational Linguistics (COLING 2012), IIT Bom- bay, Mumbai, India, 2012. Association for Compu- tational Linguistics.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Machine transliteration of names in arabic text",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Al-Onaizan",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Knight",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the ACL-02 workshop on Computational approaches to semitic languages",
"volume": "",
"issue": "",
"pages": "1--13",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Y. Al-Onaizan and K. Knight. 2002. Machine translit- eration of names in arabic text. In Proceedings of the ACL-02 workshop on Computational approaches to semitic languages, pages 1-13. Association for Computational Linguistics.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "English-hindi transliteration using multiple similarity metrics",
"authors": [
{
"first": "N",
"middle": [],
"last": "Aswani",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Gaizauskas",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC 2010)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "N. Aswani and R. Gaizauskas. 2010. English-hindi transliteration using multiple similarity metrics. In Proceedings of the Seventh International Confer- ence on Language Resources and Evaluation (LREC 2010), Valetta, Malta.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Identifying bilingual multi-word expressions for statistical machine translation",
"authors": [
{
"first": "D",
"middle": [],
"last": "Bouamor",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Semmar",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Zweigenbaum",
"suffix": ""
}
],
"year": 2012,
"venue": "LREC 2012, Eigth International Conference on Language Resources and Evaluation",
"volume": "",
"issue": "",
"pages": "674--679",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "D. Bouamor, N. Semmar, and P. Zweigenbaum. 2012. Identifying bilingual multi-word expressions for sta- tistical machine translation. In LREC 2012, Eigth International Conference on Language Resources and Evaluation, pages 674-679, Istanbul, Turkey, 2012. ELRA.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Base noun phrase translation using web data and the em algorithm",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Cao",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Li",
"suffix": ""
}
],
"year": 2002,
"venue": "Proceedings of the 19th international conference on Computational linguistics",
"volume": "1",
"issue": "",
"pages": "1--7",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Y. Cao and H. Li. 2002. Base noun phrase translation using web data and the em algorithm. In Proceed- ings of the 19th international conference on Com- putational linguistics-Volume 1, pages 1-7. Associ- ation for Computational Linguistics.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Introduction to Algorithms",
"authors": [
{
"first": "T",
"middle": [
"H"
],
"last": "Cormen",
"suffix": ""
},
{
"first": "C",
"middle": [
"E"
],
"last": "Leiserson",
"suffix": ""
},
{
"first": "R",
"middle": [
"L"
],
"last": "Rivest",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Stein",
"suffix": ""
}
],
"year": 2001,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein. 2001. Introduction to Algorithms. The MIT Press, 2nd revised edition, September.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Towards automatic extraction of monolingual and bilingual terminology",
"authors": [
{
"first": "B",
"middle": [],
"last": "Daille",
"suffix": ""
},
{
"first": "\u00c9",
"middle": [],
"last": "Gaussier",
"suffix": ""
},
{
"first": "J",
"middle": [
"M"
],
"last": "Lang\u00e9",
"suffix": ""
}
],
"year": 1994,
"venue": "Proceedings of the 15th conference on Computational linguistics",
"volume": "1",
"issue": "",
"pages": "515--521",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "B. Daille,\u00c9. Gaussier, and J.M. Lang\u00e9. 1994. Towards automatic extraction of monolingual and bilingual terminology. In Proceedings of the 15th conference on Computational linguistics-Volume 1, pages 515- 521. Association for Computational Linguistics.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Automatic extraction of bilingual terms from a chinesejapanese parallel corpus",
"authors": [
{
"first": "X",
"middle": [],
"last": "Fan",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Shimizu",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Nakagawa",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the 3rd International Universal Communication Symposium",
"volume": "",
"issue": "",
"pages": "41--45",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "X. Fan, N. Shimizu, and H. Nakagawa. 2009. Auto- matic extraction of bilingual terms from a chinese- japanese parallel corpus. In Proceedings of the 3rd International Universal Communication Sympo- sium, pages 41-45. ACM.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Finding terminology translations from non-parallel corpora",
"authors": [
{
"first": "P",
"middle": [],
"last": "Fung",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Mckeown",
"suffix": ""
}
],
"year": 1997,
"venue": "Proceedings of the 5th Annual Workshop on Very Large Corpora",
"volume": "",
"issue": "",
"pages": "192--202",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "P. Fung and K. McKeown. 1997. Finding terminol- ogy translations from non-parallel corpora. In Pro- ceedings of the 5th Annual Workshop on Very Large Corpora, pages 192-202.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Bilingual lexicon extraction from comparable corpora using indomain terms",
"authors": [
{
"first": "A",
"middle": [],
"last": "Ismail",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Manandhar",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the 23rd International Conference on Computational Linguistics: Posters",
"volume": "",
"issue": "",
"pages": "481--489",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A. Ismail and S. Manandhar. 2010. Bilingual lexi- con extraction from comparable corpora using in- domain terms. In Proceedings of the 23rd Inter- national Conference on Computational Linguistics: Posters, pages 481-489. Association for Computa- tional Linguistics.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Learning to classify text using support vector machines: Methods, theory and algorithms",
"authors": [
{
"first": "T",
"middle": [],
"last": "Joachims",
"suffix": ""
}
],
"year": 2002,
"venue": "",
"volume": "186",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T. Joachims. 2002. Learning to classify text using sup- port vector machines: Methods, theory and algo- rithms, volume 186. Kluwer Academic Publishers Norwell, MA, USA:.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Machine transliteration survey",
"authors": [
{
"first": "S",
"middle": [],
"last": "Karimi",
"suffix": ""
},
{
"first": "F",
"middle": [],
"last": "Scholer",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Turpin",
"suffix": ""
}
],
"year": 2011,
"venue": "ACM Computing Surveys (CSUR)",
"volume": "43",
"issue": "3",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "S. Karimi, F. Scholer, and A. Turpin. 2011. Ma- chine transliteration survey. ACM Computing Sur- veys (CSUR), 43(3):17.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Machine transliteration",
"authors": [
{
"first": "K",
"middle": [],
"last": "Knight",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Graehl",
"suffix": ""
}
],
"year": 1998,
"venue": "Computational Linguistics",
"volume": "24",
"issue": "4",
"pages": "599--612",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "K. Knight and J. Graehl. 1998. Machine translitera- tion. Computational Linguistics, 24(4):599-612.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "An algorithm for finding noun phrase correspondences in bilingual corpora",
"authors": [
{
"first": "J",
"middle": [],
"last": "Kupiec",
"suffix": ""
}
],
"year": 1993,
"venue": "Proceedings of the 31st annual meeting on Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "17--22",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Kupiec. 1993. An algorithm for finding noun phrase correspondences in bilingual corpora. In Proceed- ings of the 31st annual meeting on Association for Computational Linguistics, pages 17-22. Associa- tion for Computational Linguistics.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Learning translations of namedentity phrases from parallel corpora",
"authors": [
{
"first": "R",
"middle": [],
"last": "Moore",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of the tenth conference on European chapter of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Moore. 2003. Learning translations of named- entity phrases from parallel corpora. In In Proceed- ings of the tenth conference on European chapter of the Association for Computational Linguistics- Volume 1, pages 259266. Association for Compu- tational Linguistics.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Bilingual terminology mining -using brain, not brawn comparable corpora",
"authors": [
{
"first": "E",
"middle": [],
"last": "Morin",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Daille",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Takeuchi",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Kageura",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics",
"volume": "",
"issue": "",
"pages": "664--671",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "E. Morin, B. Daille, K. Takeuchi, and K. Kageura. 2007. Bilingual terminology mining -using brain, not brawn comparable corpora. In Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics, pages 664-671, Prague, Czech Republic, June. Association for Computa- tional Linguistics.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "A comparison of alignment models for statistical machine translation",
"authors": [
{
"first": "F",
"middle": [
"J"
],
"last": "Och",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Ney",
"suffix": ""
}
],
"year": 2000,
"venue": "Proceedings of the 18th conference on Computational linguistics",
"volume": "",
"issue": "",
"pages": "1086--1090",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "F. J. Och and H. Ney. 2000. A comparison of align- ment models for statistical machine translation. In Proceedings of the 18th conference on Computa- tional linguistics, pages 1086-1090, Morristown, NJ, USA. Association for Computational Linguis- tics.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "A systematic comparison of various statistical alignment models",
"authors": [
{
"first": "F",
"middle": [
"J"
],
"last": "Och Och",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Ney",
"suffix": ""
}
],
"year": 2003,
"venue": "Computational Linguistics",
"volume": "29",
"issue": "1",
"pages": "19--51",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "F. J. Och Och and H. Ney. 2003. A systematic compar- ison of various statistical alignment models. Com- putational Linguistics, 29(1):19-51.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Multi-word expression-sensitive word alignment",
"authors": [
{
"first": "T",
"middle": [],
"last": "Okita",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Guerra",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Graham",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Way",
"suffix": ""
}
],
"year": 2010,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T. Okita, A. Maldonado Guerra, Y. Graham, and A. Way. 2010. Multi-word expression-sensitive word alignment. Association for Computational Linguistics.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Term extraction, tagging, and mapping tools for under-resourced languages",
"authors": [
{
"first": "M\u0101rcis",
"middle": [],
"last": "Pinnis",
"suffix": ""
},
{
"first": "Nikola",
"middle": [],
"last": "Ljube\u0161i\u0107",
"suffix": ""
},
{
"first": "Inguna",
"middle": [],
"last": "Dan \u015e Tef\u0203nescu",
"suffix": ""
},
{
"first": "Marko",
"middle": [],
"last": "Skadi\u0146a",
"suffix": ""
},
{
"first": "Tatiana",
"middle": [],
"last": "Tadi\u0107",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Gornostay",
"suffix": ""
}
],
"year": 2012,
"venue": "Proc. of the 10th Conference on Terminology and Knowledge Engineering (TKE 2012)",
"volume": "",
"issue": "",
"pages": "20--21",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M\u0101rcis Pinnis, Nikola Ljube\u0161i\u0107, Dan \u015e tef\u0203nescu, In- guna Skadi\u0146a, Marko Tadi\u0107, and Tatiana Gornostay. 2012. Term extraction, tagging, and mapping tools for under-resourced languages. In Proc. of the 10th Conference on Terminology and Knowledge Engi- neering (TKE 2012), June, pages 20-21.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Identifying word translations in nonparallel texts",
"authors": [
{
"first": "R",
"middle": [],
"last": "Rapp",
"suffix": ""
}
],
"year": 1995,
"venue": "Proceedings of the 33rd annual meeting on Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "320--322",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Rapp. 1995. Identifying word translations in non- parallel texts. In Proceedings of the 33rd annual meeting on Association for Computational Linguis- tics, pages 320-322. Association for Computational Linguistics.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Treetagger-a language independent part-of-speech tagger",
"authors": [
{
"first": "Helmut",
"middle": [],
"last": "Schmid",
"suffix": ""
}
],
"year": 1995,
"venue": "Institut f\u00fcr Maschinelle Sprachverarbeitung",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Helmut Schmid. 1995. Treetagger-a lan- guage independent part-of-speech tagger. Insti- tut f\u00fcr Maschinelle Sprachverarbeitung, Universit\u00e4t Stuttgart, page 43.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "A comparison and analysis of name matching algorithms",
"authors": [
{
"first": "C",
"middle": [],
"last": "Snae",
"suffix": ""
}
],
"year": 2007,
"venue": "International Journal of Applied Science. Engineering and Technology",
"volume": "4",
"issue": "1",
"pages": "252--257",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "C. Snae. 2007. A comparison and analysis of name matching algorithms. International Journal of Applied Science. Engineering and Technology, 4(1):252-257.",
"links": null
},
"BIBREF23": {
"ref_id": "b23",
"title": "Cross-lingual document similarity calculation using the multilingual thesaurus eurovoc",
"authors": [
{
"first": "R",
"middle": [],
"last": "Steinberger",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Pouliquen",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Hagman",
"suffix": ""
}
],
"year": 2002,
"venue": "Computational Linguistics and Intelligent Text Processing",
"volume": "",
"issue": "",
"pages": "101--121",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Steinberger, B. Pouliquen, and J. Hagman. 2002. Cross-lingual document similarity calculation using the multilingual thesaurus eurovoc. Computational Linguistics and Intelligent Text Processing, pages 101-121.",
"links": null
},
"BIBREF24": {
"ref_id": "b24",
"title": "Dgt-tm: A freely available translation memory in 22 languages",
"authors": [
{
"first": "R",
"middle": [],
"last": "Steinberger",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Eisele",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Klocek",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Pilos",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Schlter",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of LREC",
"volume": "",
"issue": "",
"pages": "454--459",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Steinberger, A. Eisele, S. Klocek, S. Pilos, and P. Schlter. 2012. Dgt-tm: A freely available trans- lation memory in 22 languages. In Proceedings of LREC, pages 454-459.",
"links": null
},
"BIBREF25": {
"ref_id": "b25",
"title": "Mining named entity transliteration equivalents from comparable corpora",
"authors": [
{
"first": "R",
"middle": [],
"last": "Udupa",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Saravanan",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Kumaran",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Jagarlamudi",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceeding of the 17th ACM conference on Information and knowledge management",
"volume": "",
"issue": "",
"pages": "1423--1424",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Udupa, K. Saravanan, A. Kumaran, and J. Jagarla- mudi. 2008. Mining named entity transliteration equivalents from comparable corpora. In Proceed- ing of the 17th ACM conference on Information and knowledge management, pages 1423-1424. ACM.",
"links": null
}
},
"ref_entries": {
"TABREF0": {
"type_str": "table",
"text": "Wikipedia term pairs processed and judged as positive by the classifier.",
"html": null,
"num": null,
"content": "<table><tr><td>Processed 11597K DE Automotive 12307K DE IT</td><td>Positive 3249 1772</td></tr></table>"
},
"TABREF1": {
"type_str": "table",
"text": "Example positive pairs for English-German.",
"html": null,
"num": null,
"content": "<table/>"
},
"TABREF2": {
"type_str": "table",
"text": "Classifier performance results on EUROVOC data (P stands for precision, R for recall and F for F -measure). Each language is paired with English. The test set contains 600 positive and 1359400 negative examples.ET HU NL DA SV DE LV FI PT SL FR IT LT SK CS RO PL ES EL BG",
"html": null,
"num": null,
"content": "<table><tr><td>P R</td><td>1</td><td>1</td><td>.98 1</td><td>1</td><td>.98 1</td><td>1</td><td>.7</td><td>1</td><td>1</td><td>1</td><td>.67 .81 1</td><td>1</td><td>1</td><td>1</td><td>1</td><td>1</td></tr></table>"
},
"TABREF3": {
"type_str": "table",
"text": "Results of the EN-DE manual evaluation by two annotators. Numbers reported per category are percentages.",
"html": null,
"num": null,
"content": "<table><tr><td>Domain IT Automotive</td><td>Ann. 1 P1 81 P2 83 P1 66 P2 60</td><td>2 6 7 12 15</td><td>3 6 7 16 16</td><td>4 7 3 6 9</td></tr></table>"
}
}
}
} |